nalyses were performed in the statistical environment R, with the lme4 package for LME analysis and the mice package for MI based on chained equations. The LME analysis used was similar to the analysis performed in the original METEOR study. Isoliquiritigenin inhibitor The levels used were defined by the participant and the carotid artery site within the participant. The repeated measure was time. The model was specified in terms of fixed effects for carotid artery site, age, sex, scan reader, ultrasound machine, treatment group, time, and the interaction between treatment group and time. This beta of this interaction term was the difference in rate of change in CIMT between Rosuvastatin and placebo and was considered as the treatment effect. Time as a continuous variable was the interval from the date of randomization to date of CIMT measurement.
Random effects within the model were intercept and slope for individual participants and for sites within participants. Subsequently, five imputed Hesperidin 520-26-3 data sets were created using MI. The imputation model included fixed effects for age, sex, treatment allocation, CIMT measurements of the same visit, and all other CIMT measurements on the same artery segment during all other visits, resulting in a total of 38 predictor variables for each missing CIMTmeasurement. CIMT measurements performed after the missing observation were explicitly included in this model because it has been shown previously that including the outcome improves the MI process. Random effects were not included in the imputation model.
After analyzing the data using the LME model for each imputed data set, the obtained estimates and variances were recombined using Rubin,s method. The pooled estimate is simply the average of the estimates of theMimputed data sets. The total variance of the pooled estimate is calculated from two components. One component reflects the within imputation variance, and the other component reflects the between imputation variance. The within imputation component preserves the natural variability within data sets, and the betweenimputation component estimates the uncertainty caused by the imputed missing data and measures how the point estimates vary across the M imputations. The estimate from the first imputed data set was stored separately, to assess the performance of SI.
The annualized difference in rate of change in maximum CIMTbetween Rosuvastatin and placebo and the corresponding 95% CI were calculated with four different LME modeling methods: standardLMEfor the complete bootstrapped data set, standardLMEon the data set with missing CIMT values, LME for the data set with missing values after first being imputed with SI, and LME for the data set with missing values after first being imputed with MI. For each missing value scenario, the mean squared error of the LME estimate was calculated. The MSE represents the squared difference between the reference effect size and the estimated effect size and incorporates both the variance of the estimator and its bias. In situations where there is no bias in the point estimates, the MSE reflects the variance. A large MSE indicates that estimated values are more biased or imprecise or a combination of both. The MSEs were very small and were therefore multiplied by 1,000. In addition, coverage o