Statistical analysis
Normality of continuous variables was quantified using the Shapiro-Wilk test and was reported as the mean (standard deviation, SD) and the median [inter-quartile range, IQR] for parametric and non-parametric data respectively. Two group comparisons of continuous data for those with and without CMD was performed using a non-paired t-test and Wilcoxon rank-sum tests respectively. Comparisons between the MBFR < 2.0 group and the MBFR ≥ 2.0 group were undertaken using a Fisher’s Exact Test.
In univariate and multivariate analysis, septal mitral annular E/e’ was used as the main outcome variable while known influencers of septal mitral annular E/e’ (age, gender, diabetes, hypertension, smoking status, Body Mass Index (BMI), fasting TG, fasting LDL, MBFR and Hs-CRP) were used as the input variables. For multivariate analysis, an initial fully saturated model was constructed using all the known predictor variables. Non-significant predictors were sequentially deleted, and the model re-run. Likelihood ratio testing was employed to quantify significant differences, or lack thereof, between model iterations. The final models only contained the significant predictors. Results are reported as regression slope (β) and its level of significance. Where required, inferential analysis was performed using standard regression techniques except in the case of a dichotomous outcome variable, where logistic regression was employed. Predictor variables were tested for co-linearity using a standard Pearson’s correlation. Variables with correlation coefficients (R) greater than ±0.80 were considered co-linear with one or other excluded largely on empirical grounds. Where appropriate, dummy variables were used when examining the significance of categorial predictors and all two-way interactions were tested for significance. Finally, all statistical models were tested for adequacy using appropriate regression diagnostics with regression residuals tested for normality using the Shapiro-Wilk test and for homoscedascity using the Breusch-Pagan test.
Statistical analysis was performed using a propriety statistical package (STATA version 15.1) with the level of significance set at p < 0.05.