In the retrospective study, the machine learning model was applied to the EHRs of more than 4,000 patients admitted to two psychiatric facilities in the Netherlands and compared to the facilities’ reports of patient violence. In both locations, the tool demonstrated “good predictive validity,” per the study’s authors, and could therefore be a helpful addition to institutions treating patients at high risk of violence.
The tool is believed to be the first to analyze EHRs’ clinical notes for inpatient violence risk. Though the model was largely successful, the study’s authors note that it performed better when trained on inpatient data from one facility and used for analysis in the same facility; when the same training data was applied for analysis across both facilities, the model was somewhat less effective, though still a helpful tool for violence prediction.
More articles about AI:
Kaiser Permanente researchers develop AI tool to predict HIV risk
In healthcare, ethical AI is a life-or-death issue: Q&A with AI Ethics Lab’s founder and director
AI develops ‘turbocharged’ flu vaccine
