• Politics
  • Diversity, equity and inclusion
  • Financial Decision Making
  • Telehealth
  • Patient Experience
  • Leadership
  • Point of Care Tools
  • Product Solutions
  • Management
  • Technology
  • Healthcare Transformation
  • Data + Technology
  • Safer Hospitals
  • Business
  • Providers in Practice
  • Mergers and Acquisitions
  • AI & Data Analytics
  • Cybersecurity
  • Interoperability & EHRs
  • Medical Devices
  • Pop Health Tech
  • Precision Medicine
  • Virtual Care
  • Health equity

Researchers Develop New Suicide Risk Prediction Model from Massive Cohort


"We demonstrated that we can use [EHR] data in combination with other tools to accurately identify people at high risk."

A study that covered more than a half-dozen states and used data from millions of patients yielded a suicide prediction model better than previous methods, according to the Kaiser Permanente researchers who conducted it.

Data was drawn from 5 of Kaiser Permanente’s covered states (Colorado, Hawaii, Oregon, California, and Washington) and also from Detroit, Michigan’s Henry Ford Health System and the HealthPartners Institute in Minneapolis, Minnesota. In all, electronic health records (EHR) information showed that more than 2.9 million people made more than 20 million mental health or primary care visits at the centers between 2009 and 2015.

>>READ: Hunting for the Heart of a Changing Community

The researchers drew from EHRs and state death certificate databases to identify suicide attempts and fatalities in the 90 days following those visits. From there, they dug into the EHRs for relevant demographic and clinical factors—more than 300 of them—that had been observed in the 5 years leading up to those visits. They compared that information with patient responses to mental health questionnaires—which are typically weighted heavily in suicide attempt risk methods.

They then created a LASSO (least absolute shrinkage and selection operator) regression model using a random selection of 65% of the visits, which was validated on the remaining 35%.

Elements of the work were based on previous models created in other health systems, but the resulting risk score considered far more factors, and delivered much higher accuracy: In previous methods that assigned risk scores, patients placed in the top 5% (highest risk) accounted for less than a third of actual suicide attempts (an non-analytics models that base their predictions solely on questionnaires and clinician interviews are far less accurate).

In the new Kaiser Permanente model, the highest scores predicted more than 40% of attempts, both for patients whose initial visit was for primary care and for those who visited a mental health specialist. More than 5% of the patients in the top 5% of risk scores attempted suicide within 90 days.

"We demonstrated that we can use electronic health record data in combination with other tools to accurately identify people at high risk for suicide attempt or suicide death," author Greg Simon, MD, MPH, said of the work. He added that some of the health centers highlighted in the research “are the best in the country for implementing suicide prevention programs…But we could do better.”

So Kaiser Permanente is working to implement risk scores like these into its population health processes. He believes that the model is replicable in other health networks around the country—and that they can be used to develop plans for caring for patients at the highest risk.

The new study, “Predicting Suicide Attempts and Suicide Deaths Following Outpatient Visits Using Electronic Health Records,” was published today in the American Journal of Psychiatry.

Related Coverage:

Social Media as an Intervention: There's Plenty of Smoke, but Can It Help People Quit?

Facebook and the Intertwining of Mental Health, AI, and Social Media

AI Identifies Patients at Highest Risk of Cholera Infection

Related Videos
Image: Ron Southwick, Chief Healthcare Executive
George Van Antwerp, MBA
Edmondo Robinson, MD
Craig Newman
© 2024 MJH Life Sciences

All rights reserved.