A multinational pharmacoepidemiology atlas of prescription medication associations with all-cause mortality across six population cohorts

Observational studies sometimes report that individual medications are associated with lower mortality, but such signals are rarely compared across drugs and health systems. We screened prescription medications for associations with all-cause mortality using time-dependent Cox models in six cohorts from the UK, US, Finland, Hong Kong and Japan (8.93 million individuals aged 40 years or older; 1.49 billion prescriptions). Most medications were associated with higher mortality, reflecting confounding by indication. A few medications showed inverse associations with mortality, notably the varicella-zoster virus (VZV) vaccine, the antimalarials atovaquone-proguanil (Malarone) and mefloquine, and the hormone replacement therapy estradiol. These associations were only modestly attenuated by adjustment for comorbidity, lifestyle and socio-economic factors, and persisted within strata defined by pre-existing disease. While adjustment for self-reported health abolished the Malarone association, probably reflecting a healthy traveler profile, the VZV association was maintained. In FinnGen, deaths from external causes, which vaccination cannot prevent, were lower among VZV vaccine users, but much less so than all-cause mortality. In a target trial emulation in All of Us, the VZV association was maintained against matched unvaccinated individuals (HR 0.61, 95% CI 0.53-0.70) and pneumococcal vaccine recipients (HR 0.64, 0.55-0.74), but was attenuated against recipients of tetanus-diphtheria vaccines, a group that also has low mortality (HR 0.85, 0.70-1.03). Healthy-user selection contributes to these signals but does not appear to fully explain the VZV vaccine association, which warrants confirmation in randomized trials. This multinational atlas provides a resource for prioritizing drug-repurposing candidates in ageing research.