Prediction of a Structured Clinical Assessment by Patient Reported Outcomes and Machine Learning Algorithms: A Comparative Study

Participants will be recruited to complete self reported surveys normally used as standards of care for screening and monitoring depression and anxiety symptom severity, provide a voice sample composed of an answer to open ended questions and then be assessed by a mental health professional using structured and clinically validated assessment tools for depression and anxiety. Their voice will be analyzed by machine learning models that predict the severity of depression and anxiety symptoms. The models' performance will be compared to the clinician assessments and how that correlation compares to a similar comparison between the clinician assessments with the self reported surveys. It is hypothesized that the performance of the machine learning models in assessing the severity of depression and anxiety symptoms is no worse than the self reported surveys when both are compared to clinician assessments. It is also hypothesized that presence or absence of the diagnoses of Major Depressive Disorder and Generalized Anxiety Disorder can be predicted better than chance by the analysis of the participant's voice sample using machine learning models.