AI-Assisted Time-to-Event Projection Case Study

This case study presents the application of large language models (LLMs) in projecting time-to-event outcomes, specifically focusing on overall survival in patients with Stage B/C colon cancer. The study aims to assess how LLMs can assist analysts in interpreting clinical data to estimate life expectancy and median survival times. The analysis is based on a randomized trial comparing levamisole with and without 5-FU to observation alone, examining overall survival as the primary outcome. The findings indicate that LLMs can provide valuable clinical insights and support the interpretation of statistical models for time-to-event data. The analysis included parametric fitting of survival curves using various statistical distributions, identifying the log-normal distribution as the best fit for this high-risk population. The study emphasizes the importance of structuring data inputs to optimize LLM responses and suggests methodologies for dealing with limitations in survival projections. The results highlight the potential role of LLMs in enhancing clinical research and decision-making.