Does neighborhood income predict emergency department visits?

Background/Objective: In a pay-for-performance (P4P) environment, primary care providers (PCPs) may be rewarded or penalized for their patients' use of the emergency department (ED). However, for diverse reasons that may not be under the control of their PCP, lower-income patients have higher ED visit rates. Thus, performance measures based on fixed targets for ED use may unfairly penalize practices with low-income patients. Neighborhood-level poverty may proxy individual income (a datum typically unavailable in P4P calculations) and could also capture structural forces that limit access to routine care and increase ED use rates. To the extent that ED utilization differs by income, using neighborhood income to risk-adjust ED performance assessments could reduce the disadvantage that practices with poorer patients would otherwise face. We sought to explore the strength of the association between ED use and neighborhood income. Methods: Our sample included 54,347 patients enrolled in a central Massachusetts managed-care network in 2009, with data on age, sex, race, payer (4 commercial insurers), and PCP. Census data were used to identify the median family income in each patient's neighborhood, as defined by Census tract (a geographical area containing an average of ~4000 residents). We used the 2009 Federal poverty level (FPL) for a family of 4 to categorize neighborhood income as follows: 200% FPL or less, 200-400% FPL, 400-600% FPL, and 600% FPL or higher. We included this categorical variable in multivariable logistic regression models predicting 1 or more ED visits in 2009. Results: About 16.7% of the sample had 1 or more ED visits in 2009. Adjusting for age, sex, race, and payer, patients in the poorest neighborhoods had an odds ratio (95% confidence interval) of 1.7 (1.4-2.0), and those in lower-middle income neighborhoods were also at increased risk (1.24 [1.1-1.4]) of an ED visit compared to those in the wealthiest neighborhoods. Low income was the second-most powerful predictor in this model (after age less than 1 year). There was twice as much ED use in the highest- versus lowest-decile of predicted risk (23.3 vs. 11.3%). Conclusions: Neighborhood-level income predicts ED use. Models predicting “any ED use” should incorporate publicly available neighborhood-level variables, such as income. Otherwise, targets for ED use—even if adjusted for traditional “case mix” variables—may be unfair. Research is needed on whether adjusting for neighborhood income fully removes the penalty for treating poor patients inherent in models that ignore income.