Performance of the STS Risk Score in Predicting Outcomes in Impella-Supported High-Risk Percutaneous Coronary Intervention: Insights From the PROTECT III Study

Background Patients undergoing high-risk percutaneous coronary intervention (HRPCI) with Impella often present with similar or worse anatomical complexity than surgical revascularization candidates, yet the performance of surgical risk models in this setting is unknown. We evaluated Society of Thoracic Surgeons (STS) score performance in predicting outcomes in the PROTECT III study. Methods PROTECT III enrolled 1237 patients undergoing elective or urgent HRPCI with Impella support from March 2017 to March 2020. Patients were stratified into low (STS 8) risk categories. Primary endpoints were 30-day major adverse cardiovascular and cerebrovascular events (MACCE) and mortality. Model discrimination was evaluated with receiver operating characteristic analysis; calibration was assessed with observed-to-expected (O/E) ratios and Hosmer–Lemeshow testing. Secondary endpoints included 90-day MACCE and 1-year mortality. Results Of 1237 patients, 728 patients were low-risk, 316 intermediate, and 193 high-risk based on STS score. Higher STS risk was associated with a higher number of comorbidities and more left main and multivessel disease. At 30 days, MACCE rose with risk (low 5.1%, intermediate 10.1%, high 17.6%; P < .001), driven by mortality (3.5%, 8.6%, 16.8%; P < .001). Patterns persisted at 90 days. The STS score showed good discrimination for 30-day percutaneous coronary intervention mortality (C-index 0.72; 95% CI:, 0.66-0.77) with acceptable global calibration (Hosmer–Lemeshow P = .13), comparable to the HRPCI-specific risk score. However, it underestimated mortality at 30 days (O/E 1.23), suggesting that existing scores do not capture the high-risk nature of this population. Conclusions The STS score provides reasonable discrimination in Impella-supported HRPCI but poorly predicts event rates, particularly underestimating mortality. These results underscore the need for modification of existing models and validation of HRPCI-specific tools.