When policy meets statistics - The very real effect that questionable statistical analysis has on limiting health care access for bariatric surgery

Hypothesis: Bariatric surgery for Medicare patients must be performed in an accredited hospital that performs at least 125 cases per year. We assessed the validity of this volume threshold and its policy implications. Design: Using the 2001-2003 National Inpatient Survey, the effect of hospital volume on in-hospital mortality was statistically modeled and the effect of a 125-case per year threshold on access to bariatric surgery was calculated. We performed Monte Carlo modeling to investigate the effect random sampling has on the apparently high mortality rate for low-volume hospitals. Setting: US inpatient hospitals. Patients: Patients with hospital discharge codes indicating bariatric surgery. Main Outcome Measure: In-house mortality. Results: The observed in-hospital mortality distribution as a function of hospital volume was similar to theexpected frequency attributable to random sampling alone. A small number of excess deaths in very low-volume facilities cause statistically significant results for volumeoutcome studies. Although 74% of all bariatric surgeries are performed in high-volume centers, 73% of all hospitals currently offering these services are now classified as low volume. Conclusions: When the results of statistical analysis are used for policy determination, the consequences for patient care may be substantial. Most studies of volumeoutcome relationships rely on statistical methods that tend to amplify the effects and few fully characterize their statistical models. Despite the weak evidence for a volumeoutcome relationship for bariatric surgery, a 125-case per year threshold has been set for center-of-excellence status, which eliminates most hospitals currently providing these services and disproportionately restricts access for the poor and underinsured.