ADVERTISEMENT
With EHR Data, Model Predicts Suicide Attempt 2 Years in Advance
Using data from electronic health records, a predictive computer model was able to flag 38% of patients who attempted suicide an average 2 years before the attempt, according to a study published online in JAMA Network Open.
“Computers cannot replace care teams in identifying mental health issues,” researcher Ben Reis, PhD, of Boston Children's Hospital in Massachusetts said. “But we feel that computers, if well designed, could identify high-risk patients who may currently be falling through the cracks, unnoticed by the health system. We envision a system that could tell the doctor, ‘Of all your patients, these three fall into a high-risk category. Take a few extra minutes to speak with them.’ ”
Researchers developed the model using structured health record data, such as diagnostic codes, lab results, medical procedure codes, and medications, for 3.7 million patients aged 10 to 90 years. Data for each patient spanned 6 to 17 years, and patients were from 5 independent and geographically dispersed health care systems in the United States.
The computer model first identified patterns associated with documented suicide attempts in half the patients. In the remaining patients, the model predicted, based on the previously identified patterns, who would go on to attempt suicide. With a specificity of 90%, the model detected an average 38% of index suicide attempts an average 2.1 years in advance across all 5 centers.
Three-Day Intensive Intervention Significantly Reduces Suicidal Ideation
The strongest suicide attempt predictors included drug poisonings, drug dependence, acute alcohol intoxication, and mental health conditions such as borderline personality disorder and bipolar disorder. Unexpected predictors included rhabdomyolysis, cellulitis or abscess of the hand, and medications for human immunodeficiency virus (HIV).
Retraining the model at individual centers yielded more accurate results.
“We could have created one model to fit all medical centers, using the same codes,” said researcher Yuval Barak-Corren, MD, Boston Children's Hospital. “But we chose an approach that automatically builds a slightly different model, tailored to suit the specifics of each health care site.”
“Our approach demonstrates the feasibility of developing scalable, interpretable risk prediction algorithms using real-world health care data,” researchers concluded.
—Jolynn Tumolo
References
