Importance of prognostic factors in the analysis of data from clinical trials

Abstract

Predict ing the future is no easy matter . Horserace bet t ing, weather forecasting, predic t ion of future trends in the national economy or the size of the p o p u l a t i o n i n none of these fields of endeavor can the pract i t ioners be said to meet wi th a great deal of success. Yet medical p r o g n o s i s t h e predic t ion of the future course of disease in a part icular pa t i en t r ema ins an impor tan t and not wholly unsuccessful part of the physic ian ' s art. The principal aims of medical prognosis are outs ide the scope of this paper . They are related to the managemen t of the pat ient ' s illness, including the p rov i s ion of advice to the pa t ien t and his relatives, the choice of therapeut ic measures and ancillary care, and the possible adminis t ra t ion of prophylact ic measures against fur ther adverse events. I am concerned here solely with the use of prognost ic in format ion in the analysis of clinical trial data, a l though I shall touch later on the related topic of the use of prognost ic in format ion in the design of a trial. A prognost ic variable or prognost ic factor is an i tem of in format ion about a part icular pa t ien t that helps to predict some aspect of the outcome of disease. Prognosis can, of course, be a t tempted at var ious stages dur ing the course of disease. The later it is den e (whether the pa t ien t is approaching cure or decline), the more successful it is likely to be. In clinical trials we are part icularly in teres ted in prognosis at the t ime of t rea tment ass ignment . The variables used are often called base l ine variables. (The term covar ia tes is often used, part icularly by statisticians.) Prognost ic variables measured dur ing the course of disease may themselves be affected by the choice of t rea tment and are therefore of l imited use except as response variables in their own right.