Use of prognostic factors in analysis of historical control studies

This paper considers the use of regression methods for the interpretation of comparative results between treatments in historical control studies. A logistic regression model is used for response rate studies and Cox's regression model for studies of disease-free or overall survival. For both models, stepwise regression procedures may be utilized to test for treatment differences adjusting for differences in patient characteristics between the treatment groups. Examples are given from two clinical studies: the first example concerned a study which compared disease-free and overall survival experience between patients receiving surgery plus radiation therapy plus FAC-BCG (5-fluorouracil, Adriamycin, and cyclophosphamide-bacillus Calmette-Guérin) as adjuvant treatment compared with surgery plus radiation therapy alone for patients with stage II or III breast cancer; the second example concerned a study of acute leukemia in which differences among complete remission rats by institution were interpreted using a logistic regression equation to adjust for prognostic factors.