The use of prognostic factors in clinical trials

Numerous examples exist of prognostic factors that have been conclusively established and that can segregate the target patient population into subgroups with vastly different outcomes. At the same time, we can expect, in the majority of trials being conducted, that if a new treatment turns out to be successful, it generally results in only a modest improvement in patient outcome. This has contributed to a major emphasis on the identification and use of prognostic factors in clinical trials. Besides the inherent descriptive information for predicting patient outcome, prognostic factors are increasingly used in trial design and analysis with the hope of reducing or correcting bias that could otherwise occur from simple patient allocation schemes and unadjusted comparisons of treatment groups, and tailoring therapy so that the best treatments are selected in patient subgroups. Although the above objectives are reasonable, numerous misunderstandings and misuses of prognostic factors regularly occur. These involve such issues as the inappropriate application of clinical/statistical criteria for identification and confirmation of such factors, the sometimes undue emphasis on selection of best subdivisions of prognostic factors, the overreliance on prognostic factors and statistical models by proponents of historical control studies, the exaggerated importance of prognostic factors in special randomization schemes, etc. However, if a thorough understanding of such issues is combined with a careful approach to the use of prognostic factors, significant benefit to the design, conduct, and ultimate scientific information from clinical trials will result.