Study
All echocardiography and CMR data were acquired in the context of the Mineralocorticoid Receptor Antagonists in End-Stage Renal Disease (MiREnDa), which was a prospective, randomized, placebo-controlled, double-blind, parallel group, three-center phase 2 intervention study investigating the efficacy and safety of spironolactone in hemodialysis patients. The study design, flow, and endpoints were reported previously [19, 20]. Briefly, patients were enrolled at 20 dialysis centers. Cardiac imaging was assessed in one of the three participating university centers (Frankfurt, Erlangen-Nuremberg, and Würzburg) according to predefined standardized procedures. The patients underwent three dialysis sessions per week. To avoid inaccuracies in measuring LVM, CMR and echocardiography were performed immediately after each other for all subjects on dialysis-free days. Data from CMR and echocardiography were collected after the placebo run-in phase and before randomization (week 0 visit). The coordinating university for the study was the University Hospital of Würzburg. The Center for Clinical Trials at the University Hospital of Würzburg was responsible for implementing the study and the statistical analyses. Blinded data assessment was performed by trained research staff [19, 20]. The data used in this analysis were collected before the double-blind treatment phase.
Echocardiography
All patients underwent a comprehensive echocardiographic study by an experienced cardiologist blinded to CMR findings. The protocol based on the current ASE recommendations utilized linear measurements derived by 2D TTE-guided M-mode approach [5, 13]. LVM was calculated from the TTE measurements using the following equations:
Where LVIDd is the LV internal diameter at end-diastole, PWTd is the posterior wall thickness at end-diastole, and IVSTd is the interventricular septal thickness at end-diastole. LVM indexation to anthropometry was calculated by body surface area (BSA) using the Mosteller formula [22]. In addition, the indexation by height2.7 was used [23]. In this case, the LVMI is marked accordingly as LVM/height2.7.
CMR imaging
CMR images were acquired using both 1.5 or 3 Tesla magnetic resonance imaging scanners. Imaging was performed using an electrocardiogram-gated, breath-hold, 2D, steady-state, free precession cine with contiguous, left ventricular short-axis stack of images acquired from above the base to below the apex of the left ventricle with a slice thickness of 8.0 mm, pixel size smaller than 1.4 mm×1.4 mm, field of view ~320 mm (adapted to patient size), and temporal resolution <50 ms.
Image analysis was performed using commercial software (Circle cvi42®). LV myocardial area slices were measured planimetrically by delineating endocardial and epicardial LV borders at end-diastole and end-systole, allowing calculation of end-diastolic and end-systolic volumes, as well as myocardial volumes. LVM was calculated by summing the short axis slice volumes at end-diastole. Papillary muscles and trabeculae mass (PMT) were included in the LV cavity volume. LVM was mainly indexed to BSA using the Mosteller formula (see above). LVH was defined as indexed LVM±2 standard deviations of CMR references [24].
Scans were analyzed by an experienced radiologist (J.D.) blinded to the results of echocardiography. To validate CMR data, a second independent radiologist analyzed a random subsample of 26 (13%) CMR scans according to the same predefined procedures [20].
Statistical analysis
Mahalanobis distance was used to detect outliers in the data set and data from one subject was excluded from analysis due to its strong statistical divergence from the other LVMI measurements in population. A further exclusion criterion was the absence of data from CMR or echocardiography. Continuous data were expressed as median (quartiles) and categorical data as frequencies (percentages). If data matched the assumption of normal distribution (Shapiro-Wilk test, p>0.05), continuous data in independent groups in paired samples were compared by Student`s unpaired or paired t test (k=2) or single factor variance analysis (k>2 independent groups), otherwise Mann-Whitney U/Wilcoxon sign rank test (k=2) or the Kruskal Wallis test (k>2) was used. Distributions for categorical data were compared using χ2 or Fisher’s exact test if appropriate. The degree of correlation and its confidence interval were computed using Concordance Correlation Coefficient [25]. The mean difference in LVMI between CMR and echocardiography was calculated according to the formula = . Agreement between the CMR technique and echocardiographic methods was analyzed by the Bland-Altman plot. The bias was estimated by the mean difference between CMR and echocardiography. In the case of constant bias, evaluated by the regression line, we specified the 95% limits of agreement. The stratification into LVMI quartiles was based on CMR measurements of LVMI. To compare the diagnostic performance of both echocardiographic formulas, we performed receiver operating characteristics (ROCs) to calculate the area under the curve (AUC). To test its significance, we used the method by Hanley et al. If z ≥1.96, the null hypothesis AUC1=AUC2 was rejected by means of the z-test with a significance level of 0.05 and the so-called true ROC areas taken as significantly different [26]. Cut-off values for classifying LVH status were calculated by the Youden index. A two-tailed p-value <0.05 was considered significant. For this purely exploratory analysis, no corrections for multiple comparisons were made. Statistical analyses were conducted using SPSS Version 19.0 (IBM Corp., Armonk, NY, USA) and Excel 2016 (Microsoft Corp., Redmond, USA).