SOURCES OF GEOGRAPHIC VARIATION IN HEALTH CARE: EVIDENCE FROM PATIENT MIGRATION

Abstract

We study the drivers of geographic variation in US health care utilization, using an empirical strategy that exploits migration of Medicare patients to separate the role of demand and supply factors. Our approach allows us to account for demand differences driven by both observable and unobservable patient characteristics. Within our sample of over-65 Medicare beneficiaries, we find that 40-50 percent of geographic variation in utilization is attributable to demand-side factors, including health and preferences, with the remainder due to place-specific supply factors. JEL: H51, I1, I11.

Keywords: Dartmouth Atlas; Health care spending; regional variation.

Figure I

Distribution of Utilization Across HRRs Notes: Map shows the distribution of level utilization in quintiles. Lower and upper limit of each quintile are displayed in the legend. The sample is all movers and non-movers (N = 16,432,955 patient-years). Histogram displays the distribution of average utilization by HRR. We first average utilization across individuals within each HRR-year, upweighting non-movers by four, and then take a simple average within HRR across years.

Figure II

Share of Claims in Destination by Relative Year Notes: Figure shows the share of a mover’s claims located in their destination HRR, among those in either their origin or their destination HRR. The sample is all movers (N = 3,702,189 patient-years).

Figure III

Distribution of Destination-Origin Difference in Log Utilization Notes: Figure shows the distribution across movers of the difference δ̂_i in average log utilization between their origin and destination HRRs. The sample is all movers (N = 3,702,189 patient-years).

Figure IV

Change in Log Utilization By Size of Move Notes: Figure shows the change in log utilization before and after move. For each mover, we calculate the difference δ̂_i in average log utilization between their origin and destination HRRs, then group δ̂_i into ventiles. The x-axis displays the mean of δ̂_i for movers in each ventile. The y-axis shows, for each ventile, average log utilization two to five years post-move minus average log utilization two to five years pre-move. The line of best fit is obtained from simple OLS regression using the 20 data points corresponding to movers, and its slope is reported on the graph. The sample is all mover years between two and five years pre-move and between two and five years post-move (N = 1,919,137 patient-years). For comparison, we also compute the average change in log utilization for a sample of matched non-movers, which we show with the “×” marker on the graph. Specifically, for each mover in our data in each calendar year, we randomly draw a non-mover in the same year in the mover’s origin HRR who shares the mover’s gender, race, and five-year age bin; the union of the selected non-mover patient-years forms the matched sample.

Figure V

Pre-move Differences in Log Utilization Notes: Figure shows the level of pre-move log utilization for movers relative to non-movers by the size of their subsequent move δ̂_i. For each mover, we calculate the difference δ̂_i in average log utilization between their origin and destination HRRs, then group δ̂_i into ventiles. The x-axis displays the mean of δ̂_i for movers in each ventile. The y-axis shows for each ventile the average of difference in log utilization between mover and matched non-mover patient-years two to five years pre-move. In Figure IV we describe the construction of the matched sample of non-movers. The line of best fit is obtained from simple OLS regression using the 20 data points, and its slope is reported on the graph. The sample is all mover years between two and five years pre-move (N = 1,048,843 patient-years).

Figure VI

Event Study Notes: Figure shows the coefficients θ̃_r(i,t) estimated from equation (6). The coefficient for relative year −1 is normalized to 0. The dependent variable y_it is log utilization; x_it consists of indicator variables for five-year age bins. The dashed lines are upper and lower bounds of the 95 percent confidence interval. We construct this confidence interval using a two-step procedure. In the first step, for each HRR j, we construct the asymptotic distribution of ȳ_j, which is a normal distribution with mean μ_j and standard deviation σ_j calculated from the data. In the second step, we bootstrap equation (6) with 50 repetitions drawn at the patient level, making a random draw from the distribution of ȳ_j for each mover’s origin and destination to construct their δ̂_i for each repetition. The sample is all movers (N = 3,702,189 patient-years).

Figure VII

Correlates of Average Place Effects Notes: Figure shows bivariate OLS regression results (left panel) and post-Lasso multivariate regression results (right panel) of HRR-level place effects on a set of HRR-level characteristics. All covariates have been standardized to have mean zero and standard deviation one. To obtain the post-Lasso estimates, we first run a Lasso regression on the full set of covariates, with the penalty level chosen by a 10-fold cross-validation to minimize mean squared error. We then run an OLS regression on the set of covariates chosen by the Lasso regression. The sample in both panels is the 96 HRRs for which all covariates are available. Horizontal bars show 95 percent confidence intervals. Hospital Compare Score approximates hospital quality using timely and effective care measures publicly reported by CMS. Specialists Per Capita, PCP Per Capita, and Hospital Beds Per Capita count specialists, primary-care physicians, and hospital beds per thousand residents, respectively. Non-Profit Hospitals is the percent of hospitals that are non-profit. Physican preference measures are drawn from survey responses of PCPs and Cardiologists from Cutler et al. (2015); physicians classified as High Follow-up or Low Follow-up recommend follow-up visits more (or less) frequently than clinical guidelines suggest; physicians classified as Cowboy recommend care more intensive than guidelines suggest, and those classified as Comforter recommend palliative care for severely ill patients. Average Age, Percent Black, and Percent Female are computed among all patients in our baseline sample of Medicare beneficiaries. Median Family Income is the median income of households across zipcodes in each HRR taken from Census data. Average Education is the percent of the 25 and over population with a high school degree as computed from Census data. The Health variables are all the estimated patient components of a series of health measures as described in Online Appendix Section 1. The Patient Preferences variables are drawn from Cutler et al. (2015) and detail Medicare beneficiaries’ survey responses to desired care in hypothetical cases; Have Unneeded Tests and See Unneeded Cardiologists are the fraction of patients who would desire such treatment regimens; Aggressive Patient provides the fraction of patients who would like aggressive end-of-life care; and Comforter Patient provides the fraction of patients who would like palliative end-of-life care even if it shortens their life.

Figure VIII

Correlates of Average Patient Effects Notes: Figure shows bivariate OLS regression results (left panel) and post-Lasso multivariate regression results (right panel) of HRR-level patient effects on a set of HRR-level characteristics. Procedure and explanatory variables are the same as in Figure VII.