Detection of formal thought disorders in child and adolescent psychosis using machine learning and neuropsychometric data.

Introduction: Formal Thought Disorder (FTD) is a significant clinical feature of early-onset psychosis, often associated with poorer outcomes. Current diagnostic methods rely on clinical assessment, which can be subjective and time-consuming. This study aimed to investigate the potential of neuropsychological tests and machine learning to differentiate individuals with and without FTD. Methods: A cohort of 27 young people with early-onset psychosis was included. Participants underwent neuropsychological assessment using the Iowa Gambling Task (IGT) and Simple Reaction Time (SRT) tasks. A range of machine learning models (Logistic Regression (LR), Support Vector Machines (SVM), Random Forest (RF) and eXtreme Gradient Boosting (XGBoost)) were employed to classify participants into FTD-positive and FTD-negative groups based on these neuropsychological measures and their antipsychotic regimen (medication load in chlorpromazine equivalents). Results: The best performing machine learning model was LR with mean +/- standard deviation of cross validation Receiver Operating Characteristic Area Under Curve (ROC AUC) score of 0.850 (+/- 0.133), indicating moderate-to-good discriminatory performance. Key features contributing to the model's accuracy included IGT card selections, SRT reaction time (most notably standard deviation) and chlorpromazine equivalent milligrams. The model correctly classified 24 out of 27 participants. Discussion: This study demonstrates the feasibility of using neuropsychological tests and machine learning to identify FTD in early-onset psychosis. Early identification of FTD may facilitate targeted interventions and improve clinical outcomes. Further research is needed to validate these findings in larger, more diverse populations and to explore the underlying neurocognitive mechanisms.