Pitfalls in prediction modeling research

European Journal of Endocrinology, vol. 194, pp. E1–E4

Abstract

This paper discusses 11 pitfalls in the design, analysis, and reporting of medical prediction modeling research. These concern pitfalls related to timing and quality of measurements (ie, incorporating future predictors, differences in the precision of measurements between development and implementation, outcome misclassification, and predictors with missing values), pitfalls related to modeling (ie, univariate pre-selection of potential predictors, overfitting the data, and simplifying the model too much), pitfalls related to reporting and model interpretation (ie, being unclear about the prediction time-horizon, ignoring treatments after time-zero, and conflating prediction and causal research), and pitfalls related to the focus on something new (ie, developing a new model).

Authors 2

1. Rolf H. H. Groenwold corresponding Aachen Department of Biomedical Data Sciences Department of Clinical Epidemiology

   Leiden University Medical Center

   Affiliation as printed

   Department of Biomedical Data Sciences, Leiden University Medical Center , 2333 ZA Leiden ,

   Department of Clinical Epidemiology, Leiden University Medical Center , 2333 ZA Leiden ,

2. Olaf M. Dekkers Aachen Department of Clinical Epidemiology Department of Endocrinology, Leiden University Medical Center

   Aarhus University · Leiden University Medical Center · Aarhus University Hospital

   Affiliation as printed

   Department of Clinical Epidemiology, Aarhus University and Aarhus University Hospital , 8200 Aarhus ,

   Department of Clinical Epidemiology, Leiden University Medical Center , 2333 ZA Leiden ,

   Department of Endocrinology, Leiden University Medical Center , 2333 ZA Leiden ,