Percutaneous Coronary Intervention Operator Profiles and Associations With In-Hospital Mortality

Abstract

Background: Percutaneous coronary intervention is performed by operators with differing experience, technique, and case mix. It is unknown if operator practice patterns impact patient outcomes. We sought to determine if a cluster algorithm can identify distinct profiles of percutaneous coronary intervention operators and if these profiles are associated with patient outcomes.

Methods: Operators performing at least 25 annual procedures between 2014 and 2018 were clustered using an agglomerative hierarchical clustering algorithm. Risk-adjusted in-hospital mortality was compared between clusters.

Results: We identified 4 practice profiles among 7706 operators performing 2 937 419 procedures. Cluster 1 (n=3345) demonstrated case mix and practice patterns similar to the national median. Cluster 2 (n=1993) treated patients with lower clinical acuity and were less likely to use intracoronary diagnostics, atherectomy, and radial access. Cluster 3 (n=1513) had the lowest case volume, were more likely to work at rural hospitals, and cared for a higher proportion of patients with ST-segment-elevation myocardial infarction and cardiogenic shock. Cluster 4 (n=855) had the highest case volume, were most likely to treat patients with high anatomic complexity and use atherectomy, intracoronary diagnostics, and mechanical support. Compared with cluster 1, adjusted in-hospital mortality was similar for cluster 2 (estimated difference, -0.03 [95% CI, -0.10 to 0.04]), higher for cluster 3 (0.14 [0.07-0.22]), and lower for cluster 4 (-0.15 [-0.24 to -0.06]).

Conclusions: Distinct percutaneous coronary intervention operator profiles are differentially associated with patient outcomes. A phenotypic approach to physician assessment may provide actionable feedback for quality improvement.

Keywords: algorithms; health services; machine learning; percutaneous coronary intervention; research.