Technology Diffusion and Productivity Growth in Health Care

Abstract

We draw on macroeconomic models of diffusion and productivity to explain empirical patterns of survival gains in heart attacks. Using Medicare data for 2.8 million patients during 1986-2004, we find that hospitals rapidly adopting cost-effective innovations such as beta blockers, aspirin, and reperfusion, had substantially better outcomes for their patients. Holding technology adoption constant, the marginal returns to spending were relatively modest. Hospitals increasing the pace of technology diffusion ("tigers") experienced triple the survival gains compared to those with diminished rates ("tortoises"). In sum, small differences in the propensity to adopt effective technology lead to wide productivity differences across hospitals.

Figure 1. Survival Rates by Year and Diffusion Quintiles, 1986–2004

Notes: The Figure reports risk-adjusted 1-year AMI survival rates annually from 1986–2004 for hospitals in each adoption quintile. Adoption quintiles were defined based on the common factor estimated from a 1-factor model of hospital use of aspirin, β blockers, and reperfusion (see Table 1). Risk-adjusted survival was derived from Medicare claims data on 2.8 million AMI admissions from 1986–2004.

Figure 2

Interpreting the Evidence on Survival and Health Outcomes: “Flat of the Curve” vs. Productivity Differentials

Figure 3. Estimated 2004 Production Function for One-Year AMI Survival: Low Diffusion (Solid), High Diffusion for Factor A (Dotted) and High Diffusion for Factor B (Dashed)

Note: The figure plots predicted one-year survival, based on regression analysis reported in Appendix Table A.2 with hospital fixed effects, for three hypothetical hospitals: A hospital that is one standard deviation below average on both adoption factors (Low Factor A, Low Factor B), a hospital that is one standard deviation below average on Factor B but one standard deviation above average on Factor A (High Factor A, Low Factor B), and a hospital that is one standard deviation below average on Factor A but one standard deviation above average on Factor B (Low Factor A, High Factor B). While the slope and shape of the production function comes directly off the estimated coefficients, the intercepts for each of these three curves are derived from a between-hospital regression of the hospital fixed effects regressed on the two factors, their squares, and their interactions.