Conformal Survival Bands for Risk Screening under Right-Censoring

2025-05-07

AI Summary

This paper addresses the challenge of quantifying uncertainty in individualized survival distributions under right-censored data. We propose the first framework integrating conformal inference, inverse probability censoring weighting (IPCW), and false discovery rate (FDR) control to construct statistically guaranteed personalized survival prediction bands. The method is model-agnostic—compatible with any survival predictor—and theoretically proven to be asymptotically calibrated. It provides interpretable FDR-type guarantees—for instance, among patients flagged as high-risk, at least 50% truly survive beyond 12 months. Extensive experiments on simulated and multicenter clinical datasets validate the calibration accuracy and practical utility of the bands for risk stratification. Our approach significantly enhances both the reliability and operational feasibility of individualized prognostic decision-making in survival analysis.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Causality
• Planning, Routing, and Scheduling: Scheduling under Uncertainty

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Large-scale security measurements

Abstract

We propose a method to quantify uncertainty around individual survival distribution estimates using right-censored data, compatible with any survival model. Unlike classical confidence intervals, the survival bands produced by this method offer predictive rather than population-level inference, making them useful for personalized risk screening. For example, in a low-risk screening scenario, they can be applied to flag patients whose survival band at 12 months lies entirely above 50%, while ensuring that at least half of flagged individuals will survive past that time on average. Our approach builds on recent advances in conformal inference and integrates ideas from inverse probability of censoring weighting and multiple testing with false discovery rate control. We provide asymptotic guarantees and show promising performance in finite samples with both simulated and real data.

Problem

• Quantify uncertainty in individual survival estimates using censored data
• Provide predictive survival bands for personalized risk screening
• Ensure reliability in flagged low-risk patients with survival guarantees

Innovation

• Uses conformal inference for survival uncertainty
• Integrates inverse probability censoring weighting
• Ensures false discovery rate control