The association between medical care utilization and health outcomes: A spatial analysis

As health care spending continues to strain government and household budgets, there is increasing interest in measuring whether the incremental dollar spent on health care is worth it. In studying this question, researchers often make two key assumptions: that health care intensity can be summarized by a single index such as average spending, and that samples of hospitals or regions are spatially independent: Manhattan and the Bronx are no more alike than are Manhattan and San Diego, for example. In this paper we relax both assumptions. Using detailed data on 897,008 elderly Medicare enrolees with acute myocardial infarction (or a heart attack) during 2007–11, we find first that the total level of health care spending has little impact on health outcomes; more important is how the money is spent. Same-day stenting, a treatment with proven effectiveness, positively predicts survival, while home health care spending does not. Second, there is strong evidence of spatial autocorrelation; without corrections this can lead to inefficient estimates and standard errors that are biased downward. Spatial autocorrelation in outcomes appears to be the consequence both of unmeasured health status and spatial correlation in new and effective technology. •We study the impact of health spending on mortality after controlling for a set of characteristics in the United States.•We adopt a variety of spatial econometric models to allow for spatial dependence.•We use data on 897,008 elderly Medicare enrolees with AMI during 2007–11, aggregated by Hospital Referral Region and year.•When predicting health outcomes, it is more informative how the money is spent rather than the level of health care spending.•There is evidence of spatial correlation that if ignored can lead to inefficient estimates and biased standard errors.