Statistical models usually have one of three different purposes, parameter estimation, prediction or model fit. These purposes are often misunderstood; the model may, for example, be developed as a prediction model, but the results presented in terms of parameter estimates such as relative risk, odds ratio or hazard ratio.
Bouwmeester et al. (1) reviewed 71 prediction studies published in the six highest impact journals during 2008. The study design was unclear in 15% of the papers. Continuous predictors were dichotomized in 32% of the studies. The number of events per predictor could not be determined in 67% of the studies. Of the rest, 53% had fewer than ten events per predictor. Many studies, 29%, relied on the p-value for selecting predictors in multivariable analyses.
The authors conclude that a majority of prediction studies in high impact journals do not follow methodological recommendations, which limits their reliability and applicability.
1. Bouwmeester W, Zuithoff NPA, Mallett S, Geerlings MI, Vergouwe Y, Steyerberg EW, Altman DG, Moons KGM. Reporting and Methods in Clinical Prediction Research: A Systematic Review. PLOS Med 2012;9:e1001221.