Statistical analysis

Statistical analysis was performed using Statistical Package for Social Sciences (SPSS), V.22.0 (IBM, Chicago, Illinois, USA). Categorical variables were reported as frequency and percentages. Comparison analyses for categorical data were performed using the χ² or Fisher’s exact tests, dependent on appropriateness. Continuous data underwent histogram plots for assessment of normality. Normally distributed data were reported as mean±SD and comparative analysis performed using independent t-tests. Non-normally distributed data were reported as median (full range) and compared using Mann-Whitney U test. Paired continuous data were analysed using paired t-test or Wilcoxon Signal Rank test, as appropriate. Non-normally distributed ECM biomarkers, NT-pro-BNP, GDF-15 and hs-TnT data were presented unadjusted; however, they were transformed logarithmically for analysis. Fold change was calculated using the mean (responder/non-responders) cohort value when comparing two independent datasets. The fold change for paired data sets was also calculated using matched means (coronary sinus/peripheral). Variation in continuous variables over three time periods was analysed using either one-way analysis of variance (ANOVA) or Friedman test, respectively. Mixed between-within subjects ANOVA was used to compare variation in continuous data in functional responders and non-responders over 6 months. Bivariate correlation analysis with either Pearson (parametric) or Spearman rank (non-parametric) estimators was performed between two continuous variables to explore relationships. Univariate logistic regression analysis was performed for functional response for predefined vascular biomarkers and established clinical variables. Those with p<0.2 were pooled as covariants for multiple logistic regression. A high alpha was set on the basis of the clinical response definition. A stepwise entry method was applied with forward selection and backward elimination. The accuracy of the model was verified with a Hosmer-Lemeshow goodness-of-fit test. P<0.05 was considered statistically significant. Given the limited data in this field, we were unable to perform a power calculation at the start of our study. We, therefore, performed this single-centre pilot study to be able to more accurately predict power for the most prominent biomarkers found enabling a larger multicentre trial to be done. Following completion of our study, we were able to perform a power calculation based on the two most important predictors noted (PINP and CTx). In order to demonstrate a significant difference between baseline concentrations in PINP and CTx as predictors of response (assuming a non-response rate 40%), a sample size of 430 was calculated and assumed an 80% power with significance level of p=0.05.