Statistical analysis
This study is the first randomised controlled trial pilot and
feasibility study using CGM sensors in pregnant women comparing a group
with CGM feedback and a group without. The actual value of standardised
effect size to be used was not known before this pilot trial. When
estimating the sample size, we used a simple method by applying the
rules of thumb by Teare et al. which recommended a pilot trial sample of
at least 120 (60 in each study arm) if the primary outcome of the trial
is binary 20.
The analyses were conducted on the intention-to-treat principle which
includes all the women who participated in the randomization and
completed the study (Figure 1). Categorical variables were summarised by
counts and proportions; continuous variables data was summarised by
means and standard deviation (SD), or by median and interquartile range
(IQR) in the case of deviations from the normal distribution. The
primary analysis used multivariable logistic regression to assess the
associations between the two study arms (unblinded and blinded) with GDM
outcomes, and multivariable linear regression to assess the associations
between CGM groups and OGTT FG, 1hPG and 2hPG concentrations. Regression
models were adjusted for covariates such as maternal age, ethnicity,
education, family history of diabetes and pre-pregnancy BMI with the vce
(robust) option without multiple imputation as the percentage of missing
covariate data was very low (<2.0%).
From the total number of participants, 206, who were randomized and
completed the study, 40 were excluded leaving 79 in the unblinded group
and n=87 in the blinded group to be included in the final analysis
examining the primary outcomes (Figure 1). These participants would have
had to wear the CGM sensors at the 4 timepoints: first trimester at 9-13
weeks, the early second trimester at 18-23 weeks, late-second to
early-third 24-28 weeks and the third trimester at 32-33 weeks (Figure
S1). The secondary analysis used all available data at the four
different timepoints, and the cross-sectional differences in the CGM
parameters (such as %TIR, %TAR, %TBR, mean glucose, SD, MAGE and
%CV) between the CGM groups was assessed using linear regression at all
four timepoints. The regression model was adjusted for maternal age,
ethnicity, education, family history of diabetes, GDM outcomes,
gestational age of CGM application and pre-pregnancy BMI with the vce
(robust) option without multiple imputation as the percentage of missing
covariate data was very low (<3.0%). The data from the CGM
with less than 50% of data captured at all the timepoints of interest
(from the first to the third trimester) was excluded from further
analysis; in total, 45 from the unblinded study arm and 58 from the
blinded study arm were included in the final analysis (Figure 2). A
two-sided p value <0.05 is considered statistically
significant, and p value <0.1 was reported as trends for both
primary and secondary outcomes. All analyses were performed by using the
statistical software STATA 13.1.