# An Effective Meaningful Way to Evaluate Survival Models

arXiv: https://arxiv.org/abs/2306.01196 · HTML: https://mlsanity.com/paper/2306.01196/

## Key facts

- Cited by 15 papers in the map (data through 14 Oct 2026).
- Builds on 8 of its 50 references that are in the map.
- First posted on arXiv 1 Jun 2023.

## Builds on (most cited first, 8 of 8)

- [Effective Ways to Build and Evaluate Individual Survival Distributions](https://mlsanity.com/paper/1811.11347/) (2018)
- [Survival Regression with Proper Scoring Rules and Monotonic Neural Networks](https://mlsanity.com/paper/2103.14755/) (2021)
- [Countdown Regression: Sharp and Calibrated Survival Predictions](https://mlsanity.com/paper/1806.08324/) (2018)
- [X-CAL: Explicit Calibration for Survival Analysis](https://mlsanity.com/paper/2101.05346/) (2021)
- [Survival Cluster Analysis](https://mlsanity.com/paper/2003.00355/) (2020)
- [The Brier Score under Administrative Censoring: Problems and Solutions](https://mlsanity.com/paper/1912.08581/) (2019)
- [Survival Mixture Density Networks](https://mlsanity.com/paper/2208.10759/) (2022)
- [Inverse-Weighted Survival Games](https://mlsanity.com/paper/2111.08175/) (2021)

## Built on by (most cited first, 15 of 15)

- [An Introduction to Deep Survival Analysis Models for Predicting Time-to-Event Outcomes](https://mlsanity.com/paper/2410.01086/) (2024)
- [Interpretable Vision-Language Survival Analysis with Ordinal Inductive Bias for Computational Pathology](https://mlsanity.com/paper/2409.09369/) (2024)
- [Conformalized Survival Distributions: A Generic Post-Process to Increase Calibration](https://mlsanity.com/paper/2405.07374/) (2024)
- [Position: Stop Chasing the C-index when Evaluating Survival Analysis Models](https://mlsanity.com/paper/2506.02075/) (2025)
- [Toward Conditional Distribution Calibration in Survival Prediction](https://mlsanity.com/paper/2410.20579/) (2024)
- [Efficient Training of Probabilistic Neural Networks for Survival Analysis](https://mlsanity.com/paper/2404.06421/) (2024)
- [MENSA: A Multi-Event Network for Survival Analysis with Trajectory-based Likelihood Estimation](https://mlsanity.com/paper/2409.06525/) (2024)
- [Bayesian Active Learning for Censored Regression](https://mlsanity.com/paper/2402.11973/) (2024)
- [Overcoming Dependent Censoring in the Evaluation of Survival Models](https://mlsanity.com/paper/2502.19460/) (2025)
- [A meaningful prediction of functional decline in amyotrophic lateral sclerosis based on multi-event survival analysis](https://mlsanity.com/paper/2506.02076/) (2025)
- [ADHAM: Additive Deep Hazard Analysis Mixtures for Interpretable Survival Regression](https://mlsanity.com/paper/2509.07108/) (2025)
- [Transfer Learning and Machine Learning for Training Five Year Survival Prognostic Models in Early Breast Cancer](https://mlsanity.com/paper/2509.23268/) (2025)
- [Budget-constrained Active Learning to Effectively De-censor Survival Data](https://mlsanity.com/paper/2510.12144/) (2025)
- [Subgroup Discovery with the Cox Model](https://mlsanity.com/paper/2512.20762/) (2025)
- [Large Language Models are Approximate Survival Estimators](https://mlsanity.com/paper/2609.38181/) (2026)

## Abstract (arXiv)

One straightforward metric to evaluate a survival prediction model is based on the Mean Absolute Error (MAE) -- the average of the absolute difference between the time predicted by the model and the true event time, over all subjects. Unfortunately, this is challenging because, in practice, the test set includes (right) censored individuals, meaning we do not know when a censored individual actually experienced the event. In this paper, we explore various metrics to estimate MAE for survival datasets that include (many) censored individuals. Moreover, we introduce a novel and effective approach for generating realistic semi-synthetic survival datasets to facilitate the evaluation of metrics. Our findings, based on the analysis of the semi-synthetic datasets, reveal that our proposed metric (MAE using pseudo-observations) is able to rank models accurately based on their performance, and often closely matches the true MAE -- in particular, is better than several alternative methods.