Echocardiography protocol for early detection of cardiac dysfunction in childhood cancer survivors in the multicenter DCCSS LATER 2 CARD study: Design, feasibility, and reproducibility

Abstract Background Cardiotoxicity is a well‐known side effect after anthracyclines and chest radiotherapy in childhood cancer survivors (CCS). The DCCSS LATER 2 CARD (cardiology) study includes evaluation of echocardiographic measurements for early identification of CCS at highest risk of developing heart failure. This paper describes the design, feasibility, and reproducibility of the echocardiography protocol. Methods Echocardiograms from CCS and sibling controls were prospectively obtained at the participating centers and centrally analyzed. We describe the image acquisition, measurement protocol, and software‐specific considerations for myocardial strain analyses. We report the feasibility of the primary outcomes of systolic and diastolic function, as well as reproducibility analyses in 30 subjects. Results We obtained 1,679 echocardiograms. Biplane ejection fraction (LVEF) measurement was feasible in 91% and 96% of CCS and siblings, respectively, global longitudinal strain (GLS) in 80% and 91%, global circumferential strain (GCS) in 86% and 89%, and ≥2 diastolic function parameters in 99% and 100%, right ventricle free wall strain (RVFWS) in 57% and 65%, and left atrial reservoir strain (LASr) in 72% and 79%. Intra‐class correlation coefficients for inter‐observer variability were 0.85 for LVEF, 0.76 for GLS, 0.70 for GCS, 0.89 for RVFWS and 0.89 for LASr. Intra‐class correlation coefficients for intra‐observer variability were 0.87 for LVEF, 0.82 for GLS, 0.82 for GCS, 0.85 for RVFWS and 0.79 for LASr. Conclusion The DCCSS LATER 2 CARD study includes a protocolized echocardiogram, with feasible and reproducible primary outcome measurements. This ensures high‐quality outcome data for prevalence estimates and for reliable comparison of cardiac function parameters.


| INTRODUC TI ON
Treatment of childhood cancer has drastically improved over the last decades, resulting in a 5-year survival of over 80%, nowadays. 1 However, as the population of long-term childhood cancer survivors (CCS) increases, awareness has risen concerning their risk of various late treatment effects. 2 Cardiotoxicity is a well-known side effect of anthracyclines and radiotherapy involving the heart region and is responsible for substantial morbidity and mortality, even decades after therapy. 3,4 Besides coronary artery disease, valvular disease, pericardial disease, and arrhythmias, 5 the most important manifestation of cardiotoxicity is cardiomyopathy, which can range from asymptomatic left ventricular dysfunction to overt or even fatal clinical heart failure. Surveillance guidelines recommend periodical echocardiographic examination for early detection of left ventricular dysfunction, with surveillance intervals based on cardiotoxic therapy exposures. 6 More detailed risk stratification through sensitive detection tools may enable late-effects clinicians to better prevent clinical heart failure in case of increased risk, or reduce surveillance burden in case of low risk. The most frequently reported systolic function parameter, left ventricular (LV) ejection fraction (LVEF), is considered the standard echocardiographic measure of LV dysfunction but is a rather late marker compared with measurements of myocardial strain. 7,8 Compared with magnetic resonance or nuclear imaging, echocardiographic LVEF comes with some disadvantage of a testretest variability of 5%-10%-points in expert echo-laboratories, depending on a biplane or 3D approach used. [9][10][11] In the last decade, effort has been put into refinement of risk estimation for the development of heart failure after cancer therapies, with little success for blood biomarkers but with a better outlook for more sensitive, early echocardiographic parameters. [12][13][14] In other populations with (risk of) cardiovascular disease, global longitudinal strain (GLS) shows very promising results for earlier identification of those at risk for heart failure and death. 7,8,15 In CCS, abnormal GLS is highly prevalent, 12,16 but its prognostic value has yet to be shown. Myocardial strain measurement has not been widely adopted in clinic, possibly hampered by the different algorithms used by different software vendors, which can be partly resolved by the use of vendor-independent software. 17 The Dutch Childhood Cancer Survivor Study (DCCSS) LATER cohort (1963-2001) part 2; clinical visit and questionnaire study provides a unique opportunity to investigate cardiotoxicity, next to various late effects in a large sample of very long-term survivors (DCCSS LATER 2 CARD study). This substudy includes solitary and combined analyses of echocardiography, electrocardiograms, and blood biomarkers. 18 The echocardiographic measurements serve as primary study outcome and as reference for the biomarker and ECG studies. Specific aims of the echocardiography study are to evaluate the prevalence and associated (treatment and lifestyle related) risk factors of subclinical cardiac dysfunction in CCS compared with identification of CCS at highest risk of developing heart failure. This paper describes the design, feasibility, and reproducibility of the echocardiography protocol.
Methods: Echocardiograms from CCS and sibling controls were prospectively obtained at the participating centers and centrally analyzed. We describe the image acquisition, measurement protocol, and software-specific considerations for myocardial strain analyses. We report the feasibility of the primary outcomes of systolic and diastolic function, as well as reproducibility analyses in 30 subjects.
Intra-class correlation coefficients for inter-observer variability were 0.85 for LVEF, 0.76 for GLS, 0.70 for GCS, 0.89 for RVFWS and 0.89 for LASr. Intra-class correlation coefficients for intra-observer variability were 0.87 for LVEF, 0.82 for GLS, 0.82 for GCS, 0.85 for RVFWS and 0.79 for LASr.

Conclusion:
The DCCSS LATER 2 CARD study includes a protocolized echocardiogram, with feasible and reproducible primary outcome measurements. This ensures high-quality outcome data for prevalence estimates and for reliable comparison of cardiac function parameters.

K E Y W O R D S
2D echocardiography, cardiac toxicity, diastolic function, myocardial strain, systolic function sibling controls, and with identify more sensitive echocardiographic markers of subclinical cardiac dysfunction.
As myocardial strain analysis is very comprehensive and results are influenced by the software and definitions used, it is of key importance to describe the methods in detail. Furthermore, the feasibility of (strain) measurements should be described, as they depend on image quality. Last, the measurements should be reproducible in research and clinical follow-up. The overall design of the DCCSS LATER 2 CARD study has recently been published. 18 Here, we describe the image acquisition, the protocol for offline conventional and strain measurements, as well as their feasibility and reproducibility, as part of the multicenter echocardiography study in the DCCSS LATER 2 CARD cohort.

| Patient population
The cross-sectional DCCSS LATER study part 2 investigates late treatment effects in a nationwide cohort of 5-year CCS, treated under the age of 18 years between 01-01-1963 and 31-12-2001. 18 This baseline cohort comprises 6,165 CCS, of which 5,455 were alive at study inception. From this cohort, the DCCSS LATER 2 CARD study aimed to include 1,900 CCS for cardiac evaluation with echocardiography, electrocardiography, and blood biomarkers. Of these, 1,600 CCS were defined as risk group 1, who received well-known cardiotoxic therapy (anthracyclines, mitoxantrone, radiotherapy on the heart region, solitary, or combined). 3 Risk groups 2, 3, and 4 were study groups of at most 100 subjects each, having received either cyclophosphamide, ifosfamide, or vincristine, respectively, without any other studied treatment. A control group of untreated siblings was recruited to account for background cardiovascular risk. Subjects were recruited from 7 Dutch pediatric oncology centers, between February, 2016 and February, 2020. For risk group 1, a protocolled echocardiography was part of standard surveillance, 6 whereas for the other risk groups and sibling controls, all diagnostics were obtained for research. For participants from risk group 1 who did not have an indication for a new surveillance echocardiogram during the study period (surveillance echocardiography was recently performed, or the participant was already under the care of a cardiologist), we obtained their most recent echocardiogram if performed no earlier than January 1, 2016. All participants gave their informed consent for the use of study, and clinical data and the medical ethic boards of all participating centers approved the study protocol.

| Echocardiography
Pediatric and adult cardiologists developed the echocardiography protocol, in collaboration with the DCCSS LATER 2 CARD steering committee and the Dutch childhood cancer cardiac consortium.

| Image acquisition
Experienced sonographers acquired images on the locally available Philips (8%) or GE (92%) stations. Table 1 summarizes the requested images. From approximately halfway the inclusion period, we also included the right ventricle (RV) focused apical four-chamber view in our protocol. For all images, three heart cycles were recorded, with these exceptions: five cycles for patients in atrial fibrillation, five cycles for tissue Doppler imaging (pulsed wave frame rate >180/s), and five cycles for strain analysis (preferred frame rate 60-100/s and ratio with heart rate ≥3:4; minimum 45/s). 19,20 For color Doppler, Nyquist limit was set at 50-70 cm/s.

| Data storage and handling
Images were locally stored and pseudonymized. Raw DICOM files were transferred to our echocardiography core laboratory at the Radboud university medical center, Nijmegen, via protected storageor exchange media. Extraction of our offline measurement results from the analysis software was automated using custom scripts, and the data were then imported in a web-based database (Castor EDC, Ciwit BV, The Netherlands).

| Measurement protocol
For reader convenience, we refer to conventional (all but strain) and strain measurements. Echocardiograms obtained for standard care were analyzed by the sonographer and an imaging cardiologist.
For "research-only" echocardiograms, only images (without online measurements) were stored. All measurements were (re)performed offline at the core laboratory by one out of two observers (RM or JL). The observers were blind for all participant information such as previous cancer diagnosis, therapy modalities and doses, cardiovascular risk factors, electrocardiographic findings, and blood tests.
Analysis of conventional parameters for research-only participants was performed upon receipt in the core laboratory, to be able to report unexpected findings (intra-cardiac tumors, congenital heart disease, valve dysfunction, myocardial dysfunction) to their own physician within six months, as indicated by the ethical committee.  Table 2.
Observers subjectively rated the overall image quality (including acoustic window, artefacts, contrast) of an examination as "good," "fair,' "moderate," or "poor," allowing sensitivity analyses for the study results. When available, left ventricular dimensions were measured on 2D parasternal long-axis images rather than Mmode, to prevent oblique measurements. 22 LVEF was calculated according to the biplane Simpson's method. 22 EchoPac's speckletracking based LVEF measurement ("autoEF") was preferred to enhance reproducibility, though manual adjustment of endocardial contours, and reference frames remained possible. As not all participating centers routinely measured 3D LVEF, this measure was not included in the protocol. Tissue Doppler velocity measurements were performed and averaged over three regular cardiac cycles.

Event timing
The cardiac cycle with the best image quality was selected. For all cardiac chambers, end-diastole was set at the QRS-peak to pursue the highest reproducibility. 23 End-systole was set on the minimum of the volume or area curve derived by the software. Unless stated otherwise, we report end-systolic strains for the LV and RV and peak

Contour drawing and tracking
Endo-and epicardial 15-point contours were manually applied to the LV wall in end-systole, tracked by the software and, if necessary, manually adjusted in end-diastole to cover the whole cardiac wall.
Tracking quality was visually confirmed based on the superimposed grayscale images and the typical curve morphology. If tracking was TA B L E 2 Standard echocardiographic measurements and derived parameters c TDI measurements are performed in three cardiac cycles. d Besides qualitative evaluation of morphology, valve opening, Color Doppler and additional measurements to determine severity. [38][39][40] inadequate, the contours were iteratively adjusted. 25 In each view, the software divided the myocardium in six segments. Not-tracked segments could be discarded to a total maximum of three segments in the three apical views and one segment in the short-axis view. 23 Per default, we report midwall (referred to by the software as myocardial) strain for the LV, 23 which has been shown to be the least susceptible to potential foreshortening. 26 Endocardial longitudinal strain of the RV was preferably measured from a RV focused apical four-chamber view. To anchor the tracking of the RV apex, the interventricular septum was tracked, but then excluded to report the RV free wall longitudinal strain.
Discarding of one of the three RV free wall segments was accepted when inadequately tracked. Endocardial longitudinal strain of the LA was measured from a non-foreshortened apical four-chamber view.
As no segmentation applies to the LA, no segments could be discarded and global strain is calculated. 24

Calculation
Left ventricular GLS was calculated as the mathematical average at end-systole of the 18 segments derived from the three apical (fourchamber, three-chamber, two-chamber) views, 27  Global circumferential strain (GCS) was measured in short-axis view at the papillary muscle (mid-ventricular) level and was averaged over six segments. 23 RV free wall longitudinal strain was averaged over three free wall segments and for the LA, longitudinal reservoir strain was reported per default (no segmentation, Figure 1A). 24 Strain measurements will be referenced as absolute values (ie, −21% is "better" than −18%).
Left ventricular mechanical dispersion, an important prognosticator for ventricular arrhythmias in multiple cardiac diseases, 28 was defined as the standard deviation of the 18 segmental time intervals from the QRS-peak to peak negative strain and expressed in msec ( Figure 1B). Segmental strain curves that showed no (clear) negative peak (ie, dyskinetic and akinetic segments) cannot be included in assessment of mechanical dispersion, as they do not allow timeto-peak calculation.

| Training and supervision
All sonographers on-site were familiarized with the protocol. The two core-laboratory observers were physicians and extensively

| Study outcomes
Primary study endpoints were previously defined, 18

| Statistical analysis
The feasibility of the primary outcomes of systolic and diastolic function and all strain analyses is presented as a percentage of all analyzed echocardiographic examinations. Differences in proportions between CCS and siblings were analyzed using Pearson chi-square test. Core laboratory inter-and intra-observer variability was tested in 30 randomly selected participants with sufficient image quality for the measurement concerned (RM, JL).
Intra-observer measurements were at least two weeks apart.
The same cardiac cycle was used to exclude any temporal vari-

All 18 segments were included in 75% of the GLS calculations in CCS
and in 74% of those in siblings (P=.64).
Supplemental Figure 1

| Core laboratory variability
The core laboratory inter-observer variability in a subset of 30 participants is depicted in Table 4 and Intra-observer analysis (

| Feasibility
The feasibility of the (advanced) echocardiographic measurements was higher in sibling controls than in the participating CCS.
Although our overall image quality rating was not different between CCS and siblings, it might not be robust enough to detect  protocol. Of note, the core laboratory measurements did not start before siblings were included, to ensure blinding.
Third, to prevent missing CCS with less frequent surveillance or with more severe cardiotoxicity, we included some additional echocardiograms as stated in the methods section.
Nevertheless, the feasibility of GLS measurement in 80% in the context of an acquisition protocol is comparable to that of the Normal Reference Ranges for Echocardiography study. 17

| Core laboratory variability
Generally, the inter-observer variability of conventional and strain measurements in the present study lies within the ranges reported in the literature. 9,12,17,31 Two measurements are of particular importance to discuss.
First, the reproducibility of biplane LVEF is comparable to that of GLS, which we attribute to the use of the semi-automated endocardial border tracking option for LVEF measurement. This functionality is based on speckle tracking and minimizes user interaction, which has been shown to reduce inter-observer variability and the need for expert readers, compared with manual contouring. 32,33 Second, we chose not to include global radial strain (GRS) in our protocol. GRS has been reported to have inferior reproducibility compared to GCS, 17,31,34 although it is acquired during the same analysis in the short-axis view. The GRS measurement is subject to Abbreviations: ICC, intra-class correlation coefficient; LA, left atrium; LOA, limits of agreement; LV, left ventricle; LVEDD, LV end-diastolic diameter; LVESD, LV end-systolic diameter; RV, right ventricle; TAPSE, tricuspid annular plane excursion; TDI, Tissue Doppler Imaging. a As a percentage of the mean measurement of the two observers the lower lateral than axial resolution of echocardiographic images.
The specific software we use offers a useful function to aid tracking by adjusting the end-diastolic contours after the myocardium has been tracked. Since these corrections entail both the endocardial and epicardial contours, we judged the pitfall of accumulating manual measurement errors, that are relatively large compared with the myocardial wall thickness, to be unacceptable. Our data support findings in the literature that GCS measurements are more difficult to reproduce than GLS measurements. 17,31 Notably, LV mechanical dispersion showed acceptable ICCs for inter-and intra-observer variability, but with wide limits of agreement (11-19msec, 27%-53%) relative to the measured values. These limits of agreement are accepted in the field, as differences of 20msec were shown relevant for predicting arrhythmias. 28

| Myocardial strain
Myocardial strain imaging has not been fully adopted by the chamber quantification guidelines, 22 but has been endorsed by the adult cardiotoxicity imaging expert consensus. 11 As there is sufficient evidence that GLS is feasible, reproducible and adds prognostic value in a variety of cardiovascular diseases, 7,8,15

| Limitations
Although LVEF measured on three-dimensional echocardiography proved to be more reproducible and thus more suitable than biplane LVEF to detect subtle changes over time, 10   analysis were performed in a random sample that not include patients with very low LVEF. However, in the setting of surveillance echocardiography the reproducibility in "borderline cases" may be of most importance.

| CON CLUS IONS
The echocardiographic substudy of the nationwide cross-sectional DCCSS LATER 2 CARD study evaluates the prevalences of contemporary systolic and diastolic function parameters in a large cohort of CCS and has a parallelly included sibling cohort for comparison. It includes a protocolized echocardiogram, with feasible and reproducible primary outcome measurements.

ACK N OWLED G EM ENTS
We thank the other members of the DCOG LATER consortium