DICE 2: A further investigation of the effects of chance in life, death and subgroup analyses

In an investigation into how chance might influence the distribution of dea ths in a randomised trial and the time of those deaths, and to highlight th e possible dangers of subgroup analyses,100 randomised controlled trials we re simulated and 50 subgroup pairs were simulated for some of these trials. Each of 580 control patients from a colorectal cancer trial was randomly c oded to simulate allocation to treatment or control, the main outcome measu re being time to death. Not surprisingly, most of the 100 trials gave non-s ignificant results. Four were conventionally significant with logrank 2p-va lues of less than 0.05, The most extreme result was associated with a logra nk 2p-value of 0.003, showing an absolute reduction in four-year mortality of 40% (SD 15) for patients allocated to treatment. One of the simulated pr ognostic factors for this trial (subgroup 13) showed that mortality for one type of patient was non-significantly slightly increased by treatment, whe reas treatment reduced four-year mortality by 64% (SD 16) among the other p atients in the trial (2p=0.00006). Similar, extreme results were found for a trial of borderline statistical significance overall. Chance can influenc e the overall results of any randomised controlled trial, regardless of how well it is conducted, and can play an even more powerful role in the resul ts of subgroup analyses. This should be borne in mind both by trialists whe n reporting their results and by readers and reviewers of those reports.