The Longitudinal Assessment of Bariatric Surgery (LABS) study is a prospective, longitudinal cohort study with three phases: LABS-1, LABS-2 and LABS-3 . LABS-1 is complete and reported 30 day postsurgical adverse outcomes in nearly 5000 participants . LABS-2 is a cohort of 2458 participants and has a primary goal of evaluating long-term efficacy of bariatric surgery . The LABS-3 Diabetes substudy comprised a subcohort from the LABS-2 study recruited specifically to better understand the physiological mechanisms of glucose metabolism improvement following RYGB.
Participants and study visits
Candidates were approached for inclusion if they were scheduled for RYGB at a participating LABS-2 site at the University of Pittsburgh, University of Washington or Oregon Health & Science University (ESM Fig. 1). Individuals with diabetes were included if they were documented to have a fasting plasma glucose ≥7.0 mmol/l and ≤10 mmol/l or HbA1c ≥48 mmol/mol (6.5%) and ≤69 mmol/mol (8.5%) and were treated with lifestyle interventions alone or metformin and/or sulfonylurea. Exclusion criteria included type 1 diabetes, creatinine >150 μmol/l, treatment with insulin or other non-metformin/non-sulfonylurea diabetes medications, treatment with weight-loss medications and individuals unlikely or unwilling to comply with follow-up visits. Participants on metformin, without a pre-treatment glucose or HbA1c measurement but meeting criteria for diabetes, were assumed to be on this therapy for diabetes prevention, weight loss or polycystic ovarian syndrome, and were excluded. Control participants (those without diabetes) were included if their fasting plasma glucose was <7.0 mmol/l and/or their HbA1c was <48 mmol/mol (6.5%) and they were not taking any diabetes medications. They were matched to participants with diabetes based on age, sex and pre-surgical BMI. All studies were approved by the investigational review boards at each site and consent was obtained from each participant before enrolment.
In the week before each research study visit, participants met with a registered dietitian to receive instruction regarding a standardised diet consisting of ~30% total energy from fat, ~10–15% from protein and ~55–60% from carbohydrates. During this time, participants with diabetes also stopped their diabetes medications and self-monitored their capillary blood glucose levels at home. All other usual medications for comorbid conditions (i.e. hypertension, gastroesophageal reflux and hyperlipidaemia) were continued, as was treatment for obstructive sleep apnoea.
At the end of this pre-study week, participants presented to the clinical research centres at their respective institutions following an overnight fast on each of 2 days, scheduled not more than a week apart. On one day, fasting blood samples were drawn for lipids and proinsulin levels after which participants were given an MM (BOOST Plus, Nestlé Health Science, Epalinges, Switzerland; 1506 kJ (360 kcal), 45 g carbohydrate, 14 g fat, 14 g protein). Blood samples were obtained for glucose, insulin, C-peptide, glucagon, GLP-1 and GIP 15 min before and 0, 10, 20, 30, 60, 90, 120, 150, 180 and 240 min after meal consumption. On the other day, they underwent an insulin-modified frequently-sampled IVGTT (FSIVGTT) .
Following these baseline studies, participants underwent identical procedures as described above at 6 and 24 months after RYGB.
Determination of glucose was performed by the hexokinase method using Roche reagents (Roche Diagnostics, Indianapolis, IN, USA). Levels of insulin and C-peptide were measured by a two-site enzymometric assay using Tosoh reagents on a Tosoh 2000 autoanalyzer (Tosoh Corp., Tokyo, Japan). Proinsulin levels were determined by radioimmunoassay using a Millipore kit (MilliporeSigma, Burlington, MA, USA). All lipid analyses were performed at the Northwest Lipid Metabolism and Diabetes Research laboratory, University of Washington, Seattle, WA, USA.
Total GLP-1, total GIP, total adiponectin and glucagon were measured by the Oregon Health & Science University Oregon Clinical and Translational Research Laboratory, Portland, OR, USA. To determine total GLP-1 and total GIP, blood was collected into EDTA vacutainers (prepared with 500 KU aprotinin and 10 μl dipeptidyl-peptidase-4 inhibitor per ml of whole blood) on ice. To determine adiponectin, blood was collected into EDTA vacutainers (prepared with 500 KU aprotinin/ml whole blood) on ice. These three analytes were measured using single-plex immunoassays from Meso Scale Discovery/Sector Imager (Gaithersburg, MD, USA). Glucagon was measured in blood collected in heparin vacutainers on ice, prepared with 500 KU aprotinin/ml whole blood, using an RIA from EMD Millipore (Billerica, MA, USA). The specificity of the glucagon assay was established by testing cross-reactivity with several gut hormones: oxyntomodulin (0.02%), 7-37 GLP-1 (none), 7-36 GLP-1 (<0.01%) and 1-37 GLP-1 (none).
Definition of diabetes and diabetes remission
Diabetes remission included both partial and complete remission, according to the American Diabetes Association Consensus Group recommendation , as HbA1c <48 mmol/mol (6.5%) (or fasting glucose ≤6.9 mmol/l if HbA1c was not available) and the absence of concurrent pharmacological therapy for diabetes.
Modelling for insulin secretory response and insulin sensitivity
Insulin sensitivity (SI), beta cell responsivity to glucose (Φ; measured as Φ total, Φ basal, Φ dynamic and Φ static) and hepatic insulin extraction (HE basal and HE total) were determined during the MM as previously described . Disposition index (DI) during the MM was derived from the product of SI and Φ total. Variables of sensitivity to insulin and secretion response during the FSIVGTT, including SI, glucose effectiveness (SG) and acute insulin response to glucose (AIRglu), were modelled as previously described . The DI during the FSIVGTT was derived from the product of SI and AIRglu.
Forty participants with type 2 diabetes and 20 without diabetes (60 in total) were estimated to provide an effective sample size for detecting modest-to-large effects on insulin sensitivity and stimulated-islet-cell secretion for within-person analyses and large-to-very-large effect sizes for between-group analyses based on prior studies [16, 17]. For continuous measures, participants in the diabetes vs no-diabetes groups were compared based on a Wilcoxon Rank Sum test to assess statistical significance of differences, unless normally distributed, in which case paired t tests were performed. For categorical measures, the frequency and percentage of each category was reported, and a Pearson χ2 test was used for statistical significance. For tables with small cell frequencies, Fisher’s exact test was used. Measures of insulin secretion and sensitivity from the FSIVGTT and MM over time were modelled using generalised linear models which included a heterogeneous autoregressive working correlation matrix to account for the correlation between successive measures. The model assumed a γ distribution for the measures to account for the highly skewed nature of the data. The mean measures at 6 months and 24 months were compared with baseline measures using a Wald t test. Missing values were treated as random occurrences. The generalised linear models used here utilised likelihood-based inference under which estimates are unbiased when data are missing completely at random.