Statistical analyses
The study was powered to detect a significant difference in weight loss at
2 mo among the five groups with a significant trend in weight loss demonstrating
a decrease in percent body weight incrementally going from the omni, semi-veg,
pesco-veg, veg, through the vegan group. Assuming a mean incremental difference
in change in body weight of 1% successively between each of the five groups
(corresponding to an effect size of 0.57), a pooled SD of 2.5%, and significance at a
¼ 0.05, a sample size of 60 participants (12 per group) was estimated to provide
94% power for the linear trend among the five groups [21]. The sample size of 12
per group provided 80% power for differences of 2.85% for linear contrasts
between two groups. Attrition was defined as a participant not completing the
main outcome of body weight at either 2 mo (for 2-mo outcomes) or 6 mo (for
6-mo outcomes).
For differences in baseline demographic characteristics, analysis of variance
(ANOVA) was used with the Tukey’s test for post hoc analyses of
continuous variables and c2 test of independence was used for categorical
variables. Change in percent weight loss among the five groups was analyzed
at both 2 mo (after the intensive intervention) and at 6 mo (to assess weight
loss maintenance) using one-way ANOVA. To test that weight loss would be
incremental among the five groups (going from the vegan group losing the
most to omnivores losing the least), an a priori linear contrast for trend was
used at each time point. Additionally, three a priori linear contrasts among the
specific groups were examined at each time point: vegan versus omni, vegan
versus semi-veg, and vegan versus pesco-veg. Because weight loss differences
between veg and vegan participants were expected to be smaller than the
other groups, this study was not powered to detect weight loss differences
between veg and vegan. Missing data for body weight was handled in two
ways: 1) baseline observation carried forward for missing values at each time
point (assuming no change) and 2) weight gain imputed at a rate of 0.3 kg/mo.
This rate of weight gain has been shown to commonly occur during behavioral
weight loss interventions [22,23] and has been used as the weight gain
amount for other large, dietary weight loss trials [24,25]. Weight gain was
extrapolated from time of attrition up through the 2- and 6-mo assessmen