Fig. 5: Applying AgentMD on the MIMIC-III cohort.
From: AgentMD: Empowering language agents for risk prediction with large-scale clinical tool learning
a AgentMD is applied to 9822 admission notes in MIMIC. b AgentMD calculation results are aggregated by the risk calculators, and patients are ranked within each tool. c The distribution of the number of selected calculators for each patient. d The distribution of the number of eligible patients for each used calculator. e Calculation result distributions from two calculators. f Sample ROC curves of clinical calculators where AgentMD outperforms GPT-4 in predicting in-hospital mortality. Source data are provided as a Source Data file.
