Project 466757

Despite the evidence supporting the effectiveness of measurement-based care in treating Major Depressive Disorder, its clinical use remains low. A key barrier to the use of measurement-based care is the time needed for clinicians to conduct assessments. However, speech analysis is an efficient, scalable measure that can provide information on depression severity. Emerging evidence supports the use of language and acoustic speech features in the prediction of depression severity, but it is unknown how this compares to the ability of a licensed expert clinician to estimate depression severity using limited information from a short sample of speech. We hypothesize that expert clinicians will accurately predict depression severity, as well as provide a greater degree of accuracy in the prediction of depression severity compared to computer-based acoustic measures provided with the same speech samples. Participants are assessed for depression severity using the Montgomery-Asperg Depression Rating Scale (MADRS) and provide a 3-minute sample of neutral, naturalistic speech. After anonymization, 1-minute isolated samples of uninterrupted speech will be scored for depression severity by expert psychiatrists and clinical psychologists using a visual analog sliding scale. We will used mixed effects linear regression to test the concordance of the expert clinician estimation to MADRS scores and compare to the prediction of a machine learning model trained from acoustic features using PRAAT software. Through the development and testing of this novel system of measuring depression severity, we aim to better understand the extent to which trained expert clinicians form rapid impressions of psychopathology in their patients.