Ante- and Post-Hoc Explanations for Prediction Models of Cisplatin-Induced Acute Kidney Injury: A Comparative Study.

Whereas cisplatin is considered as an effective anticancer drug, it can cause cisplatin-induced acute kidney injury (Cis-AKI) as a side-effect. We trained a prediction model of Cis-AKI to predict the incidence of Cis-AKI using Generalized Additive Models (GAMs) and GAMs plus Interactions (GA2Ms), which can provide us with ante-hoc (or model-based) explanations for their decisions. Furthermore, we trained XGBoost and used SHapley Additive exPlanations (SHAP) in order to obtain post-hoc explanations. We then compared these ante-hoc and post-hoc explanation methods in terms of consistency with current medical insights on Cis-AKI.