Accuracy of left ventricular ejection fraction by contemporary multiple gated acquisition scanning in patients with cancer: comparison with cardiovascular magnetic resonance

The study sample consisted of consecutive patients with cancer who had both MUGA and CMR within 30 days between January 2007 and September 2016 at the University of Minnesota Medical Center, Minneapolis, Minnesota, USA. The institutional MUGA database was cross-matched with the University of Minnesota Cardiovascular Magnetic Resonance Registry, an ongoing observational registry including all patients that undergo CMR at the University of Minnesota, to identify the study patients. Patients were excluded if their records indicated any of these intervening clinical events that could potentially impact cardiac function: acute myocardial infarction, heart failure hospitalization, administration of potentially cardiotoxic cancer treatment, or acute systemic illness such as sepsis. An electronic database was created to include demographic information, medical history including reasons for the studies, co-morbidities and medications, and MUGA and CMR findings for each patient. This retrospective cross-sectional study was approved by University of Minnesota’s Institutional Review Board with a waiver of informed consent.

MUGA scans were performed per standard recommendations to determine LVEF [6, 7] before or after cancer treatment. The UltraTag RBC kit (Mallinckrodt, Inc., St. Louis, Missouri, USA) was used. Erythrocyte labeling was performed using modified in vitro method with technetium 99 m-labeled red blood cells with an activity of approximately 11 to 13 MBq/kg. Images were acquired with a Siemens e-cam dual-head gamma camera (Siemens, Erlangen, Germany) equipped with a parallel hole, low energy high-resolution collimator, with energy window of 15% symmetrically placed over a photopeak of 140 keV. Data were acquired in electrocardiogram-synchronized frame mode using 24 frames per cardiac cycle, with 128 × 128 matrix of 16-bit pixels. Acquisition times were adjusted to achieve a minimum of 200,000 counts per frame. Patients were resting and supine, and the best septal view was individually adjusted from 45° left anterior oblique position with 10°–15° caudal tilt.

Experienced nuclear medicine technologists performed LVEF analyses. Scintigrams were smoothed off-line using standard algorithms, and background correction was performed. The LV regions of interest, as well as background activity, were selected automatically by the computer program (E. Soft; Siemens Medical Solutions, Erlangen, Germany) with manual correction by the interpreting technologists as necessary. Left ventricular time-activity curves were constructed, and LVEF was calculated as ([background-corrected end-diastolic counts − background-corrected end-systolic counts])/(background-corrected end-diastolic counts) × 100. Since our aim was to evaluate “real-world” MUGA data, clinically reported LVEFs were used by design, and analyses were not repeated for this study.

CMR was performed on clinical 1.5 T Siemens scanners (Avanto and Aera) using phased-array receiver coils according to standard recommendations [8, 9]. A typical protocol was as follows: First, localizers were acquired to identify the cardiac position, and the standard long- and short-axes of the heart, and then cine images were acquired in multiple short-axis (every 10 mm to cover the entire LV from the mitral valve plane through the apex) and three long-axis views (2, 3, and 4 chamber) using a steady-state free-precession (SSFP) sequence (repetition time msec/echo time ms, 3.0/1.5; temporal resolution, 35–40 ms; slice thickness, 6 mm; inter-slice gap, 4 mm; flip angle, 60°; in-plane resolution, approximately 1.6 mm × 1.6 mm). Other sequences, including perfusion and delayed-enhancement imaging were performed as clinically indicated.

The LVEF was determined by quantitative analysis according to standard recommendations [10]. To allow use as a reference standard, all CMRs were re-analyzed, blinded to MUGA and clinical data, by a single investigator (C.S.) with 9 years of experience in CMR. Short-axis cine images were used for manual tracing of LV endocardial and epicardial contours at end-diastole and end-systole. Papillary muscles were excluded from the LV myocardial mass (i.e., included in the LV blood volume) to match the methodology used in MUGA. This is an acceptable approach [10], allowing quicker quantitative analyses [11] and is more practical in the routine clinical setting. The LVEF was calculated as ([LV end-diastolic volume − LV end-systolic volume])/(LV end-diastolic volume) × 100.

Statistical analyses were performed on Stata version 13 (StataCorp LP, College Station, Texas, USA). Continuous variables were expressed as means and standard deviations, or medians and inter-quartile ranges (IQR) for data that were not normally distributed. MUGA clinical and CMR reference LVEFs were compared using Student’s paired t-test, Bland-Altman analysis and Lin’s concordance correlation coefficient [12]. Lin’s concordance correlation coefficient (r _c ) provides a measure of reliability based on covariation and correspondence, unlike Pearson’s correlation coefficient (r) that provides a measure of linear covariation without accounting for the degree of correspondence between the two sets of values. Pearson’s correlation coefficient (r) was used to examine the correlation between the time interval between the two studies and differences between MUGA and CMR LVEFs. We studied the two most common absolute thresholds of LVEF used in cardiotoxicity clinical studies and practice: 50 and 55% [13]. The kappa statistic was used to assess agreement between MUGA and CMR classification of normal vs. abnormal LVEFs. Kappa values were interpreted as widely accepted in the literature [14] – a kappa value of <0 would be considered as less than chance agreement, between 0.01 and 0.20 as slight agreement, between 0.21 and 0.40 as fair agreement, between 0.41 and 0.60 as moderate agreement, between 0.61 and 0.80 as substantial agreement, and between 0.81 and 0.99 as almost perfect agreement. Logistic regression was used to determine univariate predictors of misclassification. All statistical comparisons were two tailed, and a p value less than 0.05 was considered to indicate statistical significance.

Eighty patients were identified as having had MUGA and CMR within 30 days of each other, of which 77 were cancer patients. Two patients received potentially cardiotoxic cancer treatment between MUGA and CMR and were excluded from the study. The remaining 75 patients formed the study cohort. Patient characteristics are listed in Table 1.

All MUGAs were performed for assessment of LVEF prior to, or after potentially cardiotoxic cancer treatment. CMRs were performed for characterization of cardiomyopathy noted on a prior imaging study (69%), evaluation of suspected obstructive coronary artery disease (15%), evaluation of suspected infiltrative cardiomyopathies (12%), abnormal electrocardiogram (3%), and in one case, evaluation of a suspected intracardiac mass (1%).

There was a median of 8 days (interquartile range 5, 10 days) between MUGA and CMR studies. In 62 (83%) patients, the studies were performed within 15 days of each other. MUGA was performed at least a day before CMR in 69 (92%) patients, the two studies were performed the same day in two (3%) patients, and MUGA was performed after CMR in the remaining four (5%) patients.

The mean MUGA LVEF was not significantly different when compared with the mean CMR LVEF (48.5% vs. 50.0%, p = 0.17). However, the random error between MUGA and CMR LVEFs was substantial, as evidenced by the Bland-Altman 95% limits of agreement (−19.4, 16.5) (Fig. 1). In 42 (56%) patients, the MUGA LVEF was lower than the CMR LVEF.

The Lin’s concordance correlation coefficient (r _c ) was 0.63. Using the cutoffs proposed by McBride [15], this indicates poor agreement between MUGA and CMR LVEFs.

There was no correlation between the time interval between MUGA and CMR and the absolute difference between MUGA and CMR LVEFs (r = −0.20, p = 0.08), or the time interval between the two studies and whether MUGA LVEF was higher or lower than CMR LVEF (r = 0.05, p = 0.65).

Using a LVEF threshold of 50%, there was misclassification of 26 of 75 (35%) patients between normal and abnormal categories (Table 2 and Fig. 2) – 19 patients that had MUGA LVEF <50% had a CMR LVEF ≥50%, and seven patients that had MUGA LVEF ≥50% had CMR LVEF <50%. Thus, MUGA LVEFs only had fair agreement with CMR LVEFs (kappa = 0.31).

Next, using a LVEF threshold of 55%, there was misclassification of 15 of 75 (20%) patients between normal and abnormal categories (Table 2 and Fig. 2) – 12 patients that had MUGA LVEF <55% had a CMR LVEF ≥55%, and three patients that had MUGA LVEF ≥55% had CMR LVEF <55%. MUGA LVEFs had a moderate agreement with CMR reference LVEFs (kappa = 0.54).

Figure 3 shows MUGA and CMR images from two study patients with misclassification between MUGA and CMR LVEFs.

On univariate logistic regression analysis, the only significant predictor of misclassification at the 50% LVEF threshold was the indexed LV end-diastolic volume (Table 3). At the 55% LVEF threshold, significant predictors of misclassification were indexed LV end-diastolic volume, atrial fibrillation and hematocrit (Table 3). Thus, a smaller indexed LV end-diastolic volume was a predictor of misclassification at both thresholds for a normal LVEF of ≥50% and ≥55%.

Our findings must be interpreted in the context of the study design. Rather than to perform a head-to-head comparison of two imaging modalities, our aim was to examine the accuracy of real world, clinical LVEFs by MUGA, which oncologists and cardiologists use every day in clinical practice to make important decisions. To achieve this, we compared clinically analyzed MUGA LVEFs with CMR LVEFs that were ascertained by a single blinded expert investigator for this study. This design allowed comparison of real world MUGA LVEFs to arguably the most accurate estimates possible, of the true LVEF (since even a necropsy cannot provide a LVEF).

Since MUGA and CMR studies were not performed on the same day in all cases except one, there is a possibility of true LVEF changes in the interim period. We limited this possibility by excluding patients with clinical events and potentially cardiotoxic treatment during or in the time interval between the two studies. Additionally, we did not find correlations between the time interval between the two techniques and the absolute difference in LVEFs, whether MUGA clinical LVEF was higher or lower than CMR reference LVEF, or whether there was misclassification between normal and abnormal categories.

While making clinical decisions on the management of cardiotoxicity, the change in LVEF is often used in conjunction with the absolute LVEF. We did not investigate changes in LVEF in this study. Finally, this is a relatively small, single-center study subject to referral bias.