SynthAgent: A Multi-Agent LLM Framework for Realistic Patient Simulation -- A Case Study in Obesity with Mental Health Comorbidities

TL;DR

This paper addresses data scarcity in obesity research with mental health comorbidities by introducing SynthAgent, a multi-agent framework that fuses NHANES, medical claims, epidemiological priors, and case literature to generate personalized, longitudinal virtual patients enriched with HEXACO, reinforcement sensitivity, and temperament profiles. A five-agent MAS (Summarizer, Generator, Augmenter, Evaluator, Refiner) coordinates data fusion, narrative generation, evidence augmentation, and quality validation to produce clinically coherent records with preserved joint dependencies among metabolic, psychological, and personality features. Four LLM engines (GPT-5, Claude 4.5 Sonnet, Gemini 2.5 Pro, DeepSeek-R1) are evaluated against a GPT-4o judge across 120 profiles, revealing GPT-5 and Claude as top performers in fidelity and diversity, respectively, with Claude offering greater cohort variety and psychological realism. The framework demonstrates privacy-preserving, scalable synthetic cohorts suitable for hypothesis testing, intervention evaluation, and AI-driven decision support, with potential applicability across medical domains beyond obesity.

Abstract

Simulating high-fidelity patients offers a powerful avenue for studying complex diseases while addressing the challenges of fragmented, biased, and privacy-restricted real-world data. In this study, we introduce SynthAgent, a novel Multi-Agent System (MAS) framework designed to model obesity patients with comorbid mental disorders, including depression, anxiety, social phobia, and binge eating disorder. SynthAgent integrates clinical and medical evidence from claims data, population surveys, and patient-centered literature to construct personalized virtual patients enriched with personality traits that influence adherence, emotion regulation, and lifestyle behaviors. Through autonomous agent interactions, the system simulates disease progression, treatment response, and life management across diverse psychosocial contexts. Evaluation of more than 100 generated patients demonstrated that GPT-5 and Claude 4.5 Sonnet achieved the highest fidelity as the core engine in the proposed MAS framework, outperforming Gemini 2.5 Pro and DeepSeek-R1. SynthAgent thus provides a scalable and privacy-preserving framework for exploring patient journeys, behavioral dynamics, and decision-making processes in both medical and psychological domains.

Figures (3)

• Figure 1: Overview of the SynthAgent multi-agent framework for obesity patient simulation. The system integrates data from multiple empirical sources, including NHANES, medical claims, epidemiological datasets, and patient case reports. After data pre-processing, the summarizer agent condenses empirical evidence into structured inputs for the generator agent, which creates initial patient profiles. The augmenter agent enriches these profiles using literature-based clinical evidence, while the evaluator agent checks for logical, temporal, and behavioral inconsistencies. The refiner agent then resolves any issues to produce the final validated synthetic patient.
• Figure 2: t-SNE visualization of clinical core embeddings across four LLMs. Segmented circles indicate comorbidity patterns (red=depression, blue=anxiety, green=social phobia, purple=binge eating, gray=obesity only).
• Figure 3: A sample simulated patient generated by the proposed Multi-Agent System (MAS) using the GPT-5 LLM engine. Several dimensions are truncated to fit within the available space, and ellipses (…) indicate that the information continues. The patient’s overall score is 81.