A case for Bayesianism in clinical trials

Cumulative meta-analysis of clinical trials builds evidence for exemplary medical care

Cumulative meta-analysis of clinical trials (a Bayesian interpretation for accumulating evidence) will profoundly affect medical care by summarizing evidence in the assessment of technology innovations. Application of the technique to the randomized control trials (RCTs) of streptokinase treatment of acute myocardial infarction, reduction of peri-operative mortality by antibiotic prophylaxis, and prevention of death from bleeding peptic ulcers has revealed efficacy years before it was suspected by any other means. Arrangement of the trials according to event rate in the controls, effect sizes, quality of the trials or according to covariables of interest has supplied unique information. If carried out prospectively the technique supplies invaluable information regarding indications for another trial, the number of patients necessary to determine the validity of past trends, and the type of patients who might be benefitted. Careful examination in a cumulative manner of the prior trials can reduce the need for future large trials.

Stopping a clinical trial early: frequentist and Bayesian approaches applied to a CALGB trial in non-small-cell lung cancer

In May 1984, the Cancer and Leukemia Group B (CALGB) opened a phase III clinical trial for patients with stage III non-small-cell lung cancer (NSCLC). The experimental design entailed randomization of 240 patients equally to one of two treatments: radiotherapy alone or chemotherapy followed by radiotherapy. The original design was a fixed sample size design with the intent to analyse the results after 190 deaths. Shortly after the trial began, it was decided to apply group sequential concepts by using a truncated O'Brien-Fleming stopping rule, implemented via a Lan-DeMets alpha-spending function. A study monitoring committee was established to review the analyses as they were produced. The study was stopped at the fifth interim analysis in May 1987 after 155 eligible patients had been entered. This paper reviews the statistical and other considerations leading to this decision and presents later follow-up information on these patients. Some Bayesian alternatives to the standard frequentist approaches are also explored and it is demonstrated how these alternatives provide a natural way to address many of the issues raised in monitoring clinical trials.