A Bayesian approach to learning curves for major adverse cardiovascular events after using a prospective national multi-registry database

Transfemoral carotid artery stenting (tfCAS) is a technically demanding procedure performed to treat atherosclerotic carotid artery disease with a potential risk of major adverse cardiovascular events (MACE). As physicians gain experience, their complication rates tend to decrease, but methods for evaluating surgical learning curves may fail to account for the diversity of patients treated or the correlation structure due to the multiple procedures performed by each physician. Using the Society for Vascular Surgery Vascular Quality Initiative registry, we will conduct a retrospective longitudinal study including tfCAS procedures performed from 2005 to 2024. This study aims to estimate physician learning curves for MACE and operating time following tfCAS using a comprehensive Bayesian modeling framework, while accounting for patient-level heterogeneity. We evaluate how patient factors modify the slope and shape of learning trajectories. Bayesian logistic regression models incorporating Gaussian process priors will be used to flexibly estimate nonlinear learning curves using each physician’s cumulative procedure count. Models will be developed for combined and stratified (symptomatic vs. asymptomatic) cohorts, adjusting for demographic and clinical covariates. A structured initial data analysis (IDA) plan will evaluate data quality, variable distributions and correlation structures, prior to final model specification. This is guided by contemporary recommendations for IDA and statistical analysis plans for observational studies for transparency and reproducibility.