Large Language Models are Approximate Survival Estimators
Survprompt is introduced, a framework that converts structured patient covariates into free-text clinical vignettes and prompts pre-trained LLMs to predict survival zero-shot, and feature ablations revealed that LLMs prioritized clinical variables similarly to specialized survival models.
Juan M. Zambrano-Chaves, Peniel N. Argaw, Risa Ueno et al.
· 0 citations