ADDRESSING GAPS IN REAL-WORLD MENTAL HEALTH DATA: FEASIBILITY OF AI-BASED ESTIMATION OF THE PHQ-9 AND CGI-I IN MAJOR DEPRESSIVE DISORDER
Author(s)
Michelle Leavy, MPH1, Zachary Bryant, PharmD1, Ombretta Palucci, MS2, Pedro Alves, MSc1, Costas Boussios, PhD1.
1OM1, Inc., Boston, MA, USA, 2OM1, Inc., Pully, Switzerland.
1OM1, Inc., Boston, MA, USA, 2OM1, Inc., Pully, Switzerland.
OBJECTIVES: The Patient Health Questionnaire 9-item (PHQ-9) and Clinical Global Impression Scale-Improvement (CGI-I) are standard measures for assessing depression severity and treatment response, but documentation of these measures is inconsistent in electronic medical records (EMRs). This limits the utility of real-world data (RWD) for large-scale studies of major depressive disorder (MDD). Artificial intelligence (AI) models have been developed and validated to estimate these scores from routinely recorded clinical notes. This study assessed the feasibility of applying the models to expand the population available for RWD studies of MDD.
METHODS: The estimated PHQ-9 (ePHQ-9) and estimated CGI-I (eCGI-I) models were applied to the OM1 MDD PremiOM Dataset, a large U.S.-based RWD source containing data on over 490,000 MDD patients receiving care from mental health professionals. For the eCGI-I analysis, patients initiating a new antidepressant with baseline and follow-up observations were included (n=182,750). For the ePHQ-9 analysis, MDD patients with or without observed PHQ-9 scores within the study window were included (n=77,871). The proportion of patients with analyzable outcome data was compared before and after applying estimation models.
RESULTS: In the eCGI-I cohort, 21% (38,252/182,750) of patients had a recorded CGI-I score in the study timeframe. Application of the eCGI-I model expanded the available sample size by 4.3x to 6.0x across antidepressant drug classes. In the ePHQ-9 cohort, 38% (29,608/77,871) of patients had observed PHQ-9 scores; the ePHQ-9 model generated estimated scores for an additional 61,794 patients, expanding the proportion with analyzable depression severity data.
CONCLUSIONS: AI-based estimation of the PHQ-9 and CGI-I substantially increases the proportion of MDD patients with analyzable mental health outcome data in RWD, with sample size gains of up to 6x. These models offer a scalable, validated approach to addressing critical gaps in structured clinical data and improving the feasibility of real-world studies of MDD treatment response and disease severity.
METHODS: The estimated PHQ-9 (ePHQ-9) and estimated CGI-I (eCGI-I) models were applied to the OM1 MDD PremiOM Dataset, a large U.S.-based RWD source containing data on over 490,000 MDD patients receiving care from mental health professionals. For the eCGI-I analysis, patients initiating a new antidepressant with baseline and follow-up observations were included (n=182,750). For the ePHQ-9 analysis, MDD patients with or without observed PHQ-9 scores within the study window were included (n=77,871). The proportion of patients with analyzable outcome data was compared before and after applying estimation models.
RESULTS: In the eCGI-I cohort, 21% (38,252/182,750) of patients had a recorded CGI-I score in the study timeframe. Application of the eCGI-I model expanded the available sample size by 4.3x to 6.0x across antidepressant drug classes. In the ePHQ-9 cohort, 38% (29,608/77,871) of patients had observed PHQ-9 scores; the ePHQ-9 model generated estimated scores for an additional 61,794 patients, expanding the proportion with analyzable depression severity data.
CONCLUSIONS: AI-based estimation of the PHQ-9 and CGI-I substantially increases the proportion of MDD patients with analyzable mental health outcome data in RWD, with sample size gains of up to 6x. These models offer a scalable, validated approach to addressing critical gaps in structured clinical data and improving the feasibility of real-world studies of MDD treatment response and disease severity.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD49
Topic
Real World Data & Information Systems
Disease
Mental Health (including addiction)