Geographic Variations in Health Care: Overcoming Bias, Expanding Generalizability, and Improving Relevance

PROJECT SUMMARY A substantial amount of research over the past 50-years involving Medicare and other health insurance claims data has focused on evaluating variation in health care use and outcomes across geographic regions. For example, over the last quarter-century, the Dartmouth Atlas Project has focused on variation in Medicare fee-

for-service health care use for diagnostically defined cohorts of patients, often conditioning on future outcomes (e.g., death) to account for variation in health status. Numerous other claims studies have also used Medicare claims data to estimate comparative effectiveness of different treatments and procedures. Almost all of these

types of research studies have used a nationwide measure of health care markets created at Dartmouth known as “hospital referral regions (HRRs).” These regional markets, and the methodology underlying their delineation, have remained largely unchanged for nearly 30-years. In addition, because Medicare primarily

covers people aged 65 and over, these regional measures—even at the time—are not representative of the whole population, leading to questions regarding the external validity of the published results especially given the tendency to equate such findings with the whole population. Motivated by the recent surge of interest in

health and health care inequities, a growing concern in algorithmic bias, availability of newer and more extensive data on younger populations, and advances in network and geospatial analysis, this project proposes to revisit the methodology, definitions, and practical applications of regional and network measures of

health care use and outcomes. These new approaches will avoid the potential biases of prior geographic measures by better capturing care patterns of underserved populations, and will facilitate geographic variations and comparative-effectiveness research that overcomes bias and possesses greater statistical power to detect

effects of interest. Secondly, this project will develop new measures that quantify heterogeneity of geographic and other variations in use and spending across population strata, including disparity indices. Thirdly, this project will evaluate the bias of Medicare estimates and develop procedures to generalize results to other

populations. Results of all analyses, including the algorithms for HRR delineation, will be used to modernize statistical and geographic approaches to characterizing health care access and health outcomes. These will be widely disseminated to research and stakeholder communities, thus empowering public health professionals

and researchers to define analysis and administrative units pertaining to their specific health care systems and needs. This project will have a major impact on the research communities engaged in the evaluation of geographic variation in health care delivery and health outcomes.