Concise Comprehensive Assessment of Psychiatric Disorder Risks Using Machine Learning

Importance. The global prevalence of mental health problems demands a short but comprehensive screening and monitoring scale of major psychiatric disorders.Objective. To use advanced machine learning techniques to develop a concise and comprehensive screening and monitoring tool for psychiatric disorder risks and evaluate its reliability and validity. Design, setting, and participants. We obtained three datasets of outpatient participants at a psychiatric hospital in China who completed the 567-item Minnesota Multiphasic Personality Inventory (MMPI). We used the first dataset (N=6,704) for model training, testing, and internal validation to obtain a shortened version (100 items) using the stacked generalization ensemble of several machine-learning techniques. We validated the shortened scale against two prospective pristine datasets (N=928, unpreregistered; N=484, preregistered). Main outcomes and measures. The Area Under the Curve of the Receiver Operating Characteristic (AUC of ROC) to measure validity and Cronbach’s Alpha to measure reliability.Results. We reduced the length of the MMPI-2 by over 82%, from 567 to 100 items. The shortened scale can measure ten target conditions with at least 85% AUC and a Cronbach’s Alpha 0.97. We implemented the shortened scaleinto a mobile-friendly web application.Conclusions and relevance. An advanced machine-learning approach can significantly reduce a long scale to a shortened one while retaining high validity and reliability. This shortened scale can not only help alleviate pressure on healthcare systems burdened by the increased number of patients with mental health concerns but also allows for regular health monitoring or screening by individuals.