Prognostic models based on literature and individual patient data in logistic regression analysis

Prognostic models can be developed with multiple regression analysis of a d ata set containing individual patient data. Often this data set is relative ly small, while previously published studies present results for larger num bers of patients. We describe a method to combine univariable regression re sults from the medical literature with univariable and multivariable result s from the data set containing individual patient data. This 'adaptation me thod' exploits the generally strong correlation between univariable and mul tivariable regression coefficients. The method is illustrated with several logistic regression models to predict 30-day mortality in patients with acu te myocardial infarction. The regression coefficients showed considerably l ess variability when estimated with the adaptation method, compared to stan dard maximum likelihood estimates. Also, model performance, as distinguishe d in calibration and discrimination, improved clearly when compared to mode ls including shrunk or penalized estimates. We conclude that prognostic mod els may benefit substantially from explicit incorporation of literature dat a. Copyright (C) 2000 John Wiley & Sons, Ltd.