Explainable AI for survival analysis: a median-SHAP approach

TL;DR

This work critiques conventional Shapley-value explanations for survival analysis, where using a mean anchor and expectation can mislead due to skewed survival times. It introduces median-SHAP, which uses the median as the central summary statistic and the median individual as the anchor, with conditional reference distributions to stay on the data manifold. The method yields more interpretable, robust explanations for predicting median survival times and is validated on the Worcester Heart Attack Study and breast cancer survival data, showing closer alignment to classification-based explanations than traditional SHAP. The approach advances clinically meaningful, person-centric explanations for time-to-event predictions, maintaining additivity and robustness to outliers.

Abstract

With the adoption of machine learning into routine clinical practice comes the need for Explainable AI methods tailored to medical applications. Shapley values have sparked wide interest for locally explaining models. Here, we demonstrate their interpretation strongly depends on both the summary statistic and the estimator for it, which in turn define what we identify as an 'anchor point'. We show that the convention of using a mean anchor point may generate misleading interpretations for survival analysis and introduce median-SHAP, a method for explaining black-box models predicting individual survival times.