Exploratory Phenotypic Clusters in Minimally Invasive Coronary Surgery: An Unsupervised Machine Learning Study

INTRODUCTION: Minimally invasive direct coronary artery bypass (MIDCAB) is an established alternative to full sternotomy for isolated left anterior descending (LAD) revascularization. However, the impact of preoperative biological heterogeneity on early postoperative recovery in homogeneous beating MIDCAB cohorts remains unclear. This study aimed to identify latent preoperative phenotypes using unsupervised machine learning and to evaluate their association with early perioperative outcomes.
METHODS: This retrospective single-center study included consecutive patients undergoing isolated beating MIDCAB with left internal mammary artery (LIMA) to LAD grafting between August 2020 and May 2025. Emergency and redo cases were excluded. Preoperative variables (age, ejection fraction, hemoglobin, creatinine, C-reactive protein, and smoking exposure) were analyzed using Gower distance and k-medoids clustering. The optimal number of clusters was determined by silhouette analysis. Early outcomes included operative time, intensive care unit (ICU) stay, and hospital length of stay. Cluster stability was assessed using bootstrap resampling and the adjusted Rand index (ARI).
RESULTS: Forty-five patients were included. Three preoperative phenotypes were identified. Age, hemoglobin, creatinine, and smoking exposure varied across clusters, whereas ejection fraction and C-reactive protein did not differ significantly. No significant differences were observed in operative time (p=0.5099), ICU stay (p=0.1166), or hospital stay (p=0.2538). Bootstrap validation demonstrated moderate stability (mean ARI=0.43).
DISCUSSION AND CONCLUSION: Unsupervised machine learning identified distinct biological phenotypes in patients undergoing isolated beating MIDCAB; however, early postoperative outcomes were comparable across groups. Larger studies are needed to assess long-term clinical implications.

Keywords: Coronary artery bypass grafting, machine learning, minimally invasive cardiac surgery, perioperative outcomes, phenotyping.