Risk-Adjusting Key Outcome Measures in a Clinical Quality PCI Registry: Development of a Highly Predictive Model Without the Need to Exclude High-Risk Conditions

Abstract

Objectives: This study sought to determine the most risk-adjustment model for 30-day all-cause mortality in order to report risk-adjusted outcomes. The study also explored whether the exclusion of extreme high-risk conditions of cardiogenic shock, intubated out-of-hospital cardiac arrest (OHCA), or the need for mechanical ventricular support affected the model's predictive accuracy. Background: Robust risk-adjustment models are a critical component of clinical quality registries, allowing outcomes to be reported in a fair and meaningful way. The Victorian Cardiac Outcomes Registry encompasses all 30 hospitals in the state of Victoria, Australia, that undertake percutaneous coronary intervention. Methods: Data were collected on 27,544 consecutive percutaneous coronary intervention procedures from 2014 to 2016. Twenty-eight patient risk factors and procedural variables were considered in the modeling process. The multivariable logistic regression analysis considered derivation and validation datasets, along with a temporal validation period. Results: The model included risk-adjustment for cardiogenic shock, intubated OHCA, estimated glomerular filtration rate, left ventricular ejection fraction, angina type, mechanical ventricular support, ≥80 years of age, lesion complexity, percutaneous access site, and peripheral vascular disease. The C-statistic for the derivation dataset was 0.921 (95% confidence interval: 0.905 to 0.936), with C-statistics of 0.931 and 0.934 for 2 validation datasets reflecting the 2014 to 2016 and 2017 periods. Subgroup modeling excluding cardiogenic shock and intubated OHCA provided similar risk-adjusted outcomes (p = 0.32). Conclusions: Our study has developed a highly predictive risk-adjustment model for 30-day mortality that included high-risk presentations. Therefore, we do not need to exclude high-risk cases in our model when determining risk-adjusted outcomes.