Assessing possible late treatment effects in stopping a clinical trial early: a case study. Diabetic Retinopathy...

Suppose a fixed-sample trial in a disease with a long response time shows a statistically significant benefit of the experimental treatment before patients have completed the planned follow-up period. The question may then arise--and did arise in the Diabetic Retinopathy Study (DRS)--whether the observed early benefit of treatment may be offset at some time in the future by the subsequent development of harmful treatment effects. If this question raises serious concerns, then the investigators are faced with a dilemma. If the trial is stopped because of the observed early treatment benefit and the treatment is administered to the untreated control group as well as to patients outside the study, and if the treatment is later found to have deleterious effects, then it may ultimately do more harm than good to patients. Moreover, the fact that the treatment is harmful may never become known. If, on the other hand, the trial is not stopped and the treatment proves to have no deleterious effects, then the control group and patients outside the study would be harmed because the treatment was withheld. We show how, in the DRS, this very problem was formulated and resolved. First a severe, delayed harmful treatment effect was postulated. Projections based on this postulation showed that the early gains were so great that they were unlikely to be offset--ever. Based in part on these projections, the following decisions were made: (a) the study protocol would be changed so as to allow treatment of the untreated control group, and (b) patients would continue to be followed in order to make possible the detection of late, harmful treatment effects, should they develop.