Comparative performance of biomarker-based and biomarker-free HFA-ICOS models for predicting anthracycline-related cardiotoxicity

Background Cardio-oncology patients face an increased risk of chemotherapy-related cardiotoxicity (CTRCD). The Heart Failure Association (HFA)/International Cardio-Oncology Society (ICOS) score is a validated tool for cardiovascular (CV) risk stratification. Although cardiac biomarkers play a key role in CTRCD risk assessment and diagnosis, their availability in routine clinical practice remains limited. Purpose We aimed to compare the performance of the full HFA-ICOS model, incorporating cardiac biomarkers, with biomarker-free models, including a simplified clinical model and a modified model integrating baseline global longitudinal strain (GLS), in anthracycline-treated patients. Methods We retrospectively analyzed 316 patients treated with anthracyclines across three centers. Baseline CV risk was assessed using the HFA-ICOS score. Three models were evaluated: (1) the full HFA-ICOS model including cardiac biomarkers (troponin and NT-proBNP), (2) the mini-HFA-ICOS model excluding biomarkers, and (3) a modified biomarker-free HFA-ICOS model including baseline GLS. CTRCD was defined according to the 2022 ESC guidelines. Discriminative performance was assessed using recever operating characteristic (ROC) curves, and comparisons between models were performed using DeLong’s test. Risk reclassification analyses were also conducted. Results Asymptomatic (27%, 39%, 46%, and 60%; p 0.017) and symptomatic CTRCD (1.3%, 4.6%, 13%, and 40%; p < 0.001) were more frequent in higher risk groups. For all CTRCD definitions, the full HFA-ICOS model showed higher discrimination than the mini model (AUC 0.640 vs 0.604; p = 0.004). When baseline GLS was incorporated into the biomarker-free model, discrimination was numerically improved compared with the mini model (AUC 0.643 vs 0.604; p = 0.0004) and was comparable to the full HFA-ICOS model (AUC 0.643 vs 0.640; p = 0.75), although differences in AUC did not reach statistical significance. When excluding mild asymptomatic CTRCD to minimize biomarker-related circularity, the modified biomarker-free HFA-ICOS model including GLS demonstrated discrimination comparable to the full and mini models (AUC 0.622 vs 0.617; p = 0.87; AUC 0.622 vs 0.605; p = 0.39). Compared with the full HFA-ICOS model, the mini model resulted in risk reclassification in 7.6% of patients, entirely driven by downgrading, whereas inclusion of GLS reduced discordance with the full model (5.1%) and reclassified 3.8% of patients compared with the mini model. Conclusions While integration of cardiac biomarkers improves CTRCD risk stratification within the HFA-ICOS framework, baseline GLS provides meaningful incremental value when biomarkers are unavailable or when biomarker-independent CTRCD definitions are applied. GLS may therefore represent a pragmatic alternative to biomarkers for refining cardiotoxicity risk assessment in routine cardio-oncology practice.