Fast forward another thirty years to 1980, and polio would be eliminated in many rich countries through vaccination. Smallpox would be eradicated worldwide. Insulin could now be manufactured by yeast in bulk in bioreactors, and emergency care would look completely different: with implantable pacemakers, AEDs, and coronary bypass surgery.

But you could still die from cancers caused by stomach ulcers, which we now know are typically caused by H. pylori infection and treatable with antibiotics. Or take hepatitis C – a deadly infection that causes liver fibrosis and cancer – which is now curable with antivirals in around 98% of patients.

Some of the most fatal conditions we’ve known, like HIV/AIDS and cystic fibrosis2, are also now highly treatable; taking early treatment for them returns people to near-normal life expectancies. Insulin treatment has advanced to the point that small wearable devices can monitor sugar levels in the blood and release insulin to stabilize them in real time.

But when I read most science journalism, hardly any of it mentions these achievements, the stream of innovation,3 or explains what is still untreatable and why. There's instead far too much hyping up of preliminary studies – what caused/cured cancer in six mice, for example – and much less about what’s changing people’s lives right now, let alone how much people’s lives have changed over the decades.

So, since last year, I’ve been writing round-ups of the biggest breakthroughs in medicine and putting them into context to give you a sense of where we are. Here’s a link to my post from 2024 in case you’re interested.

Subscribe now

I’ve split this into three sections – the first section only includes the results of large randomized controlled trials for new drugs published in papers this year, which are closest to becoming available. The second section is about record-breaking achievements and the first patients to receive new treatment methods. And the third section is about lab research that could soon have a big impact in medicine. Finally, a personal note.

This post is long and won’t fit in an email. As usual, I’ll pay you if you spot an error in this post, except for minor typos or grammatical errors. Please let me know if you find one, so I can fix it.

Clinical breakthroughs

• Suzetrigine (‘Journavx’) became the first non-opioid painkiller for surgical treatment in decades. In a phase 3 trial of 2,000 patients, it reduced pain as effectively as hydrocodone and paracetamol, but had fewer side effects and doesn’t appear to be addictive. Michelle Ma has written a great article about the history of pain medication leading up to it.

• The second chikungunya vaccine (‘Vimkunya’) was approved in the US and EU; it’s a recombinant ‘virus-like particle’ vaccine. Chikungunya is a disease spread by mosquitoes that is similar to dengue: both can cause weeks to months of joint pain and in rare cases, paralysis. I recently had the vaccine myself because I travel somewhat frequently to India, where chikungunya outbreaks are common; all I got was a slightly sore arm.4

• Several new lipid-lowering treatments are succeeding in large clinical trials. Some involve designing small interfering RNA (siRNA) to block specific genes from producing their proteins, with just a single injection that has effects for half a year or more.

  • Obicetrapib. It cut LDL cholesterol levels by an additional 30% in people with a history of heart disease or familial hypercholesterolemia, who were already on the maximum doses of statins and other cholesterol lowering drugs, in a phase 3 trial of over 2,500 people. The drug inhibits the CETP protein, which moves cholesterol between lipoproteins and the blood.

  • Enlicitide. It cut LDL cholesterol by an additional 59% in people with hypercholesterolemia who were already taking statins or other lipid-lowering drugs, in a phase 3 trial of over 300 people. Merck is now testing its effect in a broader population. The drug inhibits the PCSK9 protein, which regulates LDL cholesterol uptake and has been a popular target for new cholesterol drugs, but enlicitide is the first to do so in pill form.

    

  • Olezarsen. It cut triglyceride levels by 50–70% and reduced the rates of pancreatitis episodes by around 85% in people with severe hypertriglyceridemia in a phase 3 trial of over 1,000 people. The drug is an antisense oligonucleotide taken by injection each month, and works by blocking the formation of the ApoC3 protein.

    

  • Lepodisiran, an siRNA drug, causes a dramatic reduction in lipoprotein(a) levels. In a phase 2 trial of over 300 people with very high levels of lipoprotein(a), a single injection of the highest dose (400mg) reduced their lipoprotein(a) levels by 94% and kept them low for months. Their phase 3 trial is ongoing and expected to be completed 4 years from now.

  

  Subscribe now

• A new long-acting treatment (‘Qfitlia’ or fitusiran) for haemophilia was approved. Patients tend to rely on regular infusions of clotting products or similar products to prevent excessive bleeding. This new treatment is also an siRNA, and is given by injection roughly once per two months. In phase 3 trials, it reduced bleeding episodes by around 70% more than standard treatment.

• A new cure for whipworm infections, moxidectin-albendazole. In a phase 3 trial of around 270 children in Tanzania, a single combination of moxidectin and albendazole cured 69% of whipworm infections, meaning they no longer had detectable worm eggs in their stool, compared to 16% with albendazole alone.

  

• A new antibody treatment for multiple myeloma called teclistamab led to large improvements in phase 3 trials. So far, CAR-T cell therapies have improved the situation for patients with multiple myeloma, but they are very expensive and time consuming. This new off-the-shelf antibody treatment could offer a more scalable option that’s still highly effective. It reduced the chances of disease progression by around 83% in patients who had relapsed into myeloma and improved their overall survival, when taken as a combination with daratumumab, which has synergistic effects, compared to the standard alternate treatment.

  

• A portable test for tuberculosis infections was developed. It’s a handheld device that detects DNA from the bacterium that causes TB and gives results in under an hour. It has higher sensitivity than other methods, and also meets specificity thresholds set by the WHO.

• The UK became the first country to offer a vaccine against gonorrhea, as a targeted roll-out for gay men. The vaccine isn’t new: it’s the meningitis B vaccine, which provides some cross protection against gonorrhea, caused by a closely related bacterium. Protection in trials so far has been moderate: one dose offered around a 26% reduction in the chances of infection, while two doses boosted that to 33–40%. But the NHS estimates that, with high uptake among gay men, it could prevent up to 100,000 cases of gonorrhea over the next decade.

• New diabetes and weight loss drugs are on their way. Orforglipron is a new GLP1 drug that can be taken orally. Although it’s less effective than other GLP-1 drugs like tirzepatide, what sets it apart is that it is not a peptide, but a small molecule drug that can be synthesized chemically. This should make it cheaper, easier to scale up, and possible to take without needing to time it with meals. In a phase 3 trial of over 500 participants with type 2 diabetes, it reduced body weight by 4.5 to 7.6%, depending on the dose, in 40 weeks.

• The first paratyphoid vaccine may be on its way. It is a live attenuated vaccine created by deleting two parts of the bacteria’s genome. In a phase 2b challenge trial of 72 participants5, recipients of the vaccine had around a 73% lower risk of infection from Salmonella Paratyphi A. Moderate side effects were somewhat common, though. The researchers want to pair it with typhoid vaccination as part of a broader prevention tool against enteric fever.

  

Subscribe now

Breakthrough patients

• A baby named KJ became the first gene-edited baby6, and the first person to be treated with a custom in vivo CRISPR treatment for his rare genetic disease. He had a deficiency in the CPS1 enzyme, which meant his liver couldn’t convert ammonia into urea, damaging his brain and liver.

  • Babies with this rare deficiency typically stay in hospital until they’re eligible for a liver transplant, and have to follow very strict diets. In this case, researchers at the Children’s Hospital of Philadelphia diagnosed him within days of birth, and spent 6 months designing and testing a personalized gene-editing therapy before fixing his deficiency with a CRISPR base editor delivered by lipid nanoparticles. His blood ammonia levels dropped and he began tolerating more dietary protein, gained weight, and reduced his medications. Now 18 months old, he recently took his first steps.

• A research team at Roche and Boston’s Children’s Hospital set a new world record for the fastest human genome sequencing and analysis – it took them under 4 hours to perform whole genome sequencing and analysis with Roche’s SBX workflow (see next section). We’ve come a long way since the 1970s, when it took weeks to sequence a single gene, the 2000s, when it took 6 months for a whole genome, or even the 2010s when it took 3 days.

  

• 22 women became the first to receive mitochondrial donation to replace the mitochondrial DNA in their eggs, and have had 8 live births so far. They carried mutations that could have otherwise caused potentially fatal inherited metabolic diseases in their children; mitochondrial donation has helped avoid those worries.

• A 42 year old man with type 1 diabetes became the first person to receive a transplant with cells that are gene-edited to hide them from the immune system, meaning he will hopefully not have to take immunosuppressants. In the first 3 months after the pancreatic cell transplant, his immune system showed no reaction to the edited cells, which survived and produced insulin without notable side effects.

• A 56 year old man with chronic granulomatous disease, which involves long-term inflammation of the intestine and makes people vulnerable to repeated infections, became the second person ever to be treated with prime editing. It’s a relatively new form of gene editing that can fix tiny but critical errors, such as a single missing base in this case, with high precision.

Subscribe now

Research breakthroughs

• Roche developed a new genome sequencing method called Sequencing by Expansion (SBX), which expands DNA molecules into larger structures called Xpandomers, which are easier to read. It could improve accuracy and bring genome sequencing costs down further. And as I mentioned in the previous section, the method was recently used to break the record for the fastest turnaround time in sequencing a human genome.

• Scientists developed a new gene-editing tool, STITCHR, to insert large pieces of DNA into genomes. It’s based on a highly active retrotransposon called R2Tg in zebra finch7, which they engineered into a programmable system that can insert edits up to 12.7 kilobases long, roughly the size of the average human gene, without errors.

• Even longer gene edits, of nearly 1 million bases, became possible with ‘bridge recombinases’, which were discovered by researchers at the Arc Institute. They use a bridge RNA which folds into two loops, one that binds to the target DNA and the other to the donor DNA, before bridging them together. This could allow scientists to edit entire genes, long regulatory elements or clustered gene families, which were previously out of research for CRISPR based tools. Here’s a video of how it works:

• Scientists combined AI models RFDiffusion and AlphaFold2 to create a ‘multi-step enzyme’ for the first time, and that enzyme has never been seen before in nature. Enzymes are extremely specific proteins that can speed up reactions; they are widely used in industrial processes, pharmaceuticals, cooking, and consumer goods like laundry detergent; synthetic multi-step enzymes could broaden their uses further. On the Hard Drugs podcast, Jacob and I talked about the amazing breadth of proteins and how AI tools are being used to improve or design new proteins never seen before.

• Nanoparticles designed so far have been symmetrical, but now researchers have designed one with different sides, which can interact with different molecules at the same time. These two-faced nanoparticles (hah) could make vaccines, medicines, research tools, and diagnostics more versatile.

• DNA vaccines may now be possible. DNA vaccines have many of the benefits of mRNA vaccines (being easy to produce, fast to update, and producing proteins that represent viral proteins closer to their natural form), but are additionally stable at higher temperatures. Until now, they’ve been hard to implement because cells tend to reject DNA entering them. But scientists have created DNA lipid nanoparticle formulations that can package and deliver vaccines, and tested them for DNA vaccines against influenza and COVID in mice and rabbits.

Subscribe now

Reading recommendations

If you enjoyed this post, you might like some of these books on the history of science and medicine, a genre that I’ve been reading a lot more of recently. Here were my favourite reads this year.

• Genentech: the Beginnings of Biotech by Sally Smith Hughes.

  • This amazing book traces the development of genetic engineering, and how Herbert Boyer and Robert Swanson turned the science into a commercial product by developing recombinant insulin. This transformed insulin treatment from the extract of tens of thousands of animal pancreases per pound to the product of yeast in bioreactors. And by doing so, they kickstarted the field of biotechnology.

• The Butchering Art by Lindsey Fitzharris.

  • Joseph Lister, the pioneer of antiseptic techniques in surgery, had a fascinating life that I didn’t know anything about until I read this book. His father, JJ Lister, was himself a famous microscopist (in his free time), and Joseph Lister took forward that interest in microscopy and microbiology. He also took inspiration from Louis Pasteur, who was his contemporary, on germ theory, but played down the theory as it was controversial at the time, instead choosing to popularize antiseptic techniques on their own.

• The Code Breaker by Walter Isaacson.

  • I mistakenly expected this to be a somewhat shallow, generic popular science book, but it was a very thorough account of how CRISPR gene editing emerged from decades of obscure basic research into bacterial immune systems. The book focuses on Jennifer Doudna, who pioneered the technique as a medical tool, and her life and work. One thing I thought was great was that Isaacson interviewed many of the scientists involved and tells their perspectives of the same events, so you can make up your mind about the ‘he said, she said’ controversies yourself (there were surprisingly many).

• Vaccinated: One Man’s Quest to Defeat the World’s Deadliest Diseases by Paul Offit.

  • A surprising number of people haven’t heard of Maurice Hilleman, one of the greatest scientists of the 20th century and someone who likely saved hundreds of millions of lives by developing vaccines against measles, mumps, rubella, hepatitis A, hepatitis B, chickenpox, adenovirus, pneumococcal disease, meningococcal disease, Hemophilus influenzae b, and many other diseases. He seems to have been a very funny, and foul-mouthed, guy, but also an incredibly strict manager of his division at Merck. Hijinks ensue.

• A History of Immunology by Arthur M Silverstein.

  • I loved this book. I enjoyed learning about the early, often wildly speculative theories of immunity, how blood groups and allergies were discovered, and the 19th century intellectual battle between the ‘humoralists’ (who believed immunity arose from the blood) and ‘cellularists’ (who believed it arose from cells), which the humoralists largely won, leaving cellular immunity comparatively neglected until the mid-20th century. Another fun point was that some scientists genuinely believed that immunology was more or less ‘solved’ in the 1960s, before the discoveries of T cells or V(D)J recombination.

  • The book was also fascinating about where the vast diversity and extreme precision of antibodies comes from. It’s a great read for anyone sympathetic to the idea that biology is more theoretical than physics.

Or how about a podcast?

And if you prefer podcasts, may I suggest Hard Drugs, a new podcast on medical innovation hosted by … myself … and my wonderful friend Jacob Trefethen. We’ve recently done episodes on:

• The first cancer vaccine

• The history of vaccines

• Will AI solve medicine?

• 100 years of insulin in 15 minutes

• Lenacapavir: the miracle drug that could end AIDS

• and more.

Subscribe now

On a more serious note.

I struggled with writing this post this year. If you’re interested in medicine and global health like I am, this has been a fairly terrible year.

The US’s largest global health program, PEPFAR, which supplies HIV treatment to over 20 million people worldwide, faced massive staffing cuts and was completely paused for months, leaving many HIV clinics to shut down.

The economists Charles Kenny and Justin Sandefur estimate that, as a result of the huge cuts to US foreign aid and shut down of USAID, around 500,000 to a million people will have died by the end of this year.

That may sound hard to believe until you realize that the US is the largest international aid donor, by virtue of its size, and that these programs reach tens of millions of people with lifesaving treatment annually. Shamefully, many other countries decided to follow step.

Then there’s the rise in anti-vaccine sentiment and policy, the disruptions to US-funded clinical trials abroad, and the massive layoffs at the FDA and CDC. But expertise takes time to gain, and programs are far easier to maintain than recreate from scratch.

I suppose this isn’t just something of interest to people who follow medicine as an abstract topic.

Medical innovation affects our own lives, the health of people we know and love, and millions of people with friends and families far away even if we may never meet them.

People often describe innovation as a set of specific, sporadic breakthroughs that happen by pure chance or sheer determination, but it’s really much more of a continuous stream every year, like the ones I’ve described here. That stream can be sped up or slowed down by policy, funding, and institutional choices; and real progress – providing the fruits of that innovation to people who need them – takes a huge number of people and institutions working together.

I think it’s up to us to turn things around and to minimize the disruption to science and medical innovation.

Earlier this year, I pledged to donate more than 10% of my lifetime income to highly effective charities, but it feels like a drop in a puddle. If this post inspires you, or someone you know, we could make a bigger difference.

I’m still optimistic about what’s possible, in part because I keep meeting people who are unusually motivated to work on these problems anyway.

But things could move much faster with more people involved, and with effort, determination, and better ideas and incentives. I still think the best days of medicine are ahead of us, but only if we choose to make them happen.

Thanks for reading, and I hope you have a happy new year.

– Saloni

Anyway, most data visualizations I had come across were not very good. A lot of graphs were (and still are) confusing, misleading, or overly simplistic. I’ve seen quite a lot – three dimensional bar charts, double-axis charts with completely different scales for the same metric, unitless charts, pizza slice charts with sizes that corresponded to nothing in the data. Even now I come across charts that are ugly in such novel ways that I wonder how much imagination it must have taken to create them.

But with time, I’ve increasingly understood the importance of good data visualization. A lot of credit goes to my colleagues at Our World in Data for inspiring me and giving me feedback during the four years I worked there. I spent time thinking more deeply about the value of charts, and when they worked better than a written description. In the end I came to the conclusion that there were several situations in which I would prefer a chart.

In this post, I want to give you a sense of why data visualization matters, and walk you through how to make it more effective, accurate, and beautiful.

This post is long and won’t fit in an email. As usual, I’ll pay you if you spot an error in this post, aside from minor typos or grammatical errors. Please let me know if you find one, so I can fix it.

Subscribe now

Why visualize data?

Visuals can help explore data

First, for exploring data. Plotting data helps spot new patterns, trends, and unusual data points that can be hard to spot in a description, especially if those descriptions are about an average or a snapshot of the data, as they often are.

A classic example you’ll learn in statistics is ‘Anscombe’s quartet’: where the same correlation can result from vastly different underlying patterns in the data. It’s a reminder of how important it is to think about what’s in a correlation – I’d recommend this great blogpost by my friend Julia Rohrer on the topic.

Another example is from my work: this chart of the sex gap in life expectancy. Before I visualized the data, I hadn’t realized that the gap had changed so much over time, or why it had happened.

Visualizing the data made me notice two things. One was the massive impact of conflicts (e.g. the two world wars). Another was the widening gap over the twentieth century, and then its subsequent narrowing.

After reading much more on the topic, I learnt that the widening gap was driven by the rise of smoking, especially among men, which raised the risks of various cancers and heart disease and early mortality. I ended up writing a whole article about it – the causes of the sex gap in life expectancy and how it has changed over time and varies around the world.

Visuals can help explain concepts

Charts and diagrams are also valuable for explaining concepts. I often still prefer finding ways to write a clear written description that people can understand. But sometimes, visualization helps people absorb concepts better or faster, or helps them make new connections.

Take this diagram I made, of the difference between period and cohort data.

The visual is a few-second summary of what’s often a long and confused explanation for a concept that is really quite simple. A concise description – as I’ve given on the diagram itself – would help too, but the visual helps make the contrast much clearer.

Below is another example. It shows death rates across ages. You can see the trend looks like a hook, which might be surprising. To explain why it has this shape, I’ve split the data into two components: the rates of dying from diseases vs the rates of dying from external causes (which include accidents, injuries, violence, and so on).

Now you can see why it is hook shaped. Risks decline steeply with age after infancy, rise suddenly in adolescence, and then rise gradually and exponentially after that.

This shape is often called the ‘Gompertz-Makeham law of mortality’, in which the risks from diseases generally reflect the Gompertz part (an age-dependent risk), while the external causes generally reflect the Makeham part of the law (an age-independent risk).

Visuals can help share information more effectively

Charts can also be much more effective for sharing information or a message. People say “a picture paints a thousand words” and perhaps it’s true – it’s far easier to read and reshare a chart someone has shared with you than it is to read & share a 1,000 word blogpost about the same topic.

Given my hatred of bad headlines though, I’d caution that you’ve got to use that power wisely. If a good chart can get a message across widely, so can a misleading chart. I think it’s quite common for people to want to become more widely read and less common for people to think about how to deal with the responsibility it comes with. Your ideas can mislead other people. It’s important to try to get things right and acknowledge your mistakes, especially when you have a larger audience. I think this goes especially for graphs, which are more easily distributed.

Caution aside, here are two of the most memorable charts I’ve seen. One is a chart showing the enormous decline in the cost of genome sequencing – a much faster decline than Moore’s law – which has enabled tons of biological research and diagnostics and some treatments too.

[I know the line is 3D and a lot of people hate that, but I don’t think it’s that bad when it’s a single line. It doesn’t make it hard to follow the trendline anyway, although if I was interested in specific data points I’d be a little confused about whether I should read off the values from the front or the back of the 3D object.]

Another chart I find memorable is the one below, showing the adoption of mobile phones by fishermen in Kerala, and the impact that had on prices of fish.

Fish prices were very volatile before, but became much more stable as fishermen could share information much more easily. The idea is that, with phones, they no longer had to guess which harbour would offer the best price or risk arriving somewhere already oversupplied. Instead, they could call ahead, coordinate with buyers, and spread out across markets. Here is a link to the study it comes from, in case you are interested.

In essence, charts can be more memorable, shareable, and quickly-understood than a written explanation. They can also help you spot patterns to look into, as well as potential errors or artefacts.

Subscribe now

Visuals can help spot patterns, potential errors, and artefacts

I had a stressful experience of this a few years ago, when I spotted an impossible result in a chart in the supplement of my PhD thesis, just a week before I planned to submit it. The impossible result was due to a coding error that affected some of the results in one of my thesis chapters as well. Thankfully, I figured out what went wrong and fixed it in time. Crisis averted, thanks to dataviz.

So I’ve hopefully now convinced you that charts are valuable. Another thing that I found quite persuasive was noticing that the difference between a good chart and a bad chart was often the difference between understanding a concept versus being extremely confused (or angry at it).

Take a look at this three-dimensional bar chart below, which is a real chart from an academic meta-analysis about the effect of compression garments on recovery from exercise. What information do you learn from looking at the chart?

After you’ve finished being distracted by the totally unnecessary three-dimensional aspect of the graph, you’ll probably note that there is a large effect size on ‘resistance’. And something is happening to ‘metabolic’ at 24 hours.

Ultimately, the three dimensionality makes it harder to understand the graph and you probably wish it was all split up so you could see the data more clearly. So that’s what I’ve done below, roughly.

Aside from being able to read each set of results more easily, I’ve now also noticed the confidence intervals are really wide – there’s a lot of uncertainty in this data, which I didn’t notice in the original plot because I was quite distracted by navigating the 3D bars.

It also made me realize that the effect size didn’t have units that I could mention on the graph. After reading the paper, I found out they were ‘standardized mean differences’, which means the researchers took the difference between the two group averages and divided it by the standard deviation of the scores. In other words, they’re expressing how big an effect is relative to the typical variability in the data, so the number has no units, allowing many different effects to be compared on the same unit-less scale. Sounds confusing? You’re not alone. Many statisticians dislike this and similar metrics because it depends heavily on how spread out the data happen to be, which can change from study to study for reasons that have nothing to do with the underlying effect. The numbers can look bigger or smaller simply because the measurement was noisier or the sample was more or less varied, making comparisons across studies seem more like comparing apples to orangutans.

When possible, I prefer visualizing data in units that are already familiar to people or practically useful (like, a length of time measured in minutes instead of standard deviations from the average). Why? Familiar or practical units are easier to interpret and also easier to sense-check and potentially spot issues with. Although I know that they are not always possible, if visualization is meant for a broader audience, it’s valuable to try using them.

Now, you’re probably convinced that some charts are very bad, some charts are quite interesting, and finally, that improving a chart can help people understand the data better. But aside from splitting 3D bar charts into panels, how can you improve data visualizations?

Practical advice to improve data visualization

Here are some guiding questions I ask myself to help improve my charts. I’ll go through them below.

• Is my chart type meaningful?

• Can I make it clearer?

• If my chart is too complicated, can I guide the viewer through it?

• Does the chart work as a stand alone, as far as possible?

• Is my chart’s presentation justifiable?

• Is my chart reproducible?

A meaningful chart helps you answer a precise question

In the process of making new visualizations, I often start with a topic – such as the baby boom – or a statistic I’m interested in, like the number of children born per woman, and then look for datasets or metrics that I can visualize. But there are often many options: there’s the total fertility rate, the birth rate, and the age-specific fertility rate, just to mention a few. Getting familiar with what the metrics actually mean is crucial for presenting the data accurately.

When I’m trying to choose, I ask myself what I’m actually trying to understand. What question am I trying to answer? I went through an example in much more detail in a previous blogpost, but to keep it short here, I started with a simple question: ‘How do causes of death vary with age?’

There were many ways to go about visualizing this. I could show the number of deaths from different causes across ages, the death rate from different causes across ages, or the relative share of deaths from different causes across ages, for example.

Here are each of them:

Each metric, however, gave me a different perspective of the data. It could be misleading if I took just one of them as the canonical answer to my question of how causes of death vary with age. Ultimately, that was because my question was too broad. After I looked back at each of the charts I made, I realized that they were helpful in answering narrower questions:

• The relative share helped answer ‘What are people dying from at different ages?’

• The absolute number showed me ‘How many people are dying from different causes at each age?’

• The death rate showed me ‘What are the risks of dying from different causes at each age?’

By exploring them one by one, I had a better understanding of the overall topic and could describe and present each chart better. Here’s a whole article I wrote about those particular charts, in case you’re interested.

Subscribe now

Clearer charts help you focus on the data itself

Ugly 3D charts are an example of how a lot of graphs are difficult to understand for no good reason. Charts are often convoluted or overwhelming in ways that slow down your understanding of the data. I’m not saying that you should make charts simpler, but rather that you should make them clearer.

In the graphic above, it’s much easier to focus on the content of the right-hand version – you probably finished reading that one first – while the left-hand version had a distracting font and might have made you tilt your head to read it. The right-hand version was much faster to read not because the information was simpler, but because it was clearer.

A key goal in data visualization, and in communication, is to help people spend less time trying to remember and understand, and more time actually following along, so they can explore and consider the implications of new information.

It’s hard to give advice for every type of chart, but below I’ll share some general pieces of advice that I think apply to many charts. They aren’t always possible or desirable to fulfill in every case. Consider them instead as recommendations that you might have to trade off with other concerns, some of which I’ll mention along the way.

Keep text horizontal

Unless you are living in Mongolia, Taiwan or Japan, you are probably used to reading text horizontally.1 Therefore, when possible, you should keep text horizontal – it’s the way people are used to reading. Compare the examples below.

Horizontal text might mean you need to shorten category names or using multiple lines, or use horizontal bars instead of vertical ones, for example.2

Label directly, unless you have too many/repeated categories

When charts have separate legends, you often have to spend time flitting your eyes between the legend and chart to see which category refers to which element of the chart. It’s time consuming and not always necessary, especially if the chart has only one instance of each category. Compare the two below as an example.

An exception is when your chart has a lot of categories, or categories that refer to many elements on the chart (as in the map below, where each constituency was won by a party). Directly labelling each one would obviously make the chart look very crowded.

Here, you can also see an example of another piece of advice I’ll describe below: the colours are matched to the familiar colours of the political parties – yellow for Liberals, blue for Conservatives and red for Labour – which makes it easy to interpret the map even if you don’t look at the legend at all.

Split charts into panels to make individual trends visible

How about the chart below? It’s quite crowded, and there are so many colours that it’s difficult to identify the line for each state from the legend. In my view, the colours don’t add much – the graph would make the same point if the lines were all shown in grey and described simply as referring to different US states.

But when you have many categories on a chart, direct labelling could look crowded as well. And, in any case, it would be hard to follow each line since so many are overlapping.

A better alternative is to split up the data into multiple panels, like in the example below. This type of chart is often called a ‘small multiple’.

Above, all the panels have the same scale, but separating them into different panels makes the trends in each country much easier to follow.

On the other hand, my small multiple chart makes it harder to compare the trends between countries, since they’re now separated from each other. You can’t easily tell if one country has a higher value than another country at a particular time.

That’s a hint as to when to pick each version – if you want to make comparisons between entities, consider keeping them together and directly labelling them. If you want to let the reader follow the trend in each entity on its own, then split them up.

Subscribe now

Show multiple perspectives of the same data when one isn’t sufficient

There are often times when you want to communicate multiple things about the same data. Not all of it has to fit on the same chart.

Take the example below, where I was interested in understanding the risks of infant mortality with age. One question I had was how the risks changed with age, which I’ve shown in the top panel. Another was how that accumulated – what share of infant deaths had happened by a certain time? I showed that in the second panel.

Here’s another example. This map of Germany shows property sales prices across the country, which is helpful to quickly spot areas with higher (or lower) property prices.

But a coloured map makes it hard to understand the distribution of prices; our brains aren’t great at mapping differences in colour to differences in number. So I added an extra histogram showing the distribution of property prices in municipalities.

Order categories logically or alphabetically

This is quite simple, but when you do have a legend, it’s helpful to order the categories with some familiar logic. If there is an inherent order in the categories, such as strongly agree, agree, disagree, and strongly disagree – don’t mix them up but sort them in ascending or descending order.

When there isn’t an inherent order, consider using alphabetical order, since this makes them easier to search or skim through.

Match colours to concepts when possible

People often have strong associations between colours and concepts. Plants are often green, meat is often red, bananas are often yellow. Also, red often means bad, and green or blue often mean good.

It’s easier to process a graph if the colours match familiar representations of those concepts. It’s harder if there’s a mismatch and the plant category is red, or the bad category is blue.

As my friend Ian says, we shouldn’t be running a Stroop test on people and doing the equivalent of asking them to read the word ‘blue’ when it’s in a red font.

Colour choices might also unintentionally make viewers think something is bad if it’s red, even if it’s actually neutral, for example. So it’s worth being thoughtful about them.

Subscribe now

Make charts more colour-blind friendly

About 4 to 5% of the population has a form of colour blindness. To make sure they can also understand your charts, here are some suggestions:

• Use an online simulator like Coblis to check your charts are clear and the categories can be distinguished

• Consider a colour blind friendly palette

• Label directly (I’ve heard from people with colourblindness that directly labelling helped them easily tell apart the categories)

Lisa Charlotte Muth has a great blogpost with additional points to consider when designing data visualization in a colour-blind friendly way.

Plain language helps readers focus on the data

It’s a shame how a lot of great visualizations reach a limited audience because they are not written in plain language.

My view on science communication is that most concepts can be understood by most people, if explained clearly, even if they don’t have years of expertise or training in a field. The bigger challenge is figuring out how to do it, particularly in a way that doesn’t involve cutting important/relevant information.

Here is an example by my former colleagues at Our World in Data of making a technical chart familiar and easy to understand for a wide audience.

The point is, we want to find a way to make things clearer without oversimplifying. If I had to guess, I’d say that experts also find lots of information easier to grasp when it is written in plain language. It’s generally easier to make connections between concepts, imagine the consequences of something, and also spot potential errors when reading familiar language.

… but sometimes jargon is important

There are exceptions: jargon is often important. If jargon refers to a specific category that you would like to distinguish from other similar categories, using plain language could confuse rather than clarify.

Consider these two examples:

1. The mean and the median might both be called the average in plain language (and people generally assume the average refers to the mean, even if you actually displayed the median).

2. Iron-deficiency anemia and anemia of chronic disease might both be called ‘anemia’ in plain language, despite having very different causes and implications for a medical audience. If the data refers to only one of them, it’s useful to clarify which one.

Here’s my recommendation in cases where distinguishing related concepts is important. Use the precise terminology, and if the chart is also useful for a wider audience, then include a brief definition or examples in plain language as well. For example, by saying ‘the average (median) value’.

Below you can see that I kept the phrase ‘external causes’ to match the official name for the category while also giving examples in plain language.

Subscribe now

Consider guiding readers through a complicated chart

Sometimes the data you want to present can’t be presented well in a simple chart. A heatmap, a density plot, or a Lexis chart are all very valuable for communicating particular aspects of some data, but are unfamiliar to most readers and take time to read even for experts. That’s not always bad: it’s a trade-off you should think about.

A great suggestion I was given at Our World in Data was to annotate the chart directly to guide people through the visualization.

Below are two fairly complex chart types I’ve used in my work, with annotations to make them easier to follow.

The first is a heatmap on a Lexis plot – meaning the axes are age and year – and the colour refers to the death rate. There are therefore three dimensions on this chart, which in this case are very helpful to see that some events affected particular age groups at particular times, like HIV/AIDS.

But you need a little bit of an introduction to read this chart in the first place, which is why I’ve given a teeny tutorial of the chart in the first panel.

Here is another, which uses a ridgeline plot. Each curve represents a birth cohort (the data from all women born in a particular year), and the height of the curve refers to the number of children born to them at different ages. This ridgeline plot is helpful for seeing how generations of women differed in terms of the ages at which they had children, and in how many children they had.

In this case, I thought it would take some time to get how the chart works, but it probably takes more effort to remember that while reading and know what to take away from it – so I’ve described the key points next to the chart, which you can read alongside it.

Subscribe now

Make sure your chart works as a standalone

I have a strong opinion that charts should generally be clear on their own, without needing to read additional pages of text first.

Why? One reason is that charts are often reshared on their own, whether you like it or not, and lots of people will not link to the original source that contains additional context. So it’s best to include relevant context on the chart itself. Second, it helps actually read the data easily. I think it’s key to help readers focus on the data, rather than ask them to retain many pieces of information as they read a new chart.

Unfortunately, many academic charts are not very good at this. They often have a paragraph of text in a separate figure caption; sometimes the key details are spread across a whole page of text on a different page from the chart.

Many journalistic charts are not very good at it either, but for the opposite reason – they provide very little context at all. It’s harder to interpret a metric about mental health accurately if you haven’t said whether the numbers come from a survey or from diagnoses, for example. Giving key details, even briefly, can make it easier to interpret a metric accurately. Not giving them is like leaving out units on a chart; the different ways of measuring actually affect how one would interpret the data.

I like to think that we had an optimal balance at Our World in Data, with key information in the chart itself, and additional context in the subtitle or caption. It’s worth adding that the more important that context is, the more important it is to make that clear in prominent places, potentially rewriting the title with more precise language, even if that makes it less catchy.

Another organization that does this well, in my opinion, is Pew Research Center. In the example below, you can see their chart showing how the demographics of their survey respondents closely match benchmark figures in the population.

It’s also a nice chart for many other reasons – the title is a key takeaway, the subtitle is a concise and plain language summary of the metric, the chart is very easy to follow. The footnote gives useful information about how large the survey sample was and the survey’s response rate, and there’s a source line telling me when the survey took place. All of this is useful because the chart is telling me about the quality of survey data, and it’s not distracting.

Subscribe now

Consider how your chart might be misinterpreted and think about making improvements

There are many ways that charts could be misinterpreted; I don’t think it’s possible to avoid all of them, and some people will misinterpret charts regardless. But occasionally, those misconceptions will be common, and I think it’s important for visualizers to try to pre-empt them.

One example I have seen is line charts with arrows instead of points.3 Unfortunately in this case it implies there is an underlying direction to the data and that the data will continue moving in that direction in the future.

Another mistake I fell for myself was accidentally colouring the land in light blue on a map (in the first map below). A few people said they were initially confused about why there was a United Kingdom-shaped lake on the map. That would indeed be very intriguing.

I could excuse myself by saying that a UK-shaped shape on a map is most likely representing a land mass, not a lake, but I can see how it would be mind-bending and confusing. Plus, it goes against the suggestion I gave before, on matching colours (like blue) to concepts (like the sea). On the right is how I’ve corrected it.

Zooming in too much or too little can both cause problems

Many people will be familiar with the problem of the extremely zoomed-in y-axis. Take the example below, of the cherry blossom chart once again.

A common response is that you should stretch the axis to zero or show the full scale. So I’ve done that below. Is it better?

Well, now it’s misleading in a different way. Although the change was meaningful, you can’t see it at all here.

It’s helpful to think back to why the original chart seemed misleading. People will often say they thought it exaggerated the size of decline. I think that’s because the lowest point on a graph is perceived as the lowest possible value, since it often actually is.

There’s a trade-off between zooming out and not showing meaningful changes. There’s nothing intrinsically correct about displaying the full year – especially if it’s not actually relevant for cherry blossoming, and perhaps they wouldn’t flower at all if the climate changed substantially.

My view is that to strike a balance, it’s helpful to include a bunch of space above and below the maximum and minimum observed data, just so that people can tell the minimum observed point isn’t the minimum value possible, but not necessarily stretch the axis to zero. Here I’ve also changed the y-axis labels to more readable dates.

If your minimum observed data is already close to zero though, you might as well stretch the axis to include zero (or whatever the minimum possible value is, in your chart).

Metrics can be misinterpreted

Below is a chart I once worked on and never published. What is wrong with this chart?

There are various things you may not like – like the labels being quite far away from the data points, or the log axis scale being confusing.

But there are two more problems. One is that the graph itself is often kind of alarming: seeing that the risks of suicide are so much higher with a diagnosed mental illness (even though the absolute risk may be low, and the vast majority of people diagnosed don’t die by suicide).

A related challenge is that people might think the risks are higher for almost everyone in that group. But the risk estimates describe the average effect for a diagnosed group. Averages can hide important differences: in this case, it’s unclear whether the risk is elevated to a similar degree for everyone in the group, or whether it’s much higher for some individuals but not for others.

This issue – distinguishing group averages from individual differences – reflects the ‘fundamental problem of causal inference’. This concept is about how we can observe only what actually happened to each person, but never the alternative outcome that would have occurred in a different condition, since we only see one version of history. So instead we generally estimate the ‘average effect’ across a group, and have to do additional research to find out whether the effect varies between people.4

A third, related challenge is that people interpret the lines as showing the range of increased risks. But as you’ll see from the footnote, they’re not ranges! They’re confidence intervals: a measure of how precisely we know where the average is, not a measure of the variability of outcomes.

One idea is to write a clearer subtitle or add notes on the chart to explain that this is an average effect, the ranges are confidence intervals, and that the increase in risks likely vary a lot between people.

But even labelling confidence intervals doesn’t help much: people tend to perceive them as a depiction of variability on a visualization, as a very interesting experimental paper recently demonstrated, and as a result, people think that the range of data points is narrower than it really is.

Subscribe now

So what is the solution? How do we make the range of data clearer?

An option the authors suggest is to display ‘prediction intervals’ instead of ‘confidence intervals’. In prediction intervals, the interval reflects how much individual outcomes vary (using the standard deviation). In confidence intervals, the interval reflects how uncertain we are about the average effect (using the standard error).

In practice, this would require individual-level data so that I could calculate the standard deviation of the outcomes. Since I don’t have that, another feasible alternative is simply to show the underlying percentages, which makes clear that the risks are higher and also shows what those risks actually are, in numbers that are familiar.

The choice whether to use this graph instead of the other also depends on the audience and, like the y-axis problem, it could conceal meaningful differences.

Risk ratios, although they’re easy to misinterpret, can still give meaningful information about how strongly a treatment or condition is associated with an outcome within comparable subgroups. They help highlight relative differences that might be important for understanding patterns, even if they don’t clearly tell us how big the actual risks are or how much those risks vary from person to person. If you do use them, you should be careful about how you communicate them.

Try to avoid optical illusions

Below is an illusion that was new to me.

The curved lines are becoming steeper on a linear scale, and it looks like the difference between them is narrowing even though the difference is actually constant. I suspect this is because I’m thinking about the closest distance between the two lines (something like the horizontal difference) rather than the vertical difference at a given point along the x-axis.

I found this illusion in a review paper called ‘The Science of Visual Data Communication: What Works’, which I highly recommend for many other tips. The authors suggest that when it is important to communicate the difference between two lines, then it’s more useful to plot the actual differences (A-B).

One way I like to plot differences is with a range plot, which helps highlight a difference while also showing the starting point and the ending point.

This range chart I made highlights the time it took between an antibiotic class being discovered to when it was first in clinical use. Penicillins, for example, took 15 years to scale up into sufficient quantities to be used widely. Some other antibiotic drug classes took even longer: over 50 years for pleuromutilins. In contrast, many antibiotics discovered during the ‘Golden Age of antibiotics’ (1940s–1960s) were rapidly translated from discovery to clinical use.

Make charts more transparent

A chart with no source isn’t much better than claiming a trend was revealed to you in a dream. But seriously, it should be possible for people to find out where the numbers in your chart come from. Transparency would ideally help people have a better sense of the credibility of the data, read about the methods, verify the data, and reuse or adapt the chart.

At minimum, I think people should list the source of the data on the chart, with footnotes that clarify details for a more interested reader. But I have a bigger dream for transparency in data visualization.

Maximally reproducible charts

My dream is that one day, every chart will have a button you can click that will open up the data behind the chart, metadata, an explanation of what was measured, and the code to recreate the chart – which you can instantly run to reproduce the chart you saw.

Sadly I don’t know how to make that happen and I’m quite busy with other work, so my current best alternative is to include a link on my charts: code.scientificdiscovery.dev (as you can see below) which takes you to my GitHub repository for this newsletter.

There, you can find a folder with each chart, the data, an explanation, and the code to reproduce it. [You can also find the code for charts I’ve made for this newsletter on the ‘Code’ tab at the top of my newsletter. For the charts I produced at Our World in Data, find the code here.]

Transparency is important; there’s a possibility that I have coding errors in my work, even though I’ve looked through it a few times. I’ve set up a ‘bug bounty’ scheme to reward people for finding errors in my work and telling me about them.

Sadly there have been few takers yet, although I don’t think this is because the errors are absent – I occasionally spot them myself. I’ve considered paying someone to go through and check it all, but in any case, I’d still think it was helpful to keep things open source as more people read my posts (and more people could potentially spot errors), and because people are occasionally interested in reproducing charts or adapting them for their own purposes.

That’s another reason I’ve found to share the data and code behind your charts: I personally really enjoy when people explore data I’m interested in further and create new visualizations with it, which I get to learn from.

Subscribe now

Conclusion

When I did data visualization during my PhD, I received no training in it. I think it’s a skill that’s often seen as a nice bonus, or simply difficult to do well and better left to other professionals.

I hope I’ve convinced you of the opposite. Data visualization matters, and it has a lot of value beyond textual descriptions. Charts can be more memorable, shareable, and quickly understood than a written explanation. You might have even scrolled past most of my text and only focused on the headings and charts themselves, and that would be fine. Sort of. Well, if you did, you might have missed some important points I made and taken the wrong messages from them and that would annoy me a little bit.

But all this shows how impactful data visualization can be, how it can influence people’s interpretation, and that it takes responsibility to get that right. Getting there is not a mystery.

There are lots of resources you can find online, and tools like Datawrapper to make charts quickly. Most importantly, you can follow my guiding questions to help improve your visualizations:

• Is my chart type meaningful?

• Can I make it clearer?

• If my chart is too complicated, can I guide the viewer through it?

• Does the chart work as a standalone, as far as possible?

• Is my chart’s presentation justifiable?

• Is my chart reproducible?

If you’d like, use this post as a reference when making your own visualizations. There are some general principles that many people will agree with, but I’ve also explained that many choices come with trade-offs. Your decisions ultimately depend on what you’re actually trying to communicate, and that’s what should be front and centre when you’re working on a new visualization.

And, if it’s too difficult to make a choice, you could show multiple perspectives in the same chart. The tyranny of choice can often be solved by simply choosing both.

Go ahead and make some beautiful charts!

Further reading

If you’re interested in more, or I didn’t answer a specific question you had, here are some of my recommended readings for more guidance on data visualization:

• Datawrapper’s blog series on Do’s and Don’ts in data visualization [blog series]

• A great review of the research on data visualization: ‘The Science of Visual Data Communication: What Works’ by Steven L. Franconeri, Lace M. Padilla, Priti Shah, Jeffrey M. Zacks, and Jessica Hullman [academic review]

• Edward Tufte’s books on data visualization [books]

• Data visualization: a practical introduction by Kieran Healy [walkthrough book using R]

  Subscribe now

Which tools do I use?

People often ask me which tools I use to create data visualizations. My workflow involves several tools; some are skippable, depending on how much time you have and what you’re trying to achieve.

Here’s my typical workflow for a chart:

1. I download the data and write up code in R, using ggplot2 to create a graph

2. I then export the graph as an .svg file, and open it up on Adobe Illustrator or Figma5 to make additional tweaks, such as adding annotations on top of the chart

For a standard type of chart – like a bar chart, a choropleth map, a line or bubble chart – I instead use Datawrapper to create interactive charts very quickly. They follow great data visualization principles, have influenced my work, and their styles look great. (Plus, most of the features are free if you make an account. I have a subscription to Datawrapper through Works in Progress, which allows us to use a customized theme.)

Datawrapper has probably improved the quality of data visualization in the media. I rarely see 90º rotated text on charts or ugly legends or distorted scales in mainstream publications anymore, and it’s partly because they’re relying on Datawrapper, which has some strong (and great) default settings.

A few personal updates

As always, I hope you enjoyed reading this. If you haven’t already, I hope you subscribe and share it with your friends! If this is your first time reading, I recommend checking out the About page.

It’s been a while since I posted, so here are some personal updates in case you missed them:

• In August, I left my role at Our World in Data and joined Works in Progress full time! I’ll now be editing more pieces, writing more long-form articles on all things science, health and medicine, and hosting the podcast Hard Drugs on medical innovation with my friend Jacob Trefethen.

  • On the Hard Drugs podcast, we’ve released episodes on: Lenacapavir – the miracle drug that could end AIDS, the history of insulin, the history of vaccines, and on whether AI will solve medicine. You can listen wherever you get your podcasts, read the transcripts, or watch on YouTube.

• Since August I’m also a volunteer advisor at Coefficient Giving (previously Open Philanthropy) on clinical trial reform, working with Matt Clancy and Jordan Dworkin, which I’m very happy about. Here’s a longer tweet about my change in roles.

• Scientific Discovery is now a Works in Progress publication! This will give me more time to write posts regularly, and I’ve also rebranded the design of this newsletter, as you can see on the welcome page and the new logo.

See you next time! :)

– Saloni

You’ll learn about proteins in our body, why cells are like cities, what proteins are really, and how they are formed. And also about the first protein structure discovered and why it was ugly; why proteins are weird blobs, the biggest and smallest proteins, the most abundant proteins on earth, how fast proteins are, and what our favourite proteins are.

You can read the transcript below, listen on Apple, Spotify or wherever you get your podcasts, or watch on YouTube. This episode will be the first of a mini-series we’re releasing on proteins, drug development and AI. If you like it, I hope you share it with your friends, family, colleagues, and everyone you know, and subscribe!

Transcript

Saloni Dattani:
In today's episode, we're going to talk about the wonderful world of proteins.
Proteins are all around our body. We use them in our daily lives, and they do amazing things to keep us going.

Jacob Trefethen:
Protein design just won a Nobel Prize and we are going to do a mini-series of episodes here to talk about AlphaFold and other AI systems used to design proteins, whether people can increasingly design dangerous proteins, not just medicines, and whether protein design can help us get cures for some of the toughest diseases that still plague humanity.

Saloni Dattani:
But first, let's start with the basics. You might remember being in high school biology and seeing a simple diagram of a cell. It probably looked a bit like a fried egg or a sunny side up. There was the nucleus, which was a bit like the egg yolk. And then there were a few other things scattered around, like mitochondria and ribosomes, but that was a massive simplification.

In reality, cells are incredibly busy. There are billions of molecules in every cell, including loads of proteins, which have different functions. So let me just think about what are the different things that the proteins are doing? Well, there are structural proteins; they provide shape and strength to cells. There are storage proteins; they store little molecules. There are signalling proteins that help cells communicate with each other. So insulin, for example, is a hormone, a protein, and it's made in the pancreas and it tells cells to take up glucose from the bloodstream, and that lowers blood sugar after eating.

There are also transport proteins that move molecules between cells. Haemoglobin, for example, is a protein in red blood cells that binds to oxygen and carries it around in the blood. There are also enzymes — enzymes speed up chemical reactions in our body, by lowering the activation energy needed for them. There are regulatory proteins that control other proteins and pathways. And there are defence proteins that protect us from attack; so antibodies are a type of protein. Snakes and spiders have venoms, which are proteins that help them disable their threats. There are so many different types of jobs that a protein might have, and many proteins have multiple jobs at the same time.

And this means that this basic diagram view, that you might've had of a cell, was quite simple. In reality, the cell is extremely busy. It's more like a bustling city, and there are literally billions of molecules, proteins, DNA, RNA, fats, sugars, and ions — all moving around, reacting and interacting with each other.

Every part of the cell has its own job and it's a bit like different districts in the city. There's a great blog post by Niko McCarty where he describes this, and I thought it would be helpful just to have a sense of what's going on. He says, “A microbe's guts are a veritable Times Square, crowded with sugars, proteins, and water molecules that ricochet and smash into each other billions of times each second. Space is limited. A bacterium's insides are 70% water by mass; the other 30% is dominated by proteins first, followed by RNA and lipids. DNA accounts for just 1%. And all of this stuff fits inside a volume that is one quadrillionth of a litre.”

Jacob Trefethen:
That's a lot of proteins and I can't even see one of them.

Saloni Dattani:
Right? They're so small. And so if you think of this city — of each cell — the nucleus is something like the city hall, it's managing the information; it has instructions for what should happen. There are mitochondria; the power stations of the cell. There are ribosomes that construct new proteins. And then there are proteins, that are the workers and the machines of the city, but they're also the structural components and the signalling molecules and all of these things.

Jacob Trefethen:
Our body is doing so much with all of those proteins. Are proteins used outside of the body too?

Saloni Dattani:
They are! In fact, if you've done any cooking, you would know, for example, that chemical reactions change the proteins that you're cooking with. So, for example, if you cook an egg white, it becomes firm when it's cooked. That's because the heat denatures the proteins — it makes them unfold — and then it makes them coagulate into a different kind of mesh, and that makes it opaque.

There's also gluten, which is a protein that gives bread its stretchy texture — that's made of two proteins. There are also lots of proteins that are used in industry and biotechnology. If you've done your laundry recently, you might have used a detergent that was made of enzymes, and the enzymes break down stains, like fat or blood. Then there are a bunch of proteins that are used in baking and brewing and textile manufacturing. Of course there are lots of proteins that are used in medicine as well. So I mentioned that antibodies are a type of protein, and lots of medicines are types of antibodies. There's also insulin, which people use in diabetes; it's a protein that is also a therapeutic drug.

Jacob Trefethen:
What actually are proteins? What do they look like and how do they form?

Saloni Dattani:
Proteins are long chains of amino acids. You can sort of think of that as like beads on a string. And then that string, or that chain, is folded into some kind of 3D shape. The string is the protein's backbone, and each bead is an amino acid. Each amino acid has unique features. So as this string folds into a structure, you can kind of imagine that maybe happening at a small scale — maybe there's like a little helix of the string in some place, or maybe there are two parallel strings next to each other. But imagine that... we have to kind of zoom out and this whole 3D shape of the protein could also be connected to another protein; it could be two proteins together, making a protein complex.

Jacob Trefethen:
How is that made? I know I eat some protein, but I think we make some too.

Saloni Dattani:
That's right. So you have lots of DNA in your cells, and the DNA, which is the code of life, is the instructions for which proteins to make and how they should look. The DNA is transcribed into RNA, which is typically this temporary molecule, and then the RNA is then translated into protein by ribosomes. They sort of form one-by-one into this chain, and then rapidly fold into a much bigger structure.

This was kind of interesting to me because when I was reading this, I was thinking, okay, how did the first protein that was ever discovered look? What did people think when they first saw it? And that was fascinating because the first protein whose structure was determined was in 1958, and that was myoglobin. This was determined by John Kendrew, a British scientist. When he discovered this, it was only four years after the discovery of DNA's structure — DNA is of course very beautiful; it has this symmetrical structure, of this helix. And he was really disappointed when he figured out what myoglobin looked like. He wrote in his paper: “Perhaps the most remarkable features of the molecule are its complexity and its lack of symmetry.”

Jacob Trefethen:
Oh no, it's ugly.

Saloni Dattani:
But in hindsight, the irregularity is exactly what makes proteins so powerful. It's not really like DNA, which has this kind of linear messaging — it has the code, and then the code just linearly turns into RNA. But a protein is actually doing multiple things. It's in the cell being bombarded sometimes with lots of different molecules, and it needs to be able to recognise these different shapes and structures, and sometimes, it has multiple functions — and this function of every protein depends on that 3D structure.

The folded shape means that there are like little pockets, grooves and surfaces that the protein uses to bind to other molecules, or carry out specific chemical reactions, or even receive signals and then change shape in response. That means the same protein molecule might be doing multiple things at once. It could be doing a chemical reaction, but also binding to something else, and then when it gets some regulatory signal, it could be changing shape and stopping that chemical reaction from happening.

Jacob Trefethen:
So there's benefits to being a weird blob. There's nothing wrong with being a weird blob.

Saloni Dattani:
I thought it would be fun if we both share some fun facts about proteins. I found these from the book Biology by the Numbers, which is a great textbook, and it's also free online. The authors create these rough estimates and pull together key numbers on lots of different things related to cell biology. Some of them are rough estimates, but they're kind of our best guess right now.

Jacob Trefethen:
Hit me.

Saloni Dattani:
Alright, first one, how many proteins are in a human cell?

Jacob Trefethen:
They're busy, so I'm going to guess a lot. And I'm going to guess it depends on the cell, but I will go with a hundred million.

Saloni Dattani:
That is a lot, and it does depend on the cell. But the estimate for the average number is ten billion proteins per cell.

Jacob Trefethen:
Oh no. Two orders of magnitude wrong, not a good start. Okay, well, I've got one. Which is bigger: the protein or the mRNA that codes for the protein?

Saloni Dattani:
Um... surely the protein is bigger, no? Why would the instructions be bigger than the protein?

Jacob Trefethen:
That's what I always think, and it's the other way around. So the mRNA is bigger — you look at them side by side - well, images of 'em - and the mRNA is like 10 times bigger. Because each amino acid is coded for by three nucleotides, and the nucleotides themselves are bigger and heavier. So it's counterintuitive to me, but you know, it makes sense, I guess, when you think about it physically.

Saloni Dattani:
That does make sense... well, I don't know if that makes sense. I feel like I need to think about this more.

Jacob Trefethen:
Yeah, it doesn't make sense from a computer science point of view, but from a physical point of view it feels like, yeah.

Saloni Dattani:
Right.
I have one. So, you know, as a small person, I wanted to find out which protein was the smallest. Do you have any guesses?

Jacob Trefethen:
The protein that's the smallest? Well, the definition of a protein... I wonder if I'm allowed to have- it's got to have at least two amino acids, so I know it's not going to be less than two, but that probably wouldn't count as a protein because it wouldn't fold into anything, wouldn't have much function. So I'm going to guess philosophically, two, and then, literally, more than two.

Saloni Dattani:
Well, you're right. I think the typical definition of a protein is something that floats on its own in water and can fold into a stable shape. If you use that definition, then the smallest ones are some 20 to 30 amino acids long. There are actually lots of really tiny proteins, and these tiny proteins are called "micro proteins", and they're less than a hundred amino acids or so. One example that's actually even smaller than 20 or 30 is somatostatin, which is a hormone that controls other hormones — so it controls growth hormone and insulin — and that's only 14 amino acids long.

Jacob Trefethen:
Oh wow, it's that small. Oh okay.

Saloni Dattani:
Right. It still has a stable shape, because parts of the chain are connected to each other. So it's not considered a typical protein, but it's a peptide and it's very small.

Jacob Trefethen:
Got it, okay. What's the biggest? I think you know the answer to this one.

Saloni Dattani:
I think I do. Is it titin?

Jacob Trefethen:
It's titin. That's the biggest human protein at least, I don't know outside of humans. But that one is 33,000 amino acids long.

Saloni Dattani:
I got one. What's the most abundant protein on earth?

Jacob Trefethen:
I am going to guess it has something to do with photosynthesis, because that seems like one of the biggest functions on earth.

Saloni Dattani:
Very good guess. So it's kind of a tie, and we're not really sure which one is more abundant, so that was a bit of a trick question.

Jacob Trefethen:
Oh wow.

Saloni Dattani:
But one of them is RuBisCO, and that is used in photosynthesis; it's used to grab carbon from the air and turn it into useful organic material. And that's used by all photosynthetic organisms. And scientists estimate that there are about five kilograms of RuBisCO per person on earth.

Jacob Trefethen:
Oh my god. What?! Wow.

Saloni Dattani:
I guess there are a lot of plants.

Jacob Trefethen:
Yeah, fair enough. They're winning. They're winning... for now...

Saloni Dattani:
There's actually the second, which might be ahead. We're not sure-

Jacob Trefethen:
Oh right.

Saloni Dattani:
-and that is collagen. That is used as a kind of structural protein, and it makes up about 30% of the protein mass in your body — so about three kilograms of collagen per person. But it's not just humans that have collagen, it's also the livestock and all animals. That means there's- well, the total number- the total mass of livestock is also enormous, right? And so this means there's roughly four to six kilograms of collagen per person on earth.

Jacob Trefethen:
Ready for another fun fact?

Saloni Dattani:
Yes.

Jacob Trefethen:
Well, enzymes are a type of protein that speed up reactions... so how much do you think enzymes speed up reactions?

Saloni Dattani:
Mmm... a thousand times, maybe? Two thousand? I feel like... a lot. But I don't know.

Jacob Trefethen:
A lot. A lot. And I bet some do a thousand, but if you're really looking at the best of the best, we're talking billions of times, and possibly trillions of times, so we're talking millions of reactions per second per enzyme in some cases, and just totally changing what is happening at the molecular level.

Saloni Dattani:
That's crazy. That means, I guess, some reactions just wouldn't happen if the enzymes weren't there.

Jacob Trefethen:
Oh, absolutely. Yeah. I mean, statistically speaking, yeah.

Saloni Dattani:
So we were talking about protein folding the other day, and I was thinking: well, how fast do proteins fold into shape? Do you have any guesses?

Jacob Trefethen:
Oh... that is a tough one. The folding has to happen quickly, otherwise they'll get distracted by other forces. So I will go with tenths of seconds, no, hundredths of seconds.

Saloni Dattani:
Pretty close. So, on average, proteins fold in milliseconds, but some proteins fold really quickly, in micro seconds, which are a millionth of a second. And I guess you're right that it really does have to happen fast, because there's so much other stuff going on in the cell. It could just be bombarded with something else before it folds.

Jacob Trefethen:
Yeah, well, no fun.

Saloni Dattani:
How fast are enzymes colliding with other molecules in the cell? Or how many collisions are there per second?

Jacob Trefethen:
Okay. I have the sense that things are just crazy up in there and everyone's sort of bumping around. So I'm going to say a thousand collisions a second.

Saloni Dattani:
Well, you were right with the idea.

Jacob Trefethen:
Oh no, I should have just said "A lot."

Saloni Dattani:
But I think the estimate is 500,000 molecules are colliding with an enzyme per second.

Jacob Trefethen:
Wow.

Saloni Dattani:
And that's a lot! And that makes me think that proteins have to be really specific in how they bind to their targets. It's like, you know, if you're at a really crowded party and you're trying to find a friend, you would just bump into so many people before you actually find your friend. So you have to actually be able to recognise them among the 500,000 random strangers around you.

Jacob Trefethen:
Yep. That's tricky. Okay, Saloni, what's your favourite protein?

Saloni Dattani:
My favourite protein is tubulin. It's part of microtubules. The microtubules are kinda the skeletons of your cells... That sounds a bit grim, actually. But they are basically formed of these hollow tubes that are made of this protein, and each of the little structures is kind of like a tiny corn kernel. That tube can sort of assemble and disassemble in response to signals, and that means that the entire skeleton can kind of assemble and disassemble... which means the whole cell can change its shape or its size and move around, because of these microtubules. The microtubules also act as tracks to move things around, so they're a bit like a cellular railway or something, which I think is just super cool. And I remember learning about this in my undergrad and just seeing some diagrams and thinking, wow, that's amazing.

Jacob Trefethen:
That's a good one. I have an even better one though, which is… gluten in bread! Woo! I'm a bread guy.

Saloni Dattani:
That's a good one.

Jacob Trefethen:
We each have our favourites.

Saloni Dattani:
This was the first of a series of mini episodes we're doing on proteins. Stay tuned for our next episode on the history of Insulin. And if you like this, share it with your friends and subscribe.

Subscribe to our podcast on Spotify

Subscribe to our podcast on Apple

Subscribe to our podcast on YouTube

Sometime last year, when I talked to Sam Bowman, he suggested starting a science podcast through Works in Progress. The only way I could imagine doing a podcast was with a co-host, so we thought about it for a while and who to work with.

The person who came to mind for me was Jacob Trefethen. He works at Open Philanthropy and leads their science and global health R&D funding team, and he’s just fantastic. I had talked to him over the phone before that, but then in September, when I was visiting San Francisco for the Progress Conference, he very kindly offered that I could stay at his place, and I took him up on it!

He’s also a really lovely person and so fun and nice to talk to. His blog, which he started just last year, includes several posts I’m jealous I hadn’t thought of — and his first post is one I still think about all the time.

One evening, when we were at a bar with his friends near his place, before I had the chance to suggest it, he spontaneously said he’d had a vision that we would start a podcast together. So we both independently thought of starting a podcast together at roughly the same time, and that just felt perfect. And now here we are.

We’ll talk about medical innovation, how to speed it up, and how to scale it up. Roughly speaking, each episode will be about a technology, how it was developed and where we go from here; or about a disease that we currently lack good tools to prevent or treat, and how to make progress against it.

Many people think of medicines as if they’re not a form of technological innovation. Once they’re around, we take them for granted and don’t remember what life was like before them.

We also find it hard to imagine that new technologies might be around the corner to treat diseases that are untreatable today. And unfortunately, many that are newsworthy are hyped up, or lacking any real basis. But some real breakthrough technologies do already exist, and are waiting to be tested in trials right now, sometimes slogging through them for over a decade. We rarely celebrate the people, incentives, or tools that made them possible. We assume that it’s not possible to speed up (or slow down) the process of doing good science and developing medical technology. But it is.

The podcast will be about how it happens. How specific medical technologies were developed, the struggles the scientists faced, where we are today, what’s blocking more progress, and some ideas for how to take things forward from here.

Both of us write about medical innovation, and we often focus on diseases of global poverty, like malaria, tuberculosis, and AIDS. We believe there’s a huge amount of progress that can be made on neglected diseases with better technology. There’s so much low-hanging fruit waiting to be picked. Sometimes the technologies to address them are out there already, but are waiting to be scaled up, to reach everyone who needs them.

But we’re also very excited by innovation at the frontier — progress in protein structure prediction and design, all the recent breakthroughs in gene-editing technology, and many more. And there’s still much progress in health that can be made in wealthier countries, in treating dementia, cancer, addiction, and other conditions that I want to see become preventable and curable in my lifetime.

In my view, there isn’t actually a big distinction between research at the “frontier” and ways to tackle ancient diseases; they come together in many ways. Breakthroughs in research tools, like electron microscopy or genome sequencing, for example, make it possible to make far more progress in testing and research against neglected diseases.

It feels ambitious to have a podcast about all those topics together, but I think they’re all part of the same picture. Innovation isn’t worthwhile if it doesn’t reach the people who need it, and bridging that gap is often the motivation for further innovation — to find better, easier, safer, cheaper, more scaleable ways to solve problems; to build the infrastructure and networks to provide it, and develop the incentives to make innovation happen faster and be delivered at scale.

All in all, we have about... fifty episode ideas on our wishlist.

The first episode is about lenacapavir — the new breakthrough HIV drug I wrote about in December. It’s given as an injection only once every six months, and almost completely protects people from infections. It could effectively function as a vaccine despite actually being a small-molecule drug. And there are even more long-acting HIV drugs in the pipeline, which could improve on it further.

We also talk about the scientific journey leading up to this point, and how HIV was transformed from a near-certain death sentence to a manageable condition.

There’s now an opportunity to scale up these new drugs widely and change the course of the AIDS epidemic worldwide — but only if we take it.

Subscribe here:

You can subscribe wherever you get your podcasts. Listen to or watch our first episode here:

• Spotify

• Apple Music

• YouTube

• Or wherever you get your podcasts

You can also read the full transcript online, and our show notes with links and other material.

Our podcast is brought to you by Works in Progress and Open Philanthropy.

We have a lot of people to thank, especially Aria Babu, an editor at Works in Progress who’s responsible for getting it all going and organizing podcasting; Adrian Bradley, our great on-site producer and editor; Anna Magpie for fact-checking; Abhishaike Mahajan (AKA the great Owl Posting) for our cover art design; David Hackett, music composer; Atalanta Arden-Miller, for art direction; and Sanela Rankovic, a researcher at the Institute for Protein Design we talked to about lenacapavir and HIV capsid inhibitors.

One example is a chart showing the decline in US cancer death rates since the 1990s, which is shown below. It’s not an enormous decline, but it’s substantial.

The age-standardized death rate from cancer has declined by around a third since 1990 in the US. This means that, on average, people of a given age in 2021 had a third lower risk of cancer death than people of the same age in 1990. In many other high-income countries, the trend is similar, with a general decline since the late 80s or early 90s.

The early 1990s were the peak of cancer mortality in many richer countries. As cigarette smoking rose during the first half of the 20th century, cancer rates followed. Lung cancer became the most common cause of cancer death, and pushed overall cancer death rates up.

From the 1960s, following the growing public consensus that smoking caused lung cancer — such as in the US Surgeon General’s report — smoking rates began to decline. Because there’s a decades-long gap between smoking and cancer death, it took a few more decades for cancer mortality to decline, which you can see in the chart.

Subscribe now

That’s the part of the story I agree with.

But here’s the part I don’t agree with — a claim I’ve heard many, many times1 — that almost all of the decline in cancer mortality is due to the decline in smoking, and the other factors account for only a small remaining part.

Maybe people believe this because lung cancer is the biggest cancer cause of death, or that smoking increases the risk of multiple cancers, including oral cancer, liver cancer, and bladder cancer, not just lung cancer. That’s true, and I’ve written about it before. But it doesn’t follow that almost all of the decline in cancer mortality is due to the decline in smoking.

In my view, without smoking, we probably would have seen cancer mortality decline much earlier.2 There are major cancer types — particularly stomach and colorectal cancer — whose death rates started declining much before the decline in smoking; and this is important because they were previously some of the most common causes of cancer death.

But there are other reasons as well. We’ve also seen substantial declines in childhood cancer mortality, even though children’s cancers aren’t caused by smoking, but typically by inherited or spontaneous genetic mutations very early in development; and the reasons for progress against childhood cancers also apply to adult cancers. We also have evidence from other studies including RCTs that some cancer therapies result in large reductions in cancer progression and death.

I’d go even further. This isn’t just about cancer mortality. For several cancers, incidence rates have been declining as well. This should be fairly intuitive for lung cancer, whose incidence has declined as smoking rates have fallen, but it’s also the case for some other types, like cervical, stomach, and liver cancers. As I’ll describe some of the causes for the declines in cancer mortality, you’ll notice that some of them have large effects on whether those cancers develop at all, not just their mortality.3

Finally, I’ll explain a deeper issue with the whole idea: we can’t neatly divide the causes of an outcome into slices that add up to 100%. Even if one factor, like the decline in smoking, has had a large effect, that doesn’t mean the remaining causes only account for a small remaining part. This misinterprets how causes work together in the real world.

As usual, I’ll pay you if you spot an error in this post, aside from minor typos or grammatical errors. Please let me know if you find one, so I can fix it.

Smoking drove up cancer incidence and mortality during the 20th century

There’s always a risk, when you write something like this, that people will take it to mean something completely different. So just to be clear, I’m not saying that smoking was unimportant.

Smoking drove a large rise in lung cancer mortality during the 20th century, and this pushed up overall cancer mortality as well.

Cancer death became more common in the mid- or late-20th century than it was before. You can also see a longer-term visualization of this from France here, where the data spans 1925 onwards.

The impact of smoking on cancer is very large. On average, smokers have a 21 times higher risk of lung cancer death in the US than never-smokers. They also have a higher risk of various cancers, including bladder, oral, kidney, and pancreatic cancers, and some cardiovascular diseases, as the chart below shows.

(Note that the raised risk of cancer depends on the type and frequency of smoking; this data is based on the average cigarette smoker in the US.)

You can see that smoking raises the risk of death from various other cancers and cardiovascular diseases, though to a lesser extent than lung cancer.

One reason is that the carcinogens in cigarettes — such as polycyclic aromatic hydrocarbons (PAHs) and tobacco-specific nitrosamines — can cause cancers in multiple organs, including the mouth, throat, larynx, lungs, esophagus, bladder, pancreas, kidney, and even the cervix. Another reason is that tumors can metastasize, meaning they migrate from their original site to other parts of the body.4

I suspect this is why people claim that the decline in cancer mortality is almost entirely because of the decline in smoking. But I disagree for several reasons.

Subscribe now

Stomach and colorectal cancer mortality began to decline before smoking did

Before lung cancer became common, stomach and colorectal cancers were among the most common causes of cancer death.

But, over the 20th century, they declined. If we look closer, we can see that stomach cancer mortality has declined more than 9-fold since 1950 in the US, which is quite substantial, and that this trend began before smoking began to decline in the 1960s.

You can see a longer version of this trend starting in the 1930s here, although the numbers differ from the ones shown here because they’re using a different reference population for age-standardization.

Most stomach cancers are caused by infection with Helicobacter pylori (H. pylori), a bacterium that can lead to chronic stomach inflammation and eventually cancer in a subset of infected people.

Scientists clearly identified H. pylori as the cause of many stomach cancers in the late 1980s and early 1990s, but stomach cancer had been declining even before that. The reasons for this are still unclear, but might have included general improvements in sanitation, food safety, and the use of oral antibiotics and antacids, since modern research suggests that H. pylori is likely transmitted through contaminated food and water, and antibiotics and antacids can eliminate H. pylori infections in the stomach.

In fact, one early clue of the bacterial cause of stomach cancers was that antibiotics and antacids (famously ‘Pepto-Bismol’) reduced stomach ulcers, which sometimes progress into cancers. It seems the reason was that they made the stomach less habitable for H. pylori — antibiotics directly kill the bacteria, while antacids reduce stomach acidity, which disrupts the specific acidic environment H. pylori relies on to colonize and damage the stomach lining.

Since the 1980s and 1990s, when H. pylori’s role was clearly identified, doctors have also been able to screen for and treat the infection directly, and thus reduce the risks of stomach cancers developing in the first place.

Separately, it seems that in several countries, colorectal cancer mortality has also declined, beginning before the decline of smoking — though my understanding is that the historical reasons are less clear.

Let’s see what this means overall with two charts below. Between 1950 and 1990, the overall cancer death rate increased only slightly because, while lung cancer was becoming more common, deaths from other major cancers like stomach and colorectal cancer were becoming less common. After around 1990, deaths from more cancer types became less common; and the overall decline became more rapid.

Subscribe now

We’ve also seen large declines in childhood cancer mortality

Childhood cancers are typically caused by genetic mutations — such as in TP53, PRB, and other “tumour suppressor” genes, or conversely, in “oncogenes” which typically drive tumour growth. These mutations can be inherited or develop spontaneously (“de novo”) very early in development.

A prominent example is acute lymphoblastic leukemia (ALL), a common form of leukemia in children.

Survival rates for acute lymphoblastic leukemia among children — that is, the share who are still alive X years after diagnosis — have increased greatly over time. In the 1960s, only around 10% survived more than 5 years after diagnosis, but in recent years, >90% survive that long, which you can see below.

Survival rates like this are a common metric for studying improvements in cancer survival, but they can be affected by changes in diagnosis — the earlier people are diagnosed, the longer they appear to survive after a diagnosis, even if their trajectory hasn’t improved. Similarly, if less severe cases are increasingly diagnosed, the average survival can appear longer.

Although I think that’s less of an issue in this case, I generally tend to prefer looking at trends in mortality (as below), because they’re less affected by this problem, as this is at a point when cases are severe enough that most are captured in statistics.

In the chart below, you can see that there has been a substantial decline in childhood cancer mortality rates. This is evident for leukemias, lymphomas and multiple myeloma, and to a lesser extent, for brain cancers as well. There’s been a very large improvement over time.

For acute lymphoblastic leukemia, one major reason for progress is the use of new treatment regimens of chemotherapies. Previously, combination chemotherapies were used for shorter periods to clear leukemias in children, but some cancerous cells would remain, and most would soon relapse with new cancers forming in their central nervous system, leading to a low survival rate.

Later on, research showed that a new treatment regimen — which involved intensive treatment with eight drugs followed by consolidation treatment — helped keep them cancer-free for much longer, and was adopted as the standard treatment protocol in many countries.

In addition, genetic research helped identify specific mutations driving cancer growth among a fraction of people with acute lymphoblastic leukemia — like a mutation that created a fusion protein called BCR-ABL1.5 This led to the development of targeted therapies that block this protein’s signalling pathway and stop the leukemia cells from multiplying.

The reasons for this progress — including better treatment regimens and new treatments targeted against tumours with particular genetic mutations — have also been applied to adult cancers.

Another major breakthrough has come from vaccinating children against infectious diseases. Children with cancer are often immunosuppressed, which leaves them more vulnerable to infections. This can be either because their cancer depletes immune cells, or because treatments like chemotherapy suppress their immune system to slow tumor progression.

Many vaccines can help protect these children, including vaccines against measles, mumps, rubella, pneumococcal, influenza, and varicella (chickenpox). By reducing the overall spread of these diseases, vaccines help shield children who might not be able to mount an immune response themselves.

One example is chickenpox, which I’ve written about before. Although people often consider it mild, chickenpox used to cause around a hundred deaths per year in the US, and children with cancers were disproportionately represented in these statistics. Since chickenpox vaccination was introduced routinely in the US, chickenpox deaths have been reduced to zero.

But this isn’t all; the use of vaccination goes beyond infectious diseases.

Subscribe now

Vaccination reduces the incidence of certain cancers

Some vaccines play a direct cancer-prevention role, which means that vaccination can reduce incidence rates of some cancers as well. That’s because some cancers are caused by infections.

This was first demonstrated in a human cancer in 1979 with the identification of HTLV1, a retrovirus that can cause T-cell leukemia and lymphoma. Earlier, I mentioned another example: the bacterium H. pylori, which can cause stomach cancer.

There are at least a dozen pathogens that we now know can cause cancers; I’ve listed some examples below.

Note that, although many of these pathogens are quite common, only a fraction of people infected develop those cancers.

• Helicobacter pylori (H. pylori): A bacterium that can cause chronic inflammation of the stomach lining, leading to stomach cancer and mucosa-associated lymphoid tissue (MALT) lymphoma.

• Human papillomavirus (HPV): A group of viruses that cause genital warts and cancers, including almost all cervical cancer cases, as well as a share of oral, oropharyngeal, penile, anal, vaginal, and vulvar cancers.

• Hepatitis B virus and hepatitis C virus: Both can lead to chronic liver infection and inflammation, which can lead to hepatocellular carcinoma (a type of liver cancer).

• Human T-cell lymphotropic virus type 1 (HTLV-1): A retrovirus that can cause adult T-cell leukemia or lymphoma.

• Epstein-Barr virus (EBV): A herpesvirus that can cause several cancers, including Burkitt lymphoma, Hodgkin lymphoma, nasopharyngeal carcinoma, and certain stomach cancers.

• Kaposi’s sarcoma-associated herpesvirus (KSHV/HHV-8): The virus that causes Kaposi’s sarcoma, especially in people with weakened immune systems, such as people with HIV/AIDS.

There’s strong evidence that these infections can cause cancer because scientists have observed the pathogens infecting the specific cells that then develop into pre-cancerous growths and then into tumours. In many cases, they’ve also identified the intermediate steps showing how these pathogens affect critical genes in our body that typically suppress cancer growth (“tumour suppressor genes”) or other genes that drive cancer growth (“oncogenes”).

For some, countries have also used large-scale programs to eliminate the infections using antibiotics6, vaccines, or other tools, which have reduced the incidence of those cancers as well.7

But how do these pathogens cause cancer?

Pathogens can cause chronic inflammation, which damages tissues and increases the risk of cancerous mutations during repair. They can also produce proteins that directly interfere with genes or proteins that protect against cancer.

Some cancer-causing pathogens seem to contribute to cancer through both pathways: by inflaming tissues and disrupting cell regulation.

Cancer-causing pathogens generally only infect some types of tissues in the body. The chart below shows cancers that are part- or all-driven by pathogens, and the the estimated share of those cancer cases that could be prevented if the pathogens were eliminated. For example, researchers estimate that human papillomaviruses cause almost all cervical cancer cases, and that almost all cases could be prevented if they were eliminated.8

These are often tissues where, if these pathogens weren’t present, it would be extremely unlikely for cancers to form — typically because they’re stable, their cells rarely divide otherwise, and they’ve been generally protected from other exposures.

Thankfully, vaccines already exist for a few of the pathogens that can lead to cancer! And they not only prevent infections but also reduce the risk of cancer later in life.

One example is the human papillomavirus (HPV) vaccine. Randomized controlled trials have shown that it reduces pre-cancerous growths. As the vaccine has been introduced more widely, countries like the UK have seen sharp declines in cervical cancer rates, to the point where the disease has been almost eliminated in younger generations.9

Another important example is the hepatitis B vaccine, which is given to newborns in many countries and reduces the risk of liver disease, chronic inflammation, and liver cancer later in life.

Since liver cancer typically develops decades after infection, it’s often difficult to directly measure the long-term impact of newborn vaccination on cancer, even though the pathway (from hepatitis B/C infection to chronic inflammation to cancer) has been understood for some time.

One landmark study has helped estimate the effect more precisely. It was a cluster-randomized controlled trial10, conducted in the 1980s in Qidong, China, in which entire towns — rather than individuals — were randomly selected to give the hepatitis B vaccine to all newborns.

Researchers later linked this trial data with national health records and found that, more than thirty years on, those born in vaccinated towns had a 84% reduction in the incidence of primary liver cancer, and 70% reduction in the mortality rate from liver diseases. Because the allocation was random, we can be more confident that the large difference wasn’t due to pre-existing differences between the towns.

Meanwhile, we still don’t have a vaccine for hepatitis C, which also causes liver cancer. But it can now be cured with antivirals in most infected people, as the chart below shows. This nice review describes some of the key scientific advances that made this historic progress possible.

As an aside, my friend Jacob Trefethen says there are likely multiple reasons for slow hepatitis C vaccine development. The virus mutates rapidly within each person, making it hard to target; clinical trials are difficult to run, especially considering that most new infections in richer countries are among people who inject drugs, who are harder to recruit and follow over time in studies. And animal models have been limited, with chimpanzees no longer used for ethical reasons.

Only a few vaccine trials have been attempted, and their results have been inconclusive. It’s possible that a vaccine could prevent both first-time infections and reinfections, and some researchers argue that human challenge trials could speed up development.

Subscribe now

Breakthroughs in cancer treatment have reduced cancer mortality

I’ve sometimes heard people claim that chemotherapy and other cancer treatments only extend life by a few weeks or months. But that’s not really true. Perhaps this impression comes from looking at the benefit of individual new drugs in isolation, especially in late-stage cancers.

When you look at the full picture — including combinations of drugs, earlier diagnosis, and better care — treatment regimens have improved over time, and many cancers now have much better survival rates than they did just a few decades ago.

It’s worth remembering that until quite recently in history, it was extremely rare for anyone to become cancer-free.

Surgery was the only potential cure, and even that came with serious risks, including infection and death, especially before anesthesia and antibiotics. Today, with modern treatments and long-term follow-up, many people have become cancer-free11; and this is now common for several types of cancer.

Another point is that there actually have been some individual cancer breakthroughs that have made a huge impact, much more than just a few weeks or months. Although such large effects are rare, they do happen. We should keep these in mind to appreciate the real progress that has been made, and to learn from both the successes and failures.

Here are some major examples.

• Imatinib (“Gleevec”) — a tyrosine kinase inhibitor approved by the FDA in 2001 that transformed the survival of people with chronic myeloid leukemia (CML). Before it was introduced, people with this cancer had poor chances of survival, and their treatments were generally limited to bone marrow transplants or interferon therapy.

  • After imatinib was introduced, survival rates rose suddenly, which you can see in the chart below. It shows survival data from all CML cases in the Italian provinces of Modena and Reggio Emilia. Before imatinib was introduced, around 40–50% survived at least 5 years after diagnosis; the year afterwards, around 80% did.

    

  • Imatinib works by blocking an abnormal BCR-ABL protein (that forms as the result of a cancerous mutation) which signals leukemia cells to divide uncontrollably. By interfering with this signal, the drug can stop cancers with this mutation from progressing. The chart below shows how survival rates of chronic myeloid leukemia have changed over time.

    

• Trastuzumab (“Herceptin”) — a monoclonal antibody that targets the HER2 receptor, which is involved in some breast cancers. It was approved by the FDA in 1998 and substantially improved the survival of people with HER2-positive breast cancers.

  • In this key randomized controlled trial from 2005, you can see that people with HER2 positive breast cancer who took trastuzumab tended to survive longer than those in the control group. Around 85% survived 4 years with trastuzumab, versus 67% in the control group.

    

If you’ve followed this blog, you’ll also have read about some recent breakthroughs, such as osimertinib and vorasidenib, which have reduced cancer progression and death in large-scale clinical trials, and were both approved last year.

• Osimertinib (“Tagrisso”) is a small-molecule drug approved by the FDA in 2024 for people with a particular type of lung cancer, EGFR-positive non-small cell lung cancer, which is more common in people without a history of smoking and especially in East Asia.

  • Osimertinib blocks the mutated EGFR protein that drives the cancer’s growth. A recent trial of people with stage three cancer of this type showed that the drug dramatically extended people’s survival without the disease progressing. Those on placebo survived a median of 5.6 months without the disease progressing, compared to 39 months on the drug — nearly three extra years. It also halved the risk of the cancer spreading to the brain. It shows that even at an advanced stage, sometimes targeted treatments can hold cancer back for years.

• Vorasidenib (“Voranigo”) is an oral drug approved by the FDA in 2024 for people with IDH1/2-positive gliomas, a type of brain cancer that’s often diagnosed in younger adults and tends to be hard to treat.

  • Many patients with these tumors have mutations in the IDH1 or IDH2 enzymes, which help the cancer grow. Vorasidenib blocks both enzymes. In a recent trial, people taking it survived 27 months before their cancer progressed, compared to just 11 months in the placebo group. This is again a large impact.

These are a few examples of small molecule drugs and monoclonal antibodies that interfere with specific mutations and cancer growth signals.

But progress in survival hasn’t only come from new medicines. There have also been major advances in radiation therapy, surgical techniques, screening, and diagnostic imaging — as well as advances in cancer genetic research that’s helped identify which mutations are driving people’s tumors, so treatments can be tailored to the patient.

But how much do they matter? After all, lung cancer is now the most common cancer. Breast cancer risks are increased by smoking as well. So can’t we still say that smoking is mostly responsible for the decline in cancer mortality overall, and the other factors are relatively unimportant? Here’s why we can’t.

Subscribe now

We live in a multi-causal world

Imagine a woman who develops breast cancer. She goes to a doctor after noticing an unusual lump on her breast, gets tested and diagnosed with stage three HER2-positive breast cancer. Soon, she’s treated with anti-HER2 therapies like trastuzumab (Herceptin), other chemotherapies, and potentially surgery and radiation therapy.

Like over three-quarters of women diagnosed with breast cancer this way in the UK, she survives at least another five years; in her case, having her tumour removed and becoming cancer-free.12

But who was responsible for her becoming cancer-free?

Is it the scientists who discovered the HER2 mutation involved in her tumour? Or the researchers who developed medicines to block those receptors? The statisticians and staff who designed and ran the trials, compiled the data, and ran the analyses? Or the healthcare workers who ran the tests that detected the mutation in her tumour? The doctors who decided to use those medicines to treat her, or the surgeons who resected parts of the tumour? Maybe it’s the hospital staff who kept the center running, or the construction workers who set up the hospital in the first place. Or perhaps it’s the woman herself, since she decided to see the doctor to begin with.

My answer is that it was all of them: they all contributed in some way to making this happen. (And I find that inspiring.)

This is an example of the multi-causal world we live in — where her survival was possible because of the influence of many individuals, whose actions together saved her life.

Without a hospital and cancer doctors nearby, it’s less likely she’d be tested and treated at all. Without research to develop HER2-positive breast cancer treatments, her chances of survival might have still been low.

Without the pharmacists who prepared her medication, the engineers who built systems to track patients’ conditions and coordinate medical supply systems, or the logistics workers who transported supplies to her hospital, her treatment might never have reached her at all.

But we can still try to estimate the number of additional lives saved by a particular factor

It’s possible that a particular breakthrough is so effective that it saves far more people from cancer than would have survived otherwise, assuming everything else stays the same. So, in a hypothetical scenario where all other factors are held constant, we could conclude that a particular factor (such as a new drug) had a large causal effect on survival.

But that doesn’t mean the other factors suddenly became less impactful.

The drug’s effect depended on many other factors being in place. Without doctors, healthcare workers, researchers, hospital infrastructure, and many other people and resources, it’s very difficult to save those lives, even if the drug itself would be effective.

This tells us two things:

• We can estimate the added impact of a particular factor, by comparing it to a counterfactual scenario where that particular factor is absent. For example, we can estimate the added impact of a new drug, by comparing the actual number of deaths to the number we would expect if the drug had not been introduced.

• We still can’t neatly divide up the answer between different factors like slices of a pie, because their effects can combine and interact.13 An obvious example is that screening can only save lives if treatments are available; simply knowing that someone has cancer doesn’t save them, so you can’t fully separate the impact of screening from the impact of treatments.14 This problem remains even if the different causes don’t interact, but each contribute to an outcome.

For the same reason, even though smoking has had a large impact on reducing cancer, this, counterintuitively, doesn’t make other factors less important. I’ve previously written an article about this concept.

This is why, if it was the case that — and I hate to make up a number — but even if it was the case that the decline in smoking resulted in 70% of the decline in cancer mortality overall, it would be incorrect to interpret that as meaning that the remaining factors were only responsible for 30%. In theory, other factors could contribute another 50%, 70%, or more.

That’s because these aren’t the type of percentages that can be added up; they don’t sum up to 100%. They are called ‘attributable fractions’ and are considered individually. Although this confusion has been described in epidemiological papers for decades, I find it striking how relevant it is to this discussion.

There are some additional takeaways I have from this concept.

1. There are often multiple ways that the same lives could have been saved. For example, people with lung cancer might be saved with new medicines, surgical techniques, or if they hadn’t smoked in the first place. In many cases, multiple things will have contributed to their life being saved, and we can’t pin it to only one factor.

2. But this also means we can’t just add up the effects of each approach as if they work independently. Many strategies may overlap: they protect or treat the same people. Once someone’s life is saved by one intervention, they might survive for a while without needing to be saved from the same threat again. So, adding up the number of lives saved by each method would double-count many of the same individuals and exaggerate the overall impact.

Conclusion

I started writing this post with a simple question in mind: what caused the decline in cancer death rates? But the more I thought about it, the more I realized that the question itself was trickier, and more interesting, than it first seemed.

It’s not just about cancer. It’s a question about how we understand causes and effects overall. We live in a world where things don’t happen in isolation; they interact and their effects combine. And while it might seem unsatisfying at first that there’s no single answer or percentage we can divide up and assign to each cause, that doesn’t mean we’re unable to say anything useful. We can still try to estimate the impact of specific changes, we just need to be careful not to treat them as separate or additive when they often work together.

This might sound like a conceptual or academic issue, but it shapes how we think about the world: how we evaluate new policies, how we interpret the impact of scientific breakthroughs, how we judge the value of public health programs and healthcare systems. I think it shapes how we understand scientific progress itself.

As always, I hope you enjoyed reading this. If you haven’t already, I hope you subscribe and share it with your friends! If this is your first time reading, I recommend checking out the About page.

This Substack is free, I have no plans to monetize it, and I’ll pay you if you spot any errors in this post and let me know. (Minor typos and grammatical errors excluded.) If that’s not a great deal, I don’t know what is.

See you next time! :)

– Saloni

P.S. Thanks very much to my friend Julia Rohrer for feedback on this post.

In recent months, US foreign aid programs have been slashed. It’s hard to wrap your head around the scale of these programs. The largest, PEPFAR, has provided HIV treatment and prevention to 20 million people annually – mostly in Africa, but also in other countries around the world – and is estimated to have saved around 25 million people since it began. It was launched by the Bush administration in 2003. Another one is PMI, also launched by Bush, which provides antimalarial bed nets and treatment to tens of millions of people each year, and is estimated to have saved more than 10 million lives since it began.

Many people don’t know how important global health programs have been to people around the world despite costing only a tiny fraction of our incomes in richer countries.

Even aside from HIV/AIDS and malaria, we’ve almost wiped out polio worldwide thanks to global vaccination campaigns. Deaths from tetanus have fallen more than seven-fold in three decades. Millions of children have been effectively treated for trachoma, a painful bacterial infection that can cause blindness. But many people haven’t heard of these achievements. Maybe that’s why it can all seem distant and abstract, not real and urgent.

The disruptions started at the end of January when the US froze all foreign aid programs to conduct a “three-month review.” But since then, many programs tackling HIV, malaria, and other global health efforts have been permanently canceled. The US also withdrew all their funding for Gavi, which provides vaccines against polio, measles, diphtheria, cervical cancer, and many other diseases, for millions of children in poorer countries.

Given the scale of these programs, the impact of the disruption was immense. With PEPFAR serving tens of millions per year, hundreds of thousands were losing access to HIV treatment per day.

What does that actually mean? In many cases, clinics had lifesaving antiviral treatment on their shelves, but were asked to shut down. USAID staff were laid off, programs were no longer being paid, and procurement of more supplies had stopped as well. Suddenly, hundreds of thousands of people – every single day – were finding out they could no longer get their next round of HIV treatment.

In Africa, over 60% of people living with HIV/AIDS are women. Around a million are pregnant. But today’s antivirals are so effective that if taken during pregnancy, the risk of passing HIV to the infant is minimal. Without these drugs, around a quarter of babies born to HIV-positive mothers would also contract the virus – and half of them would die before the age of two.

The freeze meant over a thousand more newborns were contracting HIV every day, because their mothers had lost access to treatment: without warning, through no fault of their own, and often, with no alternatives.

The economists Charles Kenny and Justin Sandefur estimated that around 6,000 people would die per day from a complete US aid freeze, which was the case during the “90-day pause”. They estimate that around 3.3 million would die if it continued for a year. I find it hard to understand how this could happen at such a scale. It’s terrifying to try to imagine.

Around a third of the local organizations running PEPFAR had completely shut down during the freeze because they were unable to secure replacement funds and staff. Less than 10% restarted providing treatment weeks later.

Then, all of a sudden on March 10th, the Trump administration announced the review was completed, in 35 business days. They “made a final decision with respect to each award, on an individualized basis,” to terminate 8,644 State and USAID awards. As the economist Charles Kenny put it, “That’s about one award every minute and 20 seconds.” It must have been quite the review process.

Several cancellations were challenged in court. In several cases, judges ruled that the administration was legally required to pay out the funds. One Supreme Court ruling compelled the Trump administration to resume around $2 billion in foreign aid funding for work already completed.

Only some of the funding has been restored since then. It seems that about a quarter of HIV program funding has been permanently terminated. But since thousands of USAID staff were laid off, even those programs whose funding was restored are still disrupted. Both USAID’s main website and PEPFAR’s data portal are still unavailable, as of April. A few weeks ago, the remaining staff at USAID were asked to “shred as many documents first, and reserve the burn bags for when the shredder becomes unavailable or needs a break”.

In some cases, research projects had already rolled out new testing methods or treatments and were collecting data – but funding for the final steps, like analysis, was cut. That means the entire study could now go to waste.

For weeks – sometimes daily – reporters at the New York Times, Washington Post, and others covered the disruptions, and I thought more people and governments would notice and act on it.

But then the UK government announced that we would also be cutting foreign aid, despite earlier campaign promises to preserve or even increase it. Instead of stepping up, we in the UK are following the Trump administration in making deeper cuts. Why?

Somehow it still feels unthinkable that we’d let millions die preventably. Or we’d suddenly take away treatment people have relied on for years without debate, or any warning. Foreign aid costs a very small fraction of national income, and the US in particular spends relatively little, but in absolute terms, it makes a huge impact in poorer countries.

For a long time, I didn’t think personal charitable donations were “enough”, so I only donated occasionally as a one-off.

The better way I thought to make a difference was by fostering the policies, institutions, and innovation that made large-scale change possible: Trade and growth-oriented policies that make it possible for countries to invest more in healthcare and public health programs. Large aid-supported programs like PEPFAR, PMI, and Gavi that support millions of people. Breakthroughs like new antivirals or malaria vaccines, and innovations that can cut the cost of treatment dramatically, making it possible to save far more lives with the same amount of resources.

But this has made me realize how easy it is to take global health and humanitarian programs for granted, and I want to help however I can. Bridging the gap is urgent.

What’s the use of developing new technologies without the infrastructure – the people and resources – to deploy them? That’s what’s at risk right now. If programs shut down now, we could lose years of built-up infrastructure and expertise, and it’ll be very hard to bring them back. Millions of people are now at risk of losing that.

Worse is knowing that we’re on the cusp of rolling out new breakthrough treatments – like a new antiviral that prevents HIV with a 96–100% efficacy and only needs to be given out once per six months – that could have helped eliminate HIV transmission; and new malaria vaccines that could save a million children over the next few years.

The gap is difficult to fill with philanthropy alone and it’s terrifying to really think about. Imagine one in ten people around you have HIV, but they’ve suddenly lost treatment to this deadly disease that could kill them in years. That’s the sudden reality that millions of people faced during the freeze, and may continue to face, in some countries.

But that doesn’t mean that we can’t do anything about it. We don’t need to see this unleash the maximum possible amount of damage on millions of people. We can reduce that as far as we can, and I think we can keep some of these programs alive.

For one, I’ve decided to donate nearly 20% of my income this year, and I’ve committed to donating at least 10% of my income over my lifetime. On an individual level, this still pales in comparison to what’s needed, but I’d like to tell you how I thought about it.

I spent time thinking about how to donate that amount most effectively this year. Should I split my donation across multiple organizations, or give it all to one?

Right now, programs like the Foreign Aid Bridge Fund and the Rapid Response Fund, are helping to fill some of the gaps caused by the cuts. I think their work is really important. Both are trying to find some of the most cost-effective programs affected by the cuts, and raise philanthropic donations to direct to them. In case you’re interested in more detail, Sigal Samuel wrote a good article at Vox about them.

But recently, I also attended a talk by people working at Project Resource Optimization, who are experienced in cost-effectiveness analysis and have compiled a live list of high-impact USAID-funded programs affected by the cuts. They’re regularly talking to USAID-funded local programs and implementing partners to find out which have shut down, and which might still be saved.

I was very moved by their work and the thoughtfulness of their approach as well as their clarity on what was urgently needed right now. Rather than choosing larger programs or ones that are already well known, they compiled a “first cut” list of all potentially lifesaving programs in global health areas that were terminated. They then narrowed this down to an “urgent and vetted” list of those that they estimated were the most impactful, and cost-effective, meaning they are likely to save the most lives with a given amount of money, and crucially, which are urgently at risk of shutting down. Their list is online here.

I chose to donate 16% of my income to one of the smaller projects on that list. This was one where I think my donation would have the highest likelihood of helping it get across the threshold to keep it alive.

The one I chose was planning to expand clean water and sanitation in parts of Nigeria – that basic global health intervention known as ‘clean water’ that reduces the spread of diseases like cholera, rotavirus, polio, and many others. These infections can be deadly and are sadly common causes of child deaths in poor countries.

It’s also one that’s close to my heart because I believe that clean water and sanitation are still rather undervalued. A recent meta-analysis I’ve written about before looked at clean water treatment programs and their effects on child mortality across RCTs in 21 poorer countries. They found that clean water treatment, such as simple water chlorination, reduced child mortality by around 30% overall in the studies in these countries.

That’s a surprisingly large impact. But perhaps it seems that way to us living in countries with large-scale sewage systems and ubiquitous taps with clean drinking water that it’s hard to imagine how much it matters to people who don’t have it.

Then I donated another 3% of my income to cost-effective charities through GiveWell, which is also working to cover the shortfalls left by the cuts. In a recent post, they explained in detail how they’re responding in filling urgent gaps and expanding their core areas, with more funding towards malaria prevention and treatment, HIV testing, and healthcare worker training.

(For disclosure: I work at Our World in Data, one of many non-profits that receives some GiveWell funding.)

Although I believe funding is particularly urgent right now, I also think it’s important to think beyond the next few months and support these programs in the long term. That’s why I also signed up for the “10% Pledge” on Giving What We Can – meaning I’ll donate at least 10% of my lifetime income to effective charities – as a long-term commitment.

I wish I’d pledged earlier. If this post inspires you, or someone you know, that will make an even bigger difference. If you can inspire others too, maybe we could collectively keep more of these programs alive.

The impact of funding cuts is especially severe right now, so the sooner people take action, the better. But at the same time, I think it’s important to be strategic by supporting impactful programs that are providing essential services right now and are at risk of shutting down, as well as stable programs that can expand into these areas.

In the longer term, I think there are many ways to contribute. Global health and humanitarian programs rely on people doing lots of different types of work – from those providing healthcare directly to people running operations, handling logistics, writing grants, raising awareness, or doing research. You don’t have to work on the front lines or donate large sums as an individual to make a difference.

Beyond that, I’d like to see more governments have the moral courage to step up right now. Personal, philanthropic donations are just a small fraction of what governments can spend at a national level. We don’t all have to sit back and watch this horror unfold.

Share

Where you can donate

Here is a brief list of some of the places you could donate.

• GiveWell – A nonprofit that identifies and recommends some of the most cost-effective, evidence-based charities in global health. In response to the aid cuts, they’ve prioritized filling urgent funding gaps in malaria prevention, HIV testing, and healthcare worker training.

• Project Resource Optimization – A team of researchers and cost-effectiveness analysts who built a live list of high-impact USAID-funded programs affected by the aid freeze. They’re talking regularly with implementing partners to identify the most urgent and cost-effective projects at risk of shutting down. On their ‘funding opportunities’ page, you can find programs directly affected and donate to them directly, or contact the research team for more information.

• Foreign Aid Bridge Fund – An emergency philanthropic fund created to support lifesaving programs suddenly defunded by the foreign aid freeze. It focuses on bridging short-term gaps to prevent program collapse, such as HIV clinics in Kenya and Zimbabwe, health programs in Ethiopia, and disaster-resilient housing in Nepal.

• Rapid Response Fund – Another emergency philanthropic fund launched by Founders Pledge and The Life You Can Save to quickly support top-rated global health and poverty programs facing sudden shortfalls. It provides emergency funding to keep high-impact interventions running during the aid cuts.

• Gavi – A public-private global health partnership that provides vaccines to millions of children in low-income countries. The US, their largest donor (12%), withdrew all their funding for Gavi, putting global vaccination campaigns at risk.

If you want to sign up to a 10% Pledge from Giving What We Can like I did – and commit to donating at least 10% of your lifetime income to effective charities – you can do that here.

They also offer a more flexible Trial Pledge which can be 1–10% income for a fixed period of time, such as a year, for people who want to dip their toes into the water.

In case it bears repeating, this post and my newsletter as a whole reflect my personal views and not my employers’.

I find that frustrating — especially when there are genuine breakthroughs that have reached the final stages of testing that most people still never hear about.

So I want to do it differently. Here I’ve highlighted five medical treatments that have shown high efficacy in (often multiple) phase three randomized controlled trials, which are large, rigorous, and receive much closer independent oversight. I’ve chosen medical treatments whose results were published in 2024.

Because of this criteria, these are just a slice of the important medical advances this year, and only include those that made it through large trials; there are likely a lot more that I’m not aware of, including at earlier stages at research, which can be more difficult to interpret.

As usual, I’ll pay you if you spot an error in this post, aside from minor typos or grammatical errors. Please let me know if you find one, so I can fix it.

Subscribe now

1. Lenacapavir: a game changer for HIV prevention

First on this list is the one I’m most excited about.

Lenacapavir is a long-acting antiviral injection for HIV. It was initially approved to treat people with resistant HIV, and recent trials show it’s also highly effective for prevention.

Most HIV prevention methods rely on daily pills, but remembering to take them can be difficult. Stigma and access barriers make it even harder, particularly for people in sub-Saharan Africa, the region most affected by HIV.

Recently, injectable alternatives like cabotegravir, which is taken every two months, have helped. But lenacapavir goes further: a single injection every six months.

The key is its potency and duration. It targets the HIV capsid — the virus’s protein shell — which is a new approach for HIV drugs. And, thanks to clever medicinal chemistry, lenacapavir is stored in fat tissue and slowly released over months. Though its half-life is two weeks, it’s effective in very small quantities, meaning its effects last at least six months.

Two recent phase 3 trials have demonstrated its effect as a preventive drug. In the first, with cisgender women in Uganda and South Africa, not a single woman on lenacapavir got HIV — a 96–100% reduction compared to the background risk. And it was over 9 times more effective as tenofovir, Gilead’s daily oral PrEP pill. Here’s the key figure:

In the second trial, involving men and gender-diverse people across six countries, only two people taking lenacapavir out of thousands contracted HIV, an 82–99% reduction in risk compared to the background rate, and again, 9-fold improvement over tenofovir.

It may be surprising that tenofovir, the daily oral pill, appeared ineffective: one major reason is that many people struggle to take it consistently, which can be due to forgetfulness, stigma, or unreliable access, a challenge in the poorest regions.

The chart below shows this: adherence to taking tenofovir declined over time, leaving people less protected. Lenacapavir avoids this problem by requiring just two injections a year, making it much easier to stay protected.

What’s the catch? Trial data shows most people only experienced mild injection site reactions, and had fewer stomach side effects than oral PrEP.

The bigger challenge is scaling it up. Gilead has agreed to allow generic manufacturers in poor countries to sell it so it can be produced and distributed affordably in poorer countries. But for middle- and high-income countries, cost and insurance coverage will be key to access, and health systems will need to find ways to make sure people who need them can actually get their twice-yearly shots.

What’s next? It’s already been in use for a few years as a treatment for people with HIV resistant to other drugs, but now it’s also being tested as a first-line treatment for HIV. A twice-yearly treatment injection, if it succeeds, would also be hugely transformative.

There’s a lot to be excited about, in my view. Innovation like this seriously changes our ability to tackle the global HIV pandemic.

So far, HIV drugs have been able to reduce the severity of HIV and slow its transmission. But preventive drugs have often fallen short because they’re hard to take consistently.

Some countries like the UK are on track to effectively eliminate HIV transmission in the next decade. But infection is still common in poorer countries. Lenacapavir changes that and should raise our expectations for what’s possible, both for the global effort against HIV and the prospect of further innovation.

Subscribe now

2. Omalizumab: finally, a step forward in preventive food allergy medication

Omalizumab is a monoclonal antibody drug, given as an injection, which helps people tolerate various food allergens, including peanuts and milk.

It was already approved to treat asthma in 2003 and later, for hives and nasal polyps. In the 2000s, some doctors noticed it was helping their asthma patients with additional food allergies. Unfortunately, insurance typically wouldn’t cover treatment for conditions a drug isn’t directly approved for, so trials were needed to confirm its efficacy against food allergies directly. But those trials hit roadblocks in the 2000s and were stopped, which you can read about in this great article about omalizumab by Sarah Zhang in The Atlantic.

Only recently were larger trials conducted. After positive phase three trial results, the drug was approved this year by the FDA for food allergies.

For people with severe food allergies, exposure to small amounts of allergens can be life-threatening. Current treatments are typically taken after exposure to the allergen, although some people also try “oral immunotherapy” to train the immune system, where people have gradually increased doses of the allergen, but this takes months and can result in severe reactions if it doesn’t work.

On the other hand, omalizumab acts quickly. The drug is injected every 2–4 weeks and blocks “IgE” antibodies, which play a big role in allergic reactions. Blocking them makes the body less sensitive to some allergens.

The new phase three trial, which included people (mostly children and adolescents) with multiple food allergies, showed that omalizumab improved tolerance of small quantities of various food allergens.

As you can see, while 67% of participants in the trial taking omalizumab could tolerate a challenge of 600 mg of peanuts, only 7% taking the placebo could. There was also protection against other food allergies, such as cashews, eggs, and milk:

What’s the catch? Unfortunately, around a third to half of people in the trials still couldn’t tolerate those amounts, even with the drug. You can see this in the same chart above. Also, the amounts of allergens tested were fairly small: ranging from 600 to 1,000 mg.

So, Novartis and Genentech, which manufacture omalizumab, continue to recommend that people taking the drug avoid consuming those foods because it may not provide complete protection. The main benefit is reducing the risk of accidental reactions.

This is still an important advance, especially for young children. It means less fear of a severe allergic reaction to trace amounts, which can be a risk in foods unlabelled or served at restaurants or homes.

I’m excited because I think this signals what’s possible in improving allergy treatment and prevention, and I’m sure there’s still a lot more innovation to come in this area.

What’s next for omalizumab? The results above are only the first part of the phase three trials. Next, researchers will test its longer-term efficacy versus the existing allergy therapy, and third, its protection after introducing food allergens into the diet.

Subscribe now

3. Xanomeline-trospium: a new era in schizophrenia treatment

This year, a new antipsychotic drug was approved: Xanomeline-trospium, also known as KarXT (under the brandname Cobenfy).

Unlike many antipsychotics, which target dopamine receptors, this drug targets muscarinic receptors in the brain, but not dopamine receptors, and could reshape future innovation.

The story of xanomeline goes back to the 1990s, when it was originally tested as a treatment for Alzheimer’s disease. During those early trials, something unexpected happened: participants experienced fewer hallucinations, delusions, and episodes of agitation. These symptoms are often seen in schizophrenia.

The results led researchers to test xanomeline as a potential schizophrenia drug. But it led to major side effects, like severe nausea, vomiting, and stomach pain, which ultimately sidelined the drug.

The issue wasn’t with the brain’s muscarinic receptors, which the drug was meant to target. Rather, xanomeline also affected muscarinic receptors outside the brain, especially in the digestive system. To address this, researchers paired xanomeline with another drug called trospium, which doesn’t cross into the brain but blocks muscarinic receptors in other parts of the body. The combination, xanomeline-trospium, allowed xanomeline to work on muscarinic receptors in the brain while trospium minimized its digestive side effects.

With this combination treatment, new clinical trials began.

Three large phase three trials have found that xanomeline-trospium effectively reduces a broad range of schizophrenia symptoms — including what’s called the “positive symptoms” of schizophrenia (e.g. hallucinations and delusions), but also “negative symptoms” (e.g. social withdrawal). You can see this below, and in their pooled results across all trials.

One major difference with previous antipsychotic medications is that xanomeline-trospium lacks movement-related side effects (such as tremors), metabolic changes, and weight-gain that were seen in previous drugs, making them easier to take regularly. See this great post by Nils Wendel for more about it.1

For people with schizophrenia, and their friends and families, this is a really important advance. It’s a step toward more effective, tolerable treatments, and hopefully the beginning of a new generation of antipsychotic medication.

If you’re interested, I recently wrote an article about the history of antipsychotic medication, which I updated after xanomeline-trospium was approved.

Now, research is ongoing on its potential effects on psychosis in Alzheimer’s disease, once again.

Subscribe now

4. Tirzepatide: weight loss, diabetes, and more

A list of medical breakthroughs in 2024 wouldn’t be complete without a GLP-1 drug. We often hear about semaglutide, but here I’ll tell you about tirzepatide, a similar drug.

Tirzepatide is a once-weekly injection that helps with weight loss and reduces the risk of diabetes, among other effects.

It works by mimicking two natural hormones, GLP-1 and GIP, which help regulate how the body processes food and sugar. By activating the receptors for these hormones, tirzepatide reduces hunger, increases the feeling of fullness, and improves how the body manages blood sugar.

There have been a series of large phase three trials of tirzepatide. In the SURMOUNT-1 trial published last month, people with obesity and pre-diabetes who took tirzepatide had a 90% lower risk of progressing to diabetes, which you can see below.

In the SURMOUNT-2 and SURMOUNT-4 trials, also published this year, tirzepatide helped people lose weight and keep it off.

The SURMOUNT-4 trial (shown below) involved people taking tirzepatide for 36 weeks (9 months), and then being randomized to continue the drug or switch to a placebo. Those on the placebo gradually regained weight over the following year, while those on tirzepatide continued to lose weight — with an average 25% reduction in body weight and waist circumference in under two years.

People with diabetes or obesity are not only affected by the conditions themselves, but also various other health problems, including heart disease, kidney disease, and sometimes sleep problems.

GLP-1 drugs, which were originally approved to treat diabetes, appear to be highly effective against all of the above.2

It’s easy to disbelieve or downplay the importance of these advances, but doing so means ignoring the alternatives and what came before.

Traditional weight-loss methods like bariatric surgery, dieting, and exercise can be effective, but they come with major challenges. Surgery is invasive and costly, while diets and exercise routines are often hard to maintain, especially for people who feel hunger more intensely than you or me.

Many previous attempts to develop obesity medication failed: some had immediate toxic side effects, others increased the risks of psychiatric problems and even suicide, and many simply didn’t work.

Stephan Guyenet wrote a great article about this history for Works in Progress two years ago. Efforts to develop GLP-1 drugs also failed for a long time, until breakthroughs in research finally helped them succeed. With multiple large trials looking at efficacy and safety, I think we can be confident that these new drugs are much safer.

There is a catch though: mild but unpleasant side effects (like nausea, diarrhea, and constipation) are common in people taking GLP-1 drugs, especially at the beginning of treatment.

More importantly, demand for these drugs vastly outstrips supply, given obesity and diabetes are common, and currently only a few manufacturers produce them. It’s going to be important to find ways to scale them up, reduce their prices, and incentivize further innovation.

Subscribe now

5. Osimertinib: life extension for people with certain types of lung cancers

Osimertinib is designed for people with EGFR-positive non-small cell lung cancer, which is a form of lung cancer driven by mutations in the EGFR gene. This type of cancer makes up about 10–15% of lung cancer cases in the US, but is much more common in East Asia. It is especially common in people with lung cancer without a history of smoking.

Osimertinib is a small molecule drug that inhibits the EGFR tyrosine kinase protein, which is mutated in certain lung cancers and signals cancer cells to grow and divide.

While it’s already shown clear effectiveness for earlier-stage cancers that can be surgically removed, this latest trial focused on people with advanced lung cancer: specifically, stage 3 non-resectable cancer, meaning surgery is no longer an option. This can be because the tumor has grown near vital organs or spread to nearby tissues or lymph nodes. In these cases, effective treatment options are limited.

Below are results of the LAURA trial, which took place in several countries including the US, India, Japan, Brazil and China.

At the end of the trial, people taking osimertinib were 84% less likely to see their cancer spread or worsen, compared to those on the placebo. Another way to understand these figures is through the time without progression. People on the placebo either died or saw their cancer worsen in an average of 5.6 months. But for those taking osimertinib, this extended to 39 months — almost three extra years.

Osimertinib also reduced the risk of cancer spreading to the brain, which can be a major concern for lung cancer patients. After twelve months, 20% of people on osimertinib developed brain tumors or died, compared to 41% of those who didn’t take it.

Like most cancer drugs, osimertinib can cause side effects. Nausea and vomiting were the most common, but these were mild, and the rate of many other complications was similar in both groups since they are common for people with advanced cancer.

Osimertinib demonstrates that cancer can be held back for years even at a late stage. It’s also another example of precision medicine in cancer treatment: by understanding the genetic mutations that evolve in tumors, scientists can develop more targeted and effective therapies.

Subscribe now

Concluding thoughts

For many, these might seem like surprising, one-off advances. In my view, they’re a few striking examples in a continuing stream of medical innovation.

Annual death rates from various conditions, including cancers, cardiovascular diseases, infectious diseases, have declined greatly over the long-run and advances in cancer therapy, cardiovascular medication, antibiotics and vaccines are just a few of the many reasons why. This progress has continued in recent decades, in rich countries. Life expectancy has risen across all ages, across many countries. Records of life expectancy have been broken repeatedly.

In a 2002 paper called “Broken Limits to Life Expectancy”, the great demographers Jim Oeppen and James W. Vaupel describe this progress:

The linear climb of record life expectancy suggests that reductions in mortality should not be seen as a disconnected sequence of unrepeatable revolutions but rather as a regular stream of continuing progress.

And I think they’re right. We’ve continued to see large breakthroughs in medical innovation, and it is interesting and worthwhile to ask how they happened, whether they could’ve happened sooner, how to improve the pace and quality of medical research, and how to make sure that innovation reaches those who need them. After all, this is biology’s century.

Several more specific things also come to mind with these examples.

One is the different ways to work on medical innovation. Medical innovation can mean improvements in efficacy, safety, duration, a wider breadth of effects. It can also mean achieving the same effects with fewer doses, fewer concurrent medications, and so on.

It’s not just the first effective treatment for a condition that’s the breakthrough. “Incremental innovation” can mean the difference between a drug that prevents HIV infections in the long-term versus one that does not reduce the risk in practice, because people don’t take it regularly.

Another takeaway is the importance of various steps in the process of medical innovation:

• Basic research has helped understand the triggers of disease and their progression, which helps identify new targets. We see this with lenacapavir (the first-approved HIV antiviral targeting the viral capsid), xanomeline-trospium (the first to target muscarinic receptors in particular instead of dopamine receptors), osimertinib (which targeted particular EGFR-mutated proteins, which research shows drive cancer progression), tirzepatide (developed on the back of understanding the effects of GLP-1 and GIP hormones), and omalizumab (targeting IgE antibodies, which play an important role in allergic reactions).

• Medicinal chemistry has helped refine drugs to make them better, safer, and longer-acting. We saw this with lenacapavir (formulated so it would be deposited in fat tissue and slowly released over months), xanomeline-trospium (where the combination helped reduce side effects). But also osimertinib (which can cross the blood-brain barrier, allowing it to reduce cancer spread to the brain); tirzepatide (which is a synthetic protein that acts on both GLP-1 and GIP receptors); and omalizumab (where the antibody structure’s gives it high specificity and safety for targeting on IgE antibodies).

• Repurposing existing drugs can be transformative. Omalizumab (originally for asthma) and xanomeline (originally tested for Alzheimer’s) were successfully repurposed for new uses, highlighting that breakthroughs don’t always require brand-new molecules — sometimes just a new application. Researchers know this, of course, and occasionally, the process of screening existing drugs leads to new repurposing.

• Sometimes it takes many years for drugs to make it to the market. Omalizumab (originally developed in the 1990s and approved in 2003) and xanomeline (also developed in the 1990s) are the clearest examples of this, though fortunately there were shorter timelines between development and approval for the other drugs. The history of antipsychotic medication also includes another pivotal example: clozapine, which was developed in the 1950s but first approved in the US in… 1990. Clozapine’s story is important because it inspired much more innovation in the field of antipsychotic medication; I’ve written about why the process took so long in an earlier article.

  • Timelines are often long especially for neglected conditions, where there’s often little financial incentive for pharmaceutical innovation; and for rare and sporadic diseases, where it’s difficult to recruit enough participants for trials or reliably test drugs’ effects. This doesn’t need to be the case, and there may be new pathways and incentives to accelerate their research.

I’ll end there, except to apologize for the long delay since my last post (eight months ago, yikes!) There’s unfortunately no one to blame but myself: I had several half-written drafts that I scrapped, got caught up in other work and hobbies on the weekends (which is when I usually write this), and found it hard to carve out the time to write.

But now I plan to write this regularly again, hopefully every week.

As always, I hope you enjoyed reading this. If you haven’t already, I hope you subscribe and share it with your friends! If this is your first time here, I recommend reading the About page.

This Substack is free, I have no plans to monetize it, and I’ll pay you if you spot any errors in this post and let me know. (Minor typos and grammatical errors excluded.) If that’s not a great deal, I don’t know what is.

See you next time! :)

– Saloni

Share

Much of my work these days is about data visualization. It still feels a little new and I’d been so used to writing that I hadn’t fully grasped why visualization mattered until recently.

But the big reason I’ve learnt to visualize data is to explore and learn about a topic in a way that would be difficult with just a written scientific description.

I get the feeling that many people underestimate how valuable this can be, so in this post I’m hoping to give you a greater appreciation for how data visualization can help explore, probe, and understand a topic more deeply.

Each chart in this post can be recreated with code I’ve shared on my GitHub.

The key thing to remember with data visualization is to ask: what question am I trying to answer?

Iʼll do this with a topic I’ve often written about — causes of death. This time I’m interested in the question: How do causes of death vary with age?

It’s a fairly broad question, and it’s not obvious how to answer it. So Iʼll start by just exploring a few different chart types on the topic. To do this, I’ve created charts using national data from death certificates across the United States, between 2018 and 2021.

This first and the simplest stacked bar chart looks at the number of deaths from each cause across age groups.

What stands out is that most deaths tend to occur at older ages, and come from a range of causes, but especially cardiovascular diseases (pinkish brown) and cancers (blue).

I also can see that ‘external causes’ (in red) are prominent in youth. This is a broad category that includes violence, accidents, suicides, overdoses, and some other causes of death.1

And infancy also contributes to a large number of deaths overall. Most infant deaths are from perinatal conditions (purple).

At the same time, there’s something strange about this chart. It looks like many of these causes of death are leveling off with age. Is that really the case?

To understand whether the risks of death are leveling off with age, I can instead look at the death rate from each cause at different ages instead.

I’ve shown this with line charts below. The data again comes from death certificates in the United States between 2018 and 2021.

As you can see, the risks don’t level off. They continue to rise exponentially with age for many causes of death. But some causes are also very deadly during infancy.

And you can see clearly that ‘external causes’ don’t fit the usual pattern: they rise very suddenly during adolescence. In fact, they are the only cause of death with mortality rates that high at that age.

I’ve written an article about this here: How do the risks of death change with age — and how have they changed over time?

But why did causes of death appear to level off with age in the previous chart, showing the number of deaths?

Well, let’s think about what the number of deaths depends on — both the risk of dying and the actual number of people who had been alive at that age and were at risk of dying.

So let’s plot the population size at each age in the United States. I’ve shown this in a frequency chart below.2

Now I can see that the population is actually smaller at older ages, which is why the total number of people dying had started plateauing as well.

Since weʼve now solved that little mystery, let’s go back to the chart showing death rates.

While I like this chart, I think it has drawbacks.

I canʼt easily answer the question ‘How do causes of death vary with age’, because I can’t compare them very easily on this chart across age groups. Each cause of death is shown with a different y-axis scale. So if I was trying to compare them, I’d have to keep flitting my eyes between the y-axes, the x-axes, and the different cause-of-death labels, to try to compare any pair of causes, which is annoying.

And since some of them are so much smaller than others, if I put them all on the same y-axis scale, some causes would look so small that I wouldn’t be able to see precisely how their risks changed with age.

A friend suggested I try something else: if I stack death rates from different causes, then I can see them all on the same chart and also see the total death rate from different causes at different ages.

I’ve shown this in the chart below. Here, the total height of the curve shows the total annual death rate, out of 100,000 people at that age.

[Stacking rates is an approach that often makes me feel queasy, but it works here because causes of death are exclusive — people only have a single ‘underlying cause of death’ coded from their death certificate.]

This chart helps clear up some of the previous confusions. It also shows that the total risk of death isn’t leveling off, but continues to rise with age. And it helps to see the relative risks of dying from different causes at each age.

On the other hand, the age-patterns within each cause of death aren’t very easy to see anymore, because of the stacking. It isn’t clear here that cancer and cardiovascular death rates continue to rise with age, for example.

As for the youngest ages, honestly I can barely tell what’s going on.

So let’s try a fourth way of exploring the data. This time, I’ll look at the relative share of deaths from each cause, at different ages. This stretches it out to fill the space.

Now I can see precisely what share of deaths at each age results from different causes.

This helps me see that most deaths in childhood and adolescence are caused by ‘external causes’.

I can also see that at older ages, diseases have gained prominence. Cancers and cardiovascular diseases have become the most common causes of death at older ages, and causes of death have also become more varied.

But what’s not helpful is that, by focusing on the relative share, this chart obscures the underlying reasons for these trends.

For example, although external causes look very prominent here, deaths from other causes at older ages actually dominate overall, as we saw in the first chart.

Or take a look at the apparent decline in cancers after the age of around 65, for example. Unfortunately, this doesn’t reflect an underlying decline in the risks of dying from cancer.

Both cancers and cardiovascular diseases actually continue to rise exponentially with age, as we saw before. (I’ve repeated the chart below.) If you look at the y-axis scales, you’ll notice that cardiovascular disease deaths have grown much more than cancers. This is why cancer deaths appeared to decline — they were declining only in relative terms.

With all these charts, I’ve now learnt a lot about different aspects of data on causes of death across the lifespan. I’ve also explained what I like about each chart and what I don’t.

But which one do I prefer?

Letʼs think about the question I had initially: How do causes of death vary by age?

Unfortunately, there wasn’t just one chart that answered it satisfyingly, and I think the question was just too broad to answer with a single chart.

But each chart helped me gain a broader understanding of the topic, and told me something meaningful on its own.

So here are the four charts again, this time titled with the specific questions I think they answer.

This first chart below — the number of deaths — simply shows how many people are dying from each cause at each age. It also helps to see the total number of deaths at each age.

The second chart below — the death rates — shows how the risk of dying from each cause varies with age.

It’s not geared towards comparing different causes, but to seeing the age patterns within each cause separately.

The third chart below, where death rates are stacked, actually helps me compare them. It also helps me see the total risk of death across causes. It answers the question: what are the risks of dying from different causes at each age?

The fourth and final chart below — the relative share — had been a confusing chart to understand at first.

Many different factors affect this figure: not just the risks from each cause, but also how they compare and displace each other at older ages.

But if we step back, the question it answers is actually fairly simple. What are people dying from at different ages?

Each chart I’ve shown answers a valuable question on its own. Of course, not every chart needs to be titled with a question. I’ve placed these questions in the titles because they help remind me of what exactly I’m trying to explore and understand.

I think exploring different types of charts has helped me understand the topic overall, and also helped understand what each type of chart is useful for.

If you want to recreate these charts, you’ll be able to find the data and code on my GitHub. There you’ll also find code to split the charts by sex. There’s also another version to split ‘external causes’ into its subcategories. (If you do decide to recreate or adapt any charts with my code, I’d love to know what you create!)

I hope you enjoyed this! As I shared earlier, this is a new section within Scientific Discovery called “Notes”, where I share thoughts on sometimes more technical topics.

Iʼll treat this as a separate blog, but you can opt out of these posts separately here.

As always, if you’ve spotted any errors in this post, I’d very much appreciate knowing. I offer rewards for it!

See you next time! :)

– Saloni

Share

Subscribe now

It’s been a while! But I’m going to get straight into it, because this post is long (and won’t fit on email).

If you like this, I hope you subscribe! As always, please point out errors if you spot any. I’ll really appreciate it, and I offer rewards for it, so you can “earn while you learn.”

Here we go!

Subscribe now

The seasonality of births

Have you had a chart stuck in your head, years after seeing it? Mine is a heatmap showing that some birth months are much more common than others; meaning that birthdays aren’t spread equally over the year.

I couldn’t track down the chart I remembered, so I found the latest data from the United States and recreated it myself. Here it is, with data between 2007 and 2022.

As the chart shows, births tend to rise between July and September, in the United States.

These seasonal differences across the year aren’t small. In 2022, for example, February had the fewest births (275,000)2, while August had the maximum (335,000). This meant a range of 60,000 births, or about 10% above and below the average.

Study: Changes in birth seasonality in Spain: Data from 1863–1870 and 1900–2021 (Adela Recio Alcaide, César Pérez López, Francisco Bolúmar Montrull, 2023)

Researchers have actually known that births are seasonal for a long time. Data shows that birth seasonality varies between countries and has changed over time.

In a recent study, demographers looked at data from Spain, and found that the seasonality of births has shifted and declined over a century. I’ll take you through this with the chart below, which shows data since 1863 (!)

There are large fluctuations in births month-by-month, as you can see, which represent the seasonality of births.

The authors studied these seasonal patterns with Fourier spectral analysis, a method that’s often used to understand wave functions and time series data to uncover periodic patterns.

You can see that fluctuations in the monthly number of births have become smaller over time, meaning the seasonality of birth has declined. This is especially the case since around the 1940s and '50s.

In fact, the authors noted three breakpoints in the pattern — 1919 (corresponding to the flu pandemic), 1940 (the end of the Spanish Civil War), and 2020 (the Covid-19 pandemic and lockdowns). At these three points, the seasonal pattern of births noticeably shifted.

Another finding (harder to see in this chart) is that births in Spain have shifted from a spring peak to a summer peak. (More on this below.)

Something else I find interesting is the actual numbers of births themselves. It surprised me to see that births stayed relatively stable across the 20th century, during a period of population growth; that they declined so much during the 1970s and 80s; and that they rose from the 1990s.

But why are births seasonal in the first place? Multiple factors seem to contribute. For example, temperature and climate — researchers have previously found that hotter temperatures lead to declines in births nine months later. See this in the chart below.

On the right, you can see the effect of each 90ºF day (i.e. 30ºC) on the monthly birth rate nine months later. On the left, you can see how this effect is stronger at higher temperatures.

This might be why there’s a notable relationship between latitude and birth seasonality around the world, which is in the next chart below.

Each dot represents a different country or state (in the US), and I’ll describe it below.

You can see how countries in Europe (which are at higher latitudes) tend to have their “peak birth month” in the summer (around May to August). In contrast, countries in the Asia/Middle East and the Caribbean tend to have their peak birth month later in the year, in autumn or early winter. This pattern is also visible within the US — southern states have later peak birth months.

Aside from climate, the authors suggest various other factors that could also be relevant.

One is agriculture. For example, parents might plan to avoid births during the harvest season, when raising a child would have been more difficult. In the past, especially in places that had a large agricultural labour force, this could have been very relevant.

And we also now have better birth control and temperature control (such as air conditioning), which could have also helped reduce the seasonality of births over time.

It seems like there’s still much left to understand on this topic. I started reading about this out of curiosity, and didn’t expect to find the history behind it so interesting. If you’re interested in some more, here’s a link to the study.

Subscribe now

So, if you think way back to the start of this post, I’d been thinking about a chart I had vaguely remembered and wanted to recreate. Now, here’s another one I’d wanted to see for a while, but hadn’t.

What do people die from, and how does this vary with age?

It’s common knowledge that cardiovascular diseases and cancers are the most common causes of death.

But this metric — “the most common cause” — is about causes of death across all age groups. Because older people have much higher risks of dying, their causes of death dominate deaths across age groups.

I hadn’t seen a satisfying way to visualize causes of death in different age groups all at once, which is important because we know that children and young people tend to die from different causes. But recently, I stumbled on this new and important paper on the causes of death we experience among our relatives.

It inspired me to make this chart, with the latest data from the United States, which shows the relative share of deaths from each cause, across all age groups.

You can see that, in childhood, in the US, the most common causes of death are ‘external causes’. This is a broad category that includes accidents, falls, violence, and overdoses, and is shown in red. But there’s also a notable contribution from birth disorders (in muted green), childhood cancers (in blue), and respiratory diseases (in cyan).

The share of deaths in childhood from cancers stood out to me. We’ve seen lots of progress against many childhood cancers over the last 50 years — notably in treating leukemia, brain cancers, kidney cancers, lymphomas, and retinoblastoma — but this is a reminder that there’s still further to go.

From adolescence until middle-age, ‘external causes’ are now the overwhelming cause of death. Around 80% of deaths at the age of 20 in the US are due to external causes. These result from causes such as accidents, violence, and overdoses.

At older ages, diseases rise in importance. Causes of death also become more varied, although cardiovascular diseases and cancers are the most common.

You might also be wondering about the brown category at the bottom, called ‘special ICD codes’. That’s a placeholder category in the system for deaths caused by new diseases — predominantly Covid-19, since the data spans 2018 to 2021.3

Subscribe now

The chart we saw above looked at the relative share of deaths from different causes.

But if we want to understand how these trends arise, it’s more helpful to look at the death rate from each cause instead.

For example, if you were wondering why cancers appeared to decline after the age of 70, based on the chart above, the answer would be that, well, they actually don’t. The apparent decline of cancers is because cardiovascular diseases rise much faster with age than cancers, and thus make up a larger relative share. You can see this below.

The chart also shows that most causes of death rise exponentially with age.4

This includes blood disorders, cancers, cardiovascular diseases, digestive diseases, endocrine diseases, genitourinary diseases, infectious diseases, and several other categories, which each become exponentially more deadly with age. Covid-19, which falls under ‘special ICD codes’, follows a similar exponential trend.

But ‘external causes’ are one exception. They rise very suddenly in adolescence, are relatively stable afterwards, and then become exponential in old age.

‘External causes’ is a category that includes accidents, violence, overdoses, and falls, which might be a clue. My interpretation is that adults of many ages are likely to be exposed to these factors — but at the oldest ages, people’s vulnerability to dying from them rises exponentially. Of course, there are also some causes, like falls, that become more likely in old age.

I’ve written more about this topic here: How does the risk of death change as we age – and how has this changed over time?

Subscribe now

Here’s a little summary of other great stuff I’ve read recently.

Research round up!

• People with HIV who begin antiretroviral therapy early and continue it over the long term now face a life expectancy5 that’s almost the same as people without HIV. (Life expectancy after 2015 of adults with HIV on long-term antiretroviral therapy in Europe and North America: a collaborative analysis of cohort studies. Adam Trickey et al., 2023)

  • This new study estimates that women who began early antiretrovirals had a life expectancy of around 77 years, compared to around 81 in the general population.

  • Early treatment with antiretrovirals has been surprisingly effective. It reduces HIV’s ability to multiply in our bodies, leading to levels so low that they are undetectable by standard medical tests. At such a low level, the virus isn’t able to spread to other people either.6

  • Because of this, some countries including the UK are on track to effectively eliminate HIV transmission within the decade, as I’ve written about previously. And already, the US has effectively achieved the elimination of HIV transmission to infants.7

    

• New treatments for cystic fibrosis have massively improved life expectancy of people with the disease. (Elexacaftor/tezacaftor/ivacaftor projected survival and long-term health outcomes in people with cystic fibrosis homozygous for F508del. Andrea Lopez et al., 2023.)

  • In the 1950s, infants and young children who screened positive for cystic fibrosis had a life expectancy of only 4–5 years. (Check out this great review if you’re interested in the history of cystic fibrosis treatments.)

  • But now, they have a life expectancy of around 72 years.

  • This estimate is based on simulation studies that use data on cystic fibrosis patients. Specifically, it looks at people who carry the most common CF genetic mutation, known as F508del, and also receive the combination treatment commonly known as Trikafta (or Kaftrio) at an early age.8

  • The combination treatment includes 3 drugs: elexacaftor, tezacaftor, and ivacaftor (ELX/TEZ/IVA). The chart below shows the percentage of people with cystic fibrosis who would still be alive at each age, based on which treatment they receive.

  • There’s a great article in The Atlantic about how this has changed people’s lives.

    

    Subscribe now

• A very rare situation where Alzheimer’s disease was infectious, like a prion disease. (Iatrogenic Alzheimer’s disease in recipients of cadaveric pituitary-derived growth hormone. Gargi Banerjee et al., 2024) See also the STAT article on this study.

  • A recent study looked at people in the UK who had received pituitary growth hormone transplants while they were young (to treat conditions of short stature). They received these medical transplants from the brains of cadavers, between the 1950s–80s. (NB: cadavers are no longer used as a source of pituitary hormone; for decades, synthetic hormone has been used instead.)

  • A fraction of all the people who received these transplants developed dementia very early — in their late 30s or early 40s. Having dementia this early is very rare, even among ‘early-onset Alzheimer’s disease’ patients, who typically develop it in their 50s or early 60s.9

  • It turns out that some batches of the growth hormone from cadavers had been contaminated with amyloid plaques, which are a precursor to Alzheimer’s disease. The researchers looked specifically at which batches of growth hormone these very-early dementia patients received, and found that they were contaminated with the plaques.

  • With these findings, and other experimental studies in mice, there’s strong evidence that the contaminated transplants had ‘seeded’ the amyloid protein in people’s brains, which eventually propagated into plaques, leading to dementia. Similar developments have been seen in the past with other prion diseases, like Creutzfeldt-Jakob disease.

  • I hope and wonder whether conceptualizing Alzheimer’s as a prion disease will help think of new ways to treat it, and to find treatments for other prion diseases as well.

• Did you know that cholesterol levels have been gradually improving in the US? (US trends in cholesterol screening, lipid levels, and lipid-lowering medication use in US adults, 1999 to 2018. Yumin Gao, Lochan M. Shah, Jie Ding, Seth S. Martin, 2023.)

  • A recent study looks at trends in cholesterol levels and statin usage in the US population, using data from the NHANES, which is a large national study of health and nutrition in the general population.

  • The authors find that levels of total cholesterol, triglycerides, and LDL-cholesterol (also known as ‘bad cholesterol’), have been declining, at a population-level, among people of the same age. You can see this below.

    

  • Why have cholesterol levels been declining? One reason is that more people take statins, which reduce cholesterol and are effective medications for cardiovascular disease. In 1999, around 15% of eligible US adults were taking statins, but by 2017, around 28% were.

  • Though this mechanism seems obvious in retrospect, the reason we can see a noticeable trend at the level of the population is because cardiovascular diseases are common in the population, and statins have strong effects in treating them, and there have also been improvements in screening and public awareness of these conditions.

    Subscribe now

There’s still a lot more I wanted to write about, but I’ll stop there because this post has become very long already, and who knows, you might have other things to do today.

Here’s some of the best stuff I’ve read & found recently.

Recommendations!

• Making the micropipette — how problems with mouth pipettes led to the invention of micropipettes, although this shift happened later than one would imagine. This sentence stuck in my memory: “While mouth pipetting one night in the lab, Suovaniemi ‘almost swallowed a piece of rat’s brain,’” That’s one way to get a ‘taste’ of neuroscience!10

• Aliens?? — “JHU scientists say that key evidence of an interstellar object was actually just a truck driving near a seismograph,” Every news story about aliens entertains me more than the last.

• ‘Failure at every level’ — a thrilling story about the downfall of Didier Raoult, who boosted hydroxychloroquine as a Covid treatment while conducting unapproved trials, ran 248 trials under a single ethics code, had 10 articles retracted, and has 50 more flagged for concerns.

• Have you ever wondered how the windpipes of different insects compare to each other? Me neither. But recently, entomologists used micro-CT scans to visualize the windpipes of 30 different insect species in 3D, which I thought was pretty cool actually.

  

• A case study of a man who got vaccinated 217 times against Covid. He didn’t have side effects, has no sign of ever being infected by Covid, and had a normal immune response to other infections. Why stop at herd immunity when you can be the herd?11

• The physics of flight — an incredible interactive explainer by Bartosz Ciechanowski, whose blog is amazing.

• ‘ERROR, a bug bounty program for science to systematically detect and report errors in academic publications’ — get paid for spotting errors in scientific publications! (Hey, you can do the same on this blog!)

• 10 technologies that won’t exist in 5 years — Jacob Trefethen’s list of technologies that are worth working on and achievable, but won’t exist in 5 years.

• The Webb Telescope’s image of the Crab Nebula. ‘The Crab Nebula, also known as M1, is a supernova remnant: all that remains of an exploded star.’ How cool is that? NASA has an explainer on how their full-colour images are made.

  

And… that’s all for now!

I have a busy few weeks ahead, but I’m trying to spend about an hour a day doing personal writing and dataviz, so hopefully you’ll see another post soon.

If you enjoyed this, I hope you subscribe and share this with your friends! As always, if you’ve seen any errors, please let me know, so I can fix them.

See you next time! :)

– Saloni

Subscribe now

Share

I started 2024 by hiccuping for a whole 40 minutes. Happy new year to you too!

New years (and birthdays) used to make me feel anxious about getting older. But nowadays I think getting older isn’t what it used to be.

There are a lot of common misconceptions about life expectancy and longevity, and sometimes I think our intuitions are outdated by many years.

Life expectancy is a complicated topic, but itʼs one that I think is important to understand. So in this post Iʼll take you through the topic with 16 charts.

Several points come from the new Life Expectancy page I worked on at Our World in Data with my great colleague Lucas Rodés-Guirao – check it out if you’re interested in exploring more.

As always, please let me know if you spot any errors – you can earn a reward for it and I’ll really appreciate it. And if you like this post, I hope you subscribe!

Subscribe now

1. We tend to live longer than our life expectancy implies

Iʼll start with one thatʼs surprising. Life expectancy isn’t a prediction of how long people will live.

When you hear the phrase life expectancy, it usually refers to ‘period life expectancy at birth’, which is a statistic calculated using data across the population in one single year.

For example, the period life expectancy at birth in France in 2021 was 82.3 years. People usually think this means babies born that year will live an average of 82.3 years.

But hold on… we don’t know what babies born today will experience throughout their lives, or what their death rates will be, compared to previous generations.

What if new medical advances will arrive, or what if they experience major wars or pandemics or other major changes? How do we know what changes we’ll see in the future, or how impactful they will be?

To know for sure, we’d need to wait until nearly everyone in the birth cohort has died – around a century from now – to find out what their average lifespans actually were. This statistic would be called their ‘cohort life expectancy’.

What happens if we compare cohort life expectancy to period life expectancy in people’s birth years? The chart below shows this.

[Side note: I’ve shown data from France in several charts in this post because it has particularly long-term mortality data and a relatively large population.]

Firstly, the chart shows that the blue line (cohort life expectancy) is incomplete – this is because we don’t know how long babies born more recently have lived.

For the most recent year with data – 1930 – you can see that period life expectancy in France was 57 years – but people born in 1930 actually lived for 69 years on average. They lived 12 years longer on average than their period life expectancy implied.

That’s a pretty big difference! What’s going on?

To understand, we’ll have to look at how period life expectancy is calculated: it’s a metric that’s based on death rates in each age group in one particular year.

Imagine death rates in different age groups in the year 2022. The period life expectancy in 2022 tells us the average lifespan of a hypothetical group of people, if they experienced the same death rates at each age of their lives, as the death rates seen in 2022 at each corresponding age group.

So, according to the hypothetical, babies born today would experience the same death rates when they’re aged 50 as the death rates seen in 50-year-olds this year.

Now you can understand why this differs from their actual average lifespans. Death rates have been declining over time, and people have faced lower death rates than older generations have when they’ve reached the same ages. You can see this below.

One reason why we’d expect the divergence between period and cohort life expectancy to continue is that even death rates seen today are partly a reflection of the past.

Death rates today are partly a result of the health conditions that people alive today have developed in earlier years – because of their environment, living standards, healthcare, behaviours and so on, which they experienced before.

If you’re more interested in more, I’ve written an article about it: Period versus cohort measures: What’s the difference?

This doesn’t mean period life expectancy isn’t useful – it’s still an important condensed statistic that summarizes mortality patterns in a population in a given year. This helps to compare populations across countries and over time. But understanding its details will help ensure you understand its implications correctly.

Subscribe now

2. The annual risk of death has declined across age groups

At each age of our lives, we have lower risks of dying than the generations before us. Let me show you this with a couple of charts.

The first chart below shows the annual probability of death in people of different ages in France.

Babies had a 15% probability of death during infancy in France in the 19th century. Now, the probability is 0.3%.

65 year olds had around a 4% probability of death at that age in the 19th century. Now it’s around 0.7%. Eighty year olds’ probability of death has declined greatly as well – from around 15% in the 19th century to around 2.7% now.

It hadn’t fully sunk in before I saw this chart that infants used to have the same annual probability of death as 80 year olds.

In the chart below, I’ve shown this in a different format, so you can see it across the lifespan.

Each line shows death rates in a birth cohort when they reach different ages. Death rates are expressed as a percentage, and the chart is on a log scale. The different lines show different birth cohorts: dark red for people born in the 1800s, green shows those born in the 1900s, and dark blue shows those born in the 1980s.

As you can see, infants and the elderly have higher death rates. Death rates drop quickly after infancy, then rise in youth, and then rise exponentially with age in adulthood.

But you can also see that, across birth cohorts, the whole curve has shifted downwards. The decline is large – it’s shown on a log scale. Each age of our lives is less deadly than it was for previous generations.

If you’re interested in more, I’ve written an article about it: How does the risk of death change as we age – and how has this changed over time?

The declines we see in old-age mortality are likely an understatement of how much has improved. This is because child mortality rates have declined especially among poorer demographics, who would have lower-than-average survival even after childhood. As more of them survive in the population, you might actually expect average mortality rates at older ages to increase if there had been no other improvements. Instead, they continue to fall – which suggests large improvements that aren’t ‘cancelled out’ by this effect.

Subscribe now

3. Life expectancy is about gains across age groups – but declines in child mortality now make smaller contributions to overall gains

Life expectancy at birth is a condensed summary of death rates across all age groups. But we can also look at life expectancy in adulthood – that is, for people who have already survived to adulthood. These are based only on death rates at older ages.

As the chart below shows, 65 year olds in France for example used to have a life expectancy of around 76 years before the 20th century. But now their life expectancy is around 86 years.

You saw the reason for this in the previous charts – death rates have declined greatly even at older ages.

The decline in older-age mortality has become a major driver for gains in life expectancy at birth in wealthier countries, in recent decades.

And, in the previous section, you also saw that child mortality rates have been reduced greatly.

Because child mortality rates are now at much lower levels, further declines make smaller contributions to overall gains in life expectancy, especially in wealthy countries.

You can see this in the chart below, which comes from a study by José Manuel Aburto et al. (2020). The data comes from 49 countries in the Human Mortality Database, which generally includes higher-income countries.

The chart shows how much life expectancy would increase (expressed as a weight, on the y-axis) if lives were saved at a particular age (x-axis). This is shown for different time periods (coloured lines).

In the words of the authors:

From the 18th to the first part of 20th century, the largest potential increases in life expectancy were concentrated in infancy. The effect on life expectancy improvements due to saving lives in midlife was higher than at older ages. This changed dramatically after 1950, when the effect of infant mortality decreased significantly. By 2010, the effect of reducing mortality by 1% at birth was the same as reducing mortality by 1% at age 71. In the 21st century, saving lives between ages 5 and 40 y had a negligible effect on life expectancy, as opposed to the relatively high impact of these ages before 1900.

– José Manuel Aburto et al. (2020)

At the same time, infant and child mortality continue to play a major role in poorer countries.

What’s tragic about this is that they tend to result from causes of death that are easily avertable with public health efforts – malnutrition, sepsis, pneumonia, malaria and HIV/AIDS – and that are now seen as diseases of the past in wealthier countries.

Subscribe now

4. Death rates have declined from cardiovascular diseases and cancers

In recent decades, we’ve seen more improvement at older ages than many people realise. Cardiovascular diseases, cancers, infections and a range of other causes are becoming less deadly.

In the chart below, you can see the age-standardized death rate from different causes in the United States.

‘Age-standardized’ means that we’re looking at death rates among people of the same age, i.e. adjusted for population aging. This is important because it helps to make a fair comparison across time and between countries – for example, to understand how much things have changed for fifty-year-olds now compared to fifty-year-olds in the 1950s.

You can see that there’s been a large decline in cardiovascular diseases and cancers since the 1950s. Cardiovascular diseases used to kill more than 500 per 100,000 people annually. Now they kill less than 150 per 100,000 – a decline of around three-quarters.

And cancer death rates have declined by around a third since their peak in the 1990s.

A significant driver of both these large reductions has been the dramatic decline in smoking, which had increased the rates of cardiovascular diseases and a range of cancers.

Cardiovascular diseases seem to rise and fall earlier than cancers. This is because smoking increases the risks of cardiovascular diseases faster than it leads to cancer, and because smoking cessation alleviates the risk of cardiovascular diseases faster than it alleviates the risk of cancers.

Even aside from smoking, there have been a wide range of other medical and public health efforts that have helped reduce mortality from cardiovascular diseases and cancers further.

If we just focus on cardiovascular technology, consider what we have now that didn’t exist a hundred years ago:

• Cholesterol-lowering and blood pressure-lowering drugs (e.g. statins, calcium channel blockers, beta blockers, ACE inhibitors, SGLT2 inhibitors, PCSK9 inhibitors, angiotensin receptor blockers)

• Surgical techniques (e.g. stents, angioplasty, coronary artery bypass grafts)

• Tests and monitoring devices (e.g. echocardiograms, coronary angiographies, at-home blood pressure monitoring and cholesterol testing kits)

• Medical devices (e.g. defibrillators and pacemakers)

Some of these – such as SGLT2 inhibitors and PCSK9 inhibitors – only arrived in the last decade.

Even with this, I think developing these technologies has just been the beginning – we’ve not only needed to discover how to save lives, but we’ve also needed to find ways to improve and scale them up so more people can benefit from them.

The lack of these technologies are some of the reasons why cardiovascular mortality tends to be higher in poorer countries.

In the chart above, you’ll have also noticed that not all of the trends have been positive. In the US, opioid deaths have been rising greatly (shown under ‘neuropsychiatric conditions and drug use disorders’). Additionally, the rate of decline of cardiovascular mortality has slowed to a halt in the US, with growing risks from obesity and poor diet. The sudden rise at the end shows the rise in respiratory infection death rates (from Covid-19).

This is part of why life expectancy in the US has stagnated over the past decade, while it continues to rise in other countries.

Subscribe now

5. Women tend to live longer than men because of lower death rates across their lives…

If we look at a global average, women have a life expectancy that’s around 5 years higher than men.

It’s a common misconception that this gap is because men are more likely to face risks from accidents and violent deaths.

But women actually tend to live longer than men because they have lower death rates across their lifespans, which you can see in the chart below.

In infancy, boys have higher death rates than girls, mainly because of a higher vulnerability to diseases and genetic disorders. In youth, boys tend to have a higher death rate due to violence and accidents. In adulthood and old age, they continue to have higher death rates due to these factors, occupational risks, and chronic health conditions, partly from higher rates of smoking, alcohol, and drug use.

… but the sex gap in life expectancy isn’t a constant – it varies around the world and has changed over time.

Since the gap in life expectancy results from a wide range of causes, it also varies across the world in different ways. You can see this in the map below.

The sex gap in life expectancy varies widely between countries.

The gap is highest in Russia, Eastern European countries, and South American countries, and lowest in West African countries, India, Scandinavia and New Zealand.

[Until recently, in India, Pakistan, Iran and Bangladesh for example, men actually had a higher life expectancy than women. In India, a major contributor to this was higher rates of female infanticide and childhood neglect. In Bangladesh, it came from higher death rates in women than men, in adulthood, through large gender disparities.]

The sex gap in life expectancy has also changed over time, as you can see in the chart below.

The gap spiked in many countries during the two World Wars, reflecting the surge in deaths especially in young men. But it also grew gradually over the twentieth century and has since been declining. A major part of this was due to the rise and fall in smoking, particularly among men.

Overall, the sex gap in life expectancy isn’t a constant, and breaking it down to understand it better can help us find more ways to prevent avoidable deaths.

If you’re interested in more, I wrote about the sex gap in life expectancy in this article: Why do women live longer than men?

Subscribe now

6. Lifespan inequality has been falling

In the same way as economists calculate economic inequality with the ‘Gini coefficient’, you can use it to measure the inequality in lifespans too.

To calculate it, you might start by picking two random people in the population and comparing their estimated ages at death. Then, you do this for the rest of the sample, by looking at the estimated age differences for all combinations of pairs in the population, and normalize the total to a unit between 0 and 1.

From this calculation, you can see that lifespan inequality has declined over time in many countries. Lifespan inequality tends to be lower in richer countries.

(The rise in the 1980s–2000s in Kenya is from the HIV/AIDS epidemic.)

Why is lifespan inequality falling?

Let’s slow down to understand why.

If mortality declines at ages younger than some threshold, then more people survive to an older age, and lifespan variation declines. But if mortality declines at ages older than some threshold, then lifespan variation increases.1 The balance between these two determines how lifespan inequality changes over time.

In a recent study, the authors look into this hypothetical using data from Sweden. You can see it in the chart below. This time we’re looking at the inverse (lifespan equality, rather than inequality).

The purple line shows a scenario of how lifespan equality would change if mortality reductions were focused on younger ages.2 The red line a scenario of shows how lifespan equality would change if there were equal reductions in mortality across all age groups.

Finally, the blue line shows what has actually been observed in Sweden over time. You can see that it lies in between the two scenarios. This shows us that lifespan inequality has fallen because of mortality declines in different age groups – but younger ages have seen greater declines in mortality, and the mortality decline hasn’t been equal across all age groups.

Subscribe now

7. The limits of life expectancy are being pushed higher

In 2021, Hong Kong was the country with the highest recorded female life expectancy: of 88 years.

That was a whole 40 years higher than the highest country in 1840 – in Sweden, which had a female life expectancy of around 46 years.

This statistic – the highest female life expectancy achieved by any country – has risen over time, which you can see in the chart below. The coloured dots represent the countries that achieved the highest female life expectancy each year.

You can see that the highest life expectancy has risen by approximately 1 year every 4 years.

The horizontal lines on the right-hand side of this chart show a list of predictions – they’re predictions of the maximum possible life expectancy a country could achieve.

The American statisticians Dublin and Lotka predicted in 1936 that the maximum possible life expectancy would be 70 years. That was broken in 1941. Later, in 1979, the UN predicted that the maximum possible life expectancy would be 80 years. Their prediction was broken the next year by Iceland. Many other predictions have already been broken, as life expectancy has continued to rise.

You can also see that Sweden, Norway, Iceland, Denmark and the Netherlands held the record in many years in the past. But in recent decades, Japan and Hong Kong have pulled ahead.

Notice that it’s not the case that they’ve merely caught up – they’ve continued to push the limits higher; they’re breaking their own records.

If you’re interested in more, the chart originally comes from a famous paper by Jim Oeppen and James W Vaupel in 2002, which you can read here.

Below is a view of the trends in each of these countries over time.

Subscribe now

What’s the message to take away from all of this?

To me, one of them is that longevity isn’t one obscure problem that needs to be ‘cracked’ in order for us to live longer. We’ve been able to reduce mortality and delay death by continuously working on a wide range of problems.

Disease outbreaks that used to be a constant – measles, typhoid, diphtheria, cholera, and even bubonic plague – are now much less common and less deadly in many countries. Mortality rates from cardiovascular diseases, chronic respiratory diseases and cancers – some of the biggest causes of death – have been reduced.

Another point is that we can live longer by reducing premature mortality and expanding the frontier. Worldwide, there’s still so much left to be done. Even now, life expectancy in Nigeria and Chad, for example, is just 53 years. That’s thirty years lower than it is in South Korea or Japan – but they had faced that same life expectancy in just the 1950s.

None of these trends are a given.

We might face new unexpected problems in the future, as we’ve faced so far. In recent history, the world has seen a dramatic rise in smoking, and major events like the World Wars and the AIDS epidemic have each caused huge surges in deaths at different times.

And we continue to face old challenges. Diseases like Alzheimer’s and dementias are still notoriously difficult to treat. Antimicrobial resistance threatens to reverse our ability to treat common infectious diseases. Many diseases that seem solved still require lifelong treatments that carry long-term side effects.

But that doesn’t mean we can’t make progress against them too, and I think it’s up to our own efforts to get there.

Share

Subscribe now

What I’ve been up to

1. I wrote a big new article: ‘What were the death tolls from pandemics in history?’ It includes this big chart!

   

   1. I also worked on several new topics with my wonderful colleagues at Our World in Data, such as tuberculosis, cardiovascular diseases, and pandemics.

2. I recorded an audio version of my coauthored article ‘Why we didn’t get a malaria vaccine sooner’, after people requested it. You can listen & download here:
   

3. I got DALL-E to create a little avatar for me. It’s a me, Salonio.

And… that’s all for now! If you liked this, I hope you’ve subscribed & shared it with your friends. (And if you’ve spotted any errors, please let me know and you could earn a reward.)

I once again have a big pile of new research that I want to share, so hopefully you’ll have another research round up in your inbox soon.

See you next time! :)

– Saloni

Share

Subscribe now

Thanks very much to my friend Nathaniel Bechhofer for feedback on a draft of this post.

Hundreds of thousands of people die from malaria each year, but it took 141 years to develop a vaccine for it. Advance market commitments could speed things up next time.

This is an audio recording of the article ‘Why we didn’t get a malaria vaccine sooner’, which was written by Saloni Dattani, Rachel Glennerster and Siddhartha Haria, and published on Works in Progress on 7th September 2023. This audio recording was recorded by Saloni Dattani.

Plus, here’s a bonus illustration with a little spoiler, created by DALL-E.

Every so often on this blog, I put together great research that I feel should be read more widely.

This blog is called Scientific Discovery – but scientific discovery is just the beginning. We not only need to understand the world and discover ways to make progress, but also find out how to scale them up so they can make a real difference to people’s lives.

In this post, I’ll cover research on: a brain cancer breakthrough, controlling mosquito diseases with bacteria, rabies elimination, women’s employment, personality and age, and more. This post is too long for email. I hope you enjoy it! If you spot a mistake, let me know and you could earn a reward.

Subscribe now

A brain cancer breakthrough

Phase III trial results: Vorasidenib in IDH1- or IDH2-mutant low-grade glioma (Ingo Mellinghoff et al., 2023)

Brain cancers aren’t the most common cancers, but they tend to be the most devastating. Brain cancer patients die about 20 years earlier than they would otherwise, on average.1

This is about a new drug that can change that. What’s even more hopeful is that, because of the way it was developed, it’s likely to be just one of many drugs in this new category.

Let me give you some context. This drug works against brain tumours from glial cells – these are cells that have many functions, including supporting neurons. Their tumours are called gliomas.

There are about 140,000 patients living with gliomas in the United States.2 This makes them one of the most common brain cancers; they’re about a quarter of all brain cancers. Patients are diagnosed with low-grade gliomas at the age of 41 on average, and then have an average survival time of around 7 years.

So far, treatments for child patients with gliomas have been quite effective. But for adults, who have different types of gliomas, the situation has been pretty dire.

Radiation and chemotherapies aren’t so effective. Brain surgery is risky, especially if the tumour is in a critical part of the brain. So, at early stages of the cancer, many patients ‘wait and watch’ – with chemotherapy and regular MRI scans to see when it worsens.

This new oral drug, called vorasidenib, is one in a new category that can change that.

How does it work? You may know that tumours develop from uncontrolled growth. This happens through many pathways – one way is by altering their metabolism, which lets them multiply and grow quickly.

This new drug blocks two key metabolic enzymes, IDH1 and IDH2, which help them do that. Their tumours become somewhat dependent on this new metabolic pathway. About 80% of glioma patients3 have a mutation in these enzymes – and it turns out, so do a share of patients with a range of other cancers – which suggests this drug could be widely useful.

Subscribe now

In this phase 3 trial, over 300 patients with recurrent glioma and mutations in IDH1 or IDH2 were given vorasidenib or a placebo.

The chart above shows that those who received the drug survived without a progression in their cancer for a median of 28 months, versus 11 months in the placebo group, which is 2–3x as long. Their side effects were mostly minor, but some patients developed liver toxicity.

Why am I excited about this? Although brain cancers still remain serious, this is a major step forward in our understanding and ability to treat them. The drug is now under fast-track for approval by the FDA. Another point that’s hopeful is that it’s one of several new drugs. The reason behind this is that scientists know what makes it work and why – it wasn’t discovered by chance, but through ‘rational drug design’.

Mutations in the IDH1 and IDH2 metabolic enzymes had been identified in gliomas over a decade ago through genetic sequencing. Since then, multiple IDH-inhibitor drugs have been developed, using structural imaging techniques and medical chemistry. The other drugs have also had promising clinical trials, but vorasidenib seems particularly effective because it’s able to target both enzymes.

The list of big cancer breakthroughs in modern history is pretty long – tamoxifen, cisplatin, vinblastine, imatinib, trastuzumab, rituximab, ipilumumab and methotrexate are just some of the biggest, and hopefully this will join the list. Each of these has an important story, and I’d like to write about them more. [For now I’ve linked to brief histories of each in case you’re interested.] There are also the new ‘CAR T-cell therapies’, which I hope to have time to write about later as well.

At the same time, cancer treatment still has a long way to go. It’s not just efficacy that we care about, but how to make drugs safer and easier to take in the long-term, without toxic side effects. This has been an under-appreciated part of the history of medical progress, and it continues to matter.

But another reason I find this exciting is because it’s an example of the modern revolution in biotech, or what people call ‘the century of biology.’ Biotech advances are rippling through the field, and it’s crucial that we streamline the process of clinical trials and regulation to help draw them out and scale them up, so everyone can benefit from them.

Subscribe now

Defeating dengue with Wolbachia

In more biotech news, let’s talk about a way to stop mosquitoes from spreading disease.

Maybe you’ve vaguely heard of Wolbachia – it’s a species of bacteria that blocks some mosquitoes from spreading some viruses to us, but causes negligible harm to us or the environment.

It came back to my attention because at a recent conference, researchers presented their findings that Wolbachia programs reduced dengue incidence by 94–97% in areas where they were highly established in three Colombian cities4, in the biggest field trials so far.

That’s huge if true, but since those results haven’t been published yet (here’s the original plan for the study), I decided to look into previous papers on the topic. Here’s a little summary.

Briefly, pilot programs have been quite effective so far – some have now been running for over a decade, and they shouldn’t be seen as a far-fetched experimental idea.

• How does it work? To begin with, researchers insert Wolbachia bacteria into the eggs of female mosquitoes, with microscopic needles. The mosquitoes then pass it onto their offspring – who maintain the bacteria in their reproductive cells throughout their lives.

• In the most common approach, Wolbachia blocks many mosquitoes from carrying dengue virus (and some other viruses), but it doesn’t shorten the mosquitoes’ lives.5 This helps them lay many infected eggs and propagate and replace the existing population quickly.

• To top it off, male Wolbachia-infected mosquitoes can’t produce eggs with female mosquitoes that aren’t infected – so Wolbachia-mosquitoes have a reproductive advantage.6 All this gradually increases the share of mosquitoes that are infected with Wolbachia, and reduces the share of mosquitoes that carry dengue.

  

• Getting to this point took lots of scientific research. Here’s a highly recommended article about it by Scott O’Neill, one of the leading scientists and founder of the ‘World Mosquito Program’, which helps run many Wolbachia efforts.7

  Subscribe now

• Where has it been tried, and for which diseases? By now, Wolbachia has been used to control mosquito-borne diseases in regional pilot projects in several countries: Australia, Indonesia, Brazil, Colombia, French Polynesia, Vietnam and more.

• So far, it’s been used for Aedes mosquitoes, which spread multiple diseases to us: dengue, Zika, chikungunya, and yellow fever.

  • In case you didn’t know, dengue and chikungunya are debilitating diseases of joint pain that often last several weeks or months. Zika is a risk to pregnant women and fetuses. Yellow fever can cause severe liver damage, and jaundice.

• How effective is it? Previous Wolbachia projects have been fairly effective at reducing dengue. A recent cluster-randomized controlled trial found that it reduced dengue incidence by around 77% in Yogyakarta, Indonesia8; another recent field trial in Niterói, Brazil led to a 69% reduction (shown below). The earliest program led to a 96% reduction in dengue in Queensland, over the span of a decade.

  • These diseases are highly seasonal, so there’s sometimes uncertainty around precisely how effective it is, especially when looking at shorter timespans – but the data overall suggests it’s fairly effective against dengue.

  • Wolbachia efforts aren’t as effective for other diseases (chikungunya, Zika, yellow fever) yet, but still look effective. This may change with more research.

    

• These programs have a big upfront cost, to prepare Wolbachia-infected mosquitoes. But in urban areas where dengue is common, they still seem very cost-effective, and the benefits grow over time as more dengue cases are prevented.9

• Wolbachia programs could lead to elimination of dengue in (some? many?) countries, if rolled out widely. In some pilot program regions, such as Queensland, Australia for example, it already has.

  • At intermediate levels of Wolbachia efficacy, it’s estimated that programs if rolled out widely could reduce global dengue incidence by at least 70%. This is according to modelling studies which find that programs have been effective enough to reduce the R number below 1.

  • But this depends on the baseline prevalence of dengue, the implementation and monitoring, which could be difficult in some areas.

One worry you might have is: could resistance evolve somehow?

Maybe Aedes mosquitoes could become resistant to Wolbachia? Or maybe dengue viruses could become resistant to Wolbachia?

• The first seems unlikely – Wolbachia has remained stable at high frequency in Aedes mosquitoes for over a decade in Queensland and in Malaysia, which had the earliest programs. They’re also fairly pervasive in the insect world, likely because of their reproductive effects.

• The second is less clear – dengue viruses evolve quickly, have many strains, and Wolbachia doesn’t give mosquitoes complete protection from them. A recent study reviews the evidence on this. This situation could change – but Wolbachia’s efficacy could be improved scientifically too.

A crucial part is to prepare with more research and monitoring, even after the programs have been successful. Overall though, it looks like this should be scaled up more widely, and I’m pleasantly surprised by how effective it seems to be.

Subscribe now

The summer drop in female employment

Working paper: The summer drop in female employment (Brendan M Price & Melanie Wasserman, 2023)

Let’s move onto something completely different, spurred by this year’s Nobel prize to Claudia Goldin, for her work on gender differences in the labour market.

I’ll talk about a new study about women’s employment and how it drops in the summer – the charts and findings are very interesting.

The authors look at this topic with data from the US Current Population Survey, which is a monthly survey of around 60,000 households, conducted by the Bureau of Labor Statistics. Below is a summary of what they find.

The authors find that women’s formal work10 drops during summers, both in terms of employment and job-seeking. And even among women who continue to work, their working hours decline – much more than men’s.

In total, women’s hours-worked shrink by around 9.8% from May to July (considering both the employed and unemployed), which is more than double the decline seen in men.

Why does this happen? The paper argues that it’s because they’re more likely to take time off to care for children on their summer breaks. They find many lines of evidence that support this idea:

The size of the decline is largest among women with school-aged children, and women without children don’t experience a decline. Also, the timing of the decline matches school closures; and the overall decline is almost fully accounted for by the share of women with children who say they’re taking care of house or family, in the surveys.

Another interesting finding is that women are more likely to work in education if they have a young, school-aged child (shown in the chart below). This is presumably because it’s much easier to match children’s time off school.

Through this, I also learnt a new statistic: in total, around 13% of working women in the US are in the education sector.

Subscribe now

More

• Do people become more conservative as they get older?

  • Older people are more likely to agree with conservative ideas, and say they’re conservative. But if you track a cohort of people over time, at least in the US, they tend to become more liberal as they grow older. How can these both be true?

  • Allen Downey explains this seeming-paradox. If people become more liberal with age, but younger cohorts are even more liberal than them, then older people appear more conservative as they age, relatively speaking.

    

• How do our personalities change with age? (Ingo S. Seifert, Julia M. Rohrer, and Stefan C. Schmukle, 2023)

  • To answer this question, it’s important to distinguish generational differences from changes with age. In this study, researchers looked data from cohorts as they grew older (in three datasets – from Germany, Australia and the Netherlands).

  • With this, they found some consistent changes – across all three datasets, people became more ‘conscientious’ and ‘agreeable’ until middle-age, and less ‘neurotic’ across their lifespans.

  • But, each of these traits is an average of answers to multiple questions (for example, agreeableness was measured with questions about kindness, sympathy, cooperation, warmth, and harshness). And as you can see below, the individual answers often showed different trajectories.

Subscribe now

• Human rabies was eliminated in Goa, India. (Andy Gibson et al., 2022)

  • Here are a few points to put this in context:

    • India is estimated to have the largest burden of rabies worldwide, making up perhaps a third of all rabies deaths globally.11 This is likely the result of a large population as well as a large population of free-roaming dogs, poverty, poor understanding of rabies, and a lack of coordinated strategy that was used here.

    • Rabies is near uniformly fatal in people if they aren’t vaccinated (vaccinations are typically given post-bite). But prompt vaccination is very effective, making fatalities rare. (This is an amazing story in its own right.)

    • At the same time, vaccinating humans has little impact on the spread of rabies. Rather, it’s dogs and other animals who are reservoirs for the disease, and happen to spread it to us, which is why animal vaccination is effective.

    • Animals can be vaccinated through injection or oral bait (which have sometimes been dropped from helicopters across European forests), and it’s estimated that only around 45% of all dogs need to be vaccinated to achieve elimination.12

  • So how was it eliminated in Goa? Between 2013 and 2019, this program involved:

    • Door-to-door (DD) vaccination (for confined dogs) and capture-vaccinate-release (CVR) (for free-roaming dogs) through GPS coordination, along with vaccination surveys;

    • School-based education to almost 700,000 children about how to avoid dog bites, what to do if bitten, and the importance of dog vaccination;

    • A widely-publicized rabies hotline, to help people request vaccinations or report sick or injured dogs (with or without rabies).

  • By the end, there was a 92% reduction in monthly canine rabies cases, and human rabies deaths had been shrunk to zero.

Subscribe now

• Giant virus-like particles in soil (Matthias G. Fischer, Ulrike Mersdorf, Jeffrey L. Blanchard, 2023)

  • You might know that the first virus (the tobacco mosaic virus) was described in the 1870s. For decades after that, viruses were described in terms of what they were not – they couldn’t be seen, they couldn’t be retained by filters, they couldn’t replicate without cells – until we finally found out what they were. (The first time viruses were seen was in the 1930s, using electron microscopes, which were new at the time and apparently called ‘übermikroskope’.)

  • But did you know that some viruses are so big that they can be seen with regular microscopes? These ‘giant’ viruses are still fairly small – only around 1 micro-metre (i.e. 1/1,000th of 1mm) – but they’ve changed our understanding of viruses.13

  • This new study is about a range of giant particles in soil that look like viruses (but may or may not be viruses). Here are some cool, spooky photos of them.

I actually had a heap of other papers I still wanted to share, but I’ll stop there! Enough fun facts already.

Even more

Here are some links from around the web that I enjoyed recently:

• ‘How to save twenty million lives’ – a very inspiring written interview with Dr. Mark Dybul, who led PEPFAR: a program that scaled up AIDS treatment in Africa and saved millions.

• A long-form podcast interview with Katalin Karikó, this year’s Nobel laureate in medicine for her work on mRNA. I hadn’t known much about her life before listening to it, and I found the episode very honest and moving.

• The Arctic Permafrost Atlas – a freely available book with loads of beautiful maps and diagrams that explain permafrost (pdf link, which is 236 Mb).

  

• A podcast episode on the destruction of Pompeii by Mount Vesuvius’ eruption in 79 AD. I recently visited Pompeii and Mount Vesuvius and listened to lots of great podcast episodes about their history while I was there – this was the best one.

• The Invention of Surgery, a book by David Schneider. Many surgeries we consider minor now were once very risky, if not fatal. This is a nice book that chronicles how that happened, & the people who pioneered surgical advances.

• Niko McCarty’s guide to synthetic biology – full of links and reading to introduce you to different parts of synthetic bio.

• ‘Large language models, explained with a minimum of math and jargon’ – by Timothy B Lee and Sean Trott. Exactly what it says!

Subscribe now

What I’ve been up to

1. I was on a podcast!

   1. I recorded a great fun, 3 hour long conversation with Fin Moorhouse and Luca Righetti on Hear This Idea. We talked about the history of malaria, the incredibly long (& frustrating) journey to a malaria vaccine, missing data & more. Listen here.

2. I discovered a mistake in an earlier post I wrote.

   1. In post #15 ‘What does it mean when fewer people die’, I inaccurately described the ‘survivorship age’ from a study. For example, the sentence: ‘It shows that women in France who died at the age of 90 in 1900 were in the top 1% of survivors in their cohort.’ was inaccurate. Rather, it should have said something like: ‘It shows that among those born in 1900, women in France who die at the age of 90 would be in the top 1% of survivors.’ That’s quite a different interpretation, and it also affected other parts of that section. I’ve now corrected it.

   2. I found this experience annoying for several reasons: I had re-read the study several times to make sure I was describing it accurately (it was ambiguous); it would have been easy to correct quickly had I known; and people with more experience in demography should have caught it easily, for methodological reasons – but it wasn’t pointed out to me, and only came to mind because I was working on a related topic once again.

3. So I’ve set up a reward system for people to point out mistakes I’ve made.

   1. You can earn £5, £15, or £40 for letting me know about a mistake I’ve made, including in this post. There are more details on my About page here. I hope this also motivates me to spend even more time fact-checking.

4. I wrote some articles:

   • How are causes of death registered around the world? – everything you wanted to know about causes of death and how they’re determined, and why this data is often missing in poor countries.

   • Why death tolls from different risk factors can’t be simply ‘added up’. This point is frequently overlooked, which means people don’t realise that the same deaths can often be prevented in multiple ways.

5. I made some pretty charts, like this one:

6. I’ve been eating a lot of gelato, as usual.

   

7. I didn’t do anything interesting for Halloween this year, but that’s because I no longer have my favourite costume.

And that’s all for now. I hope you enjoyed this, learnt something new and subscribed if you haven’t already.

This newsletter is completely free, I’ve turned off subscription pop-ups, and I guess I’m now offering rewards to readers too – if you let me know about mistakes I’ve made. So why not share it with your friends too, and ‘earn while you learn’?

See you next time!

– Saloni

Subscribe now

Share

I’ve been holed up over the last few months co-authoring a long piece called ‘Why we didn’t get a malaria vaccine sooner’ with two fantastic economists, Rachel Glennerster and Siddhartha Haria. It’s the lead piece in the latest issue of Works in Progress.

If you haven’t read it already, I think you’ll really enjoy it:

Read the piece here

The response so far has been incredibly positive and we’re very grateful that so many people read, enjoyed and learnt from it. Some very kind reviews so far: a ‘phenomenal writeup of the history of malaria vaccine efforts’, according to Alexander Berger, an ‘incredible read on the history, science, economics, and politics involved’, from Shruti Rajagopalan, and ‘just as good as everyone is saying: a deep, hugely informative story’, from Stuart Ritchie. Thank you!

Here’s my favourite chart I made for the piece. It’s a timeline to show how many vaccines we have, and when each one was introduced.1

The malaria vaccine licensed in 2021 is the first vaccine against a parasite in humans. (There had already been numerous parasite vaccines for livestock and pets.)

It’s also striking to see the long 80+ year gap between the first and second vaccine. When Edward Jenner developed the smallpox vaccine, he didn’t know it was caused by a microorganism – let alone a virus – because germ theory hadn’t developed yet. When it was, vaccine discovery finally took off.

Even with germ theory though, viruses were too small to be seen under microscopes at the time. The first virus was observed all the way in the 1930s. This also seems surprising from the chart. Viruses were theorized – and extracted, purified, named and used to develop vaccines – before they were observed.

To develop the 1885 rabies virus vaccine, for example, Louis Pasteur and Émile Roux isolated an unseen material from the spinal cords of rabid animals. They purified it by passing it through thin porcelain filters that would remove bacteria, which are much larger.

Subscribe now

In this post, I’ll add some thoughts on what it was like to work on a 9,000+ word article; how to fact-check historical claims; and a few more bonus things I learnt about the story that didn’t quite fit in.

How to write a 9,000+ word piece

First, make sure you have a couple of great coauthors and editors. I’m only half-joking. I wrote the ‘science and history’ part of this piece, while my coauthors wrote about the economics and policy ideas. So it would’ve been very different and lacking in ‘so what?’ if I’d written this alone. And my colleagues Ben and Sam are fantastic editors whose questions really strengthened the piece. It was a real team effort.

Behind the scenes though, this was the 5th or so complete rewrite of the piece. It’s such a hugely important story, and such a complicated one, that it was hard to figure out how to cover everything and make it very readable and hopefully enjoyable.

Here are a few things that helped.

• Tell the story to people, in-person.

  • By the second or third redraft of my part, I felt pretty stuck. There were so many different aspects that delayed the malaria vaccine, that I couldn’t figure out how to arrange it all in a way that made sense and was easy to follow.

  • So I had a half-hour roundtable conversation with some friends and writers who don’t have a science background, and told them about the story and what I was struggling with.

  • It surprised me which parts they found surprising – that malaria was caused by a parasite, not a virus, and why this would make a difference. Or, how animal models and challenge trials worked. So I decided to spend more time elaborating on these parts.

• Explore, ask, and write a list of things you don’t yet know.

  • One of the biggest struggles I had was that I’d tried to write my part of the piece while I still had unanswered questions. There were so many different rabbit holes I could’ve gone down and started working into the piece, but only some would actually be relevant.

  • First, I just vaguely explored the topic – I read existing articles and reviews on malaria vaccines, and formed a basic understanding of the problems.

  • Then I talked to several experts with questions I still had. Talking to experts isn’t necessarily about getting precise answers – conversations are rarely so detailed. But I found it very useful to hear pointers on what I should be looking into.

    • For example, some experts explained that adjuvants were critical in making malaria vaccines effective – so I read much more about them. Some mentioned the WHO’s decision in 2015 to ask for more pilot studies – so I dug into what happened then.

    • Every expert I spoke to said that the main reason we didn’t get a malaria vaccine sooner was because it primarily affected the poor. This was painful to hear; it was also a big motivation for me to find out the details and get this piece finished (which we’d been working on for several months).

  • At this point I tried writing my part of the article. It still didn’t work – my writing went into dead-ends; I over-explained some parts, while not knowing enough about the others, and often had no idea how to progress from one section onto the next one.

• Sometimes the most ‘obvious’ structure of a piece is the best one.

  • It didn’t occur to me until quite late that the best way to write this article would be chronologically. In retrospect, that seems obvious and natural – but it wasn’t at the time.

  • Because the answer to the question has so many layers, I eventually realised it would be logical to build up the piece progressively, chronologically, as a story that developed as scientific understanding grew – first, with the basic facts, and then with the major complications that people faced, in order.

  • This meant the piece needed much more historical context than I initially expected, to follow along the scientific story. Still, it made the piece more interesting than it would’ve been otherwise.

• List questions and find out what you still don’t know.

  • At this point – still fairly stuck – I wrote down a list of questions I still hadn’t answered.

    • For example: Why did the malaria eradication program fail? When did DDT resistance evolve? Why did the researchers pick the CSP protein for the subunit vaccine? How did malaria vaccine funding change over time? How many malaria vaccine trials had been ongoing since 2001? What are the barriers to doing clinical trials in sub-Saharan Africa? How were the sequential trials done?

  • Then I closed the tab with my google doc draft, and kept it closed for days, while I went and tried to find the answers to each of them. I wrote notes with links and references below each question I’d had.

  • Now, finally, even the parts that weren’t in my memory were at least written down accessibly. I could finally start writing the piece properly and fill in the remaining details, fact check and adjust anything afterwards.

Subscribe now

Fact-check as much as you can.

The worst thing about ‘fun facts’ is how many of them are not facts at all. They often come out of misunderstandings, misrepresentations, or simply trace back to nowhere. Zombie anecdotes and statistics are very common, including in academic reviews by subject-matter experts.

In my view, it’s not surprising that this happens. We misremember numbers and details. We like fun and interesting stories, and we don’t have inherent reasons to distrust popular historical anecdotes that lots of people believe.

But this means errors compound over time. It’s easier for science to be correcting when we continue to face the same problems, and accurate details of a specific concept continue to be important – like knowing how many genes are in the human genome. This seems less the case for historical anecdotes; they seem more like the game telephone (or ‘Chinese whispers’), where people insert new misunderstandings into the story as it gets passed along over time, and people cite newer and newer reviews on the topic, rather than primary sources.

I learnt a lot about how to go about fact-checking historical claims from talking to the historian of innovation Anton Howes, who is also a great friend of mine. He recently wrote a fantastic post after our conversation called ‘Does history have a replication crisis?’ which I really recommend.

What helped?

• Tracking down & citing the primary source if I could.

  • Not a review from the 2000s – if something supposedly happened in 1956, I should go and find a reference from then, from official sources or the person involved in the research themselves. Google Scholar and Archive are really helpful for doing this – you can filter search results by year, and actually read documents from the time.

  • It’s often fun, inspiring and eye-opening. Biology papers from a century ago are surprisingly perceptive, and help you place yourself at the time – you learn what people struggled with, which technologies they had access to, and which unanswered questions they had. This was also really helpful to write the piece in a style that brought people along on the journey.

  • Here, for example, is a book (in French) by Charles Laveran on his research on the parasite that causes malaria. I searched through this and other documents from the time to find out whether it was really true that he showed Louis Pasteur and Robert Koch a sample under a microscope in person and immediately convinced them, as one review put it (without a citation). But I found absolutely nothing documenting this from the time. So I removed it, along with anything else that seemed unsupported. If we really want to understand what problems people have faced, to learn from them, accuracy is far more important than a fun story.

• Finding niche, in-depth reviews of the precise topic, with precise details.

  • When I wanted to learn about the story behind how the rodent models were discovered, I read a book called Rodent Malaria.

  • Other sources about animal models lacked citations and details, and also misrepresented the story – they made it sound like rats affected by malaria were found immediately after researchers looked for them in the Congo. Instead, the book explained that it took two years to find one: researchers had tested hundreds wild rats in the Kisanga province, until they found one positive for the malaria parasite. I then tracked down the original bulletin and paper describing the finding, by searching for the authors and restricting the time-range, which corroborated this.

  • I think this is a useful ‘red flag’ to recognise a zombie claim – it lacks names of specific people, places and dates. That’s not always good enough, because specific names might just be specifically wrong. Still, this heuristic is useful to know when to dig into claims that could just be hearsay, that lack details or citations because the person doesn’t know them.

  • I went into a similar rabbit hole trying to trace the story of ‘the cobra effect’ a while ago, and wrote a thread about it here.

• Corroborating the narrative. Does it make sense, given what else you know about the time?

  • Take the finding that DDT was harmful to animals, birds and fish. It’s often thought that there wasn’t research into this until the 1950s or ‘60s, or until Silent Spring was published in 1962.

  • However, I found quite a few papers describing its harms from the 1940s – this review from 1946 had a great summary of what was known at the time. It makes a lot more sense considering the pesticide was an insect nerve poison developed during World War Two, and much of the early toxicity research was conducted by the US army.

It’s a shame that all this was necessary, as it’s very time consuming. In ideal circumstances, we’d be able to outsource our understanding of specifics to other people with expertise. Anton had some great ideas on how to fix this problem, but it’s likely there’s a lot of different ways to improve the situation.

Subscribe now

What I learnt about scientific discovery

I used to find it a little funny that this blog was called ‘scientific discovery’ even though not much of it focused on the process of discovery, or the history of fields I was interested in. From now on, I’d like to make that a much bigger focus.

I think the buzz around the ‘room-temperature superconductor’ is an example of how the science-hype cycle is often coverage of speculative technologies that might become practical years from now, assuming they work out at all. Meanwhile, there are just passing mentions of breakthroughs that have finally succeeded and why. There are rarely retrospectives on how much they’ve made a difference to people’s lives.

Both successes and failures can be incredibly informative. Why did some ideas work but not others? That sounds like a fuzzy question that’s difficult to answer, but if you look closer there are often very specific obstacles that scientists faced – that needed to be overcome, or avoided in some way.

This piece – on the story of the malaria vaccine – is about an enormous achievement, but it’s also about many disappointments and failures along the way.

Here are some things it demonstrated to me about the process of discovery.

• Iteration. Parts of the story show how discovery is often iterative, once the ‘active ingredient’ is known. Once chemicals with a particular structure had been discovered as insecticides, several more were identified, including DDT. After one vaccine with the CSP protein was shown effective, another successful one was developed with a similar formulation.

• Non-linearity. The process of discovery is not necessarily linear, as Jason Crawford has written previously. It doesn’t always go from basic research to engineering, but goes back and forth between tinkering and genuine understanding.

  • The first vaccines were developed before the development of germ theory, and before the discovery of viruses. But once this knowledge had grown, with the discoveries of specific pathogens and how they caused disease, it finally led to a stream of new vaccines.

• Multiple causes. The story shows how the same steps could have been sped up in multiple ways. The discovery of a useful animal model could have been found through closer attention to the original findings; or if more researchers had been working on the problem; or pure luck, stumbling upon a more practical animal model first.

  • This was also true of the final part of the journey – malaria trials could have been sped up if there had been more existing trial infrastructure in sub-Saharan Africa, if there was more philanthropic funding, with Advance Market Commitments, and so on. As we explain in the piece, AMCs are one of the best ideas in this situation because they get the technology past the finish line and help scale them up.

  • I also really liked this thread by Jacob Trefethen on other policy ideas and implications from the story.

• People, goals, times and places. Fundamentally, the malaria vaccine story is about how discovery doesn’t happen in the abstract, but is driven by people motivated to work on the problem, who live in a particular time and place. They’re limited by their knowledge and access to technology and resources, and influenced by the economy, international efforts and programs, and sometimes even war and conflict.

And that’s all! I hope you enjoyed reading this and the piece, of course. Here it is again if you’d like to read it now.

Read the piece here

I’ll be back soon with a roundup of great new scientific research I’ve been reading, so I hope you’ll subscribe if you haven’t already.

See you next time :)

– Saloni

Subscribe now

Hello! It’s been almost six months since I’ve done a round up of great research I’ve read, so this post is long & won’t fit on email – but it still only covers a fraction of the things I wanted to share. I hope you enjoy it! If you spot any errors, let me know so I can fix them.

Subscribe now

Recently I’ve been thinking about charts that have helped me see something in a new light, and that illustrate an important concept. Here’s one of them.

The contours of death

Okay, I just enjoy giving spooky titles to my subheadings, but this chart’s very interesting.

It shows death rates in the US across two dimensions: time and age.

The shades represent death rates, and splodges show when death rates have risen. The contour lines around the splodges are like an elevation map: they show how much death rates have risen (or fallen).1 You can see how they’ve risen and fallen over time (on the x-axis) and across age groups (on the y-axis).

You’ll see that events with wider splodges occurred over longer time periods. Events with taller splodges affected many age groups.

• For example, look at ‘poisonings’, which includes substance use and overdoses. Here, it shows the opioid crisis, and you can see it rising gradually across adult age groups.

• Or look at the HIV/AIDS panel: as an exponentially-growing epidemic, it surged over a very short period of time, especially among young and middle-aged adults.

• Or look at traffic accidents, which were especially deadly among young people and the elderly, but have become less so over time.

Suddenly, the plots I’m used to seeing – line charts where something rises for some age groups – seem quite flat, even when they show the same data.

The shapes show how events permeate across a population and time. I think they also give visual clues about the nature of the underlying events: Were they constrained to only affecting a particular age group? Did they grow exponentially over time? And maybe these tell us something about how and why they happened.

I really liked this chart, and could look at it for a long time. Are there other types of charts that have helped you understand some data better?

Here are some little dives into other great research I’ve read!

The twin baby boom

This new database, called the Human Multiple Births Database, shows national data on twin births from 24 countries. In the chart below you can see trends in having twins, measured as a rate per 1,000 deliveries.

Twin rates rose from the 1970s across many countries. You can see that they approximately doubled in some countries, from around 10 twins per 1,000 deliveries to 20 or more.

The main reason for the rise is because women have been having children later on average and using reproductive technologies that were invented then. This includes IVF (where multiple eggs are implanted) and ovarian stimulation (where multiple eggs are released) to make sure that at least one is successful. These can result in having fraternal twins.

But since the early 2000s, twin rates have plateaued and fallen. This has happened in a range of countries, as they introduced new standards for reproductive technologies. (Many now countries measure the success of these technologies by their effect on singleton2 live births, because pregnancies with multiple babies tend to be riskier.)

I’d learnt about this at school, but seeing the trend helped me understand the magnitude of twin trends in the population. You can also explore the data for other countries on this interactive website by the researchers.

Subscribe now

What makes a good conversationalist?

Surprisingly, the pandemic provided an opportunity to study this question in detail.

Volunteers in this new study had unscripted freeform conversations with strangers over Zoom in 2020, in the United States. They were asked to ‘talk about whatever you like, just imagine you have met someone at a social event and you’re getting to know each other.’

The conversations were video recorded, and analyzed for patterns of speech and behavior. Then, their speech patterns were compared to people’s feelings before and after the conversation, and their ratings of their conversational partner. In total, the dataset had 1,656 conversations each lasting at least 25 minutes.

Since the dataset was large and detailed, it could answer a lot of questions. First, how long are the gaps between each person’s speech in a conversation? Answer: Less than a second on average, which suggests that people predict when their conversational partner’s sentences are going to end and how they’ll respond. Some people spoke too soon and their speech slightly overlapped or they interrupted the other person.

What are the most common ‘backchannel words’? (These are words people use to show they’re following the other person’s speech.) Answer: Yeah and mhm. This study was done in the US, and it’s likely these vary in other countries.

Were people happier after a freeform conversation with a stranger? Yes, on average! You can see this across age groups below. Volunteers were asked to rate their happiness on a 9-point scale immediately before and after the conversation.

Although I wonder how much this came from the sample – people who like having conversations were probably more likely to join the study, and it happened during the pandemic, when many people were probably very bored at home.

Finally, what makes a good conversationalist?

Volunteers were asked: ‘Imagine you were to rank the last 100 people you had a conversation with according to how good of a conversationalist they are. ‘0’ is the least good conversationalist you’ve talked to. ‘50’ is right in the middle. ‘100’ is the best conversationalist. Where would you rank the person that you just talked to on this scale?’

People who were rated higher by their conversational partners tended to speak fairly quickly, and with more emotional intensity. Also, they tended to use more head movement (nodding for yes and shaking for no) while listening, and showed more facial signs of happiness.3

I really liked this paper because it described many basic aspects of conversations, and compared the analysis across different pattern-recognition software.

Subscribe now

Here’s something else cool.

Are speakers of tonal languages better at discerning music?

You might know that I grew up in Hong Kong, where most people speak Cantonese.

It’s considered a difficult language to learn to speak because it’s a ‘tonal language’ – each syllable can be spoken in 6 different tones.4 This means the pitch of a syllable can be either rising, falling, stable, or falling-then-rising; and can start at different levels.

The tone provides information about the word: if you say a word with the wrong tone, people might think you’ve said a completely different word. Here’s an example:

I listen to Cantopop and Vietpop music sometimes5, and I’ve occasionally wondered about the relationship between music and tones: Do tones constrain the compositions that songwriters can make? Are tonal speakers better at matching a pitch?

A new study looks at this topic. The researchers created a music game (here’s the link) to test if people could notice slight changes in melody, tuning, and beats. It was played by 493,100 participants around the world, who spoke 54 different languages in total.

They found that people who spoke tonal languages – like Mandarin, Cantonese and Vietnamese – generally tended to do better at distinguishing melodies, but weren’t as good as others at perceiving beats. You can see this below.

The authors suggest maybe this is because they place more weight on cues that are relevant to their language.

When I read studies like this, I try to imagine the different kinds of people who might be participants – could the participants vary on skills relevant to the game? Could their skills vary between countries or languages, for other reasons?

Here the study was an online game. So I’m thinking: maybe people from some cultures are more likely to play musical instruments, or the game got shared on music forums in some countries. Also, maybe people speaking some languages understood the game instructions better, or were more familiar with the melodies or rhythms in the game.

This study adjusted and explored several of these factors. You can see below that people who have taken music lessons performed better on all three parts of the game, across language types.

The researchers say that “in each case, the effect size associated with being a tonal language speaker was roughly half as large as the effect of receiving music lessons, indicating an effect of substantive practical significance.”

I think that’s a good way to put the advantage into context! It’s also cool to use a large online game in research, to get enough people from different backgrounds, and where you can fine-tune parts of the experiment.

Subscribe now

Do blood pressure drugs’ effects vary between people?

Lots of research is focused on understanding the average effects of treatments.

In a randomized controlled trial, each person is allocated to only one of the groups at a time. So we don’t get to see whether they, specifically, would have gotten better anyway. We only get to see the change for people on average.

When we try to understand whether something works, it can be challenging to tell if it works for everyone to the same degree or if it works more for some people and less for others.

The main reason for this is that we’re only able to see one version of history: When someone makes a decision, we can see what follows, but we can’t see what would’ve happened if they had made a different decision.

This distinction – between averages and individuals – is blurrier when people naturally experience a lot of fluctuations. For example, there’s a lot of day-to-day variation in people’s blood pressure – so maybe for some people who’ve taken a medication, it was just a good day, and for others it was a bad day.

But there are ways to get understand how much the effect of a drug varies between people.

One way is by understanding the mechanism of the drug. In cancer treatment, for example, many cancer patients receive treatment for their specific cancer mutations, after genetic testing on their tumours.

Another way is by testing the effect in different subgroups, although you might need very large sample sizes to find the precise differences in the drug’s effect in the different subgroups.

A third way is to track people’s fluctuations while theyʼre using the same drug, and adjust for this fluctuation. This helps to see the probability that people would do better anyway, without the drug.

That’s what this new RCT did. 280 patients with hypertension were randomly allocated to each of four different drugs (one after the other), and then two of the same drugs again. Their blood pressure was measured 14 times a day. The drugs were lisinopril, candesartan, hydrochlorothiazide and amlodipine; each drug is used to reduce blood pressure, but comes from a different drug class.

Let’s see the results! (This chart looks complicated but I’ll explain it below.)

• Each panel compares two drugs to each other. The blue diagonal line compares people’s blood pressure on one drug to their blood pressure on another drug. The black diagonal line shows where it would be if there was no difference between the drugs.

  • You can see that for some pairs of drugs, the two lines are more apart, and there are differences in drug efficacy. For example, patients taking hydrochlorothiazide have a 5.1 mmHg higher blood pressure than those taking amlodipine, on average. Lisinopril was the most effective, on average.

• Each dot represents a participant. Some dots have a range – this is because the participant was given the same drug twice, and the range shows the variation in their blood pressure over the repeated drug. You can see that there’s often a lot of variation within the same people.

• Because of the design of the trial, the researchers could also test whether different people did better on some drugs than others. With this, they estimated that if the participants received personalized treatment with a single drug that worked better for them, they’d have a 4.4 mmHg lower blood pressure on average. They also estimated that personalized treatment would be better than if all of them received the most effective drug (lisinopril).

Subscribe now

Even more

Here’s a more condensed summary of some other great stuff I’ve read.

• Can we save Tasmanian devils from contagious cancers?

  • Something you may not have heard of. The population of Tasmanian devils has dropped by around two-thirds in the last 25 years. They’ve been affected by facial cancers, which spread when they bite each other during fights.

    

  • There are two types of these cancers, called Devil Facial Tumour 1 and 2. They each cause very similar disease, but a recent study shows that each type evolved independently. (This BBC article has more background and photos in case you’re interested.)

  • Now, field trials are about to begin to test vaccines that could protect them from these cancers.

  • The vaccines will be rolled out as oral bait, similar to how oral rabies vaccines have been given to wildlife across Europe (which happened by dropping them from helicopters and aircraft). I hope the devil cancer vaccines succeed!

• A new study shows how estimates of people’s beliefs in conspiracy theories are often overestimated (Seth J. Hill and Margaret E. Roberts, 2023).

  • Estimates are usually based on surveys where people are asked whether they agree or disagree with a conspiratorial belief. Unfortunately people tend to prefer to agree across surveys, which is called ‘acquiescence bias’. This leads to overestimating how many people actually believe something.

  • The researchers show this bias exists by asking the same questions in multiple different styles, like randomly flipping the phrasing of the question so the agree/disagree responses would be flipped.

  • I also liked this quote from the study: “The consequences of bias increase in the rarity of the target belief. The smaller the fraction of the population that holds a belief—e.g., when asking about outlandish conspiracies—the more likely agreeable responses are due to acquiescence bias rather than actual belief.”

• You might have heard of ‘left-digit bias.’ This is the idea that customers tend to treat prices like $3.99 as being much lower than $4.00.

  • Here’s an example from a recent paper (Strulov-Shlain, 2022). The author looks at how the demand for products changes as the price rises, by using data on price changes and sales of the same products over time, from around 3,600 stores in the US.

  • You can see this below. There are sudden drops in demand as prices reach a new dollar. The author estimates that customers treat a 1 cent rise at the boundary (from $3.99 to $4.00, for example) as a 20 cent rise.

    

• Did the 1918 flu pandemic cause a baby boom two years later, in 1920 Europe?

  • According to a 2004 study using data from Norway, the end of the 1918 Spanish flu pandemic led to a baby boom.

  • A new study challenges this view, and argues that it was actually the end of WWI (or rather the post-WWI economy) that led to the baby boom.

  • The authors look at more neutral countries, not just Norway, and show that communities with the biggest baby busts in 1919 also had baby busts in 1920. This is evidence against a rebound.

    

  • Another important point is that, while the 1918 flu pandemic was a global phenomenon, the 1920 baby boom was largely concentrated in Europe.

    

• Have you heard of ‘multipartite viruses’? These are viruses that are made up of different segments that spread separately. But, when they infect new cells, they can only replicate their entire genome if all the different segments have infected the same cell.

  • This seems like a huge disadvantage – since most of the time, there won’t be all the segments in the same cell, so most lineages die out. So why have multipartite viruses evolved several times independently? Why don’t they all eventually die out?

  • A new study uses theoretical simulations to argue that they’ve evolved by cheating, and that there are more advantages to this than previously thought.

    

Subscribe now

What I’ve been upto

• At Works in Progress, we recently launched a new platform for updateable books – specifically, Stewart Brand’s in-progress book ‘Maintenance: Of Everything’.

  • You might have read his fantastic piece The Maintenance Race last year. That was only the first chapter: here you can read the rest.

  • This platform lets you make comments and suggestions on the chapters, which he can incorporate into new versions later on. My colleague and friend Nick Whitaker did an amazing job making this happen.

• At Our World in Data, I’ve been writing about the key things you need to know about big topics in global health, like influenza, diarrheal diseases, data on mental health, and suicides.

  • For example, did you know that there are multiple flu seasons each year in several tropical countries near the equator, or that they can have flu present all-year-round?

    

  • I also wrote about one of the most common causes of death in children globally, that most people don’t know about: diarrheal diseases.

    • It’s estimated that 500,000 children under five die from them in a year, and three-quarters of these deaths are caused by 10 pathogens.

    • Diarrheal death rates are 50x higher in poor countries than wealthy countries, and most deaths could be prevented with measures like better sanitation, rotavirus vaccination, and oral rehydration therapy.

• Aside from work, I’ve been eating lots of gelato – so much so that today my phone automatically created a collage of gelato selfies I had taken. I like this picture because it looks like I’m saying ‘Cheers!’ with a gelato.

  

• I also recently went to the Hunterian museum. It’s about the history of surgery and anatomy, and is my favourite museum in London. It recently reopened, after being shut for 6 years for renovation. It’s amazing, but also quite disturbing, and maybe not recommended for people who get queasy around body parts (which might be most people).

Anyway, that’s all for now! I hope you liked reading this and learnt something new. If you haven’t already, you should subscribe! I have lots more planned to write.

See you next time! :)

– Saloni

Share

Subscribe now

Thanks very much to my friend Nathaniel Bechhofer for reading a draft of this post and pointing out some clumsy phrasings.

In case you’re there too, I recently joined Threads (and Bluesky). I don’t know, maybe none of these will work out, but maybe they will, and I’m a little fed up with twitter.

27th January 2024: I improved phrasing in several parts of this post to make it easier to read & understand.

Hello! It’s been a while because I’ve been pretty busy at my new job at Our World in Data. But I’ve also been reading lots of great research, and I’ve tried to summarise some of it here.

If you like this, I hope you subscribe. And if you spot any errors, please let me know so I can fix them. (By the way, this post is too long for email, so it might cut off later on.)

Subscribe now

Children who died & the people who survived them

Max Roser has a great new post on the decline of child mortality, which I found very inspiring. It explains that until around two hundred years ago, around 1 in 2 children died before they were 15 years old.

Even in 1950, around 27% of children worldwide died by that age.

I think tragic statistics can often be difficult to actually absorb. It’s often easier to think of big numbers as something abstract and distant. It’s also common to think about the effects they’ve had at the broader level of population statistics, rather than the actual people involved.

But each of these children had an impact on people’s lives, many of whom are still alive today.

How many children lost a sibling? How many parents lost a child? How many people lost both – at different times in their lives?

Two new studies look at these questions, using data from the Demographic and Health Survey (DHS) Program, which has conducted very large surveys on child mortality in many low- and middle-income countries over decades.

The authors found that in the 1960s, over 50% of women lost a child or a sibling before they were five years old, on average in these countries.

• In the 1960s, in Latin American countries, 21% lost a child under five and 35% lost a sibling under five.

• In South Asian countries, 22% lost a child under five and 21% lost a sibling under five.

• In Western African countries, 50% lost a child under five and 34% lost a sibling under five. This means every other woman surveyed had lost a child under five years old; and one in three had lost a sibling under five.

Many lost both. In total, between 20–40% of women surveyed had lost a sibling and a child under five in most low- and middle-income countries, in the 1960s.

Losing a sibling was related to having a higher risk of losing a child, as you can see in the chart below.

Women who had lost a sibling under five years old were around 40% more likely to lose a child later on. In other words, disadvantaged families in one generation tended to be the families at a disadvantage in the next generation as well.1

But look back at how the first chart shows a decline over time, as well.

Globally, 27% of children died before they were 15 years old in 1950.

But that figure became 4.3% in 2020, which is a 6-fold decline in seventy years. For some wealthy countries, the decline has been more than 100-fold over a couple hundred years.

This experience and the even-lower rates in rich countries today, are why we know the global average can be driven down much further. I think we don’t have to imagine what a future could look like without this immense loss. We can already see that it’s possible.

Subscribe now

The survival of a cohort

I want to show you another visualisation that helped me understand how these changes have permeated across the population.

It comes from researchers who used data from the Human Mortality Database (HMD), which has national mortality data from many high-income countries.

The authors show the survival of people relative to others who were born the same year.

This chart looks complicated but there are so many fascinating things about it, so I’ve tried to break it down underneath, focusing on the data from France.

The survivorship estimates are ‘period measures’: they estimate how long people would survive if they experienced the corresponding death rates throughout their whole lives as those seen in each age group in that particular year. (This has been corrected.)2

• x(.01): At the top of the chart, you can see the top-percentile x(.01), which is the longest-lived percentage of women. It shows that among those born in 1900, women in France who die at the age of 90 would be in the top 1% of survivors. But among those born in 2018, they would only be in the top 40%. Now, the top 1% are those who would die over the age of 100.

• x(.50): Women born in 1900, who would die at the age of 60, would be the median survivors of their cohort. But by 2018, the median survivor would die in their mid-80s.

• x(.90): In 1900, 90% of newborns would survive at least until the age of 2 in France and the age of 5 in Sweden. The survival of the 90th percentile was lifted up several times over the century, in particular during the mid-1940s, like the survival of many other percentiles around them. For those born in 2018, 90% would survive at least until their mid-60s.

• x(.99): At the lowest end are those who would die before 99% of their peers. In 1900, they would die in infancy, which means 99% would survive only at least a few months of age. But in the mid-to-late-twentieth century, their survival would be suddenly lifted up. By 2018, 99% would survive until almost the age of 40.

The study has many other fascinating insights as well, but the reason I’ve highlighted it is because it tells us something else important too.

It tells us that people nowadays have many more of their friends and peers alive with them as they age.

So here is a different way to describe this chart.

30 year olds born in 1900 France would have 74% of their birth cohort still alive with them. 30 year olds born in 2018 would have over 99% still alive with them.

80 year olds born in 1900 France would have only 10% of their birth cohort still alive with them. 80 year olds born in 2018 would have 75% still alive with them.

What does it mean when fewer people die?

There’s a tendency to see statistics on big topics like mortality as just numbers, and treat them as being separate from stories of people. We struggle to make sense of very large numbers because we don’t know how they relate to us as individuals.

It’s easy to forget that these statistics are distilling something shared about the lives and experiences of billions of people.

By looking at them from different angles like this – the different ways people experience them, and all the people they affect – maybe we can begin to understand what they mean.

This time, we’re all part of the story – it’s about all of us, our parents, grandparents, children and friends, and families we’ve never met. It means fewer parents lose their children, fewer children lose their siblings, and we grow old with more of our friends still alive.

Subscribe now

What explains the massive changes we’ve seen in survival across the population?

Of course, the list is long – with factors like nutrition, clean water and sanitation, neonatal healthcare, vaccinations, medicines, reductions in conflicts and famine, and many others, playing major roles at different times.

Something often overlooked is that, even as people age, the health advantages they’ve already had then carry forward – because they reduce the chances of developing comorbidities that would make them vulnerable to new problems later on.

Below, I’ll focus on one of the major factors: clean water, and explain how it’s an easy way to reduce sickness and save lives.

Why does clean water matter?

Contaminated water can contain pathogens like rotavirus, cholera, adenovirus, E. coli and shigella, which cause diarrheal diseases. Or in tropical regions, it can contain parasitic worms that cause schistosomiasis, Guinea worm disease3, trachoma, among others.

All of these waterborne diseases are much more common in countries without safe drinking water and sewage, and diarrheal diseases in particular are one of the top most common causes of death in children worldwide.

Take a look at the scale on the map below. Across age groups, diarrheal deaths are about 50 times more common in low-income countries than high-income countries, on average.

This is noteworthy because we know from research how to prevent so many of them with tools like rotavirus vaccines, oral rehydration therapy, cholera vaccines, sanitation and clean water.

There are many simple and effective treatments to clean water – such as chlorination, water filtration, and solar disinfection. These can kill pathogens in drinking water.

What’s less obvious though is how effective they are, and how much of a difference they make. To answer these questions, a good source of evidence are randomized controlled trials.

Recently, a large RCT in Malawi tested out a clean water program: a random selection of households were given free coupons, which they could swap for chlorination solution at local shops, to add to their drinking water.

The trial was conducted over a period of 18 months with over 2,300 households.

In the chart, you can see that the chlorination coupon program reduced children’s diseases by around 21% over the next six months, compared to the control.

This is a big reduction. But note that I’ve said the chlorination coupon program reduced children’s diseases, not chlorination. This is an important distinction.

The result is an average across all the households in the group. The authors found that only 20–40% of households were actually using the chlorination treatment at intervals in the study.4

So, this doesn’t directly answer how effective chlorination is in reducing disease. It answers how effective the chlorination coupon program was.

If you focused on the households that consistently used chlorination, you would see they had an even larger reduction in disease. But that may not be the relevant question.

When researchers test out ideas and treatments like this, they often want to test how effective the actual program would be, if it was implemented in the real world.

This is one reason researchers use what’s called ‘intention-to-treat’ analysis56, which is when they compare the groups based on what they were allocated to receive, not what they actually received.

It also reminds us that other programs can have different effect sizes. For example, if there were piped clean water systems installed (so people didn’t have to remember to add chlorination solution regularly), I’d expect the benefits to be greater.7

To put this trial into perspective, let’s look at other trials for clean water treatments.

Clean water treatments, at scale

The authors of a big recent meta-analysis looked at clean water treatment programs and their effects on child mortality across RCTs in 21 countries.

In each trial, randomly-selected households or villages were given clean water treatments (e.g. chlorination, water filtration or spring protection). These were given through chlorine dispensers, coupons, solution, or filtration devices.

The trials took place in countries including Bangladesh, India, Pakistan, Uzbekistan, Kenya, Malawi8, Rwanda, the Democratic Republic of Congo, Zimbabwe, Zambia, Guatemala and Bolivia, all between 1998 and 2021.

It’s usually difficult for a single study to see the effect that something could have on reducing mortality.

This is because deaths within a study tend to be rare to begin with. So trying to identify a reduction in deaths, starting from a small number, is tricky, and the effect sizes can be unstable. But with a very large study or meta-analysis9, identifying the effect becomes possible.

By pooling the data from all these trials, the authors estimated that clean water treatment programs reduced mortality by around 30% in children under five years old, across trials. You can see this below.

Although there is some uncertainty10 around the precise figure, this was a surprising result to many people.

Using data on the costs of existing programs, the authors also estimated that a scaled-up version would translate to around $2,974 USD per death averted in children under five years old, which would place it among the simplest and cheapest ways to prevent death that people know of.

I want to note something here though, which is that this summarizes the effect across a group of places, treatments and times.

For example, across the trials, an average of 69% of households actually used the clean water treatments, but this ranged from 27–87%.

Another point is that the baseline level of pathogens can vary too. In more developed areas, where pathogens have already been filtered out, the benefits of such a program will be lower. But where basic water and sewage systems are lacking, the benefits of clean water treatment will be large.

Remember that these studies tell us about the results of clean water programs tested within the last 25 years. They show us that even today, our potential for making progress on something as basic as drinking water is very large.

This is one of many reasons why I believe there’s still lots of low-hanging fruit in development. Progress is not just about the frontier with new technology and invention, but also about how much we can do with what we already know, even now.

Subscribe now

New Works in Progress

• We released a new issue of Works in Progress magazine last week, and it includes many awesome pieces like:

  • Every grain of rice – on the future of rice biotechnology, by Niko McCarty.

  • The end of acid rain – on the story of how countries came together to stop acid rain, by Hannah Ritchie.

  • Why Britain doesn’t build – on the entire history of the UK’s housing crisis, by Samuel Watling.

  • Thomas Edison, tinkerer – on the great inventor who turned inventions into practical systems, by Eric Gilliam.

  • Taming the stars – on how to lay the groundwork to build low-cost, safe nuclear energy plants, by John Myers.

  • How DC densified – on how Washington, DC avoided the housing problems that so many American cities faced, by Emily Hamilton.

  • Olivine weathering – on a fascinating way to sequester carbon from the sea, and thus from the air, by Campbell Nilsen.

Recommendations

Here are some other things I’ve enjoyed recently that you might like too.

• The book The Butchering Art, on the gruesome history of surgery and how it was transformed by Joseph Lister. (I’ve been listening to the audiobook, which is very nice.)

• Some incredibly realistic medical anatomy illustrations by Sana Khan.

• An eight hour long instrumental playlist to help you work and keep you awake, by me.

  • Pseudoerasmus once described it as “The Goldilocks allegro ma non troppo instrumental playlist. Not too andante, not too presto or vivace. And not too dissonant.”

• The film And the Band Played On, about the discovery of AIDS, how it transmitted, and how it was caused by HIV.

  • Did you know that credit for the discovery of the cause of AIDS led to fierce legal battles between American and French research teams, and then US President Ronald Reagan and French Prime Minister Jacques Chirac settled terms on who would get credit and patents for it? (It was only later recognised that the French team of Françoise Barré-Sinoussi and Luc Montagnier had identified it quite early on, and they won a Nobel prize for their discovery.)

• Two great podcast episodes of Tides of History, on the history of Venice and the Venetian empire. By the way, history podcasts are amazing to listen to while you’re on holiday in the place itself.

• Three online textbooks:

  • Telling Stories with Data – on statistics, programming and communicating data science, by Rohan Alexander

  • Human Genetic Variation – on computational genetics, by Rajiv McCoy, Stephanie Yan and Kate Weaver

  • Cell Biology by the Numbers – on the maths of cell biology, by Ron Milo and Rob Phillips

    • I also loved the post Biology is a burrito, on why thinking about biology in numbers is so important, by Niko McCarty.

And that’s all!

I hope you enjoyed this, learnt something new, shared it with your friends, and subscribed if you haven’t already.

As always, if you spot any errors, please let me know so I can fix them.

Have a great rest of your week! :)

– Saloni

Subscribe now

Share

P.S. you might have noticed I’ve switched over this newsletter to a new URL, but it’s still on Substack and all the old links will redirect to the new one.

P.P.S. I’ve been recovering from a cold, so my voice might sound a little stuffy on the audio voiceover.

This week’s post is about what I believe is the most underrated problem in global health: missing data. But to illustrate it, I’m going to tell you about snakebites.

If you like this, I hope you subscribe! And if you spot any errors, please let me know so I can fix them.

Subscribe now

How many people die from snakebites?

Study: Trends in snakebite deaths in India from 2000 to 2019 in a nationally representative mortality study (Suraweera et al., 2020)

Here’s a statistic that might surprise you: Around 1 in 270 people in India die from snakebites by the age of 70.

In total, around 50,000 to 60,000 people in India die from snakebites each year. In Australia, which is home to the most-venomous snakes in the world, that number is 2. Just two.

Many more people live in India than Australia, so here’s another comparison. In India, the death rate from snakebites is around 5 per 100,000 people per year. In Australia, it’s 0.13, about thirty-eight times lower.12

If you aren’t from India, these statistics are probably new to you. But they are also surprising to many people from the country.

Deaths from snakebites are highly ‘spatially concentrated’. In urban areas of India, many people won’t know anyone who’s died from a snakebite. But in poor agricultural areas, many will.

Why do so many people die from snakebites in India?

Because snakebite deaths are caused by snakes and poverty together. Snakes are more common in places with grain agriculture, at low altitudes, where snakes live and feed on rodents and amphibians.

People who die from snakebites tend to live in poor rural parts of these areas, often work in farms, and lack access to hospitals, doctors and treatment.

Antivenom

Snakebites have a treatment: antivenom. But it might surprise you to learn that antivenom still comes from a method that hasn’t changed much in over a hundred years.3

First, venom is milked from the fangs of venomous snakes. Then, it’s injected into horses or other big animals. They produce antibodies to it, and these are extracted from their blood and purified into antivenom for humans.

Subscribe now

Recently, there’s been an increase in research funding to create synthetic antivenom. But experts believe this isn’t enough, or isn’t happening fast enough: the current timescale when these might be available is thought to be 10–30 years away.

Current-day antivenom is very expensive. It’s also difficult to store and transport.

Here’s an excerpt by Mathias Kirk Bonde on the topic, in his excellent piece Advancing antivenom:

For over 34% of Indian snakebite victims, it takes more than six hours to receive treatment. If the clinic lacks a cold chain – a coordinated system of temperature-controlled environments – it is limited to using antivenom that can survive room-temperature storage. Without equipment to diagnose the species of snake the patient was bitten by, doctors must use polyvalent antivenom designed to work against multiple species of snake, which is typically more expensive and less effective than species-specific monovalent antivenoms.

Antivenom is no miracle cure and adverse immune reactions to it are common. Patients need to be monitored for up to multiple weeks after receiving antivenom in case of delayed serum sickness, which can be fatal.

In India, snakebite deaths happen more often in summers and during monsoon seasons. This is when more people are involved in agriculture, and when water floods into underground burrows and snake habitats. This leads them up into higher ground and buildings, and into contact with humans.

The flooding and disruption from monsoons is also likely to mean life-saving antivenom is even further out of reach for victims.

Even so, 50,000 to 60,000 snakebite deaths in India each year is a big number.

Only about 1,000 snakebite deaths are recorded per year by the Indian government.4 In 2008, a meta-analysis estimated around 11,000 deaths per year instead, based on how many people took insurance claims from a government scheme.

Well, this revised estimate was 58x the recorded number, and 5x the second estimate which was meant to capture underrecording.

Why were they so far off?

Missing deaths

We take it for granted in much of the world that we know how many people are dying each year, and what they are dying from.

But that’s not the case worldwide.

In many countries, there isn’t a functioning civil registry, where births and deaths are regularly recorded. This means many deaths aren’t registered officially with a death certificate. But even among the deaths that are registered, many lack a ‘cause of death’ on the certificate.

In 2015, when it was estimated that around 56 million deaths occurred worldwide, only around half were registered with a cause of death.

Only around a fifth of all deaths worldwide in 2015 were reported to the WHO with a meaningful cause of death – which excludes ambiguous causes like ‘sudden death’, ‘chest pain’, ‘event of undetermined intent’ and others.

Missing deaths are common in poorer countries because of poorly functioning civil registries, because many people die outside of hospitals, because many don’t have medical records, because many lack a doctor or nurse to fill in their death certificate, and because many countries lack coroners to investigate unexpected deaths.

Subscribe now

Because official records are limited, estimating the number of deaths from each cause can sometimes be difficult. In fact, the problem has several layers:

1. First, the cause of death is not recorded for many deaths.

2. People who die outside of hospitals are even less likely to have their deaths recorded.

3. They tend to die from different causes, because they have different demographics and different risk factors.

4. And finally, hospitals affect whether people die, and what they die from.

In India, around 94% of snakebite deaths occurred in rural areas, and 77% occurred out of hospital. This is why the previous estimates were so far off.

People who died from snakebites in these areas didn’t just lack access to the cure – they also lacked their deaths being counted.

This doesn’t mean we’re completely in the dark about what people are dying from around the world. Sometimes, data is available from some regions within a country, or from estimates from separate one-off studies.

But it means our understanding is often incomplete and irregular, and sometimes inaccurate. Missing deaths are a tragedy that I think don’t get the attention they deserve, and snakebite deaths are a dramatic example of it.

Subscribe now

The Million Death Study

So where are the numbers in my post coming from? How were these revised estimates made?

Rather than using death records from hospitals (which we’ve seen are limited for causes of death nationally), they come from a national study in India called the Million Death Study. This was an important study because it deliberately collected data that had been missing.

It involved structured interviews of 1.4 million households across urban and rural areas of the country, every six months over a period of 15 years, about their relatives who died and the symptoms they had before their death. This method is called a ‘verbal autopsy’.

These households came from a random selection of thousands of small areas across the country, such as villages or urban blocks of about a thousand people each. The study was also inexpensive, with a budget of 2 million US dollars per year (about $1 per household).5

People’s answers to the interviews in the study were then coded into causes of death by multiple medical doctors, for each death. For snakebite deaths, they agreed on the cause of death about 92% of the time.

What are the limitations of this study?

When I read research, I try to imagine myself as a participant. What might make it more difficult for researchers to interpret the data they get from me?

In this case, the study involves interviewing relatives, so what comes to mind is: if I didn’t know or remember the symptoms before the person’s death. Or if the person who died didn’t have medical records or testing for diseases. These problems might make doctors misdiagnose the cause of death from my answers, especially if the symptoms I describe are shared with many other causes.

But that’s much less of a problem for deaths caused by injuries like snakebites, which are noticeable and often happen shortly before a death, making them memorable and easier to attribute.

On the other hand, when snakebites cause long-term complications that eventually lead to a death, those deaths could be harder to correctly attribute to snakebites, which suggests they may be underestimated.

As the study continued over a long period, we also have an idea of how death rates changed over time. For example, they found large declines in child mortality overall.

Snakebite death rates declined nationwide as well, and the absolute risk of dying from snakebite in India by the age of 70 declined from 0.4% to 0.37% between 2001 and 2014.6

But the changes varied between states: In the states that initially had a higher burden, the death rates slightly worsened (you can see this below).

What about in other countries? Here’s a chart showing mortality rates from a range of external causes including venom, in a large verbal autopsy project that was conducted across 20 sites in Africa and Asia.

Because of these studies and others, the WHO now estimates that a total of 4.5 to 5.4 million people get bitten by snakes per year; that 1.8 to 2.7 million people develop illness from it, and that 81,000 to 138,000 die from its complications worldwide each year.7

Subscribe now

Snakebite deaths are not a typical example, but they help to illustrate the challenge of missing data – how it arises, and how it can affect our understanding of big problems that affect people around the world.

I think missing data is one of the most, if not the most, underrated problems in global health. It doesn’t mean we’re completely in the dark, and it’s not an excuse to not act. But knowing precisely and accurately what people are dying from matters.

It helps put many other numbers in context: To calculate numbers per capita, rates per 100,000 people, percentages, and more. It’s essential in planning and decision-making, whether that’s by governments, NGOs, private companies, or individual people making decisions about their lives.

It helps us answer questions like: Which conditions are going untreated? Where are doctors needed? How many doses of medicines and vaccines are needed? How much progress are we making? Are new problems emerging?

By highlighting what we don’t know, we can bring attention to areas where more data and research would help the most. This is one of them.

Subscribe now

Share

I hope you liked this, learnt something new and subscribed if you haven’t already! I’ve double-checked and sometimes triple-checked most of the numbers, but if you spot anything I’ve gotten wrong, please let me know.

Links

If you’re interested in more on this topic, here’s some recommended reading:

• The lifesaving power of … paperwork? – a piece on the importance of death registration around the world, by Jeneen Interlandi at the New York Times.

• Advancing antivenom – a piece on the past, present and future of antivenom, by Mathias Kirk Bonde at Works in Progress.

• Snakebite data information portal – where in the world do snakes live? The WHO has a new data portal with lots of maps about this.

• The tale of snakebite’s fleeting spotlight—and why it encapsulates all that’s wrong with global health – by Robert Fortner in the BMJ.

• Challenges and prospects of snake antivenom supply in sub-Saharan Africa – by Abdulrazaq G. Habib et al., in PLOS Neglected Tropical Diseases.

• Snakebite – a summary of the Wellcome Trust’s research funding projects on developing new antivenom.

That’s all for now! I’m sorry my posts haven’t been regular – I started this newsletter hoping to publish posts once a week, but with all my other work I’ve had to choose between speed and quality, and I’m prioritising the latter.

Hopefully I’ll find ways to do both better in the future.

See you next time :)

– Saloni

Subscribe now

Hello again! I passed my PhD viva two weeks ago! I’m very happy about it and I guess I’m officially a “Dr.” now, which feels strange.

But I also now have a long backlog of posts I want to write, so I’m going to get back into it. If you like this, I hope you subscribe! And if you spot any errors, please let me know so I can fix them.

Subscribe now

The untold success of chickenpox vaccination

In most of the world, if you’re older than twenty, you probably remember having chickenpox as a child. It’s estimated that 85–90% of people have had it by the age of 15.

If you’re older, you might have also had shingles – a painful condition caused by the same virus getting reactivated later on.1 The virus lays dormant in the periphery of the spinal cord, and can reactivate months, years or decades later, resulting in rashes, an itchy burning sensation, and in some cases, long-term nerve pain.

But people tend to think of chickenpox as just a mild disease or a ‘rite of passage’ for children. Chickenpox vaccines are often considered not worth the cost, and some people worry that vaccination during childhood will be pointless, as it will simply delay disease until adulthood when the disease is more severe.

The last few decades have seen substantial evidence against these ideas, and I think they have a story that everyone should know.

Chickenpox vaccines were first developed in the early 1970s by a Japanese virologist named Michiaki Takahashi, and went through testing and further development over the next decade. They were first licensed in Germany and Sweden in 1984, and in the US in 1995.

They are highly effective. With two doses, they reduce the chances that a child would get chickenpox by 95%, or severe disease by 99%, over the next 10 years.

In 1995, the same year they approved the vaccine, the United States Centers for Disease Control and Prevention (CDC) made the then-controversial decision to offer the vaccines routinely for all children at no cost. It was the first country to do so, and as a result has the longest experience with it.

Four years later, chickenpox vaccination became required for entry in childcare and school, with some medical and religious exemptions. [Since 2005, chickenpox vaccines don’t have to be given as an additional injection, because some formulations include it within the MMR vaccine, as an “MMRV vaccine”.] From 2007 onwards, a second dose was recommended and routinely given out as well.

Before vaccination, chickenpox was common, usually painful, sometimes severe, and a heavy burden on parents taking care of children at home.

The worst outcomes were rare, but visible because the disease was so widespread. Around 5 in 100,000 infants and 1 in 100,000 children who caught the disease then died because of it – it could cause secondary infections and pneumonia, neurological complications and haemorrhage.

Children who died tended to have immune conditions, or had leukaemia and were on track to be cured by highly-effective chemotherapy. Some weren’t recommended to take the live vaccine and had to pause treatment while they took it, or simply had to try avoiding the virus from other children and adults, which I’d imagine is pretty hard when chickenpox is so common.

Subscribe now

In the early 1990s, before the vaccines were introduced, the virus caused around 4 million cases of disease, 12,000 hospitalisations and 100 deaths each year in the United States. Children were most affected, making up >90% of cases, >60% of hospitalisations and around 40% of deaths.

I’ve used past tense, because here’s what happened after that.

25 years after routine vaccination in the United States, rates of chickenpox had declined by over 97%. Hospitalisation rates declined by 98% in children and 84% in adults.

Deaths, which were rare, declined by 93% among people aged under 50. In under-20s, they shrunk to zero. Chickenpox was eliminated as a cause of death in this age group.

And despite the worries, they didn’t worsen the burden in older age groups.2 Instead, adults also benefitted from the reduced spread of disease among children, and were less likely to develop the disease, be hospitalised or die from it.

In the 25 years since vaccines were routinely provided, it’s estimated that a total of more than 91 million cases of chickenpox, 238 000 hospitalizations, and almost 2000 deaths were prevented in the US, as a result of vaccination.

The decision was also cost-saving: this conservative estimate puts it at a net societal savings of 23.4 billion US dollars so far, with savings coming from the reduction in sickness, fewer work-days lost to childcare, and avoiding expensive treatment including antibiotics and antivirals for other infections that could arise as complications of the disease. That is, chickenpox vaccination also reduces the need for antibiotics, and thus helps against antibiotic resistance.

With a simple calculation, each case of chickenpox averted saves around 260 US dollars on average.3 This includes the costs of producing the vaccines, administering them, the potential of side effects, and the time and cost taken to travel to get vaccinated.

In the years to come, the total benefits of the decision will grow, as more children avoid the earlier levels of spread, and as the herd immunity that has built up continues to benefit future generations.

Subscribe now

America’s success with chickenpox vaccines isn’t unique. Many other countries – including Japan, Canada, Germany, Brazil, Australia, Costa Rica, Taiwan, Uruguay and Qatar, among others – have also routinely offered vaccines to all children at no cost for years and seen big successes.

Rather than worsening disease in older people, they’ve seen the opposite – huge declines in rates of hospitalisation and deaths from chickenpox across age groups.

But many countries around the world do not yet offer chickenpox vaccines routinely4. Here they are on this map, in yellow.

As I mentioned earlier, chickenpox is just the first part of the problem.

Around 30% of people develop shingles in their lifetime, from the same virus that had given them chickenpox years before. Shingles causes an itchy burning sensation, rashes, and can lead to complications with vision.

You can see the per-year rates of developing shingles in the chart below. The rates are low early in life, but start to grow quickly around the age of 45.

Shingles can also lead to long-term nerve pain, called “post-herpetic neuralgia”. People experience this differently, but it can feel like either a continuous burning sensation, or recurring electric shock-like pains, or pain in response to touch.

How often does that happen? This systematic review estimates it happens in around 5–30% of people in their lifetimes, and that 30–50% of them continue to experience it for at least a year. Here’s a chart that shows this with data from the Netherlands.

If you’re like me and weren’t vaccinated for chickenpox as a child and caught the disease, you can hopefully get a shingles vaccine, which reduces the risk of shingles by >90%. In many countries, it’s available after the age of around 50. An older version of the shingles vaccine contains the same formulation as the chickenpox vaccine, but at a higher dose; while a newer one is a recombinant vaccine.

It’s maybe not surprising that America’s chickenpox vaccination programme reduced shingles as well, since it’s caused by the same virus reactivating. Before it, rates of shingles had been growing in the US for decades. After it, the growth began to slow down, plateau and eventually fall across age groups starting with the youngest.5

In total, the routine vaccination programme was highly successful in many ways. It massively reduced the rates of chickenpox, and hospitalisations and deaths from the disease. It resulted in a net societal savings, and likely helped against the development of antibiotic resistance to other diseases often affect people with chickenpox. And it slowed down the rates of shingles, which had been growing.

(The cost-effectiveness estimate I described above doesn’t include the effect on shingles, which means it underestimates the benefits.)

At the start of this post, I didn’t even describe the typical symptoms of chickenpox. I assume most of you have memories of what it was like – at least I do. I remember having it while visiting my family in India on vacation, which I just had to spend at home with incredibly itchy rashes across my body and face, while I kept vomiting for a week.

Younger generations won’t remember anything like this, I hope. But what it will take to get there is our generation treating it like a problem that can be avoided.

Subscribe now

Share

I hope you enjoyed reading this! I’ve proofread and double-checked numbers, but if I’ve made any errors, please let me know.

Next week I’ll be back with another post about one of the most underrated topics in global health – missing data. That’s based on a talk I gave at EAGxCambridge on Saturday (slides here). Eventually, I’ll also get around to catching you up on other stuff you missed since December.

See you next time!

– Dr Saloni (haha)

Under the hood

Here’s something new I’m trying – explaining the data quality and assumptions underlying these studies and sharing my reasoning on this topic.

I hope this helps other people learn how to read studies and evaluate their quality, or at least know what I look out for. And I also hope it helps me hold myself to a high-standard, and that it allows you to see if I’ve made any mistakes.

Let’s talk first about where the numbers come from.

I’ve cited many studies in this post. The main focus is the decline in cases, hospitalisations and deaths from chickenpox in the United States.

For cases, the data comes from 4 states which had surveillance before 1995, when routine vaccination began. Although these states may not be representative of the prevalence across the country, their data is useful because other states lacked consistent surveillance beforehand and hence don’t allow for an easy comparison of before and after. Other states have consistent data from 1995 onwards, as part of the ‘active surveillance’ that began with the routine vaccination programme.

For hospitalisations, the data comes from the ‘National Inpatient Sample’, a very large nationwide dataset that includes about 20% of all discharges from hospitals in the US. It’s nationally representative in terms of a range of hospital characteristics, such as the number of beds, the urban/rural location, type of ownership and several other factors.

For deaths, the data comes from the ‘National Center for Health Statistics’, which includes deaths across the entire population in the country, using data from death certificates. The data includes deaths categorised by their ‘underlying cause of death’ and other contributing causes of death listed on the death certificate; I’ve included charts showing these separately and together in the post.

The cost-effectiveness analysis looked at a range of scenarios, to derive conservative estimates of the net costs of the vaccination programme even if the estimates used to derive it were mistaken. This included varying the rates of incidence, hospitalisation and mortality, the costs of each case, the costs of vaccine administration, and the costs of side effects; each by +/-20%. It also looks at a worst-case scenario where each one is 20% worse – even in this worst-case scenario, the savings are estimated to be 9.2 billion. The range of estimates are shown in the paper. (Note that the authors exclude the effect of vaccines on shingles, which is one reason why they are an underestimate of the savings.)

Aside from those studies: Data on efficacy comes from RCTs that compare chickenpox vaccines to MMR vaccines (which are an ‘active placebo’ for chickenpox vaccines). Data on lifetime rates come from seroprevalence studies for chickenpox (since case-surveillance only happens in a few countries), and large longitudinal studies for shingles. I’ve also used multi-national systematic reviews for other estimates when I’ve found them, to avoid picking data from individual countries, in case there’s a lot of variation and I happen to pick an odd estimate.

Second, how much of the decline should we attribute to the vaccines?

Actually estimating this is tricky, but there are things we can say.

The efficacy rates of these vaccines are very high – around 95% against chickenpox and 99% against severe disease over the next 10 years – and the vaccination coverage in the US was also high – around 90% among infants from 2006 onwards. Together, this is a simple benchmark that implies the impact could eventually result in the entire decline. (The speed of this impact would depend on how quickly those rates are achieved across age groups who account for the hospitalisations and deaths.)

But there are other things to consider as well:

1. The efficacy figures don’t account for the effects of herd immunity – higher vaccination rates also reduce the transmission of the disease, which means the reduction in severe disease could be greater than a simple calculation of efficacy multiplied by vaccination rate.

2. The median estimate by the CDC researchers assumes that the rates in 1990–95 would have remained at that level, without the vaccines. But the rates of shingles, which track chickenpox because they’re caused by it, had been rising across age groups in the US before the chickenpox vaccination program.

   • Unfortunately, before routine vaccination for chickenpox, there hadn’t been national surveillance for chickenpox cases in the US, so data on cases before that comes from only a few states, shown in the first chart.

   • It’s also worth noting that, in other countries that don’t have routine vaccination but do have data, including the UK, cases of chickenpox and shingles have continued to rise over decades, especially among older age groups. This heavily implies that you wouldn’t see such a large decline due to other ongoing factors.

   • In the US, shingles cases have declined in the youngest, and risen but slowed down in older age groups who aren’t eligible for shingles vaccines.

   • Overall, there’s a lack of global data on this topic, and data is most-often missing from countries that don’t have routine vaccination programs.

   • An alternative is to use an epidemiological model – which could account for the ‘herd immunity’ generated by vaccines. But these can depend heavily on estimates of the transmission reduction, the duration of immunity, levels of contact between people, and other factors which are often hard to estimate precisely. You might be interested in these other studies that make these estimates for some other countries.

So, while other hypothetical factors could aid the decline too, I think the vaccines can account for most or all of the decline.

Here’s a little bonus, in case you’re interested.

In the 1960s, a scientist named Robert Edgar Hope-Simpson discovered that chickenpox and shingles were caused by the same virus. How did he figure that out? You can read the whole explanation in a lecture he gave in 1965 here.

It’s quite readable – and as someone that spends time thinking about how people can identify cause and effect without RCTs, or were able to even before they existed, I enjoyed it.

Update 20/3/2023: I added a couple of lines to clarify that the cost-effectiveness analysis didn’t include the effects of chickenpox vaccination on reducing shingles, which means it underestimates the benefits.

I corrected a description of the shingles vaccine, which mistakenly said that it had a lower dose than the chickenpox vaccine. I also added links to learn more about the shingles vaccines.