Towards end-to-end LLM-based censoring-aware survival analysis
The article discusses a new framework called LLMSurvival that enables censoring-aware survival analysis using large language models (LLMs). This framework reformulates time-to-event prediction and demonstrates improved performance over traditional models in clinical tasks. LLMSurvival shows high portability and supports local deployment, making it a promising tool for medical prediction.
- ▪LLMSurvival allows for censoring-aware survival analysis with unmodified LLMs on tabular clinical data.
- ▪The framework improves overall concordance over Cox proportional hazards modeling by 3.1% for ICU mortality and 0.5% for fracture risk.
- ▪It demonstrates superior performance over established deep learning survival models and expert curated scores.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.25399 (cs) [Submitted on 25 May 2026] Title:Towards end-to-end LLM-based censoring-aware survival analysis Authors:Yishu Wei, Hexin Dong, Yi Lin, Jiahe Qian, Yi Liu, Yifan Peng View a PDF of the paper titled Towards end-to-end LLM-based censoring-aware survival analysis, by Yishu Wei and 5 other authors View PDF HTML (experimental) Abstract:Objective: Survival analysis is central to medical prediction, yet large language models (LLMs) are rarely used as end-to-end survival models because censoring prevents straightforward supervised fine-tuning. Here we present LLMSurvival, a framework that enables censoring-aware survival analysis with unmodified LLMs operating directly on tabular clinical data.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
