Statistical analysis
Continuous variables are expressed as mean ± SD. Categorical data are summarized as frequencies and percentages. Baseline clinical characteristics were compared between patients with ICD, low risk CRT-D and high risk CRT-D, using Wilcoxon ranked sum test for continuous variables and Chi2 - test for dichotomous variables, as appropriate.
Step 1- covariate selection: We included 29 potential clinical, electrocardiographic, and laboratory binary risk factors for VTA (Listed in eTable A in the Online Appendix). Numeric variables were made binary by the use of cut points with the goal of finding a simple, easily implemented predictors to be derived from them. Thresholds for categorization of numeric variables were based on the mean or clinical relevance. Univariate relationships between candidate covariates and a further event were assessed by t tests (2 for binary responses). The covariates with values of P<0.20 were further evaluated by carrying out a best-subset regression analysis, examining the models created from all possible combinations of predictor variables, and using a penalty of 3.84 on the likelihood ratio 2 value for any additional factor included (corresponds to a P of 5% for a 1-df 2 test).
Step 2 – Creating the groups: For the main analysis, CRT-D patients were grouped in two groups based on the presence or absence of RF. In a secondary analysis, we further grouped CRT-D patients into 3 groups; (i) No RF, (ii) One RF, and (iii) two or more RF. Since the identified CRT-D RF were not associated with outcomes in the ICD group (online supplemental Figure A), ICD patients were not further grouped and were included as a control group
Step 3 – Outcomes by score analysis: Kaplan–Meier estimates for any VTA, VTA or death, and appropriate shock in patients with ICD and CRT-D, stratified by their risk, were determined and statistically evaluated with the log-rank test. Multivariate Cox proportional hazards regression analyses were carried out in the subgroups for the assessment of the primary and secondary end points. The following covariates were included in the Cox regression models: age, gender, QRS width, New York Heart Association class NYHA, creatinine, left ventricular ejection fraction, diabetes, and ischemic origin.
All statistical tests were two-sided, a p-value of <0.05 was considered statistically significant. Analyses were carried out with SAS software (version 9.4, SAS institute, Cary, North Carolina).