Circulating Biomarkers in the prediction of cancer therapy related cardiac dysfunction and adverse long-term outcomes in an Irish patient population treated with Trastuzumab for stage I-III Breast cancer

Despite improvements in breast cancer survival, treatments such as anthracyclines and trastuzumab carry substantial risk for Cancer Therapy–Related Cardiac Dysfunction (CTRCD). Transthoracic Echocardiography (TTE) or Multigated Acquisition (MUGA) scans are used to monitor Left Ventricular Ejection Fraction (LVEF). However, TTE and MUGA are resource intensive and relatively insensitive. More sensitive, low-cost solutions are needed. Circulating biomarkers such as high sensitivity troponin I (hs-cTnI), N-Terminal proBNP (NT-proBNP), high sensitivity C-reactive protein (HsCRP), creatinine (Cr) and galectin-3 (Gal-3) may predict CTRCD and/or major adverse cardiovascular and cancer-related events (MACCE), but prospective studies are lacking. To address this gap, we conducted a prospective multicenter study (Cardiac Dysfunction in Patients Treated with Trastuzumab for HER-2 Positive Breast Cancer: CADY). The objective was to prospectively evaluate whether changes in cardiac biomarker levels predated cardiac dysfunction and/or MACCE in stage I-III HER2-positive breast cancer treated with trastuzumab and develop biomarker-based predictive models. Biomarker samples and clinical assessments were obtained every 6 weeks in early-stage HER2+ breast cancer patients treated with trastuzumab. Four definitions of CTRCD were considered as well as “any” CTRCD. MACCE outcomes were evaluated 16 years from study initiation. For each outcome, Cox models were developed, then models incorporating time-varying and time to event models (joint) or machine learning (long short-term memory, LSTM) techniques. Between 2007-2015, 483 women were enrolled from 14 sites in Ireland. Of 455 with complete data, CTRCD occurred in 5.9-22% depending on the definition. NTpro-BNP demonstrated the strongest associations with CTRCD. Failure of NT-proBNP values to drop by >20pg/mL or 20% at 150 days conferred >2-fold increased risk for CTRCD. Cox, joint and LSTM models incorporating NT-proBNP and clinical covariates provided superior prediction for CTRCD (AUC 0.77-0.85) versus clinical covariates alone (0.6). NT-proBNP was robust to changes in CTRCD definition. Of 295 patients with mortality data, 44 died (14.9%). Of 289 women with complete outcome data, 38 died at a median of 5±2 years from baseline over 12.4±5 years: 65.5% from breast cancer recurrence and 1 patient from a hemorrhagic cerebrovascular accident. Death was associated with CTRCD and hs-cTnI. CTRCD was associated with more than double the risk for mortality. A machine learning model incorporating hs-cTnI and clinical covariates provided an AUC of 0.94 for mortality from 80-120 days. Post-CTRCD, hs-cTnI>11ng/L was associated with >4-fold risk for mortality. An exploratory post-CTRCD Cox model incorporating hs-cTnI and clinical variables demonstrated significant differences in mortality between tertiles of risk, outperforming single threshold and clinical covariate models. In conclusion, biomarker-based models provide early and accurate prediction of CTRCD and mortality in early-stage breast cancer patients treated with trastuzumab. With further validation, the models developed from the CADY study data would allow those at high risk to avoid interruption of trastuzumab, and reduction in surveillance for those at low risk. These findings support future translation of biomarker-based risk stratification for CTRCD and mortality into clinical practice to improve cardiovascular and cancer-related outcomes.