Interpretation of clinical trials

Publisher Website
Google Scholar
Abstract
THE EDITOR—SIR, "Retrospective" is a dirty word in the terminology of clinical trials, and rightly so. Historical controls may turn out to be deceptive, as in the case of the first Interferon trial (Strander 1982). Nevertheless, we would like to suggest that a particular type of retrospective trial may be admissible. Consider the case of the fully randomised trial, of the value of hyperbaric oxygen in the radiotherapy of carcinoma cervix, stage III (Watson et al, 1978) which showed a 12% improvement in survival (p = 0.04). Analysis revealed that within the 292 patients a subgroup existed of 38 patients who were anaemic and received blood transfusion before treatment, and in which radiotherapy in hyperbaric oxygen was vastly superior (p = 0.0003) to radiotherapy in air (Dische et al, 1983). Clearly in a case like this, it would be unwarranted to suggest that anaemia and transfusion caused treatment in hyperbaric oxygen to be so effective, but, one may ask, is such anaemia associated with a beneficial response to treatment in hyperbaric oxygen? There is a danger in picking out subgroups retrospectively. If the clinical material was subdivided in scores of different ways, it might be expected that one of these subgroups was, by chance, associated with success or failure of any treatment. But suppose we only examine one subgroup: the anaemic patients. Originally, 292 patients were randomised into two groups; 132 were allocated to hyperbaric oxygen and 160 to air. With groups as large as that, we would be unfortunate if the anaemic patients were not randomly allocated (and the distribution turned out to be 15 women to hyperbaric oxygen and 23 to air). If we find that anaemia and transfusion is associated with success of treatment in hyperbaric oxygen with a very low p value, then are we justified in rejecting the idea that this association is not due to chance, just because it was discovered by retrospective analysis of a trial designed for another purpose? Note that the usual criticisms of retro- spective trials, such as, that the course of a particular disease may have altered in the course of time, do not apply in the case we have outlined above.

This publication has 4 references indexed in Scilit: