Proper Scoring Rules for Right-Censored Survival Data

2026-06-04 · Source: Takara TLDR - Daily AI Papers · Field: Science & Research — Mathematics & Computational Sciences, Health & Medical Research · Depth: Expert, medium

Summary

A framework is proposed for proper scoring of right-censored survival outcomes, addressing the inapplicability of conventional scoring rules when event times are partially observed. This framework operates by mapping the predictive distribution through the censoring mechanism, then applying the underlying proper score on the induced observed-data law. This construction yields localized scores for fixed censoring times and marginalized scores for random or partially observed censoring. It recovers familiar right-censored likelihood and IPCW-type criteria, while also providing right-censored versions of the CRPS, pinball loss, Brier score, and energy score. The marginalized score is proven proper under conditional independent censoring and strictly proper on the identifiable region. The principle also extends to censored engression, a sample-based learning objective for multivariate right-censored survival modeling. Experiments show these scores correctly rank oracle forecasts across various censoring regimes, outperforming forecast-dependent plug-in weighted scores that can exhibit ranking reversals.

Key takeaway

For AI Scientists developing survival models with right-censored data, this framework offers a robust solution to a common evaluation challenge. You should adopt these proper scoring rules, such as the right-censored CRPS or Brier score, to ensure accurate and reliable ranking of probabilistic forecasts. This approach avoids the ranking reversals seen with naive plug-in weighted scores, leading to more trustworthy model selection and improved performance in real-world applications. Consider implementing censored engression for multivariate scenarios.

Key insights

A new framework enables proper scoring rules for right-censored survival data by mapping predictions through censoring mechanisms.

Principles

• Conventional scoring rules fail with right censoring.
• Marginalized scores are proper under conditional independent censoring.
• Forecast-dependent plug-in scores can cause ranking reversals.

Method

Map predictive distribution through the censoring mechanism, then apply the underlying proper score on the induced observed-data law to derive localized or marginalized scores.

In practice

• Apply right-censored CRPS, pinball loss, Brier, energy scores.
• Use censored engression for multivariate survival modeling.

Topics

• Proper Scoring Rules
• Right Censoring
• Survival Analysis
• Probabilistic Forecasting
• Censored Engression
• CRPS

Best for: AI Scientist, Research Scientist

Related on AIssential

• Staying Alive: Uncensored Survival Analysis with Tabular Foundation Models — Tabular Foundation Models can perform uncensored survival analysis without dataset-specific training.
• How to Select Variables Robustly in a Scoring Model — Robust variable selection prioritizes stability across data subsets over performance on a single training set.
• Still Using predict_proba? Here’s Why Your Probabilities Are Lying to You — Scikit-learn's `predict_proba` outputs are scores, not calibrated probabilities, leading to critical decision-making errors.
• A consensus–but it’s a consensus of uncertainty. — Consensus in statistical analysis can appropriately include agreement on uncertainty or "not knowing."
• Improving the sharpness in neural network-based parametric post-processing of ensemble forecasts — A penalty term in neural network loss functions can improve forecast sharpness without sacrificing accuracy.
• Prediction-Intervention Games and Invariant Sets — Stable blanket predictors consistently outperform or match causal parent predictors in prediction-intervention games.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.