[Skip to Content]
[Skip to Content Landing]
February 2, 2005

Surgical Mortality, Hospital Quality, and Small Sample Size

JAMA. 2005;293(5):553. doi:10.1001/jama.293.5.553-b

To the Editor: Dr Dimick and colleagues1 recently emphasized the challenge of measuring surgical performance in the face of the pervasive problem of small sample size. We additionally note that classic statistics, developed to test differences between random samples, do not apply to hospital populations. The 95% confidence interval (CI), a frequent measure of precision that reflects variability among samples, is the range within which means are expected to fall 95 times if sampling is repeated 100 times. However, there is never a way of repeating a 1-year experience for an operation 100 times without changing conditions, population, and physicians. The CI for the specific operative mortality of an individual hospital is usually wide because of small numbers, so that even in a recognized example of poor performance (the Bristol case for pediatric cardiac surgery, based on a careful analysis of data from many years2), the CI for a specific operation at the problematic unit would often cross the mean mortality rate in other hospitals. Used as such, the CI can actually give the false confidence of adequate performance.