Statistical Analysis
Patient demographic data and baseline clinical history were summarized
using median (IQR) for continuous variables and frequency (percentage)
for categorical variables. Continuous variables were compared using the
Wilcoxon rank-sum test, whereas categorical variables were compared
using the Pearson’s chi-squared test. We also summarized the number
(percentage) of patients who remained in the study at three months.
Glucose levels were analyzed as a dichotomized variable (IFG) if levels
were greater than 100 mg/dL. For the continuous glucose outcome, we used
the linear mixed effects model with fixed time and group (baseline IFG
status Yes/No) effect and random subject effect to model the trajectory
of glucose levels over time. Specifically, time was modeled as piecewise
linear with change points at 1 and 3 months. The random effects model
takes into account the correlation between the observations within the
same subject. Model estimates of glucose levels and their 95%
confidence intervals were summarized for both the IFG and the normal
fasting glucose group at 1 and 3 months. In addition, we tested for the
difference in glucose levels for 0 versus 3 months and reported the
p-values.
For the binary impaired fasting glucose (Yes/No) outcome, we used the
generalized linear mixed effects model with the logit link function to
model the proportion of IFG patients over time. The covariates structure
was the same as above. The estimated proportion of IFG patients and
their 95% confidence intervals were reported. Like before, we also
tested for the difference in proportions for 0 versus 3 months and
reported the p-values. A p-value < 0.05 was considered
statistically significant. R version 3.2.3 (www.r-project.org) was used
for the analyses.