Using Machine Learning to Improve Pharmacotherapy in the Military

Author: George Carpenter

Purpose: This study aims to determine whether Psychiatric Electroencephalography Evaluation Registry (PEER) Interactive (an objective, adjunctive tool based on machine learning of quantitative EEG features and a 10,000 patient registry of outcomes) is more effective than the current standard of care in treatment of subjects suffering from depression. Patients and methods: This is an interim report of an ongoing, 2 year prospective, randomized, double blind, controlled study to evaluate PEER Interactive in guiding medication selection in subjects with a primary diagnosis of depression vs standard treatment. Subjects in treatment at two military hospitals were blinded as to study group assignment and their self report symptom ratings were also blinded. Quick Inventory of Depressive Symptomatology, Self Report (QIDS-SR16) depression scores were the primary efficacy endpoint. One hundred and fifty subjects received a quantitative electroencephalography exam and were randomized to either treatment as usual or PEER informed pharmacotherapy. Subjects in the control group were treated according to Veterans Administration/Department of Defense Guidelines, the current standard of care. In the experimental group, the attending physician received a PEER report ranking the subject’s likely clinical response to on-label medications. Interim Findings: Physicians who followed PEER recommendations had more statistically significant improvement than physicians who followed standard of care treatment based on VA/DOD Treatment Guidelines: • 144% greater improvement in QIDS-SR16 depression scores • 75% greater improvement in CHRT-7S scores for suicidal ideation • 139% greater improvement in PCL scores for PTS We found significantly greater improvements in depression scores (QIDS-SR16 P 0.03), reduction in suicidal ideation (Concise Health Risk Tracking ScaleSR7 P 0.002), and post- traumatic stress disorder (PTSD) score improvement (PTSD Checklist Military/Civilian P 0.04) for subjects treated with PEER recommended medications compared to those who did not follow PEER recommendations. Conclusion: This interim analysis suggests that an objective tool such as PEER Interactive can help improve medication selection. Consistent with results of earlier studies, it supports the hypothesis that PEER guided treatment offers distinct advantages over the current standard of care.

Co Author/Co-Investigator Names/Professional Title: Dan V iosifescu, MD, Professor, Psychiatry and Neuroscience, Icahn school of Medicine at Mount Sinai, New York, NY Robert J Neborsky MD, School of Medicine, UCSD, UCLA, Medical corps, US Navy Robert J Valuck PhD, Pharmacy, Professor, Epidemiology, and Family Medicine, University of Colorado, Denver, CO, Center for Pharmaceutical Outcomes research, University of Colorado, Denver, CO