How Machine Learning to Be Used to Prevent Suicide Attempts?

Introduction

One of the leading causes of death in the United States is suicide. An Electronic Health Record (EHR) suicide risk assessment tool can forecast and also mitigate clinician burnout in a non-psychiatric medical facility, according to a recent study published in JAMA Network Open.

How Does the Machine Learning Tool Help?

Prevention of suicide begins with risk identification and prediction, which generally require face-to-face screening and provider interaction. However, behavioral health issues are an ignored area of the healthcare industry and this area is not funded adequately.

Although most suicide risk models are connected to face-to-face screening data, it is tough for a provider to depend on or enhance the clinical workflow, the study authors wrote. In some settings, universal screening may reduce the risk of downstream suicidality. However, the team has highlighted that in-person screening takes attention and time, and clinicians conduct such screenings with varying qualities.

Researchers of Vanderbilt University Medical Center (VUMC) have created a Machine Learning (ML) algorithm that uses EHR data to forecast the suicide attempt risk. This model has recently undergone a prospective trial at the institution.

During the trial, the researchers observed over 115,905 suicide predictions for 77,973 patients over 296 days. Overall, the solution recorded 129 suicide attempts for 85 individuals. During the initial five months, suicide forecasts were not accurate. However, after integrating logic recalibration utilizing data from those five months, the tool improved in the next five months.

“This study validated performance of a published suicide attempt risk model using real-time clinical forecasts in the background of a vendor-supplied EHR,” wrote the study authors.

Implementing this tool into the EHR could influence clinical decision-making, screening practices, and care coordination stated the research team.

False positives and negatives screenings were regarded as a major weakness of EHR suicide risk tools.

“Here, we note very low false-negative rates in the lowest risk tiers both within (0.02%) and without (0.008%) universal screening settings,” the study authors explained.

Three things to know:

• The numbers were required to screen for suicide risk were practical for an algorithmic screening and required no added data collection or in-person screening, the report states.
• Risk models can be applied with accurate performance, but it is not equal in all clinical settings. It will need the model to be recalibrated before being deployed.
• Model data used to forecast suicide risk includes race, age, medication, gender, past healthcare utilization, patient zip code and previous medical conditions.

The next step is to pair the model with low-cost preventive strategies to build a preventive model in a pragmatic trial to be assessed for efficiency in preventing suicidality in the future,” the report states.

Data analytics and AI tools are being used increasingly in the mental healthcare space. A recent study published in JAMA Psychiatry has shown that a universal screening tool could control predictive analytics algorithms to precisely determine an adolescent’s suicide risk. The algorithm could also alert providers with information on patients in need of follow-up interventions.

Conclusion

Big Data can be used to full advantage by implementing predictive models. Across 5 diverse health care systems, a computationally efficient approach that leverages the full range of structured electronic health record data was able to identify the risk of suicidal behavior in patients. This approach could enable the development of clinical decision support tools that notify risk reduction interventions.