High-impact complications after breast cancer surgery in the Dutch national quality registry: evaluating case-mix adjustment for hospital comparisons

Abstract

Background: Comparison of quality indicators can improve quality of care. However, case-mix adjustment is deemed essential. The aim of this study was to develop and validate case-mix adjustment models and to evaluate the effect of case-mix adjustment for the quality indicators related to complications after breast cancer surgery.

Methods: Multivariable logistic regression with backward selection (P < 0.1) was used to develop case-mix models in patients undergoing breast cancer surgery (all types, breast-conserving surgery, mastectomy with or without immediate reconstruction) in the Netherlands (NABON Breast Cancer Audit). High-impact complications were defined as Clavien Dindo grade ≥3. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), corrected for optimism with bootstrap validation. Observed-to-expected plots were used to visualize the difference between unadjusted and case-mix adjusted hospital performance (hospital shifts).

Results: In total 32 084 patients from 72 hospitals treated in 2021-2022 were included. A between-hospital variation in complication rates was observed for all surgeries (interquartile range 2.4-6.0%), breast-conserving surgery (interquartile range 1.4-3.4%), and mastectomy with (interquartile range 9.4-9.1%) and without reconstruction (interquartile range 3.3-9.7%). Of the considered variables, body mass index, smoking, multifocality and neoadjuvant therapy were weakly associated with complications. However, surgery type was strongly related to complications (AUC 0.70), resulting in noticeable hospital shifts in the quality indicator scores comprising all surgeries. After stratification for surgery type, no evident hospital shifts were observed after case-mix correction.

Conclusion: For valid comparison of complication rates after breast cancer surgery between hospitals, stratification by surgery type is crucial. Subsequently, the evaluated patient and tumour characteristics have a negligible effect on the hospital variation.

Fig. 1

Hospital variation (n = 72) in case-mix variables in Dutch patients who underwent breast cancer surgery in 2021 and 2022 (n = 32 084) The plots show the distribution of median age and BMI, along with the mean percentages of other potential case-mix variables. It is presented in a plot showing the 25th, 50^th, and 75th percentiles, along with the minimal and maximal values (range). For the variable ‘immediate reconstruction’ the plot is presented as percentage of all mastectomy patients instead of all surgeries.

Fig. 2

Observed–expected plots for all surgically treated patients and stratified for type of surgery Each dot represents a hospital. On the x-axis the O/E before case-mix adjustment is presented, on the y-axis after case-mix adjustment. As a result, deviation from the diagonal shows the effect of case-mix adjustment. As an example, in the first plot an arrow points to one of the hospitals. This hospital had an O/E rate of 1.29 before case-mix adjustment and 0.91 after case-mix adjustment. This hospital is highly influenced by patient and tumour characteristics and shows a large deviation of the diagonal line.