Analyses of Racial/Ethnic Disparities in U.S. Inpatient Mental Health Treatment

Objective: To determine whether there were racial/ethnic disparities regarding inpatient total charges associated with US inpatient mental health treatment, for discharges aged between 18 and 64, from hospitals in US counties using different measurement approaches. Methods: Descriptive analyses and single-level and three-level regression analyses (discharges within hospitals within counties) using secondary data of 92,027 discharges within 328 hospitals within 228 US counties from the 2003 Nationwide Inpatient Sample and Area Resource File. Results: Descriptive analyses showed minorities generally had higher charges than whites. Both single- and three-level regression analyses revealed that at the discharge-level, Asians/Pacific Islanders had higher charges than whites and Hispanics had lower charges than whites. The single-level model indicated that charges were higher in hospitals with a higher percentage of Hispanic discharges, and were lower in hospitals with a higher percentage of black and Native American discharges. The single-level model indicated that an increase in the number of Hispanics, Asians/Pacific Islanders, and other race/ethnicity in a US county resulted in increased charges; an increase in the number of Native Americans in a US county resulted in decreased charges. In the three-level regression model, race/ethnicity was not significant at the hospital-level and was only significant for other race/ethnicity with increased charges at the county-level. Conclusion: The single-level regression model findings would suggest allocating financial resources to mitigate physician stereotyping, physician bias, or statistical discrimination; this would not be suggested using the three-level regression model. The single-level regression model findings would suggest increasing each minority’s availability of and access to mental health services; this would not be suggested using the three-level regression model. Although we believe from a statistical perspective that the three-level regression model is the preferred model as it can provide more precise estimates of all regression coefficients, future research is necessary to consider replicating our findings and to mitigate omitted variables bias.