Ying Lin1, Shuai Huang1, Gregory E Simon2, Shan Liu3. 1. Department of Industrial and Systems Engineering University of Washington, Box 352650, Seattle, WA 98195, United States. 2. Group Health Research Institute, 1730 Minor Ave, Suite 1600, Seattle, WA 98101, United States; Psychiatry and Behavioral Sciences, University of Washington, Box 356560, Seattle, WA 98195, United States. 3. Department of Industrial and Systems Engineering University of Washington, Box 352650, Seattle, WA 98195, United States. Electronic address: liushan@uw.edu.
Abstract
BACKGROUND: Depression is a common, complex, and dynamic mental disorder. Mitigating depression has become a national health priority as it affects 1 out of 10 American adults and is the most common mental illness seen in primary care. The emerging use of electronic health record (EHR) provides an unprecedented information infrastructure to understand depression trajectories. OBJECTIVE: We aim to effectively analyze patterns in the collected depression trajectories of a treatment population and compare several methods to predict individual trajectories for monitoring treatment outcomes. METHODS: Our data includes longitudinal Patient Health Questionnaire (PHQ)-9 scores over 4 years for assessing depression severity from the Mental Health Research Network. We analyzed > 3,000 patients with at least six PHQ-9 observations who have ongoing treatment. We used smoothing splines to model individual depression trajectories. We then used K-means clustering and collaborative modeling (CM) to identify subgroup patterns. We further predicted the individuals' PHQ-9 scores based on depression trajectories learnt from individual growth model (IGM), mixed effect model (MEM), CM, and similarity-based CM (SCM), and compared their predictive performances. RESULTS: We found five broad trajectory patterns in the ongoing treatment population: stable high, stable low, fluctuating moderate, an increasing and a decreasing group. For prediction, the root mean square error (rMSE) in the testing set for IGM, MEM, CM, and SCM are 12.53, 5.91, 5.18, and 3.21. LIMITATIONS: Our EHR data provide limited information on patients' demographic, socioeconomic, and other clinical factors that may be relevant to improve model performances. CONCLUSION: We established a trajectory-based framework for depression assessment and prognosis that is adaptable to model population heterogeneity using EHR data. Collaborative modeling outperformed other established methods.
BACKGROUND:Depression is a common, complex, and dynamic mental disorder. Mitigating depression has become a national health priority as it affects 1 out of 10 American adults and is the most common mental illness seen in primary care. The emerging use of electronic health record (EHR) provides an unprecedented information infrastructure to understand depression trajectories. OBJECTIVE: We aim to effectively analyze patterns in the collected depression trajectories of a treatment population and compare several methods to predict individual trajectories for monitoring treatment outcomes. METHODS: Our data includes longitudinal Patient Health Questionnaire (PHQ)-9 scores over 4 years for assessing depression severity from the Mental Health Research Network. We analyzed > 3,000 patients with at least six PHQ-9 observations who have ongoing treatment. We used smoothing splines to model individual depression trajectories. We then used K-means clustering and collaborative modeling (CM) to identify subgroup patterns. We further predicted the individuals' PHQ-9 scores based on depression trajectories learnt from individual growth model (IGM), mixed effect model (MEM), CM, and similarity-based CM (SCM), and compared their predictive performances. RESULTS: We found five broad trajectory patterns in the ongoing treatment population: stable high, stable low, fluctuating moderate, an increasing and a decreasing group. For prediction, the root mean square error (rMSE) in the testing set for IGM, MEM, CM, and SCM are 12.53, 5.91, 5.18, and 3.21. LIMITATIONS: Our EHR data provide limited information on patients' demographic, socioeconomic, and other clinical factors that may be relevant to improve model performances. CONCLUSION: We established a trajectory-based framework for depression assessment and prognosis that is adaptable to model population heterogeneity using EHR data. Collaborative modeling outperformed other established methods.