ipeval: Interventional Prediction Evaluation

Abstract

Provides methods to evaluate predictive performance of models that estimate risks under hypothetical intervention scenarios (interventional/causal/counterfactual predictions) with observational data subject to treatment-outcome confounding. Inverse probability of treatment weighting (IPTW) is used to construct a pseudopopulation in which all individuals receive a specified intervention, enabling assessment of agreement between predicted risks under the intervention and observed outcomes in the pseudo-population corresponding to that intervention. Package supports binary and time-to-event outcomes under binary interventions made at a single time point. Performance measures supported are AUC (Area Under the receiving operating characteristic Curve), Brier score, observed-expected ratio, and calibration plots. Methods implemented in this package are based on work by Keogh and Van Geloven (2024) <doi:10.1097/EDE.0000000000001713>.

Authors 3

1. Jasper W. A. van Egeraat Aachen

   Leiden University Medical Center

   Affiliation as printed

   Leiden University Medical Center [ fnd ]

2. Nan van Geloven Aachen

   Leiden University Medical Center

   Affiliation as printed

   Leiden University Medical Center [ fnd ]

3. Ruth H. Keogh Aachen

   Leiden University Medical Center

   Affiliation as printed

   Leiden University Medical Center [ fnd ]