Nomogram for Predicting Risk of Cancer Therapy–Related Cardiac Dysfunction in Patients With Human Epidermal Growth Factor Receptor 2–Positive Breast Cancer

Abstract

Background

Cancer therapy–related cardiac dysfunction (CTRCD) is an important treatment‐limiting toxicity for patients with human epidermal growth factor receptor 2 (HER2)–positive breast cancer that adversely affects cancer and cardiovascular outcomes. Easy‐to‐use tools that incorporate readily accessible clinical variables for individual estimation of CTRCD risk are needed.

Methods and Results

From 2004 to 2013, 1440 patients with stage I to III HER2‐positive breast cancer treated with trastuzumab‐based therapy were identified. A multivariable Cox proportional hazards model was constructed to identify risk factors for CTRCD and included the 1377 patients in whom data were complete. Nine clinical variables, including age, race, body mass index, left ventricular ejection fraction, systolic blood pressure, coronary artery disease, diabetes, arrhythmia, and anthracycline exposure were built into a nomogram estimating risk of CTRCD at 1 year. The nomogram was validated for calibration and discrimination using bootstrap resampling. A total of 177 CTRCD events occurred within 1 year of HER2‐targeted treatment. The nomogram for prediction of 1‐year CTRCD probability demonstrated good discrimination, with a concordance index of 0.687. The predicted and observed probabilities of CTRCD were similar, demonstrating good model calibration.

Conclusions

A nomogram composed of 9 readily accessible clinical variables provides an individualized 1‐year risk estimate of CTRCD among women with HER2‐positive breast cancer receiving HER2‐targeted therapy. This nomogram represents a simple‐to‐use tool for clinicians and patients that can inform clinical decision‐making on breast cancer treatment options, optimal frequency of cardiac surveillance, and role of cardioprotective strategies.

Nonstandard Abbreviations and Acronyms

CTRCD

cancer therapy–related cardiac dysfunction

HER2

human epidermal growth factor receptor 2

SBP

systolic blood pressure

Clinical Perspective.

What Is New?

• Strategies to estimate individual risk of cancer therapy–related cardiac dysfunction are needed to help guide clinical decision‐making for patients with human epidermal growth factor receptor 2–positive breast cancer given the variable toxicity profiles of currently available treatment regimens.

• Age, race, body mass index, baseline left ventricular ejection fraction, systolic blood pressure, coronary artery disease, diabetes, arrhythmia, and anthracycline treatment were variables associated with cancer therapy–related cardiac dysfunction during human epidermal growth factor receptor 2–targeted treatment and used to develop a nomogram to predict 1‐year risk of cancer therapy–related cardiac dysfunction.

What Are the Clinical Implications?

• This risk nomogram, based on 9 readily available variables, may help clinicians to tailor therapy for human epidermal growth factor receptor 2–positive breast cancer that balances antitumor benefit with cardiovascular risk and inform the appropriate cardiac surveillance and prevention strategies during and after treatment.

• The proposed risk nomogram offers a comprehensive and easy‐to‐use tool for clinicians to predict 1‐year risk of cancer therapy–related cardiac dysfunction during human epidermal growth factor receptor 2–targeted breast cancer therapy and warrants further validation.

Cancer therapy–related cardiac dysfunction (CTRCD) is a primary treatment‐limiting adverse effect for patients with breast cancer receiving cardiotoxic cancer treatment, such as anthracyclines or human epidermal growth factor receptor 2 (HER2)–targeted therapy. CTRCD, manifested by left ventricular systolic dysfunction or clinical heart failure, is a leading cause of premature interruption of cancer treatment, adversely impacts cancer outcomes, and results in long‐term impairment of cardiorespiratory fitness. ¹ , ² , ³ , ⁴ , ⁵ Multiple treatment‐ and patient‐specific variables known to influence risk of CTRCD have been identified. Anthracyclines are known to cause CTRCD in a dose‐dependent manner, and CTRCD risk is further increased when anthracyclines are administered sequentially with HER2‐targeted therapy. ⁶ , ⁷ , ⁸ Cardiovascular risk factors, such as age, hypertension, and obesity, are recognized to be important predictors for adverse cardiovascular outcomes among patients with breast cancer. ⁹ , ¹⁰ , ¹¹ Our group and others have noted that the risk of CTRCD varies by race, even after adjusting for differences in cancer treatments and traditional cardiovascular risk factors. ¹² , ¹³ Although these factors have been associated with CTRCD risk, a comprehensive and easy‐to‐use tool for estimating individual CTRCD risk is currently unavailable. Such a tool could be used to guide individualized selection of cancer treatment, inform the appropriate cardiac surveillance strategy during and after cancer treatment, and identify patients who may benefit from primary prevention with cardioprotective agents.

A nomogram is a simple graphical depiction of a statistical model that provides an overall probability of a specific clinical outcome for an individual patient. ¹⁴ For example, instead of considering anthracycline chemotherapy as a CTRCD risk factor associated with a hazard ratio (HR) of 2.7, the nomogram integrates this factor along with other clinical variables to estimate an individual's absolute risk. Nomograms are frequently used by oncologists and have the advantage of combining multiple variables to estimate risk of a clinical outcome for an individual patient. ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ The purpose of this study was to develop and validate a nomogram using readily accessible clinical variables to predict the probability of CTRCD at 1 year in women with HER2‐positive breast cancer treated with HER2‐targeted therapy.

METHODS

The data for this study are available from the corresponding author on reasonable request. The study was approved as a retrospective research protocol by the Memorial Sloan Kettering Cancer Center (New York, NY) Institutional Review Board under a waiver of informed consent.

Study Population

Consecutive patients with HER2‐positive breast cancer treated at our institution between January 2004 and December 2013 were identified. Inclusion criteria for this study included female sex, stage I to III disease, treatment with trastuzumab‐based therapy (with or without anthracyclines), and availability of key clinical data elements, including baseline vital signs and ≥2 echocardiograms during treatment (baseline and ≥1 follow‐up).

Clinical Data Collection

Demographic and clinical characteristics were collected from the electronic medical record, including age at treatment, race (self‐reported), cancer treatment, baseline cardiovascular risk factors (ie, hypertension, diabetes, hyperlipidemia, and current or former smoking), and baseline concomitant cardiac medications. Self‐reported race was categorized as Black or non‐Black. Participants that self‐identified as American Indian, Asian, Pacific Islander, White, or other were grouped together as “non‐Black”. To gain insight on the role of hypertension as a risk factor for CTRCD, vital signs measured within 30 days of starting trastuzumab were also collected.

The primary outcome of interest was CTRCD, defined as an absolute decline of left ventricular ejection fraction (LVEF) ≥10% to <53% or ≥16% from baseline (pretreatment). ¹⁵ , ¹⁹ , ²⁰ CTRCD events were identified on the basis of oncology and cardiology clinic assessments, review of LVEF measurements from any cardiac imaging modalities available in the electronic medical record (ie, echocardiogram, multigated acquisition scan, or cardiac magnetic resonance imaging), and detailed review of pharmacy administration records to identify cardiac‐related interruptions of anthracycline or trastuzumab therapy. CTRCD events were further characterized on the basis of presence or absence of clinical heart failure (New York Heart Association class III or IV) symptoms. Cases of CTRCD were adjudicated through detailed review of all available data by a board‐certified cardiologist (AFY).

Statistical Analysis

This study followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis reporting guideline for prognostic studies. ²¹ The primary end point was 1‐year CTRCD‐free survival, which was defined as the time from start of HER2‐targeted treatment to a CTRCD event. Patients with no CTRCD at 1 year were censored. No deaths were observed during the 1‐year follow‐up for patients in this cohort. The probabilities of CTRCD were estimated using the Kaplan‐Meier method. Univariate and multivariable Cox proportional hazards models were used to evaluate the association between clinical variables and 1‐year risk of CTRCD. Baseline characteristics included in the model were chosen a priori based on clinical relevance, including age at treatment, body mass index (<30 versus ≥30 kg/m²), anthracycline treatment, hypertension, diabetes, coronary artery disease, arrhythmia (ie, atrial fibrillation/flutter, supraventricular tachycardia, sick sinus syndrome, or frequent premature ventricular contractions), smoking, baseline pretreatment LVEF (>60% versus ≤60%), race (Black versus non‐Black), and baseline systolic blood pressure (SBP; <130 versus ≥130 mm Hg). Baseline SBP was defined as the median SBP within 30 days of the trastuzumab start date. Clinical variables included in the final regression model were chosen on the basis of clinical and statistical significance. A total score was calculated by summing the weighted points corresponding to each covariate, and the 1‐year predicted CTRCD probability was calculated for each patient using the Cox regression model underlying the nomogram. Each patient was subsequently classified for CTRCD risk by quartile of the total score.

The concordance index was estimated for the entire population to assess discriminative ability of the nomogram. ²² The concordance index represents a probability that given a randomly selected patient pair, the patient who developed CTRCD had a higher predicted probability of CTRCD. ²³ Interpretations of the value are identical to the area under the receiver operating characteristic curve, with 1.0 corresponding to perfect concordance, 0.5 corresponding to random chance, and 0.0 corresponding to perfect discordance for all patients. ²⁴ The area under receiver operating characteristic curve was used to assess the accuracy of the prediction model in 1‐year CTRCD prediction.

The predictive model was validated using 200 bootstrap samples to reduce the potential bias from overfitting. ²⁵ Specifically, a model was built on a bootstrap sample and then evaluated on the original cohort. Two model performance indexes were calculated on the basis of the bootstrap sample and original cohort. The process was repeated 200 times, which produced 200 indexes. The difference between the index built on the original cohort and the average value of the 200 indexes was considered as the bias‐corrected concordance probability, ²⁶ which addresses how well the nomogram would perform to discriminate among patients in the future.

The calibration of the nomogram was evaluated using a calibration curve, which was created by plotting the predicted CTRCD probability at 1 year against the observed probability, as calculated by the Kaplan‐Meier method. Two hundred bootstrap samples were used to prevent against overfitting. If the points were on or close to the 45‐degree line, the model was considered to have good calibration. If the points were lower than the 45‐degree line, the model was considered to overestimate outcome probabilities. If the points were higher than the 45‐degree line, the model was considered to underestimate outcome probabilities.

The net benefit of the nomogram was estimated by decision curve analysis. ²⁷ , ²⁸ Net benefit is calculated as the benefit of an intervention among the true positives minus harm of the false positives. Intervention refers to any clinical action that would be considered for a patient at high risk from the nomogram. In the current study of patients with breast cancer, possible interventions include increased cardiac surveillance, prophylaxis with cardioprotective medications, or use of alternative non‐cardiotoxic cancer treatments. A decision curve is a plot of net benefit (y axis) as a function of a threshold probability (x axis) of intermediate‐ to high‐risk disease that would prompt an intervention. To assess the clinical utility of the risk nomogram, the decision curve for the nomogram was compared with 2 default strategies: (1) “treat all,” in which all patients receive the intervention regardless of risk; or (2) “treat none,” in which no patients receive the intervention regardless of risk. The nomogram was deemed to be clinically useful if the decision curve demonstrated a higher net benefit compared with the alternative treatment scenarios.

All statistical analyses were performed using SAS software version 9.4 (SAS Institute, Inc) or R software version 4.1.3 (R Foundation for Statistical Computing) with the Hmisc, rms, and dcurves packages. ²⁶ All statistical tests were 2 tailed, and P<0.05 was considered statistically significant.

RESULTS

Clinical Features and Characteristics

Of 1440 women with breast cancer treated with HER2‐targeted therapy during the study period, 63 were excluded from analysis because of incomplete data (19 for insufficient baseline clinical characteristics and 44 for insufficient echocardiogram data). The final sample size of 1377 patients was included in this analysis. Baseline characteristics are presented in Table 1. The mean age of the study population was 51.8±11.2 years. A total of 1071 (77.8%) patients received anthracycline‐based chemotherapy (1066 for a current breast cancer diagnosis and 5 for a prior breast cancer diagnosis), and all received HER2‐targeted therapy for stage I (36.9%), II (39.6%), or III (23.5%) invasive breast cancer. The median (interquartile range) trastuzumab dose was 106 (104–110) mg/kg. A total of 941 (68.3%) patients received breast radiotherapy.

Table 1.

Clinical Features of the Study Population

Characteristic	Overall cohort (N=1377)
Age, mean±SD, y	51.8±11.2
Body mass index, mean±SD, kg/m2	26.8±6.0
Body mass index, n (%)
<25 kg/m2	632 (45.9)
25–29 kg/m2	429 (311.2)
≥30 kg/m2	316 (22.9)
Estrogen receptor positive, n (%)	872 (63.3)
Progesterone receptor positive, n (%)	660 (48.0)
Stage, n (%)
I	508 (36.9)
II	545 (39.6)
III	324 (23.5)
Anthracyclines (prior/current), n (%)	1071 (77.8)
Trastuzumab dose, median (IQR), mg/kg	106 (104–110)
Radiation therapy, n (%)	941 (68.3)
Baseline LVEF, median (IQR), %	66 (63–70)
Baseline LVEF, n (%)
<60%	163 (11.8)
≥60%	1214 (88.2)
Race, n (%)
Black	164 (11.9)
Non‐Black	1213 (88.1)
Hypertension, n (%)	323 (23.5)
Baseline systolic blood pressure, n (%)
<130 mm Hg	1029 (74.7)
≥130 mm Hg	348 (25.3)
Diabetes, n (%)	99 (7.2)
Hyperlipidemia, n (%)	236 (17.1)
Coronary artery disease, n (%)	22 (1.6)
Arrhythmia, n (%)	17 (1.2)
Smoking (current or former), n (%)	463 (33.6)
Cardiac medications at baseline, n (%)
β‐Blockers	112 (8.1)
Calcium channel blockers	77 (5.6)
Renin‐angiotensin‐aldosterone antagonists	2206 (15.0)
Statins	180 (13.1)

IQR indicates interquartile range; and LVEF, left ventricular ejection fraction.

The median (interquartile range) LVEF at baseline was 66% (63%–70%). Cardiovascular risk factors were prevalent: 323 (23.5%) had hypertension, 99 (7.2%) had diabetes, 236 (17.1%) had hyperlipidemia, and 463 (33.6%) were current or former smokers. Preexisting cardiovascular conditions such as coronary artery disease and arrhythmia were uncommon, present in 22 (1.6%) and 17 (1.2%) patients, respectively. A total of 177 (12.9%) women developed CTRCD: 143 (80.8%) presented with significant LVEF decline without heart failure, and 34 (19.2%) presented with clinical heart failure (New York Heart Association class III/IV). The median (interquartile range) time interval between start of trastuzumab and CTRCD was 176 (93–269) days.

Derivation of the Nomogram for Prediction of CTRCD

Multivariable HRs for prognostic factors used to build the nomogram were calculated, shown in Table 2. Treatment with anthracycline chemotherapy increased the risk of CTRCD (HR, 2.75 [95% CI, 1.74–4.38]). Risk of CTRCD was associated with increasing age (HR, 1.02 [95% CI, 1.00–1.04], per 1‐year increase), baseline LVEF <60% (HR, 1.85 [95% CI, 1.27–2.69]), and baseline SBP ≥130 mm Hg (HR, 1.52 [95% CI, 1.09–2.13]). Coronary artery disease (HR, 2.42 [95% CI, 1.16–5.04]) and history of arrhythmia (HR, 3.07 [95% CI, 1.22–7.74]) were also associated with higher CTRCD risk. Black women had a near 2‐fold increased risk of CTRCD (HR, 1.85 [95% CI, 1.27–2.70]) compared with other women.

Table 2.

Multivariable HRs for the Relationship Between Prognostic Risk Factors and 1‐Year CTRCD

Variable	No. of patients	No. of CTRCD events	HR (95% CI)	P value	Points assigned
Age (per 1‐y increase)			1.02 (1.00–1.04)	0.01	1.4×(Age−20)
BMI, kg/m2
≥30	316	63	1.31 (0.94–1.82)	<0.001	19
<30	1061	114	Reference
Race*
Black	164	37	1.85 (1.27–2.70)	0.001	44
Non‐Black	1213	140	Reference
Anthracyclines
Current or prior	1071	153	2.75 (1.73–4.38)	<0.0001	73
None	306	24	Reference
Baseline systolic blood pressure, mm Hg
≥130	348	66	1.52 (1.09–2.13)	0.01	30
<130	1029	111	Reference
LVEF, %
<60	163	35	1.85 (1.27–2.69)	0.001	44
≥60	1214	142	Reference
Coronary artery disease
Yes	22	8	2.42 (1.16–5.04)	0.02	63
No	1355	169	Reference
Arrhythmia
Yes	17	5	3.07 (1.22–7.74)	0.02	80
No	1360	172	Reference
Diabetes
Yes	99	22	1.19 (0.73–1.94)	0.49	13
No	1278	155	Reference

BMI indicates body mass index; CTRCD, cancer therapy–related cardiac dysfunction; HR, hazard ratio; and LVEF, left ventricular ejection fraction.

^*Race was determined by self‐report. Participants that self‐identified as American Indian, Asian, Pacific Islander, White, or other were grouped together as “non‐Black”.

A nomogram based on these findings is shown in Figure 1. Total points for each patient are calculated for 9 baseline variables from the nomogram, which corresponds to a predicted 1‐year probability of CTRCD. To use this nomogram, all 9 variables must be available. As an example, a 60‐year‐old (56 points), obese (19 points), Black (44 points) woman with a baseline LVEF of 65% (0 points), without history of coronary artery disease, diabetes, or arrhythmia (0 points), and SBP of 150 mm Hg (30 points) treated with an anthracycline‐based chemotherapy regimen (73 points) has about a 35% risk of developing CTRCD by 1 year (total points, 223). The 1‐year probability of CTRCD by quartile of total points was 3.2% (quartile 1), 9.9% (quartile 2), 14.6% (quartile 3), and 23.8% (quartile 4) (Table 3). Individuals comprising the lowest quartile of total points represented 6% of CTRCD events, whereas those in the highest quartile of total points represented 46% of CTRCD events.

Figure 1. Nomogram of risk model for predicting 1‐year probabilities of cancer therapy–related cardiac dysfunction (CTRCD) after human epidermal growth factor receptor 2–targeted therapy.

To estimate risk of CTRCD, an individual patient's baseline characteristics are plotted on each variable axis, and a vertical line is drawn to the points’ axis to determine the points for that variable. The points for all 9 variables are added, the sum is located on the “Total Points” axis, and a vertical line is drawn to the bottom axis to obtain the 1‐year probability of CTRCD. The total score can be calculated as follows: {1.4×[age−20]}+{19×[body mass index (BMI)≥30 kg/m²]}+{44×I[Black]}+{73×I[current/prior anthracyclines]}+{30×I[baseline systolic blood pressure (SBP) ≥130 mm Hg]}+{44×I[baseline left ventricular ejection fraction (LVEF) <60%]}+{63×I[coronary artery disease (CAD)]}+{80×I[arrhythmia]}+{13×I[diabetes]}, where I[] denotes the indicator function equal to 1 if the condition within the brackets is present, and 0 otherwise.

Table 3.

Risk of CTRCD According to Quartiles of the Nomogram Total Points

Quartile	Nomogram total points, median (range)	No. of people at risk	No. of CTRCD events	1‐y CTRCD event rate (95% CI)
1	86.2 (18.0–102.6)	345	11	3.19 (1.60–5.63)
2	113.8 (102.7–122.98)	344	34	9.88 (6.94–13.54)
3	136.9 (122.99–156.6)	343	50	14.58 (11.02–18.76)
4	181.5 (156.7–322.7)	345	82	23.77 (19.37–28.62)

CTRCD indicates cancer therapy–related cardiac dysfunction.

Validation and Net Benefit of the Nomogram

Performance of the nomogram was evaluated by calibration and discrimination. The CTRCD risk nomogram was internally validated with 200 bootstrap resamples. The 200 repetitions of bootstrap sample corrections provided an estimated concordance probability of 0.687. This concordance index demonstrates 69% probability that of 2 randomly selected patients, the patient who has higher risk of CTRCD will have a higher probability of developing CTRCD than the patient with lower risk of CTRCD. The calibration plots for predicting 1‐year CTRCD using the risk nomogram in the validation data set is shown in Figure 2. The plot of the predicted CTRCD outcome versus the observed CTRCD outcome approximates a 45‐degree diagonal, demonstrating good model calibration. The performance of the nomogram appears to be accurate, within 3% of the actual outcome. The receiver operating characteristic curve for the nomogram prediction model is shown in Figure 3. Area under receiver operating characteristic curve of the nomogram in 1‐year CTRCD prediction was 0.68.

Figure 2. Calibration plot of 1‐year cancer therapy–related cardiac dysfunction (CTRCD) nomogram.

The x axis displays the nomogram‐predicted probability of 1‐year CTRCD‐free survival. Patients were grouped by quartiles of predicted risk (first quartile, <103 points; second quartile, 103–129 points; third quartile, 130–156 points; and fourth quartile, >156 points). The y axis represents the observed 1‐year probability of CTRCD‐free survival, as estimated by Kaplan‐Meier method. Gray line indicates ideal agreement between observed and predicted probabilities of 1‐year CTRCD. Black line denotes actual prediction from the nomogram. Dots denote apparent predictive accuracy, which was calculated by plotting the mean Kaplan‐Meier estimate for each quartile vs the mean nomogram‐predicted probabilities for patients in each quartile. X denotes the bootstrap‐corrected estimates. Vertical bars indicate 95% CI.

Figure 3. Receiver operating characteristic (ROC) curve for 1‐year cancer therapy–related cardiac dysfunction of the nomogram.

AUC indicates area under ROC curve.

Decision curve analysis for the risk nomogram is presented in Figure 4. The 1‐year CTRCD risk nomogram demonstrates a positive net benefit for threshold probabilities between 5% and 30%. In this range, use of the risk nomogram is superior to the default strategies of intervening on all (treat all) or none (treat none) of the patients. For clinical settings in which the threshold to intervene is <5% or >30%, use of the risk nomogram has limited net benefit.

Figure 4. Decision curve of the nomogram for 1‐year cancer therapy–related cardiac dysfunction (CTRCD).

Net benefit curves show the clinical usefulness of intervention guided by the CTRCD risk nomogram (blue line) relative to a strategy of intervention for all patients regardless of risk (red line) or intervention for no patients irrespective of risk (green line). The net benefit (y axis) is plotted as a function of threshold probability (x axis), defined as the probability of disease that would prompt intervention. The 1‐year CTRCD risk nomogram has positive net benefit for threshold probabilities of 5% to 30%.

DISCUSSION

In this study, we developed and validated a nomogram to predict 1‐year CTRCD risk among women with breast cancer treated with HER2‐targeted therapy. The nomogram incorporates an extensive set of treatment‐specific (eg, anthracycline chemotherapy) and patient‐specific risk factors (eg, age, baseline LVEF and SBP, race, and cardiovascular disease history) that are widely accessible for all patients before initiation of breast cancer therapy. The risk nomogram showed good discriminative ability to estimate 1‐year CTRCD risk, good model calibration based on agreement between predicted versus observed outcomes, and clinical utility across a wide range of threshold probabilities based on decision curve analysis.

The development of clinical and molecular risk stratification tools has allowed cancer clinicians to tailor therapy for HER2‐positive breast cancer using a broad range of HER2‐targeted treatment regimens. ²⁹ More important, given the heterogeneity of cardiovascular toxicity profiles among different treatment regimens, strategies to estimate risk of CTRCD are needed to help guide clinical decision‐making for individual patients. The nomogram in this study may be a valuable tool for this purpose, so that safer and less cardiotoxic treatment options can be considered for patients at high risk for CTRCD. Furthermore, identification of patients at high risk for CTRCD using this nomogram could indicate a role for early referral to cardio‐oncology so that cardiovascular risk factors can be optimized and more intensive surveillance can be established (eg, incorporation of biomarkers, myocardial strain imaging, or both to standard 2‐dimensional echocardiograms). Decision curve analysis, a recommended strategy for assessing the impact of implementing risk prediction tools into clinical practice, ³⁰ , ³¹ demonstrated a positive net benefit of the nomogram between probability thresholds of 5% to 30%. Intuitively, the value of the nomogram is limited when the threshold to intervene is very low or high, because modifications of risk estimates at these extremes would be unlikely to inform clinical decision‐making.

Prior risk models have been proposed for prediction of both early and late adverse cardiovascular events. A 7‐factor semiquantitative risk score was proposed by Ezaz et al to stratify the 3‐year risk of heart failure or cardiomyopathy among women aged 67 to 94 years with stage I to III HER2‐positive breast cancer; however, this risk score was developed using claims‐based data from Surveillance, Epidemiology, and End Results Medicare and may not be generalizable to younger women (aged <67 years). ³² Incidence of CTRCD is notably higher in claims‐based studies compared with clinical trials or real‐world observational studies, ³³ suggesting that algorithms used in these studies to identify heart failure or cardiomyopathy that were developed in noncancer populations may not be applicable in the cardio‐oncology setting. A cardiac risk score was derived from the NSABP B‐31 (National Surgical Adjuvant Breast and Bowel Project B‐31) trial of women with HER2‐positive breast cancer treated with sequential anthracyclines and trastuzumab and predicts probability of heart failure or cardiac death within 5 years of treatment, but is based on limited inputs of age and baseline LVEF and lacks discriminatory power to be used in clinical practice. ⁶ Furthermore, this cardiac risk score may not be generalizable in real‐world clinical practice because of the lower prevalence of cardiovascular risk factors (3.8% diabetes, 7.5% hyperlipidemia, and 20.6% hypertension based on medication history) and exclusion of patients with preexisting cardiovascular disease. Finally, a cardiovascular risk stratification tool proposed by the Heart Failure Association of the European Society of Cardiology and International Cardio‐Oncology Society incorporates patient‐ and treatment‐specific risk factors to predict the likelihood of cardiovascular complications during cardiotoxic cancer treatments, including HER2‐targeted therapy. However, this tool was developed on the basis of expert consensus. ³⁴ Furthermore, in a validation study of 931 patients with HER2‐positive breast cancer, the Heart Failure Association of the European Society of Cardiology and International Cardio‐Oncology Society score had a low sensitivity (14.8%) and was unable to identify patients at low cardiotoxicity risk (<5%) or discriminate between low‐ and medium‐risk cohorts. ³⁵

The multivariable nomogram from this study addresses some of the limitations from prior risk models. First, we include clinical characteristics that were abstracted directly from the electronic health record and CTRCD events that were adjudicated by a cardiologist on the basis of clinical data, in contrast to previous claims‐based studies. Second, derivation of the risk nomogram with variables assigned point values was based on statistical significance using multivariable regression rather than expert consensus. Third, our risk nomogram focuses on CTRCD within 1 year of starting breast cancer therapy as the end point of interest, so that estimates from this nomogram can be used to generate individualized predictions and inform early clinical decision‐making on cancer treatment regimens, cardioprotective strategies, and modality or frequency of cardiac surveillance. More important, CTRCD typically occurs within the first 3 to 6 months of the HER2‐targeted treatment period, and late occurrence beyond the 1‐year treatment period is uncommon. Finally, we incorporate race and baseline SBP as important risk factors in our proposed nomogram that may help to further discriminate patients with increased CTRCD risk.

Limitations

This study was performed using retrospective data. Because of limitations of the study sample size and the number of CTRCD events, validation with a larger external data set is warranted. Beyond baseline LVEF, other imaging (eg, global longitudinal strain) or circulating biomarkers (eg, cardiac troponin or natriuretic peptides) that may have improved performance of the nomogram were not available for inclusion in our risk model, and LVEF assessments were performed at the discretion of the treating provider. Future nomograms may consider inclusion of these or other emerging biological variables as our understanding of the mechanism of CTRCD increases. The consensus definition of CTRCD as proposed by the International Cardio‐Oncology Society was not used in this retrospective study given absence of global longitudinal strain and other cardiac biomarker data; however, future studies of CTRCD using contemporary data should rely on these consensus criteria. ³⁶ All patients were treated with trastuzumab, and most received anthracycline chemotherapy; alternative HER2‐targeted agents were not available for routine clinical use during the study period. The generalizability of this risk nomogram to patients receiving novel HER2‐targeted agents is unknown. Among long‐term survivors of breast cancer, cardiovascular disease mortality is recognized as a key competing risk that can exceed cancer‐related mortality, especially among older women or those with a history of preexisting cardiovascular disease. ¹⁰ , ³⁷ However, the proposed risk nomogram focuses only on CTRCD events within 1 year of treatment and, thus, cannot be applied for risk stratification in the survivorship setting. However, risk models for prediction of late cardiovascular events up to 10 years after cardiotoxic treatment exposure among women with breast cancer have previously been proposed. ³⁸

CONCLUSIONS

We developed and internally validated a 9‐variable nomogram for individualized estimation of CTRCD risk within the first year of treatment for women with HER2‐positive breast cancer. More important, variables included in the nomogram are easy to obtain and readily available before the initiation of cancer treatment. This nomogram offers a simple‐to‐use tool for clinicians and patients that may help to inform clinical decision‐making on breast cancer treatment options, cardiac surveillance, and cardioprotective strategies.

Sources of Funding

This work was funded by a grant from the National Institutes of Health (K23 CA218897) awarded to Dr Yu. This work was funded in part by the National Institutes of Health (P30 CA008748).

Disclosures

Dr Yu reported personal fees from Glenmark, Genentech, and Ichnos Sciences. Dr Dang reported research funding and personal fees from Genentech/Roche and Puma Technology. Dr Steingart reported personal fees from Pfizer and Celgene. The remaining authors have no disclosures to report.