Computational Strategies to Tailor Existing Interventions for First Major Depressive Episodes to Inform and Test Personalized Interventions

ABSTRACT Major depressive disorder (MDD) is a chronic, recurrent illness impacting 20.6% of the U.S. population, causing significant disability and an economic impact of $326.2 billion annually. One of the largest risk factors for depression chronicity and disability is inadequate antidepressant response, defined as less than a 50% improvement in depressive symptoms after starting antidepressant treatment. Antidepressants are recommended as a first-line depression treatment and taken by 70% of patients with depression. Inadequate antidepressant response is experienced by 50-60% of patients starting an antidepressant and is responsible for 47% of the economic impact and disability caused by MDD. As such, identifying risk for inadequate antidepressant response early, during a patient’s first clinical presentation for a depressive episode, would be an innovative, urgently needed first step towards preventing recurrent depressive episodes, reducing depression chronicity and disability, and improving MDD outcomes. This step aligns with The National Institute of Mental Health (NIMH) Strategic Plan Objective 3.2 to “develop strategies for tailoring existing interventions (antidepressants) to optimize (depressive episode) outcomes.” While previous studies identified separate predictors of antidepressant response, no study to date has focused on integrating known and novel predictors of inadequate antidepressant response during a patient’s first depressive episode. This knowledge gap exists as no large studies in diverse populations have integrated comprehensive clinical, demographic, genetic, AND behavioral information in one model to predict inadequate antidepressant response prior to first antidepressant treatment. Such information is crucial to improve patient care, reduce depressive disorder chronicity and disability, and tailor existing patient interventions to optimize MDD outcomes. Utilizing electronic health record data from three large, integrated healthcare systems representing over 6.9 million members (Kaiser Permanente (KP) Northern California, KP Washington, and HealthPartners), we aim to quantify inadequate antidepressant response risk at the time of a patient’s first clinical presentation for a depressive episode by integrating clinical, demographic, genetic, and behavioral information in one predictive model. To accomplish this aim, we will use translational machine learning and predictive modeling, internal and external model validation and testing, prospective validation, and existing genome-wide genotypic data. Further, we will examine barriers and facilitators to clinical applications of predictive models for MDD to facilitate clinical translation and implementation of the predictive model, reducing the time between research innovation and clinical application. Our long-term goal is to develop a clinical tool informing decision making and promoting MDD treatment optimization.