Towards a CONCORD Between Patient Perspective and Ecological Case Formulation: Conversational Large Language Models for Personalised Mental Health

Understanding the mechanisms that maintain mental disorders at the individual level is essential for personalised psychological treatment, yet existing approaches rely on burdensome questionnaires or statistical models that infer causal relationships indirectly. Here we present CONCORD, an AI–supported conversational framework that continuously constructs individualised causal case formulations from natural patient conversations. Eleven people with posttraumatic stress disorder engaged in daily conversations over three weeks while also completing ecological momentary assessments and causal surveys for validation, generating 1,629 interactions across methods. CONCORD recovered individualised causal structures that closely matched patient-reported causal formulations while substantially exceeding the complexity captured by conventional correlational network models. The resulting networks closely conformed to a leading cognitive theory of posttraumatic stress disorder, accurately identifying trauma-related triggers as primary causal drivers and maladaptive coping strategies as downstream consequences. Because each inferred relationship is traceable to the underlying conversation, the framework remains transparent and clinically interpretable despite relying on AI. These findings demonstrate that natural conversation can provide a practical route to continuous, theory-guided case formulation, offering a scalable framework for ecological patient monitoring and personalised precision mental health.