The Trial That Stopped Early
A randomised controlled trial of a new drug for a rare aggressive cancer was designed to enrol 400 patients over four years, with overall survival as the primary endpoint. An independent data monitoring committee reviewed results after 180 patients had been enrolled and followed for a median of 14 months. The interim analysis showed 62 deaths in the placebo arm and 41 in the treatment arm. The pre-specified stopping boundary for benefit was a p-value of 0.005 at this interim look; the observed p-value was 0.004. Median survival was 11.2 months on placebo and 16.8 months on treatment. Grade 3 or higher adverse events occurred in 38% of treated patients versus 21% on placebo. The trial was stopped early and the results were published eight months later.
1. Explain why stopping early for benefit can make the reported effect size misleading, and what the publication should convey.
Read the worked answer
When a trial stops at the first interim look that crosses a boundary, the estimate of benefit is systematically inflated. The reason is that the trial stops precisely when the observed difference is unusually large; if the true effect were smaller, the trial would have continued and the extreme early result would have regressed toward the truth. The published median survival difference of 5.6 months is therefore likely to overstate the real benefit.
The inflation is compounded by the small number of events: 103 deaths in total means the survival estimates at later time points rest on few patients, and the confidence interval around the hazard ratio will be wide. The p-value of 0.004 crossing a 0.005 boundary is a legitimate statistical result but a fragile one - a slightly different boundary or timing could have produced the opposite decision.
A responsible publication should report the stopping rule and the number of interim analyses performed, present the effect estimate with its confidence interval rather than the point estimate alone, and explicitly caution that the effect size is likely optimistic. It should also report the adverse-event imbalance prominently, because the toxicity burden is part of the risk-benefit calculation and is not inflated by early stopping in the same way.
2. A clinician reads the paper and considers offering the drug to a patient who is frailer than those enrolled. What reasoning should guide the decision?
Read the worked answer
The clinician should recognise three gaps between the trial population and the patient.
First, generalisability. The trial enrolled patients fit enough to consent and tolerate treatment; a frailer patient may have been excluded. The 38% rate of grade 3+ adverse events in a relatively fit population is likely to be higher, not lower, in someone with less physiological reserve. The benefit of 5.6 months may shrink or vanish if the patient cannot complete treatment.
Second, the inflated estimate. The clinician should mentally discount the reported benefit, because the trial stopped early. A reasonable approach is to treat the published effect as an upper bound and ask whether the patient would still consider the trade-off worthwhile at a smaller benefit - say 2–3 months.
Third, what matters to the patient. A survival gain of months, weighed against a substantial chance of serious toxicity, is a values question, not a statistical one. The clinician should present the absolute numbers - the chance of serious adverse events, the likely range of survival benefit - and elicit the patient's priorities.
The defensible decision is shared and documented, not a reflex either to withhold the drug because the evidence is imperfect or to offer it because it was published.