Does Practice make Perfect? Evidence from Cardiac Surgery
- Presenter:
Chair: Didem Bernard; Discussant: Bill Encinosa Mon June 5, 2006 17:15-18:45 Room 313
Background: A number of studies in the health economics literature document the presence of a correlation between provider (hospital and individual physician) volume and patient outcomes for a variety of procedures. However, few studies attempt to translate this correlation into a well established causal relationship between procedure volume and outcome. It is important to unambiguously determine causality, as the two hypotheses have contrasting implications for policy.
Contributions: This study makes a contribution to the literature by using a novel instrumental variables technique to rule out alternate explanations (the “selective referral” hypothesis) and arrive at a consistent estimate of the impact of individual learning-by-doing on quality. The use of individual (surgeon) level data also allows me to test whether skills acquired by surgeons are equally effective across hospitals.
Data: The data for this study come from the Hospital Inpatient Data Files provided by the Florida Agency for Health Care Administration (AHCA) for the years 1994-2003. I restrict my analysis to Coronary Artery Bypass Grafts (CABG) performed in this period.
Methodology: The volume term in a regression of outcomes on provider volume is endogenous if one takes into account the fact that physicians of higher quality (i.e. with better outcomes) attract more patients. An ideal instrument should help explain variation in physician procedure volume, but have no causal relationship with physician quality (i.e. outcomes of a physician’s patients). This paper uses physician exit as an exogenous identifier. The estimation strategy works as follows: if a hospital has a set of practicing surgeons, and one of them exits for exogenous reasons (say retirement) at time t, this leads to an exogenous increase in volume for the remaining physicians at time t, as the volume of the exiting physician at time t-1 gets redistributed among the remaining physicians at time t. Assuming this shock to physician volume is not correlated with unobservable determinants of physician quality, I can use it to estimate the impact of provider volume on patient outcomes.
Findings: First-stage regression results indicate that the instrument is strongly correlated with the endogenous predictor variable. The second-stage results reveal that a one-unit increase in physician procedure volume leads to a statistically significant decline in mortality of .03 percentage points, which translates to nearly a 1% drop in mortality. This is indicative of a strong learning-by-doing effect. Also, learning is not firm-specific: a one-unit increase in physician volume at any hospital has equal impact on mortality at all hospitals where the physician operates.
Conclusions: The proposed instrument helps to unambiguously establish the presence of a learning-by-doing effect for individual surgeons in CABG procedures. The findings are in favor of federal and state regulations that support consolidation of providers.
