Researchers at Lawrence Berkeley National Laboratory (Berkeley Lab) https://wwwlbl.gov are applying deep learning and analytics to EHR data to help the Veterans Administration (VA) https://www.va.gov address a host of medical and psychological challenges affecting veterans.
Berkeley Lab, a Department of Energy (DOE) https://www.energy.gov Science Lab managed by the University of California as part of a collaboration between DOE and the VA, is combining the VA’s EHR system with DOE’s high performance computing, artificial intelligence, and data analytics resources.
DOE’s Berkeley Lab first became involved with the VA’s Million Veteran Program project in 2018 to help address the statistic that suicide is the 10th leading cause of death in the U.S and significantly higher in the veteran population.
The Lab’s goal is to identify patients at risk for suicide by using new patient-specific algorithms to provide tailored and dynamic suicide risk scores and make the resources available to VA caregivers and patients.
In 2018, five college students as interns signed on to do research on the suicide prevention project. The students developed algorithms to statistically analyze EHRs, to look for key factors related to suicide risks, and then apply deep learning methods to the large and complex datasets.
The students worked on a publicly available dataset containing medical record information on about 40,000 patients from one Boston hospital’s intensive care unit. Then they searched for patterns that might point to suicide risk.
EHR datasets contain both structured data such as demographics, information on prescribed medications, current lab work, procedures, plus studied unstructured data such as doctors’ and discharge notes. As a result, the team’s early efforts focused primarily on finding patterns in the complex information.
The team agrees that EHRs are hard to interpret so they are working to gain the trust of the physicians so the team can begin working on developing models that are interpretable. Since last summer, additional members of the team are working to fine tune how natural language processing can be used to sift through the structured and unstructured data collected on EHRs.