Breast cancer surgery profiling by generalized additive mixed models

Breast cancer surgery is evolving from radical mastectomy to breast conservation surgery (BCS) due to the ascertainment of the therapeutic efficacy of BCS with radiation for specific cancer stages, and to the introduction of screening programmes and consequent earlier diagnosis. Despite these facts many patients are still being treated by mastectomy, making comparisons of BCS rates relevant to compare the quality of breast cancer care between hospitals. Generalized Additive Mixed Models (GAMMs) represent an effective methodological tool to study the probability of undergoing BCS for a given patient, including suitable hospital effects and, at the same time, controlling for individual factors and allowing for nonlinear age effects. We analyze data for 7021 patients treated in 19 hospitals of the Friuli Venezia Giulia Region (Italy) from 2001 to 2007. A Bayesian approach is employed for estimation, and a broad array of sensitivity analyses are performed by means of Integrated Nested Laplace Approximation (INLA). The results indicate a significantly increasing BCS rate over time, a positive effect of screening and a substantial heterogeneity between treating centers, with a moderate evidence of age-center interaction. The combination of Bayesian MCMC methods and INLA appears to be a powerful methodology for the kind of analyses explored here.