DynSurvX explains changing survival predictions using patient data

DynSurvX: Time-Resolved Explanations for Dynamic Survival Predictions

Machine Learning

Summary

Predicting how long patients might survive based on their changing health information is helpful but often hard to understand. The authors created DynSurvX, a method that explains which patient details most affect survival predictions and how their importance changes over time. This approach is faster than previous methods and works well on both simulated and real clinical data. It helps doctors better understand predictions as new patient information comes in.

What this means in practice

  • For hospital data teams: Generate understandable, time-updated survival risk explanations for clinicians using patient records with both static and time-changing data.
  • For healthcare software developers: Build patient monitoring tools that deliver fast, detailed insights on changing survival predictions as new health data arrives.$Commercial implications: Enables development of medical decision support software that explains dynamic survival risks, improving clinician trust and patient care.

Authors

Van Tuan Nguyen, Agathe Guilloux

Abstract

Recent survival models integrating longitudinal data achieve strong predictive performance, but their predictions are often difficult to interpret, limiting their clinical use. We propose DynSurvX, a model-agnostic explainability framework for dynamic survival prediction with mixed static and longitudinal covariates. Based on Shapley values, DynSurvX introduces a hierarchical two-phase strategy that first identifies influential longitudinal features and then adaptively refines their temporal importance, enabling fine-grained explanations while reducing computational cost. The framework provides time-resolved explanations that evolve as new patient data become available. Experiments on simulated and two clinical datasets show varying agreement with theoretical importance across prediction models, with closer agreement for better-performing models, alongside clinically interpretable insights in real-world settings. It remains robust across prediction settings and reduces computation time by 41--80\% relative to the temporal explanation baseline, making dynamic explanations practical.