Risk adjustment for interhospital comparison of primary cesarean rates

Abstract

Objective: To create a method of controlling for case mix so that inferences could be made about variation in cesarean rates among hospitals. Methods: A total of 160,753 births from 1991 Illinois birth certificate data were analyzed. A multivariate model of characteristics independently associated with cesarean delivery was developed from a random 25% sample, validated on the other 75%, and used to create a probability of cesarean delivery for each woman. The validated model was used to calculate a predicted primary cesarean delivery rate for the 154 hospitals in Illinois that did at least 100 deliveries per year. Results: The final model included both medical and sociodemographic risk factors and predicted primary cesarean rates accurately over a full range of rates. Thirty-five hospitals (23%) had actual rates that were higher than their individual predicted 95% confidence interval (CI). Eighty-nine hospitals (58%) had actual rates within predicted CIs. Thirty hospitals (20%) had actual rates that were lower than the predicted 95% CI. Twenty-three percent of hospitals with actual rates greater than predicted rates were not in the top quartile of actual rates. Twenty-seven percent of hospitals with actual rates in the top quartile were doing cesarean deliveries appropriate for the risk status of the population served. Conclusion: Risk adjusting for hospital case mix more accurately identifies outlier hospitals than raw, unadjusted primary cesarean delivery rates. We believe that risk adjusting should be the first step in understanding variations in primary cesarean delivery rates.