Why are Massachusetts opioid prescribing rates higher in rural versus urban areas?

Aims To examine the relationship between geographical density and opioid prescribing in a model that simultaneously accounts for both supply and demand side factors. Design Using the Massachusetts All-Payer Claims (MAPC) Database, we capture individual patient-level characteristics and combine this with county-level data on economic conditions and health care delivery systems to examine the relationship between geographical density and opioid prescribing. Population and Setting Commercially insured population residing in Massachusetts between 2010 and 2014. Methodology Using a logistic regression framework, we determine the likelihood that an individual patient with one of the three conditions (back pain, joint disease, or car accident) will receive an opioid prescription and how this varies by geographic density between patients. We also perform a two-way Blinder-Oaxaca decomposition to understand whether it is the endowments (mean levels) of these factors or the strength of the relationship (coefficients) of those factors that is more important in explaining the differences in opioid prescribing across metropolitan and non-metropolitan areas. Finally, we also explore interactions between the demand and supply side factors. Findings We find that patients in non-metropolitan areas were 10 percentage points more likely to receive an opioid prescription. About half of this differential is attributable to the underlying health of the local population. Focusing on three prevalent conditions for which an opioid is commonly prescribed (back pain, joint disease, and car accidents), we find that a little less than half of the remaining gap can be explained by supply side factors, such as differences in the health care delivery system. On the demand side both demographics, particularly veteran status, and health insurance type were important factors. Roughly 80 percent of the difference in opioid prescribing rates can be explained by the inclusion of both sets of covariates. Allowing for the interaction of some demand-side (e.g., working in a physically demanding occupation) and supply-side (e.g., healthcare delivery system) variables further reduces this differential to be less than half of a percentage point and statistically insignificant. Conclusions Our findings suggest economic conditions, such as the type of working conditions that patients might experience, interact with the healthcare system in unforeseen ways and that more targeted interventions can reduce the persistent gap in opioid prescribing among more and less densely populated areas, with possible downstream impacts on overdose and mortality.