A screen of hundreds of medicines in 8.9 million people found the shingles vaccine linked to lower mortality
Researchers compared the prescription records of 8.93 million people in five countries with who died. A handful of medicines were linked to lower death rates, the shingles vaccine most robustly, but part of that is likely because healthier people get vaccinated. An observational preprint that generates hypotheses, not answers.
Could a medicine that already exists, prescribed for something else, help people live longer? The question comes up often in aging research, usually one drug and one country at a time. That makes it hard to tell whether a promising signal is real, or how it compares with all the medicines nobody looked at.
This study looked at hundreds of medicines at once. It is a preprint, so it has not yet been peer-reviewed, and it is an observational study: it can show which medicines were linked to lower death rates, not which ones caused them. The authors call their atlas "hypothesis-generating" and say that its results need randomized trials.
A trap called the healthy-user effect
Studies like this have a known weakness. People who get preventive treatments, such as vaccines or travel medicines, tend to be healthier, richer and more careful about their health than people who don't. They would live longer even if the treatment did nothing. Hormone replacement therapy is the best-known example: early observational studies suggested benefits that later randomized trials did not confirm. Much of this study is about testing how far that bias can explain what was found.
What they did
The team pooled records from the UK Biobank, All of Us and NHANES in the United States, FinnGen in Finland, a public hospital database in Hong Kong and an insurance database in Japan: 8.93 million people aged 40 or older and 1.49 billion prescriptions. The groups were very different, from working-age Japanese employees (0.4% of whom died during follow-up) to older Hong Kong hospital patients (29.5%).
For each medicine, people counted as "unexposed" until their first prescription and "exposed" from then on. Death rates in the two states were compared with a hazard ratio: 1 means no difference, 0.5 means half the death rate. The comparison was then adjusted step by step for age and sex, existing illnesses, and smoking, body weight and social status.
What they found
- Most medicines were linked to higher death rates. That is expected and is not evidence of harm: medicines are given to sick people, and the illness carries the risk. Statins, which lowered deaths in randomized trials, showed no benefit here, a reminder that this kind of screen can miss real effects.
- A few were linked to lower death rates. The shingles vaccine (against the varicella-zoster virus, which causes chickenpox and later shingles) stood out, with a pooled hazard ratio of 0.44. The diphtheria-tetanus-pertussis vaccine, two malaria tablets (atovaquone-proguanil and mefloquine) and the hormone estradiol (hazard ratio 0.77 in women) were also linked to lower mortality.
- Adjustment barely changed these links. After accounting for illness, smoking, weight and social status, the shingles vaccine figure moved only from 0.44 to 0.45.
- The malaria link looks like a healthy-traveller effect. People taking malaria tablets rated their own health much better than others, and adjusting for that removed the link. A typhoid vaccine, given to the same travellers but with no plausible effect on death in rich countries, was linked to a similarly low death rate.
- The shingles vaccine link survived most checks, but not all. In Finland, vaccinated people also died less often from accidents and other outside causes (a sign of healthy-user bias), but much less so than from all causes. In a target trial emulation in All of Us, vaccinated people had a 39% lower death rate than matched unvaccinated people (hazard ratio 0.61) and 36% lower than people who got the pneumococcal vaccine. Against people who got a tetanus-diphtheria booster, the group most similar to them, the difference shrank to 15% and was no longer clear-cut (0.85, with a range that included no effect).
What to keep in mind
- Associations, not effects. The authors stress that their numbers "should not be interpreted as causal effects", and that higher-mortality links are not evidence of harm either.
- Healthy-user bias is still there. Death rates separated soon after shingles vaccination, with the strongest link in the first 90 days, when a vaccine is unlikely to prevent deaths; the authors read this as a sign of who gets vaccinated. Compared with similar vaccine users, they estimate that a modest link, roughly 15–25% lower mortality, remains, which would fit a smaller true effect.
- Exposure was crude. Anyone who had ever received a prescription counted as a user from then on, whatever the dose or duration. That can flatter one-off treatments such as vaccines. Vaccines given in pharmacies, and medicines bought over the counter, may have been missed.
- One cohort for the key test. The target trial emulation used a single cohort with limited follow-up, and the analysis plan was not registered in advance.
- Not yet peer-reviewed.
- Not advice. Estradiol and the other medicines in this screen are prescription drugs with known side effects and risks; the estradiol link contrasts with the large randomized hormone trials, which the authors note tested different hormones in women who were mostly more than a decade past menopause.
Why it matters
Instead of one more promising drug from one database, this study offers a ranked list of signals from six countries, each with a note on how well it holds up against the usual biases. Most of what looks like protection here turns out to be about who takes a medicine. The shingles vaccine is the signal that held up best, and the authors argue it deserves to be tested in randomized trials for effects on survival.