Applying analytics to sociodemographic disparities in mental health

Mental health services and treatment are unfortunately subject to sociodemographic disparities. To address this issue, recent studies have begun to apply analytics methods—that is, artificial intelligence in general, machine learning and deep learning in particular—toward the identification of such disparities and, where possible, mitigation of bias within models used in mental health research. However, it is difficult to understand the scope and status of such research as it is spread across many journals and contexts of study. Here we conducted an analysis of articles in this area. We identified 40 articles from 2017 to July 2023 related to the use of analytics in the context of sociodemographic disparities in mental health. We find that prediction, clustering/grouping and fairness models were most often applied in the articles analyzed. A number of mental health-related sociodemographic disparities were identified in these articles, for example, associated with race/ethnicity, gender, age and socioeconomic status, but such findings were typically context dependent. Thus, we also provide suggestions in this Analysis on how to both enhance generalizability and embrace context-dependent findings, especially via the identification of heterogeneous treatment effects, model bias mitigation, use of generative artificial intelligence, incorporation of data from devices, and translation of findings into practice. In this study, the authors analyzed articles examining the use of artificial intelligence, machine learning and deep learning analytics for identifying sociodemographic disparities, such as in race/ethnicity and age, to make recommendations for improving models and generalizability.