Towards end-to-end LLM-based censoring-aware survival analysis
Researchers have developed LLMSurvival, a novel framework that enables the use of large language models for survival analysis in clinical settings. This approach reformulates time-to-event prediction as a pairwise ranking problem, allowing unmodified LLMs to handle censored data. LLMSurvival demonstrated improved performance over traditional Cox models and existing deep learning methods on ICU mortality and fragility fracture prediction tasks. AI
IMPACT Enables LLMs to perform complex medical predictions on censored data, potentially improving clinical decision-making.