BNP was measured using Triage BNP assay (Biosite Inc, San Diego, California). The interassay percentage coefficient of variation was 8.8% to 11.6%. The detection limit was 5 pg/ml and upper measuring limit was 5,000 pg/ml. hs-TnT was measured using a highly sensitive assay on an automated platform (Elecsys E170, Roche Diagnostics, Indianapolis, Indiana) with lower limit of blank (3 ng/l) and

interassay percentage coefficient of variation ≤10%. Cardiac magnetic resonance was performed at baseline and at 36 months on a 3-T Magnetom Trio scanner (Siemens, Erlangen, Germany) using body array and spine matrix radiofrequency coils as described in TSA HDAC detail previously 9 and 10. CMR images were analyzed offline by an independent, blinded, magnetic resonance physicist (S.J.G.) Wnt inhibitor using commercial software (Argus, Siemens Multi-modality Work Platform, version VB 15, Siemens). Electronic region-of-interest contours were placed around endocardial and epicardial LV borders on all CMR image slices at end-diastole and end-systole that were identified to contain 50% or more full-thickness myocardium. Papillary muscles were included in the LVM if the muscle structure was indistinguishable from the myocardial

wall, but otherwise assigned to the LV blood pool. The process of contour placement was repeated such that every patient dataset at both time points was analyzed twice to optimize the measurement precision. The intraobserver variability was 2.02% at baseline and 1.97% at follow-up. Data for continuous variables are presented as mean ± SD for normally distributed data and median and interquartile range for nonnormally distributed data. Categorical data are expressed as numbers (%). Comparisons between continuous variables were analyzed using the Student t test or Mann-Whitney U test, whereas categorical variables were analyzed using chi-square test or Fisher exact test. The primary outcome measure was change in left ventricular all mass (ΔLVM) from baseline to follow-up at 3 year. The study population was divided into 2 groups depending on the rise or fall

in LVM at follow-up compared with the LVM at baseline. We also divided the study cohort into tertiles based on BNP levels according to a prespecified protocol in 2 ways. The first was by dividing the 50 patients into the cohort’s own BNP tertiles. The second was to use the tertile BNP levels of the original 300 patients: this latter approach was used to avoid bias in the way the patients in this substudy were selected from the full cohort (n = 300). The significance level for the trend across the tertiles was calculated by Jonckheere-Terpstra test and chi-square test. Multivariable models were used to identify the predictors of ΔLVM and to calculate c-statistics, and area under the curve was compared by the DeLong method.