SEQUENTIAL METHODS IN CLINICAL TRIALS

Journal/Book: Reprinted from AMERICAN JOURNAL OF PUBLIC HEALTH Vol. 48 No. 10 October 1958. 1958;

Abstract: Peter Armitage Ph.D. Copyright by the American Public Health Association Inc. 1790 Broadway New York N. Y. Sequential methods are used frequently in clinical trials and are applicable in other problems o f interest to health workers. This paper discusses how to control more rigorously the risks o f error in such methods. THE NUMBER of published reports of clinical trials involving sequential analysis could be counted an the fingers of both hands. Yet if we define a sequential experiment as one in which the decision whether or not to stop at any stage depends an the observations so far made there is little doubt that many published trials were in fact carried out sequentially even though this was not admitted in the final report. It is worth considering why this should be so. The main reason is I believe an ethical one. The point has frequently been made that many interesting clinical trials could never be performed as the physicians in charge would be convinced that one of the proposed treatments was better than another and would be unwilling to randomize patients between these two treatments because to do so would be to withhold what they believed to be a superior treatment from half the patients under their care. This difficulty is well known to anyone working in this field. But there is a closely related difficulty which is not so frequently discussed. If during the course of a trial which was started in good faith an appreciable difference emerges between the effects of different treatments sufficient to convince the physicians in Charge that the difference is real then they will be faced with the same ethical problem. Many investigators must therefore have examined the results col1ected at various stages of the trial and some may well have decided to stop because the differences in response were unlikely to be due to random fluctuations. We have here a problem very similar to that of optional stopping in parapsychological experiments. ... ___MH

Zurück | Weiter