Regional Inequities and Cultural Sensitivity in AI Models for Psychiatric Care

The prevalence of psychiatric diagnoses is increasing globally, with major depressive disorder (MDD) among the most prevalent and a leading contributor to the growing strain on mental healthcare systems. Addressing these challenges requires large-scale, high-quality health data; therefore, we selected Denmark as a case study where rich register data is available. In Denmark, the average wait time is two years for access to psychiatric care. Artificial intelligence (AI) for healthcare is considered a potential solution for triage and resource allocation. Although Denmark is often considered a homogeneous society, its sociodemographic and socioeconomic factors differ across the country, including immigration rates. There are also regional differences in the prevalence of psychiatric diagnoses. However, we have yet to understand how AI models account for such regional inequities. To examine this, we study MDD, the most costly mental health diagnosis for the Danish healthcare System, as a use case. Our analysis consists of two parts. First, we conducted an observational association analysis to characterize the regional, sociodemographic, and socioeconomic factors that influence MDD diagnosis in Denmark. We find disparities in the odds of obtaining an MDD diagnosis according to region of care and sociodemographic factors. Secondly, we develop two MDD prediction models, utilizing low and high levels of information about individual heritage, respectively. We assess the performance of the two predictive models across the five administrative regions in Denmark, and their generalizability when fine-tuned on data from only a single region. We observe that the AI models varied in model performance across the different administrative regions, immigration status, and heritage. We also find that including more nuanced variables that account for immigration status, heritage, and socioeconomic status (SES) lead to an improved model performance on all regions, for both immigrants and people of Danish origin. These findings illustrate the importance of assessing the fairness of AI models deployed in high-risk domains such as psychiatry and approaching sociodemographic identity with cultural sensitivity. Lastly, we call on researchers to acknowledge the inequities that persist across geographical regions and to account for these disparities in future work.