Prediction models: systematic reviews and clustered study data

Abstract

In the first part, I validated and updated two search filters for finding clinical prediction studies in Medline. I considered predictor finding and model development studies. The latter were accurately traced by the existing search filters. However, predictor finding studies were hardly found by the existing and updated search filters. Further, I conducted a systematic review. I investigated the reporting and methods of prediction studies, focusing on aims, designs, participant selection, outcomes, predictors, statistical power, statistical methods, and predictive performance measures.I retrieved 71 papers for full text review: 51 were predictor finding studies, 14 were prediction model development studies, 3 addressed an external validation of a previously developed model, and 3 reported on a model’s impact on participant outcome. Study design was unclear in 15% of studies and a prospective cohort was used in most studies (60%). Description of the participants and definitions of predictor and outcome was generally good. The number of events per predictor as a measure of statistical power could not be determined in 67%; of the remainder, 53% had fewer than the commonly recommended value of 10 events per predictor. Methods for a priori selection of candidate predictors were described in most studies (68%). A substantial number of studies relied on p-values of