Large Language Models are Approximate Survival Estimators

An arXiv research paper that proposes the Survprompt framework and systematically evaluates the capability of large language models to perform medical survival prediction under zero-shot conditions, benchmarking them against traditional methods such as random survival forests.