A large language model for clinical outcome adjudication from telephone follow-up interviews: a secondary analysis of a multicenter randomized clinical trial

Abstract Automated adjudication of clinical outcomes from telephone follow-ups is crucial for reducing workload and increasing data quality in large-scale trials. Here, we show that a domain-specific large language model (Fu-LLM) effectively automates the preadjudication of key clinical events—including death, hospitalization, and medication use—based on 1,046 vignettes of follow-up telephone interviews conducted across three centers in a randomized clinical trial (China CT-FFR Study 3). Fu-LLM outperforms not only state-of-the-art general-purpose LLMs (e.g. GPT-3.5-turbo, GPT-4o, DeepSeek-v3, Claude 3.5-Sonnet, and Gemini-2.0-Pro) and conventional machine learning models (Support Vector Machine), but also human adjudicators in a silico human−model comparative study. It also shows greater robustness than different versions of GPT-4 do in temporal drift tests. Our findings demonstrate that Fu-LLM can significantly streamline outcome identification in clinical trials, offering a scalable and accurate tool for automating labour-intensive adjudication processes.