Clinical utility of the HFA-ICOS risk tool in real-world HER2-positive breast cancer patients receiving therapy

BACKGROUND: The HFA-ICOS baseline cardiovascular risk stratification framework is widely recommended to guide surveillance in patients receiving potentially cardiotoxic cancer therapies, yet real-world evidence supporting its clinical risk separation remains limited. METHODS: We conducted a multicenter retrospective study of adults with HER2-positive breast cancer treated at tertiary centers from 2016 to 2023. Baseline cardiovascular risk was classified by the HFA-ICOS framework (low to very high). The primary outcome was composite cardiotoxicity, defined by a decline in left ventricular ejection fraction (LVEF), deterioration in global longitudinal strain (GLS), ECG/arrhythmia events, or elevated cardiac biomarkers. Cardiovascular risk rates and patterns were compared across categories. Discrimination and calibration were explored, with logistic regression and Kaplan-Meier analyses performed. RESULTS: Among 687 patients, 273 (39.7%) developed composite cardiotoxicity during follow-up. Cardiotoxicity occurred across all HFA-ICOS risk categories, with overlap in event rates and timing. Discrimination analyses showed modest performance for LVEF, GLS, ECG, and composite outcomes, with AUCs of 0.51-0.55. Sensitivity was low (0.28-0.37), while specificity was moderate (0.73-0.74). Calibration analyses indicated acceptable risk prediction. Multivariable models adjusted for age, comorbidities, baseline LVEF, prior anthracycline, and radiation therapy showed that baseline HFA-ICOS category was not independently associated with cardiotoxicity (aOR: 0.88; 95% CI: 0.56-1.37). Kaplan-Meier curves showed overlapping event-free survival across risk groups. CONCLUSIONS: In this real-world cohort of HER2-positive breast cancer patients, baseline HFA-ICOS stratification showed limited ability to clearly distinguish cardiotoxicity risk across categories. These findings suggest that baseline risk assessment alone may be insufficient for individualized prediction and should be complemented by dynamic, on-treatment surveillance strategies.