New Model Predicts Suicide Risk with Health Record Data

April 28, 2020
Kenny Walter

Kenny Walter is an editor with HCPLive. Prior to joining MJH Life Sciences in 2019, he worked as a digital reporter covering nanotechnology, life sciences, material science and more with R&D Magazine. He graduated with a degree in journalism from Temple University in 2008 and began his career as a local reporter for a chain of weekly newspapers based on the Jersey shore. When not working, he enjoys going to the beach and enjoying the shore in the summer and watching North Carolina Tar Heel basketball in the winter.

Investigators are currently working on a new algorithm that can present a time window in which an individual might develop suicidal behavior.

Ben Reis, PhD

The 2020 American Psychiatric Association (APA) Annual Meeting was cancelled this year, with plans made to convert the world-leading psychiatry conference into a two-part virtual session and educational platform for attendees.

In lieu of regular on-site coverage, HCPLive® will be running a series of interviews, insights, and reporting on topics that frequently headline the APA meeting—featuring familiar experts.

A team, led by Yuval Barak-Corren, MD, Computational Health Informatics Program, Boston Children’s Hospital, developed a new machine-learning algorithm that is based on a patient’s electronic health records (EHR) to identify whether an individual is at an increased risk of attempting suicide in the future.

In the prognostic study, the investigators used a supervised learning approach that was applied to structured EHR data from more than 3.7 million patients across 5 diverse US healthcare systems.

The investigators evaluated the generalizability and cross-site performance of a risk prediction methods using readily available structured data from health records in predicting incident suicide attempts.

The model included 6-17 years’ worth of data up to 2018.

They trained the models using naïve Bayes classifiers in each of the 5 systems and cross-validated the models in independent data sets at each site. They also calculated performance metrics.

The investigators sought a primary outcome of suicide attempts as defined by a previously validated case definition using International Classification of Diseases, Ninth Revision codes, with accuracy and timeliness of the predication measured at each site.

While predictive features varied by site, the most common predictors reflected mental health conditions and substance use disorders, such as drug poisonings, drug dependence, and acute alcohol intoxication.

Some other predictors included rhabdomyolyses, cellulites, and HIV medications.

Model performance was similar across sites, with areas under the curve ranging from 0.71 (95% CI, 0.70-0.72) to 0.76 (95% CI, 0.75-0.77). The new models detected a mean of 38% of cases of suicide attempts with 90% specificity a mean of 2.1 years in advance.

The findings ultimately suggest a computationally efficient machine-learning approach that leverages the full spectrum of structured EHR data could detect the risk of suicidal behavior in unselected patients to facilitate new clinical decision support tools that inform risk reduction interventions.

The researchers are currently working to develop more elaborate temporal models that can identify these effects to better inform clinicians in a specific time window, rather than just provide a general risk assessment.

“In most medical situations, you'd like to identify risks as early as possible, and that includes mental health,” Ben Reis, PhD, director of the Predictive Medicine Group, part of the Computational Health Informatics Program (CHIP) at Boston Children's Hospital, and co-senior author on the paper, said in an interview with HCPLive®. “As far as deciding on appropriate interventions, it is very helpful to clinicians if a specific time window can be identified.”

Reis explained that the model is based strictly on the data that is included in an individual’s medical records, meaning it would not include other pieces of information such as whether the patient has been involved in military combat or has experienced some sort of tragedy in their personal lives that is causing them mental health issues.

However, the model provides the psychiatrist with an individual’s suicide risk and in turn that doctor can use the patient’s circumstances to develop a therapy plan.

Reis also said the study has passed all the required ethical reviews and the data used were de-identified. The tool is intended strictly for a patient’s doctor and not for any other third party.

“We believe this research has a chance to save lives,” Reis said.

Suicide is currently a leading cause of mortality, with suicide-related deaths increasing in recent years. Suicide is also the second most common cause of death for adolescents in the US, with suicides increasing 30% between 2000-2016.

The study, “Validation of an Electronic Health Record—Based Suicide Risk Prediction Modeling Approach Across Multiple Health Care Systems,” was published online in JAMA Network Open.