Psyche-D: Predicting Change in Depression Severity Using Person-Generated Health Data

Background: In 2017, an estimated 17.3 million adults in the US experienced at least one major depressive episode, with 35% of them not receiving any treatment 1 . Under-diagnosis of depression has been attributed to many reasons including stigma surrounding mental health, limited access to medical care or barriers due to cost 2 .Low burden personal health applications, leveraging person-generated health data (PGHD), could represent a possible way to increase engagement and improve outcomes. Methods: Here we present the development of PSYCHE-D (Prediction of SeveritY CHange - Depression), a predictive model developed using PGHD from more than 4000 individuals, that accurately forecasts long-term increase in depression severity. PSYCHE-D uses a two-phase approach: the first phase supplements self-reports with intermediate generated labels; the second phase predicts changing status over a 3 month period, up to 2 months in advance. We optimized across a range of algorithms and PGHD-derived feature inputs. Findings: The best performing model was based on, in phase 1, Extreme Gradient Boosting (XGBoost) algorithm with input from a range of PGHD features including objective activity and sleep, and self-reported changes in lifestyle; the second phase based on the Light Gradient Boosting Machine (LGBM) Dropouts meet Multiple Additive Regression Trees (DART) algorithm, with a range of PGHD features and the intermediate generated labels. PSYCHE-D achieved a sensitivity of 88.2% in predicting long-term deterioration in depression status. We also demonstrate how PSYCHE-D could be deployed, by demonstrating that a pre-trained model can also robustly predict changes in depression severity for individuals that the model is naive to (87.3% sensitivity). Interpretation: These results demonstrate that low burden PGHD can be the basis of accurate and timely warnings that an individual's mental health may be deteriorating. We hope this work will serve as a basis for improved engagement and treatment of individuals suffering from depression. Funding: This work was entirely self-funded by Evidation Health Inc. Declaration of Interest: MM is a paid intern at Evidation Health and is completing her Master's degree at EPFL. RK, MF, JM and IC are employees of, and hold stock options in, Evidation Health. IC has received payment for lecturing on Digital Health at ETH Zurich and FHNW Muttenz. He is an Editorial Board Member at Karger Digital Biomarkers and a founding member of the Digital Medicine Society. MJ has no competing interests to declare. Ethical Approval: Data used to develop the model was derived from the Digital Signals in Chronic Pain (DiSCover) Project (Clintrials.gov identifier: NCT03421223): This study received expedited review and IRB approval from WCG IRB (IRB Study #: 1181760; Protocol #: 20172916; Initial Approved Date: December 21, 2017).