Assessing variance in inpatient hospitalization rates for three major chronic conditions

Introduction: A chronic condition is broadly defined as a health condition lasting one or more years that requires relatively consistent medical attention and/or imposes limitations on tasks and activities considered part of "daily living." Chronic and mental conditions comprised 90 percent of healthcare spending in 2019 coming to approximately $3.69 trillion. Given that the average inpatient (IP) hospitalization cost was $11,700 in 2017, investigating the demographic source of chronic IP hospitalizations is of high interest to reduce costs. However, current research is not consistent in how IP hospitalization rate is measured and does not control for known disparities in condition prevalence. In this study, rates of IP hospitalization are measured and compared across three levels of residential status (metropolitan, micropolitan, and rural) for chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), and obesity; these three conditions are further assessed to understand the degree to which residential status influences whether an individual who has one of the given conditions is hospitalized by controlling for condition prevalence by county. Methods: There were two main datasets utilized in this study, a primary individual-level dataset that includes primary and secondary diagnoses for individual IP hospitalizations and a county-level dataset that presents prevalence and hospitalization frequencies for each of the conditions. Using USDA rural-urban continuum codes, each county in Missouri was assigned a residential status value of metropolitan (1), micropolitan (2), and rural (3); this file was uploaded to the MUHC SQL Shared Analytics Server. A query was designed to pull individual-level IP hospitalization data for 17 variables of interest from a database within the Shared Analytics server such as demographic, diagnosis, and location information for all ICD-10 codes considered "chronic", per the AHRQ chronic condition indicator. IP hospitalization frequencies for each condition were extracted and input into a CDC county-level prevalence dataset. IP hospitalization frequencies were divided by prevalence frequencies for each county in Missouri to attain an IP hospitalization rate, which is defined in this study as the ratio of people who were hospitalized in the IP setting for a given condition out of the total number of people with that condition by county. Two univariate regression models (linear and logistic) were developed to assess the statistical significance and degree to which IP hospitalization rate for is impacted by residential status; these models were performed three times, once for each condition Results: Although the average prevalence rates for all three conditions was highest in the rural residential status, the average IP hospitalization rates were lower across the board in rural residential areas. Univariate logistic regression analyses found that rural residential status (reference = metropolitan) was significant at p [greater than] 0.05 with p = 0.0142 (parameter estimate = -3.3042); the linear regression analysis of COPD was not significant at p = 0/06. Analysis of regression outputs for CKD and obesity found no statistically significant p-values, although p-values were lower in both conditions (for both regressions) for rural than micropolitan, indicating a relationship with population. The other responses in logistic analyses, although not significant, trend in the direction of lower IP hospitalization rates for the rural residential status, aligning with the average rates. Linear univariate regression analyses found no statistically significant correlations. Discussion, Conclusion, and Recommendations: Whether or not residential status has implications for chronic disease prevalence is no longer largely contested in the current literature; however, the existence of disparities in IP hospitalization rates is still open for debate. While the findings here point to the fact that IP hospitalization rates for those with CKD, COPD, and obesity are lower across the board for patients of rural residential status, despite higher prevalence rates in Missouri during 2019, only rural status (reference = metropolitan) for COPD had a statistically significant difference at p [greater than] 0.05 with p = 0.0142. The major takeaway from this study should be that there is great need for a standardized measurement system for IP hospitalization rate, not just for CKD, COPD, and obesity, but likely for all conditions. The field of health informatics and organizations governing the research therein should collaborate to identify the optimal measure based off the evolution of knowledge that has been collected -- it makes little sense to continue pushing out research identifying 1) statistical significance with either no basis for normalization against population, and 2) normalization on population alone. The first and foremost recommendation is that a standardized measure of hospitalization rate be identified and implemented into recognized ontologies for future research by an overarching organization such as the AHRQ or CDC to better guide researchers and to make the methodology standardized in the United States. Second, as it is also known that access to healthcare, particularly specialized care, is lower in rural areas, research should be continued into the costs of CKD, COPD, and obesity and how rural residents might adversely be affected. Third, and perhaps mostly importantly from a long-term research goal perspective, strategies should continue to be developed that identify cost-reduction strategies for IP hospitalizations, especially for chronic conditions.