Obesity predisposes anthracycline-treated survivors of childhood and adolescent cancers to subclinical cardiac dysfunction

Introduction

There are currently about 500,000 survivors of childhood cancer (age 0-18 years at diagnosis) in the United States, and as incidence and survival rates continue to increase, the long-term sequalae of treatment are becoming apparent [1, 2]. Cardiac late effects are of particular concern in the childhood cancer survivor population as approximately 60% of pediatric patients with cancer will receive an anthracycline, agents known to be associated with risk of cardiomyopathy, during their chemotherapy regimen [3]. As such, identifying subgroups of childhood cancer survivors at particularly high risk of cardiac late-effects is important for long-term surveillance. The Childhood Cancer Survivor Study (CCSS) Cardiovascular Risk Calculator was developed and externally validated to predict the risk of adverse cardiovascular events prior to age 50 in survivors of childhood and adolescent cancer. The base version of this risk calculator is intended for use at least 5 years post-treatment and includes age at diagnosis, current age, sex, and treatment information [4]. Recently, an extension of the CCSS Calculator was created to consider diabetes, hypertension, and dyslipidemia status in patients aged 20 to 39, as these comorbidities are synergistic with cardiotoxic chemotherapy [5]. However, neither the base nor extended CCSS Calculator consider how preexisting cardiovascular risk factors at the time of diagnosis affect subsequent cardiac dysfunction. This is a notable gap, as a recent 35-year prospective study of approximately 38,000 patients from the general pediatric population enrolled in the International Childhood Cardiovascular Consortium (i3C) found that patients with at least one cardiovascular risk factor in childhood had a significantly increased risk of adverse cardiac events in mid-life [6]. Further, a meta-analysis of literature published in the adult oncology population has found that baseline obesity, diabetes mellitus and hypertension status at the time of diagnosis are each associated with an increased risk of anthracycline mediated cardiotoxicity [7].

Pediatric surveillance guidelines to monitor for cardiac dysfunction following anthracycline administration recommend periodic conventional 2-D or 3-D echocardiograms to monitor for changes in left ventricular ejection fraction (LVEF) [8]. However, LVEF has poor sensitivity for subtle changes in cardiac function due to significant test-retest variability and interobserver variability [9]. As cardiac imaging has continued to advance, strain imaging, a form of speckle-tracing echocardiography, has emerged as a promising technique to monitor for early myocardial dysfunction [10, 11]. In the adult population, global longitudinal strain (GLS) has been shown to be a major predictor of cardiac events and mortality in both the low-risk general population and among patients with existing cardiovascular disease [12-14]. A recent guideline statement from the European Society of Cardiology (ESC) recommends strain measurements be used as a routine component of cardiac surveillance among all adult cancer survivors [15-17]. While previous studies have described a greater than 4-fold higher incidence of abnormal strain versus abnormal 3D LVEF among survivors of childhood and adolescent cancers, strain has yet to be recommended as a routine component of cardiac surveillance in this population [8, 18].

This study seeks to leverage the increased sensitivity of strain imaging by calculating strain measurements on surveillance echocardiograms among a retrospective cohort of pediatric patients treated with anthracyclines. These strain values will be used to address the previously identified gap in risk-stratification by identifying potentially modifiable cardiovascular risk factors present at the time of diagnosis that contribute to early cardiac dysfunction. We hypothesize that left ventricular (LV) strain measurements are more sensitive to sub-clinical cardiac dysfunction than LVEF alone, and that modifiable cardiovascular risk factors present at the time of diagnosis will be associated with decreased cardiac dysfunction appreciable by strain echocardiography.

Methods

Patient Selection:

This study was approved by the Duke University Hospital Institutional Review Board with a waiver of consent. Patient selection included consecutive pediatric patients with cancer who were treated with an anthracycline at Duke University between 2013-2019, were under the age of 18 at the time of cancer diagnosis and had at least 1 echocardiogram within the Duke University Health System. Clinical echocardiograms were acquired with GE Vivid E9, GE E95, Phillips iE33, and Phillips Epiq cardiac ultrasound systems (General Electric Vingmed Ultrasound, Horten Norway; Koninklijke Philips, Amsterdam, Netherlands). Archived studies were stored and accessed using Philips Xcelera software (Koninkijke Philips, Amsterdam, Netherlands). Basic demographic data collected from the electronic health record (EHR) included: age at time of diagnosis, race and ethnicity (as documented in the patient’s chart), BMI percentile if ≥2 years of age (CDC Extended BMI-for-Age Growth Chart), weight for length percentile if <2 years of age, sex assigned at birth, past medical history, and family history of cardiovascular disease in a first-degree relative. Treatment data for each patient was also collected from the EHR, including primary cancer type, chemotherapeutic agents received, cumulative anthracycline dose, radiation dose and location, and if a patient was given dexrazoxane prior to anthracycline administration. Cumulative anthracycline dose was calculated by the doxorubicin isotoxic equivalent dose in mg/m². Equivalents were calculated by multiplying the total dose of doxorubicin by one, multiplying the total dose of daunorubicin by 0.5, multiplying the total dose of epirubicin by 0.67, multiplying the total dose of idarubicin by 5, and multiplying the total dose of mitoxantrone by 4 [19, 20].

The dataset was expanded to include all existing conventional 2-D echocardiograms for patients within the dataset, including BMI-for-age percentile (if 2-18 years of age), length/weight-for-age percentile (if <2 years of age), and comorbid medical conditions present at the time of each echocardiogram. Patients with cancer relapses, a history of structural heart disease, evidence of reduced cardiac function (defined as LVEF <50% of strain >−16%) on a pre-treatment echocardiogram or no adequate post-treatment echocardiograms were excluded.

Patients over 2 years of age were categorized as obese at the time of diagnosis if they had a BMI > 95^th percentile-for-age, and patients under 2 years of age were categorized as obese at the time of diagnosis if they had a weight for length >98^th percentile. This operational definition of obesity was selected in accordance with the guidelines set forth by the Center for Disease Control (CDC), and in to be consistent with definitions used in pediatric clinical practice in the United States [21, 22]. Patient anthracycline dose was categorized as either high-dose (≥250 mg/m² doxorubicin equivalents) or low dose (<250 mg/m² doxorubicin equivalents).

Strain Data Collection:

Peak longitudinal LV strain was retroactively calculated on all adequate studies via the vendor-independent TOMTEC AutoSTRAIN software (TomTec-Arena 1.2, TomTec Imaging Systems, Unterschleissheim, Germany). Strain values were calculated on all adequate studies by a single, trained reader (IAG), and a subset of ~10% of these studies were randomly assigned to a blinded pediatric cardiologist (AWM) to assess inter-reader variability prior to analyses. Inter-reader variability between measurements was assessed via the interclass correlation coefficient (ICC), with a minimum ICC value of 0.6, indicating “good” reliability, necessary for further analyses [23]. Peak longitudinal LV strain calculated via the 4-chamber view was used in place of GLS for all analyses due to a very high incidence of inadequate 2-chamber or 3-chamber images (~70% of studies in this cohort), which are necessary for calculating GLS.

Studies were categorized as having abnormal strain if strain was calculated to be >−16%. There is insufficient research defining normal strain values obtained via TOMTEC in the pediatric population, so −16% was chosen as the cut-off value as it is the typical measurement used in the adult population [24, 25] and it corresponds to approximately 2 standard deviation (1.88 SD) lower than the average pre-anthracycline strain measurements in our cohort. The method of defining abnormal strain by standard deviations from the mean has been described in previous pediatric studies [18]. Studies were categorized as having reduced LVEF if the value was <50%.

Statistical Analysis:

Descriptive statistics were used to characterize the eligible patients. Two distinct time-to-event analyses were performed, one with the event defined as the first incidence of strain >−16% following the conclusions of anthracycline treatment, and the other with the event defined as the first incidence of LVEF<50% following conclusion of anthracycline treatment. Kaplan Meier curves were created to plot the incidence of abnormal strain values (>−16%) and decreased LVEF (<50%) during the study period, with the following comparisons across sub-populations performed via the log-rank test: high versus low dose anthracycline exposure; obese versus non-obese; and presence vs absence of family history of CVD in a first degree relative. Univariable cox-proportional hazards models were created via the coxph function in R studio statistical software. The univariate models explored the effect of the following variables on the incidence of abnormal strain values: anthracycline dose, age at diagnosis, obesity status, sex assigned at birth, primary cancer type (solid versus liquid), race and ethnicity, exposure to radiation, treatment with dexrazoxane, and family history of CVD in a first degree relative. A final multivariable cox-proportional hazards model was created. The proportional hazards assumption was verified graphically and statistically via scaled Schoenfeld residuals. Inter-reader variability of strain measurements was visualized graphically via a Bland-Altman plot and was quantified via the calculation of the ICC of absolute agreement via a two-way mixed-effect model.

Results

Of the 229 patients initially identified via our inclusion criteria, 114 patients were excluded. Among the excluded individuals, 38 subjects were removed due to a history of a cancer relapse, and 12 patients were removed due to preexisting structural heart disease. 64 patients were subsequently removed due to inadequate post-treatment echocardiogram follow-up, which included patients who experienced mortality prior to completing treatment (n=4), were lost to follow-up without a post-treatment echocardiogram (n=38), or had no post-treatment echocardiograms of sufficient resolution or correct orientation for longitudinal strain imaging (n=22). 115 patients were included in the final cohort, with a collective sum of 686 surveillance echocardiograms. There were 495 studies (72%) that were adequate for 4-chamber LV strain analysis (Figure 1).

Fig. 1. Schematic of Cohort Creation.

All identified patients were screened, and those with cancer relapses or a history of structural heart disease were excluded. Of those screened patients, only patients with at least one post-chemotherapy echocardiogram of adequate quality underwent full strain analysis.

The basic demographic characteristics and treatment data for the cohort are shown in Table 1 and Table 2, respectively. The study consisted of a largely pediatric cohort with a young age at diagnosis (9.51 ± 5.92 years) and an average longitudinal follow-up period (defined as time from end of anthracycline treatment to a patient’s last recorded echocardiogram) of approximately 3.9 years (SD=2.10). There was an average of 4 studies (SD=2.28) adequate for strain analysis per patient. The average BMI percentile at the time of diagnosis was 58.1 (SD=29.4), and 14 patients (12.2%) were identified as obese at the time of diagnosis. There were 4 patients in the final cohort who had insufficient data in the chart to calculate a BMI or weight/length percentile at the time of diagnosis. The average BMI percentile at the time of anthracycline administration was 56.8 (SD=32.0). There was no significant difference in patient BMI percentile at the time of diagnosis compared to at the time of anthracycline administration identified via a paired samples t-test (p=0.21 Supplementary Table 1). At the time of last follow-up, there were 24 obese individuals and 89 non-obese individuals. There were 4 individuals who were classified as obese at the time of diagnosis who would be reclassified as non-obese at the time of last follow-up, with a mean BMI percentile of 89.2 ± 7.5% The mortality rate among the 115 patients included in the study was 7.8%. Of the 9 patients who died, no deaths were attributed to a cardiac cause. Finally, there were 26 patients (22.6%) in the study with a first-degree family member with a recorded history of CVD at the time of their diagnosis. Hypertension was the most common disease in first degree relatives (92.3%), followed by coronary artery disease (CAD) (7.7%), heart failure (7.7%), and atrial fibrillation (1.8%).

Table 1:

Demographics of Cohort with Strain Data

	Overall (N=115)
Sex Assigned at Birth
Male	61 (53.0%)
Female	54 (47.0%)
Race And Ethnicity
American Indian or Alaskan Native	0 (0%)
Asian	3 (2.6%)
Black or African American	25 (21.7%)
Hispanic or Latino	16 (13.9%)
Native Hawaiian or Pacific Islander	0 (0%)
White	66 (57.4%)
Other/Unknown	5 (4.3%)
Age at Cancer Diagnosis
Mean (SD)	9.5 (5.9)
Median [Min, Max]a	12 [0, 18]
BMI Percentile-For-Age at Time of Diagnosis
Mean (SD)	58.1 (29.4)
Median [Min, Max]	54.7 [0.1, 99.8]
BMI Classification at Time of Diagnosis
Obese	14 (12.2%)
Not Obese	97 (84.3%)
Number of Studies Adequate for Strain Per Patient
Mean (SD)	4.3 (2.1)
Median [Min, Max]	4 [1, 14]
Duration of Follow-up (Years)
Mean (SD)	3.9 (2.28)
Median [Min, Max]	3.9 [0.1, 10.9]

^aAge at cancer diagnosis in years, rounded to the nearest whole number.

Table 2:

Treatment Characteristics of Cohort with Strain Data

	Overall (N=115)
Solid Tumor
Yes	50 (43.5%)
No	65 (56.5%)
Anthracycline Cumulative Exposure in Doxorubicin Equivalents (mg/m2)
Mean (SD)	197 (123)
Median [Min, Max]	151 [18.0, 457]
Category of Dose
Low Dose (<250 mg/m2)	82 (71.3%)
High Dose (≥250 mg/m2)	33 (28.7%)
Dexrazoxane
Yes	26 (22.6%)
No	89 (77.4%)
Radiation
Yes	32 (27.8%)
No	83 (72.2%)
Radiation Cumulative Dose (Gy)	(N=32)
Mean (SD)	35.4 (23.6)
Median [Min, Max]	23.2 [10.5, 117]
Mediastinal Radiation Dose (Gy)	(N=8)
Mean (SD)	23.3 (6.36)
Median [Min, Max]	21.0 [21.0, 39.0]

Both GLS and 4-chamber longitudinal strain values were obtained on 163 studies, and correlation was assessed via a linear regression model with a slope of .87 and an associated R² value of 0.79, indicating strong correlation between the two measures (Supplementary Figure 1). The inter-reader variability of strain measurements on 52 studies read by both the primary reader, IAG, and a pediatric cardiologist, AWM, can be visualized in the Bland-Altman plot (Supplementary Figure 2). It was found that IAG had a bias of reading strain on studies at 0.90 percentage points higher (less negative) than AWM. The interclass correlation coefficient calculated from these 52 studies was 0.81 (95% CI [0.62, 0.89], p=9.7e⁻⁸), indicating good reliability [23].

Following the conclusion of anthracycline treatment, reduced LVEF (<50%) was identified in 5 patients (4.3%), while abnormal strain (>−16%) was identified in 17 patients (14.7%). All 5 patients who had reduced LVEF recorded on a study also had an abnormal strain value recorded on a study. Of these 5 patients, the abnormal strain value preceded reduced LVEF in 3 cases. Due to the low incidence of reduced LVEF in the cohort, all survival analyses were performed with strain >−16% as the endpoint of interest.

Kaplan Meier curves were created to determine the cumulative incidence of abnormal strain within the cohort and to compare strain abnormality-free survival across strata of potential risk factors. Patients were censored at the time of their last recorded echocardiogram or at the time of death. Notably, strain abnormality-free survival was found to be significantly lower via a log-rank test among patients who received high dose (≥250 mg/m² Dox Eq) versus low dose anthracyclines (p=0.02, Figure 2), patients who were obese (BMI ≥95^th percentile or weight for length ratio >98^th percentile) versus non-obese (p<0.01), and patients with versus without a history of CVD in a first-degree relative (p=0.04).

Fig. 2. KM Curves of Strain Abnormality-Free Survival Probability Stratified by Sub-Populations.

(a) Anthracycline dose category, (b) obesity status at diagnosis, and (c) family history of cardiac disease in a first-degree relative were used to stratify the population into binary classes. Survival functions were created with the event of interest defined as the first instance of strain < −16% following the conclusion of anthracycline treatment and were used to plot KM curves. Vertical lines indicate censoring, occurring at the time of death or at the time of a patient’s last recorded echocardiogram. A log-rank test was used to evaluate if the KM curves in each plot were significantly different from one another. The p-values for the survival curves stratified by dose (a), obesity status (b) and family history of CVD (c) are p=0.02, p=0.002, and p=0.04, respectively.

Further investigation with univariable Cox proportional hazards models (Table 3) found that high anthracycline dose (Hazard Ratio (HR) 2.98, 95% CI [1.15-7.72], p=0.03) and positive obesity status at the time of diagnosis (HR 4.14, 95% CI [1.53-11.24], p<0.01) were significantly associated with abnormal strain, but family history of CVD was not significant. In our multivariable Cox proportional hazards model, both high anthracycline dose and positive obesity status at the time of diagnosis remained significantly associated with abnormal strain, with HRs of 2.79 (95% CI [1.07-7.25], p=0.04) and 3.85 (95% CI [1.42-10.48], p<0.01), respectively. A two-sample t-test was performed to ensure that obesity status and anthracycline dose were not co-linear variables, and there was not found to be any significant difference in BMI percentiles between patients receiving high or low dose anthracycline treatment (p=0.86, Supplementary Table 2). Further, it was found that three of the six individuals who were obese at the time of diagnosis and received high dose anthracycline treatment were noted to have an abnormal strain value on a post-treatment echocardiogram, compared to three of the eight patients who were obese but treated with a low dose anthracycline (Table 4). Sex assigned at birth, exposure to radiation, tumor type, and treatment with dexrazoxane were not shown to be significant in univariable Cox proportional hazards models. Univariate models for race and ethnicity, and patient age <5 years at time of diagnosis resulted in undefined results due to insufficient events.

Table 3:

Univariable and Multivariable Cox Proportional Hazards Models of Abnormal Peak Longitudinal Strain-Free Survival

Variables	Univariable Modelsa			Multivariable Modelb
	Hazard Ratio	95% confidence interval	P-value	Hazard Ratio	95% confidence interval	P-value
High Dose Anthracycline	2.98	1.15-7.72	0.03	2.79	1.07-7.25	0.04
Obesity	4.14	1.53-11.24	0.005	3.85	1.42-10.48	0.008
Family History of CV Disease	2.56	0.97-6.74	0.06	-	-	-
Female Sex Assigned at Birth	1.43	0.55-3.72	0.46	-	-	-
Exposure to Radiation	1.76	0.67-4.65	0.25	-	-	-
Treatment with Dexrazoxane	1.57	0.55-4.48	0.39	-	-	-
Solid Tumor	1.01	0.39-2.67	0.98	-	-	-
Age <5 at Diagnosis	undefined	undefined	0.99	-	-	-

^aModels and p-values were generated via the coxph function in R Studio statistical software. P-values <0.05 indicate that a variable has a significant effect on the hazard ratio.

^bThe multivariable model includes anthracycline dose and BMI percentile. Overall model p-value = 0.004 via the likelihood-ratio test. The multivariable model was restricted to 2 variables to prevent overfitting due to a low number of observed events (n=17).

Table 4:

Percentage of Patients Identified to Have Abnormal Strain Stratified by Obesity Status at Diagnosis and Anthracycline Dose

	Abnormal Strain Identified	No Abnormal Strain Identified	Percent With Abnormal Strain (%)
Obese (n=14)a	6	8	42.9%
High Dose (n=6)b	3	3	50%
Low Dose (n=8)	3	5	37.5%
Non-Obese (n=97)c	11	86	12.8%
High Dose (n=27)	6	21	22.2%
Low Dose (n=70)	5	65	7.7%

^aObese defined as BMI-for-age >95^th percentile or weight for length >98^th percentile if <2 years of age.

^bHigh dose defined as >250 mg/m² doxorubicin equivalents.

^cNon-obese defined as BMI-for-age >95^th percentile or weight for length >98^th percentile if <2 years of age. Note that 4 patients from the overall cohort (n=115) are excluded from this analysis due to insufficient data for BMI or weight for length calculation at time of diagnosis.

Discussion

Although improved treatment has contributed to increased cure rates for children and adolescents diagnosed with cancer, many survivors have chronic health conditions and early mortality due to chemotherapy-induced cardiac toxicity [26]. In addition to lifelong cardiac monitoring, there is a need to identify modifiable factors that may optimize cardiac health in these survivors and mitigate their risk of cardiac morbidity. In this study we observed an association between obesity at the time of diagnosis and a significantly increased risk of subclinical cardiac dysfunction among survivors of childhood cancers. This is an important finding as it represents a risk factor that can be identified at the time of diagnosis and has well-established lifestyle and pharmacological interventions. Further, it is not currently a variable used in the predominant risk calculators aimed at identifying patients at risk of future cardiac events. It also is not a factor considered in the post-treatment surveillance guidelines for adult survivors of childhood and adolescent cancers in the 2022 ESC Guidelines for Cardio-Oncology [15].

Obesity is a particularly concerning risk factor in this population as childhood obesity rates continue to rise. While obesity was once a rare pre-treatment diagnosis among pediatric patients with cancer, the prevalence of childhood obesity has increased by over 300% since 1976 [27]. Prior to the Covid-19 pandemic, nearly 20% of children in the US were obese, and the rate of BMI increase among patients 2-19 years of age further doubled during the pandemic [28]. This rapidly accelerating obesity epidemic at younger ages is disturbing, and childhood cancer survivors find themselves in a particularly vulnerable position due to decreased activity levels, high doses of steroids during treatment, and chemotherapy and radiation-related weight gain [29-31]. Therefore, it is becoming increasingly likely that patients will be overweight at the time of their diagnosis or soon after. While it certainly is not feasible to delay treatment in overweight patients until they undergo a period of weight loss, this risk highlights the need for lifestyle interventions during and after treatment.

Further, baseline obesity is a known risk factor for the development of hypertension and dyslipidemia among survivors of pediatric cancers, which despite their known association with adverse cardiac events, have been chronically underdiagnosed and undertreated among overweight and obese survivors of [32-34]. The development of the field of cardio-oncology and the creation of multidisciplinary survivorship clinics has allowed for more comprehensive care of the aging survivor population. However, while the American Heart Association (AHA) recommends structured exercise and dietary interventions as preventive treatments among survivors of adult cancers, supervised interventions remain cost and time intensive, and patient behavior has proved challenging to change [35]. Further, a 2023 systematic review from The International Late Effects of Childhood Cancer Guideline Harmonization Group did not find any studies that investigated the effectiveness of physical activity, lifestyle interventions, and treatment of traditional CVD risk factors to prevent future cardiomyopathy in survivors of childhood and young adult cancers [8]. The results presented in this study highlight this critical gap in research and underscore that despite the upfront financial and time cost required for lifestyle interventions, the risk conveyed by inaction may prove much more costly.

The results of our retrospective analysis found an over 3-fold higher incidence of cardiac dysfunction identified by abnormal strain than by LVEF alone. While these findings represent a significant increase in sensitivity through strain imaging when compared to 2D LVEF, our findings are consistent with findings from much larger studies, such as the St. Jude Lifetime Cohort Study, in which 28% of the 1820 pediatric cancer survivors were identified to have abnormal strain values despite an incidence of abnormal LVEF of only 5.8% [18]. This increased sensitivity of strain echocardiography represents a dramatic increase in the power of surveillance echocardiography. By identifying subtle changes in cardiac function that precede clinical symptoms or dramatic decreases in LVEF, providers may be given a warning sign of patients who are at risk of future cardiac dysfunction. The limited follow-up of our study precludes us from a definitive conclusion on the predictive value of abnormal strain in the post-treatment period. Nevertheless, early (within 5-years of cancer diagnosis) sub-clinical LV dysfunction characterized by a mid-range EF of 40-50% has been shown to convey an 8-fold higher risk of heart failure at 10-years post-treatment in adult survivors of childhood and adolescent cancers [36]. Further, a prospective study of 2,625 adults who were treated with anthracycline drugs found that 98% of cardiotoxicity cases, defined by a drop in LVEF, occurred within the first year following treatment, and that cardiotoxicity was fully reversible with the early initiation of heart failure therapy in 11% of patients and partially reversible in 71% of patients. However, similar prospective studies documenting the onset of cardiotoxicity in the pediatric population are lacking, as are both prospective and retrospective studies documenting the trajectory and clinical manifestations of this early subclinical cardiac dysfunction with and without treatment [37].

There are several limitations to the current study. While our retrospective cohort design allowed us to include a large number of patients and echocardiograms, there was limited standardization with respect to the timing and frequency of echocardiograms within the cohort, as well as a variable length of follow-up. While the average follow-up duration was 3.9 years and the maximum duration was greater than 10 years, the minimum follow-up duration was only 0.1 years, which limited the ability to analyze cardiac function in some patients to only the acute post-treatment period. Therefore, while we were able to identify that obesity at the time of diagnosis was a risk factor for abnormal strain values within this cohort, we are unable to determine if these subclinical changes in cardiac function truly indicate a higher risk of future overt cardiac dysfunction. Previous studies in adults with heart failure with preserved ejection fraction or heart failure with recovered ejection fraction have found that abnormal GLS values predict both higher rates of future composite cardiac outcomes and more rapid deterioration of LVEF [38, 39]. However, there is very limited data in the pediatric population and the cancer survivorship population on the prognostic utility of strain. Therefore, there is limited data at this time to support medical intervention for isolated abnormal strain measurements in this specific population.

Further, strain echocardiography represents a powerful tool to monitor for cardiac dysfunction but has known limitations with regards to inter-reader observability and the impact of imaging quality on the ability to perform reliable calculations [40]. Fully automated strain calculations using vendor-independent software have shown great promise at reducing this variability and the time required for strain calculations, however, these software programs require high quality images obtained in the correct orientation for reliable measurements. While fully automated software can calculate strain on >99% of studies, quality is highly variable, requiring manual rearrangement on >40% of studies for accurate results [24]. Among our retrospective cohort of pediatric patients, we found this particularly challenging due to a combination of patient movement during studies, port-placements requiring off-axis apical views, and poor-imaging quality resulting in inadequate visualization of the endocardium throughout the full cardiac cycle. For this reason, we maintained strict inclusion criteria such that we sacrificed the final number of patients in favor of higher quality studies and more reliable strain-measurements. Further, we ensured internal consistency by having a single reader perform strain analysis on all 495 studies, and inter-reader agreement and reproducibility via the inclusion of a second trained, blinded reader who evaluated 52 randomly selected studies with an associated ICC of 0.81, indicating good reliability. However, ultimately it is important to note that the inherent inter-reader variability of imaging studies does pose an increased risk for false positive results. Indeed, 8/17 (47%) patients in this study who were classified as having abnormal strain were in the borderline region with strain values between −15% and −16%. While these borderline values are not unexpected in this largely asymptomatic cohort, it does highlight the need for further standardization of strain measurement software and methods to decrease inter-reader variability such that false positives can be minimized.

Finally, previous studies have similarly found retrospective analysis of GLS to be challenging, with some multicenter studies reporting success rates as low as 10% due to a high incidence of poor imaging quality and artifact [41]. To address this, it has been previously demonstrated in adult populations that 4 chamber longitudinal strain does not vary significantly from strain values obtained via GLS [42, 43]. However, there are notable limitations to the use of 4-chamber longitudinal strain. Strain obtained by a single view is less sensitive to significant isolated wall motion abnormalities than GLS and has higher inter-reader variability, leading to greater levels of disagreement for the detection of cardiotoxicity when studied in adults [42]. Therefore, it is important to acknowledge that the use of 4-chamber longitudinal strain among a pediatric cohort of patients in this study is not without its limitations, and further highlights the need for future prospective studies which are able to capture both high-quality single-view longitudinal strain and GLS measurements in pediatric cancer survivors.

Conclusions

We found that obesity at the time of diagnosis was associated with subsequent abnormal strain on surveillance echocardiography among pediatric patients with a history of cancer treated with an anthracycline. Importantly, current models of cardiotoxicity risk-prediction do not incorporate cardiovascular risk factors present at the time of diagnosis, and instead rely on non-modifiable factors such as treatment exposures or cardiovascular risk factors later in the survivorship period. Obesity represents a potentially modifiable and easily identifiable risk factor. A better understanding of whether reduction in BMI attenuates risk of abnormal strain is needed, as are follow-up studies assessing the association of early development of abnormal strain and the development of clinically relevant cardiac disease.