Predictive modeling for depressive relapse in youth

Author: John Strauss

Background: Major depressive disorder (MDD) is the fourth leading global cause of disability and is highly prevalent (6.8-18.4%) in youth, where it is associated with elevated suicide risk and long term impairments in relationships, school and work functioning. It is episodic, but relapse prevention studies using medication and psychotherapy have shown to be insufficient to prevent depressive relapse in youth. Several risk factors for onset and relapse of depression have been reported. The principal aim of the current study is to: estimate the extent to which some risk factors can be used as predictors of imminent relapse amongst youth with major depression. This will include modeling risk factors for depressive relapse risk prediction algorithms. Methods: Participants-- In this two-year longitudinal observational study, we will recruit 150 youth ages 12-21 with MDD from four hospital sites of the University of Toronto Department of Psychiatry. Exclusion criteria will include a primary substance use disorder or a secondary substance use disorder that is severe. Other exclusionary criteria include epilepsy, hypothyroidism, B12/folate/iron deficiency, autism spectrum disorder, multiple sclerosis, paraplegia or spinal cord injury, juvenile rheumatoid arthritis or other major autoimmune disease, chronic renal failure, inherited metabolic disorder, active cancer. Youth may or may not be in psychiatric treatment. Measures: Where applicable, The gold standard semi-structured diagnostic interview, the K-SADS-PL, will be used for ages 6-18; and the SCID-5 for ages 19-21. Multiple psychometric instruments will be obtained quarterly for symptom (CDRS-R, TCI, ASEBA), cognition (CANTAB), trauma (LISREY Adult and Child) and environment measures. The GENEActiv wearable will be used to to quantify movement, light exposure and body temperature. Phone sensor data will be used via Purple Robot (Android) or Sensor Data (iOS); phone SMS and iMessages will also be used. Analyses: Descriptive statistics will be provided for all psychometric and other measures. Logistic regression for the binary relapse/no relapse dependent variable will initially be done univariate, and later multivariate. Survival analysis will be used. Growth mixture and trajectory modelling and/or standard GLM with regression models will be used. Exploratory analyses using machine learning will complement the statistical modelling above. Two data sources will be examined for utility in prediction, namely textual data from participants' mobile phones and data from the CAMH EMR. Signal processing methods will be employed. We have active collaborations with the University of Toronto Department of Electrical and Computer Engineering and Cerner Math, for assistance with some analyses.

Co Author/Co-Investigator Names/Professional Title: Deepa Kundur, Professor, Department of Electrical and Computer Engineering, University of Toronto Payal Agarwal, Innovation Fellow, Women’s College Hospital Doug McNair, Senior Vice President and President of Cerner Math,Cerner Corporation Marco Battaglia/ Associate Chief, Division of Child & Youth Psychiatry, CAMH

Funding Acknowledgement (If Applicable): Cundill Centre for Child and Youth Depression