You are a doctor in Ibadan. A patient with moderate heart failure sits across from you, and you reach for the prescribing guidelines on an ACE inhibitor whose pivotal trial enrolled patients from Stockholm, Rotterdam, and suburban Ohio. The label says 22% reduction in hospitalisation. You write the prescription. What the label does not say is how much of that 22% was generated by people who share your patient's genetic background, his diet, his history of undiagnosed kidney damage from a childhood illness the trial designers never thought to ask about. You prescribe with the confidence the data appears to offer. That confidence has a geography, and it ends somewhere around the Mediterranean.
The world that trials actually recruit from
Most large-scale clinical trials, particularly the pivotal Phase III studies that regulators use to approve drugs, draw the overwhelming majority of their participants from a small cluster of high-income countries. North America, Western Europe, Australia, Japan, and South Korea account for a disproportionate share of global trial enrollment relative to their share of global disease burden. Sub-Saharan Africa, South and Southeast Asia, and Latin America, where the majority of the world's people live and where many of the diseases being studied cause the most harm, contribute a fraction of the participants.
This isn't a conspiracy. It's a logistics problem compounded by economics, and it has been that way for decades. Running a trial requires reliable infrastructure: hospitals capable of GCP compliance, ethics review boards, cold-chain storage for investigational compounds, a patient population that can attend follow-up visits. All of those things cluster in wealthy cities. A pharmaceutical sponsor managing a 5,000-patient trial across 40 sites is not being negligent when it gravitates toward Boston, Hamburg, and Toronto. It's being practical. The problem is that practical and representative are not the same thing, and no one is penalised for the gap between them.
How biology and environment bend the data
Genetic variation across human populations is real and medically relevant. Variants in the CYP2C19 enzyme, which metabolises a wide range of drugs including the antiplatelet agent clopidogrel, differ substantially in prevalence between populations of East Asian and European ancestry. Studies have found that so-called poor metabolisers, who process the drug more slowly and may therefore derive less benefit from standard doses, are present at roughly two to three times the frequency in East Asian populations compared to populations of European descent. If the pivotal trial for a clopidogrel-class drug enrolled predominantly European patients, the standard dose that emerged from that trial may be genuinely suboptimal for a large portion of the patients who later receive it. That is not a rounding error.
The distortions go beyond genetics. Consider what epidemiologists call effect modification by environment. A blood-pressure drug tested primarily on patients in cool-climate, low-altitude urban settings, eating Western diets, will produce an efficacy estimate that may not transfer cleanly to a patient in Addis Ababa at 2,400 metres elevation, eating a teff-based diet with very different sodium and potassium profiles, managing hypertension complicated by a history of malaria-related kidney damage. The drug might still work. But the magnitude of effect, the side-effect profile, and the optimal dose could all differ in ways the original trial was never designed to detect, in the same way that a weather model calibrated entirely on North Atlantic data will give you increasingly unreliable forecasts the further south you push it.
Back to Adekunle in Ibadan, our 58-year-old with moderate heart failure, started on the same ACE inhibitor at the same dose as Marcus in Stockholm. Marcus has the profile that largely generated the 22% hospitalisation figure. For Adekunle, with a different genetic background, a gut microbiome shaped by fermented foods, and a history of chronic kidney disease that went undiagnosed for years, the honest answer is that the trial provides limited evidence either way. His profile simply wasn't meaningfully represented in the data. He still gets the drug, at the same dose, with the same apparent confidence.
What people get wrong about this problem
The common assumption is that more diverse trials would simply mean adding participants from underrepresented regions as a kind of ethical gesture, a box to tick. That framing misses the scientific point entirely.
Diversity in enrollment isn't a courtesy extended to populations that feel left out. It is the mechanism by which you find out whether your finding is a universal biological truth or a local one. A trial that shows Drug X reduces mortality by 18% in a homogeneous population has discovered something real but bounded. A trial that shows the same effect across five genetically and environmentally distinct populations has discovered something far more robust. Conversely, a trial that shows the effect in four populations but not a fifth has discovered something even more valuable: a signal that there is a subgroup for whom the drug works differently, which should drive further investigation and not be quietly buried in a footnote.
The other thing people get wrong is assuming the problem is primarily about race as a social category. Race is a proxy, and a crude one. What actually matters are the underlying biological, environmental, and social variables that population labels imperfectly index. Two people who identify the same way ethnically may have radically different pharmacogenomic profiles. The goal is not to check demographic boxes but to enroll populations that vary along the dimensions that actually affect drug response.
So if you are a clinician applying trial results to a patient population that looks nothing like the trial cohort, ask yourself: why does the precision on that label feel so settled? Being sceptical of the estimate is not obstructionism. It is the correct scientific posture.
The structural weight holding the problem in place
Several forces make this genuinely hard to fix. Regulatory agencies in America and Europe are the gatekeepers that matter most commercially, so sponsors optimise their trial designs for those agencies' requirements and patient pools. An agency that approves a drug based on European data is not demanding global representativeness, because its mandate is its own jurisdiction. The result is a set of incentives that point consistently toward the same trial sites, the same patient profiles, and the same geographic footprint, trial after trial, decade after decade.
There are counter-pressures. Some regulators have begun issuing guidance encouraging broader enrollment. Academic research networks in Africa and Asia have grown substantially in capacity and GCP compliance. The growth of decentralised trial infrastructure, including remote monitoring and community-based recruitment, opens possibilities that didn't exist when the current norms calcified.
The burden of proof still falls on the patients whose populations weren't studied. A drug reaches the market, gets prescribed globally, and only years later, through pharmacovigilance data or a targeted sub-study, do the gaps become visible. By then, the standard of care is set, the label is printed, and the confidence interval nobody questioned has quietly hardened into clinical fact.
The medicine we have is genuinely good, often remarkable. It was also built on a foundation of evidence that describes some of humanity better than others, and the people it describes least well are, by almost every measure, the ones who need reliable medicine most. That asymmetry is not a law of nature. It is a design choice, made slowly, by accumulation, and the only interesting question now is who is willing to pay the extra cost of unmaking it.