Improving psychosis detection and management using precision medicine.

Psychosis is a debilitating disorder that imposes substantial societal costs and disrupts the lives of predominantly young people. Comprehensive management across all clinical stages—including indicated prevention during the prodromal phase, early intervention in first-episode psychosis, and long-term care for chronic patients—is essential to improve prognosis, symptoms, and functioning. Advances in precision medicine have shown promise in enhancing detection of at-risk individuals and optimizing treatment selection. This thesis expands current knowledge on precision psychiatry by introducing scalable dynamic detection methods for individuals at risk (Part A) and novel approaches to integrating shared decision-making into precision treatment rules for first-episode psychosis (Part B). Part A focuses on improving identification of individuals at risk of psychosis using dynamic survival modelling. Chapter 1 introduces indicated prevention and the concept of Clinical High Risk for Psychosis, outlining the need for effective detection strategies and key challenges. It also reviews clinical prediction modelling in psychiatry, emphasizing dynamic survival models and their clinical relevance. Chapter 2 presents a study developing and internally-externally validating a dynamic risk calculator using electronic health records from 158,139 patients in secondary mental health care. A Cox Landmark model was constructed and compared with a static approach using multi-level meta-regression methods. The dynamic model improved discrimination performance by 0.035 in Harrell’s C (95% CI 0.031–0.043, p < 0.001), corresponding to a 20% error reduction. Calibration and clinical utility also improved, particularly for later predictions. These findings advance scalable and systematic detection strategies and enhance the translational potential of risk calculators. Part B develops precision medicine methods incorporating shared decision-making to improve treatment selection in first-episode psychosis. Chapter 3 reviews psychopharmacological treatment in early psychosis, compares existing decision support systems for antipsychotic selection, and discusses barriers to individualized care. It introduces pragmatic precision psychiatry and causal inference methods for observational data to optimize treatment decisions. Chapter 4 presents the first development and validation of precision treatment rules that integrate patient preferences. Using electronic health records from 1,709 patients in early intervention services, innovative rules were constructed combining causal forest methods with rankings of patient preferences, effectiveness, and side effects. Across preference scenarios, aripiprazole was recommended as optimal for 80% to 98% of patients. Implementation of these rules would significantly reduce most side effects, although extrapyramidal symptoms may increase. No significant effects were observed for hospitalization rates or medication changes. Part C summarizes the findings, discusses their implications for psychiatry and precision medicine, and highlights future research directions.