Research map / Learning theory and algorithms / Prediction and decision losses
Time-to-Event Prediction with Neural Networks and Cox Regression
Håvard Kvamme, Ørnulf Borgan, Ida Scheel
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use non-proportional Cox-Time instead of Cox-MLP (CC) | Cox neural network · KKBox churn | integrated Brier score: 0.119 -> 0.107 | Improved |
| use non-proportional Cox-Time instead of Cox-MLP (CC) | Cox neural network · KKBox churn | time-dependent concordance: 0.844 -> 0.861 | Improved |
| use Cox-Time instead of classical linear Cox | Cox model · SUPPORT | time-dependent concordance: 0.598 -> 0.629 | Improved |
| use Cox-Time instead of classical linear Cox | Cox model · METABRIC | time-dependent concordance: 0.628 -> 0.662 | Improved |
| use Cox-Time instead of DeepHit | survival network · FLCHAIN | integrated Brier score: 0.120 -> 0.102 | Improved |
| tune DeepHit by concordance (small alpha, more ranking loss) | DeepHit · KKBox churn | integrated Brier score: 0.147 vs 0.107 | Hurt |
| tune RSF with Ctd instead of mortality concordance (ablation) | RSF · KKBox churn | time-dependent concordance: 0.855 -> 0.870 | Improved |
| tune RSF with Ctd instead of mortality concordance (ablation) | RSF · SUPPORT | time-dependent concordance: 0.628 -> 0.634 | Improved |
| case-control sampling loss (Cox-MLP CC) instead of batch partial likelihood (DeepSurv) (ablation) | proportional Cox MLP · SUPPORT | time-dependent concordance: 0.611 -> 0.613 | No clear effect |
| sample more controls (1,2,4,8) in Cox-SGD loss (ablation) | Cox-SGD · simulation | mean partial log-likelihood convergence | No clear effect |
| sample more controls in Cox-SGD on small datasets (ablation) | Cox-SGD · simulation (n=100) | mean log-likelihood vs classical Cox | Improved |
Closest papers by meaning
- Orthogonal Survival Learners for Estimating Heterogeneous Treatment Effects from Time-to-Event Data (NeurIPS 2025 · similarity 0.83)
- XTSFormer: Cross-Temporal-Scale Transformer for Irregular-Time Event Prediction in Clinical Applications (AAAI 2025 · similarity 0.82)
- Stabilized Neural Prediction of Potential Outcomes in Continuous Time (ICLR 2025 · similarity 0.82)
- Dynamic Survival Analysis with Controlled Latent States (ICML 2024 · similarity 0.82)
- MOTOR: A Time-to-Event Foundation Model For Structured Medical Records (ICLR 2024 · similarity 0.81)
- FastSurvival: Hidden Computational Blessings in Training Cox Proportional Hazards Models (NeurIPS 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
