Artificial Intelligence and Machine Learning in Mental Health Services: A Literature Review

Key Message

Thirty-four studies were identified that were relevant for this report. Eight studies were systematic reviews (SRs), three were randomized controlled trials (RCTs), and 23 were non-randomized studies. No relevant evidence-based guidelines were identified. The studies included a variety of populations, including individuals with bipolar disorder, schizophrenia, MDD, postpartum depression, post-traumatic stress disorder, and individuals who have suicidal ideation or have attempted suicide. No specific information on subgroups (such as immigrant, refugee, ethnocultural, or racialized individuals; or First Nations, Métis, Inuit; or lesbian, gay, bisexual, transgender, queer or questioning, and two-spirited [LGBTQ2+]) were found. Two studies focused on young children (ages three to seven), and one study used the National Health and Nutrition Examination Survey (NHANES), which includes children and adults. No effectiveness or accuracy information was found on adolescents or older adults with mental health conditions, as the majority of studies focused on adults over the age of 18 and under the age of 65. Intended users of these AI technologies were primarily clinicians (for diagnosis), but three studies examined models that were intended for use by patients. The primary purpose of the AI or ML models was to differentiate patients who have or do not have mental health conditions or to assist in treatment. Diagnostic accuracy of AI or ML models was generally moderate to high when compared with physician assessment, and AI-based applications for the treatment of patients significantly reduced depression symptoms and increased the use of crisis resources in studies that compared various versions of electronic applications for mental health.

This literature review was commissioned by The Mental Health Commission of Canada to address the role of AI in mental health services. This report is a companion to an Environmental Scan, which provides more information on the types and trends of AI either emerging or currently in use for the prevention, diagnosis, or treatment of mental health problems and illnesses, research and development initiatives, and the professional groups and organizations involved in the development or use of these technologies in Canada and internationally [Artificial Intelligence and Machine Learning in Mental Health Services: An Environmental Scan. Ottawa: Canadian Agency for Drugs and Technology in Health (CADTH), Mental Health Commission of Canada (MHCC); 2021 June.]