Statistical analysis

Statistical analyses were performed using SAS (version 9.4) with the type I error rate fixed at 0.05 (2-tailed). Categorical variables are summarized using frequencies and percentages. Continuous variables with normal distributions are presented as mean (±standard deviation); continuous variables with non-normal distributions are presented as medians and interquartile ranges. Differences in baseline characteristics among the RYGB, LAGB, and LWLI groups were examined using the Pearson’s chi-square test or Fisher’s exact test for categorical variables and analysis of variance or Kruskal–Wallis test for continuous variables.

Changes in continuous outcomes from baseline to 12, 24, 36, 48, and 60 months were analyzed using mixed effects models with covariate adjustment for randomization stratification factors (gender and baseline BMI). Percent change in weight was adjusted for baseline weight. Inferences focused on the overall treatment effect, time, and treatment by time interaction. Pairwise comparisons were made between treatment groups at 60 months. For each outcome, 3 pairwise comparisons were made among the 3 treatment groups at 60 months, with the significance level set at .0167 (=.05/3) using the Bonferroni correction to account for multiple comparisons. Least-square means, along with their standard errors, were obtained from the models. Intent-to-treat analyses were conducted using multiple imputation using SAS PROC MI and PROC MIANALYZE. For each outcome, 10 datasets were imputed, and results were combined. For the primary endpoints of partial T2DM remission and complete remission, as well as other categorical data such as medication category usage, the Fisher’s exact test was used to compare differences among groups at given time point. The intention-to-treat approach was used in analyses of the primary endpoint assuming no remission for participants that did not return for follow-up.