Case-mix-adjusted benchmarking of long-term breast cancer-specific mortality: Insights from a nationwide Swedish registry.

backgroundMeaningful comparisons of breast cancer (BC) survival across hospitals require appropriate case-mix adjustment. While such adjustment is routine for short-term outcomes, it remains less established for long-term BC-specific mortality (BCSM). We developed case-mix models to evaluate whether robust hospital-level benchmarking of 10-year BCSM can be achieved using routinely collected registry data.

methodsUsing the nationwide Swedish BC registry linked to population registers, we developed full and reduced logistic regression models for 10-year BCSM. The full model included age, tumour, socioeconomic, and comorbidity variables while the reduced model used routinely collected age and tumour variables. Model performance was assessed via discrimination and calibration. The model impact on hospital benchmarking was evaluated through observed-to-expected mortality ratios at the hospital level, comparing expected mortality estimates and shifts in outlier status.

resultsThe study included 10,888 patients. Age, stage, and grade were the strongest predictors, while comorbidity and socioeconomic variables added limited incremental predictive value. Both the full and reduced 10-year BCSM models demonstrated similar discrimination and near-perfect calibration. Furthermore, the full and reduced models yielded highly comparable expected mortality rates across 12 hospitals (mean absolute difference 0.6 percentage points), with consistent identification of hospital outliers in all but one hospital.

conclusionsRoutinely collected data on age and tumour characteristics appear to offer a feasible foundation for fair benchmarking of long-term BCSM in a publicly-funded healthcare system. While external validation in contemporary and international cohorts is an important next step, these findings might be relevant to other countries with strong cancer registry infrastructure.