Developing a Crisis Surveillance System for Mental Health Using Machine Learning Models
It had observed that the prompt identification of persons at risk of mental health decline improves illness outcomes. In the current study, researchers constructed a machine learning model that used people’s electronic health information to monitor them for 28 days for any danger of a mental health crisis. A crisis had defined as the emergence of symptoms requiring healthcare services. In the hospital’s database, time series of events have represented with timestamps and event attributes in separate SQL tables. Based on data from 2012 to 2019, models are trained, tweaked, and chosen.
The system forecasts the probability of a crisis occurring during the following 28 days (whereby the algorithm is queried every week for every patient). The dashboard displays the patients with the most significant estimated risk with essential indicators, patient notes, and a questionnaire form that the doctor completes for each patient. Classifiers based on machine learning approaches such as decision trees, ensembles, probabilistic, and deep learning were evaluated. The researchers compared the XGBoost (extreme gradient boosting) model to two baseline variables, namely the clinical practice diagnosis-based baseline model and prospective study data. In the case of organic illnesses, the general model performed much better with an area under the receiver operating characteristic (AUROC) of 0.89 compared to its overall performance of 0.797%.
The lowest performance was reported for mood disorders, schizophrenia, schizotypal disorders, and delusional disorders. Researchers think this research might encourage mental health professionals to change from reactive to preventive treatment.
