Approximately 35.1% of live births for the state of Georgia were delivered by the cesarean delivery procedure with significant variation among hospitals. The purpose of this research was to develop a population-based hospital profiling methodology for study of the cesarean delivery procedure. This was a retrospective, observational design, using a 2012 linked dataset that included maternity deliveries from all nonfederal hospitals. The research was guided by Robson 10 Group Classification System, propensity score methodologies, and ethical precepts, for the development of hospital profiles and the study of variations in the cesarean delivery procedure. Key research questions aimed to determine whether hospital profiling methodologies differed according to risk adjustment methods and statistical techniques. Propensity score matching with stratification methods aimed to determine whether there were differences in patient treatment effects on the cesarean delivery outcome. Findings suggested there was a significant difference in hospital ranks and model effects according to the statistical technique and the risk adjustment methods applied. Propensity score matching with stratification demonstrated an increased risk of the cesarean delivery procedure across strata, with the majority of high risk patients situated in the 90th percentile ranges and questionable utilization practice among other strata. Applying profiling methodologies at the facility and population level could advance statewide quality improvement programs for the timely reduction in the variation of inappropriate utilization of the cesarean delivery procedure.

The aim of this thesis was to explore options for hospital profiling in the Slovak republic. Sacrificing breadth of the study in favor of depth, the scope of the analysis was narrowed down to one quality indicator only - mortality. In the first step a mortality prediction model was constructed in order to predict expected probability of death on the basis of a set of risk factors in order to filter away variation in hospital outcomes that is caused by other factors than quality of care. Validation of the model was performed on a validation sample of 25% of data. Discriminative ability of the final model is very high - c-statistics over 0.9. Furthermore, we verified that hospitals differ in the risk structure of their patient populations significantly - mean predicted probability of dying for hospitals differed from 0.02% to 33%. In the second step hospital profiling was performed. Standardized mortality ratios were calculated for each hospital as a difference between observed and expected number of deaths. After introduction of risk-adjustment and calculation of confidence intervals 43% of hospitals were re-classified. 30-day mortality was selected as the best indicator for hospital profiling.