Who Counts in a Clinical Trial?

AFor most of the twentieth century, the volunteer who supplied the evidence for a new medicine was young, male and healthy. The pattern was not accidental. A 1977 regulatory guideline in the United States advised that women of childbearing potential be kept out of early-phase drug studies, a rule framed as protection after several drugs were found to harm unborn children, and industry applied it far more broadly than its wording required. Reversal came in 1993, when legislation obliged trials funded from the public purse to enrol women and members of minority groups and to report their results group by group; commercially funded studies, which by then produced most of the applications reaching regulators, were left outside the statute. Three decades on, the enrolment figures have moved, though unevenly, and the argument has shifted ground. It is no longer about whether under-representation exists. It is about what, precisely, the numbers are evidence of, and about what changing them would buy.

BAt the centre of the debate sits a confusion between two different tasks. The first is enrolling a sample that mirrors the population which will eventually take the drug; the second is enrolling enough people within each group to say whether the drug behaves differently in one of them. Only the first is a matter of proportions. The second is a matter of statistical power, and the arithmetic is unforgiving: detecting a difference between two subgroups generally requires around four times as many participants as detecting an effect of the same size across the trial as a whole. A study that reproduced national demography exactly, participant for participant, would in most cases still be far too small to answer the subgroup question it appears to have been designed to answer. Proportional representation is thus a necessary condition for that answer, never a sufficient one — a distinction that survives poorly in summaries written for the public.

CRosalind Achterberg, a clinical pharmacologist at the Fenwick Institute, has assembled the fullest audit of the gap. Her group examined 512 trials registered between 2009 and 2019 in support of newly approved medicines, and found that participants over sixty-five made up 14 per cent of those enrolled, although the same age band accounted for 42 per cent of prescriptions once the drugs reached the market. For one class of anticoagulant her team went further, showing in later pharmacokinetic work that the compound left the blood roughly 30 per cent more slowly in patients past seventy-five; three of the drugs concerned had their recommended doses lowered after approval, as reports of bleeding accumulated. Achterberg is exact about the reach of the claim. "We have measured the mismatch for the classes we examined," she writes, "and nothing in our data licenses the assumption that it holds everywhere."

DTobias Marchetti, a trial statistician at the Braithwaite Centre for Trial Methods, accepts the 14 per cent and disputes the remedy drawn from it. Enrolment targets, he argues, are met where they are cheapest to meet. A site instructed to recruit more older participants will find them among the people already attending its own specialist clinics — mobile, urban, taking fewer other medicines — so the subgroup that arrives in the dataset is not the subgroup the target was written to represent. He also puts a price on the instrument: across sixty trials he examined, sponsors working under compulsory enrolment targets took a median of five months longer to finish recruiting, and a medicine delayed is not a neutral outcome for the patients waiting on it. Marchetti's position is not that diverse trials are unnecessary. It is that diversity has been turned into bookkeeping at the enrolment desk when it is a question of where trials are placed.

EEvidence about placement is beginning to accumulate. The Halloway programme, begun in 2021, moved trial visits out of teaching hospitals and into community pharmacies and general practices, and reimbursed both travel costs and wages lost to appointments. Over three years the share of enrolled participants drawn from the groups it targeted rose from 11 per cent to 27 per cent, while withdrawal before the final visit fell from 24 per cent to 15 per cent; the cost of each participant rose by roughly a third. The programme also surveyed those who declined to take part, and the reason offered most often was not distrust of researchers, as its designers had expected, but the sheer number of visits the protocol demanded. Distrust did appear on the list, ranked fourth. That finding has since been quoted approvingly by both camps, which suggests that at least one of them has read it carelessly.

FRegulation, meanwhile, has moved faster than the evidence. Since 2022, sponsors in the United States have been obliged to file a diversity action plan before beginning a pivotal trial, setting enrolment goals by age, sex and ethnicity and explaining how each will be met; the obligation may be waived where the disease itself falls overwhelmingly on a single group. What the statute penalises, however, is the failure to file a plan rather than the failure to reach the numbers inside it, and no application has yet been refused on these grounds. Both sides read their own design into the arrangement. To Achterberg it is a first instrument that will be tightened once the reporting shows how wide the gaps run; to Marchetti it is an admission that the regulator knows its targets cannot be reached by the means it has recommended. The asymmetry underneath is a familiar one. Counting who was enrolled is cheap. Showing that it mattered is not.