Abstract
The monitoring of safety in cardiac surgery is a complex process, which involves many clinical, practical, methodological and statistical issues. The objective of this thesis was to measure and to compare safety in cardiac surgery in The Netherlands using the Netherlands Association for Cardio-Thoracic Surgery (NVT) database. The safety of
... read more
care is usually measured using patient outcomes. If outcomes are not available, the process and structure of care may be used. Outcomes should be adjusted for differences in patient risk and the type of intervention, and coding errors in these risk factors could affect benchmarking results. Statistical methods are presented that allow the central monitoring of risk factor frequencies: Shewhart control charts, exponentially weighted moving average (EWMA) charts and cumulative sum (CUSUM) charts. The surveillance of the overall expected risk in addition to the separate risk factors resulted in a high sensitivity to coding errors. Mortality is the most commonly used outcome measure for safety in cardiac surgery. However, there is no consensus on the appropriate period to measure mortality after cardiac surgery. Survival up to one year after cardiac surgery was studied using data from the national Cause of Death Registry. The survival curves showed a steep initial decline followed by stabilization after approximately 60 to 120 days, depending on the intervention performed. Therefore, follow-up should be prolonged to a minimum of 120 days. The comparison of safety across hospitals requires risk-adjustment. The EuroSCORE is a commonly used risk-adjustment model in cardiac surgery. A comprehensive review using 67 studies showed that the discriminative performance of the EuroSCORE was good, but that the model overestimated mortality. In high risk patients the additive model actually underestimated mortality. Therefore, the EuroSCORE should not be used for benchmarking in its original form. Hospitals are often directly compared against each other, such as seen in ranking lists. The accuracy of ranking lists was studied in the NVT database. Ranking lists of three consecutive years showed considerable reshuffling and a large overlap of the confidence intervals of hospital ranks. The use of ranking lists in cardiac surgery is strongly discouraged. The comparison of risk-adjusted outcomes may be affected by upcoding or undercoding of risk factors. We simulated upcoding and undercoding of selected variables to examine the impact on benchmarking. Upcoding in random patients (i.e. nondifferential misclassification) required substantial misclassification in order to change an underperformer into an average performing center. In high-risk patients (i.e. differential misclassification) a limited extent of upcoding was sufficient. This study emphasizes the need for careful monitoring of risk factors. Administrative data are frequently used to compare mortality across hospitals. A risk model based on data from hospital administration databases (developed using the Dutch Hospital Standardized Mortality Ratio method) was compared to a model based on the clinical data from the NVT database (the logistic EuroSCORE). The administrative model was inferior to the clinical model with respect to discrimination and calibration, which affected the benchmarking result of two hospitals. Therefore, risk-adjustment models including procedure specific clinical risk factors are recommended.
show less