Predicting Suicide Risk with Machine Learning
A new predictive model can help clinicians intervene at the right time
Using anonymized data from nearly 3 million patients and almost 20 million visits, researchers from Kaiser Permanente, HealthPartners, and Henry Ford Health System developed a predictive model that more accurately identifies patients at risk of suicide than existing assessments.
The new model uses data from Epic, including results from depression assessment questionnaires, indications of substance abuse, prescriptions for psychiatric medications, and other documentation. Critical to all this was machine learning, which was used to recognize patterns that help analyze indicators corresponding to suicide risk.
Looking retroactively at patient data, the new model accurately identified nearly half of the patients who attempted or died by suicide within 90 days of a primary care or mental health visit. Suicide risk assessments that are currently in use at most hospitals and clinics use fewer data points from patients’ health records, and identified only a quarter to a third of later suicide attempts and deaths.
“Risk predictions can supplement clinical judgment and direct clinicians’ attention to where it’s most needed,” said Greg Simon, psychiatrist and mental health researcher from Kaiser Permanente. “Predictions don’t replace a clinical assessment, but they can help providers intervene with the right patients at the right time.”
Read more here.