Statistical Analysis
Data analysis was executed using R. Two-sided p-values with alpha=0.05 were used. Distributions of characteristics were tabulated using percentages for categorical variables and means with standard deviations for continuous variables. Six multivariable linear regressions were fit for the six outcomes using all risk factors as independent variables. Models estimated beta coefficients (β) and 95% confidence intervals (CI) representing associations between risk factors and outcomes.
Linear regression assumptions were evaluated using plots and hypothesis tests. QQplots verified the assumption of normality. To test for heteroskedasticity, residual plots were generated along with a non-constant variance test. There was strong evidence of heteroskedasticity. To correct this, all outcomes employed a Box-Cox transformation. Lack of multicollinearity was confirmed by estimating variance inflation factors.