Pitfalls of subgroup analysis in randomized multicentre intervention trials

Results of randomized multicentre trials in patients with acute myocardial infarction form the basis of day-to-day therapy in intensive coronary care units all over the world. The results of subgroup analysis from these trials are applied by physicians to individual patients and hence, it is important to understand the limitations of subgroup analysis as performed in many of the trials. Performing multiple analyses increases the chance of making a type I error. Small sizes of subgroups increase the chance of a type II error. The use of univariate methods may show that patients with inferior wall infarction may not benefit from thrombolysis. A closer look may show that the number of patients with diabetes, hypertension, both or neither in the two subgroups are different, and the difference in response to thrombolysis may have been a reflection of this inequality rather than the site of infarction. TD minimize this fallacy we suggest an alternative strategy for subgroup analyses using prognostic scores which could assess cumulative risk of death resulting from the interactions of various risk factors for each patient. Comparison of subgroups with identical prognostic scores can rescue subgroup analysis from confounding bias ahd make it more meaningful. This could cut down the size and administrative costs incurred in conducting such multicentre intervention trials and also facilitate extrapolation of the results to individual patients.