More Equal than Others - Regional Variation in Supply of Health Care and Mortality

Do higher health care expenditures translate into better health outcomes? Answering this question has proven difficult, as variation in (local) health care spending is likely due to a combination of demand and supply factors.  For example, poor patient health may increase expenditures through higher demand for health care.

In this paper, we implement a two-step approach to account for endogeneity in health care expenditures, using detailed patient data covering the entire Norwegian population in 2008-2012. First, we follow Finkelstein et al (2014) and use migration data to decompose regional variation in health care expenditures into variation due to patient demand and local variation of supply of health care. Formally, we estimate the following model:

            y_ijt=a_i + g_j + T_t + R_r(i,t)+ X_itb+e_ijt,

where y_ijt is individual i’s yearly utilization of health care, a_i are patient fixed effects, g_j are hospital area fixed effects, T_t and R_r(i,t) are fixed effects for calendar year and relative year since move, respectively, and X_it is a set of 5-year age categories. Patient and hospital area fixed effects are identified by having migration of patients between hospital areas. Our model hinges on the assumption that trends in utilization of health care do not vary systematically with the migrants' origin or destination. This assumption is evaluated directly in an event-study framework where we find that estimates are consistently zero for years prior to the move, lending support to our model. 

Norway's centralized, single-payer system, with hospital physicians employed on a fixed salary (rather than on fee-for-service or capitation-based contracts), may lead to less variation in health care supply compared to the US system. Even so, we find that supply factors account for roughly one third of total variation in health care expenditures, while the rest is explained by patient characteristics.

In the second step, we link the estimated supply indicator to data on cause-specific mortality. We show that higher expenditures significantly reduce age-adjusted mortality. A one standard deviation increase in hospital expenditures reduces mortality in the full population by 0.15 per 1000, corresponding to a 1.8% reduction compared to the mean. The negative effects are concentrated among persons who have been in contact with a hospital during the period of study. When disaggregating results by cause-of-death, the estimated effects are stronger for conditions where a high share of patients are hospitalized (cancer), than for to conditions where hospitalization rates prior to death are lower (accidents and other violent deaths). Overall, our results indicate that there is significant, supply-driven geographical variation in health care costs, which has statistically significant effects on health outcomes.