Preintervention Clinical Determinants and Measured β-Cell Function as Predictors of Type 2 Diabetes Remission After Roux-en-Y Gastric Bypass Surgery

Obesity and type 2 diabetes increase morbidity and mortality (1). Given their complex pathophysiology and rising prevalence, they remain challenges for clinicians. Weight loss with lifestyle modifications is modest and usually unsustainable (2). Surgical weight loss results in large sustained weight loss and improved glycemic control with diabetes remission in 60–80% of cases (3). The anatomical rearrangement of the gastrointestinal tract that occurs with Roux-en-Y gastric bypass surgery (RYGB) has profound glucoregulatory effects (4). Patients show postoperative improvement in glucose tolerance beyond what is explainable by weight loss or caloric restriction alone (5). Mechanisms that may explain this phenomenon include altered nutrient sensing (6), enhanced incretin release (4–7), altered bile acids metabolism (8), and/or change in the microbiome (9).

Preoperative clinical predictors of postoperative diabetes remission include age (10), sex (10), BMI (11), HbA_1c (12,13), known duration of diabetes (11–14), number of oral diabetes medications, and/or insulin use (11,13). Clusters of preoperative clinical indicators as defined in the diabetes remission scores DiaRem (15) and advanced (Ad)-DiaRem (16)—age, HbA_1c %, insulin use, and oral diabetes medications—have been validated as strong predictors of diabetes remission (17–19). Biomarkers of insulin resistance (20), β-cell function (21), and/or the glucagon-like peptide 1 (GLP-1) postprandial response (22) have also shown association with diabetes remission (23).

Better elucidation of the predictors of postoperative diabetes status will improve clinical management. Our primary goal was to assess whether preoperative measures of β-cell function add to a well-defined clinical predictive model of short-term diabetes remission after RYGB. Our secondary goals were to 1) characterize which clinical factors showed the strongest association with baseline β-cell function in individuals with severe obesity and type 2 diabetes, and 2) compare the predictive value of preoperative β-cell function measured after an oral or intravenous (IV) glucose challenge on postoperative type 2 diabetes remission.

To this effect, 73 individuals with severe obesity and type 2 diabetes underwent standardized measurement of β-cell function after both oral and an IV glucose challenges before RYGB and were followed for diabetes remission phenotype at 1 year after RYGB. We hypothesize that 1) measures of preoperative β-cell function will add to a clinical prediction model of diabetes remission and that 2) β-cell function calculated after oral glucose, greater than β-cell function calculated after IV glucose, will be a better predictor of diabetes remission.

This is an ancillary study with retrospective analysis of data from participants from three longitudinal studies aimed at studying the mechanisms of type 2 diabetes remission after RYGB. Some data from these previous studies have been published (4,7,8,24). All three studies shared similar inclusion criteria: patients scheduled to undergo RYGB, BMI ≥35 kg/m², age 18 to 60 years, both sexes, all ethnic groups, and diagnosis of type 2 diabetes before RYGB based on the American Diabetes Association criteria (25). All individuals were under the care of the same bariatric team and underwent RYGB at St Luke’s-Roosevelt Hospital (New York, NY). All participants followed the same pre/postoperative care, including dietary counseling. A single investigator (B.L.) was involved in study design and research procedures.

Cohort 1, studied from 2007 to 2012, consisted of individuals with known diabetes duration of <2 years, HbA_1c <7.5% (58 mmol/mol), and not treated with insulin. For cohort 2 (2010–2014), and cohort 3 (2014–2019), there were no restriction criteria for diabetes duration or insulin therapy, but HbA_1c at baseline had to be ≤8.5% (69 mmol/mol). Follow-up was up to 7 years for cohort 1 and up to 2 years for cohorts 2 and 3. Data at the 1-year time point were used for the diabetes remission phenotype. A subset of individuals was followed up to 7 years, and their data are presented in Supplemental Table 4.

After surgery, participants were in full diabetes remission (F-REM; HbA_1c <5.6% [38 mmol/mol], fasting glucose <5.6 mmol/L, 120-min glucose <7.8 mmol/L, along with no antidiabetic drug use), in nonremission (N-REM; HbA_1c ≥6.5% [48 mmol/mol], or fasting glucose ≥7.0 mmol/L, or 120-min glucose ≥11.1 mmol/L), or in partial remission (P-REM; HbA_1c 5.7–6.4% [39–47 mmol/mol], fasting glucose 5.6–6.9 mmol/L, and 120-min glucose 7.8–11.0 mmol/L). Participants were divided into F-REM, N-REM, and P-REM based on the 1-year diabetes remission status post-RYGB. Diabetes remission at the 1-year follow-up was determined based on HbA_1c, oral glucose tolerance test (OGTT) data, and diabetes medications.

Participants for all cohorts underwent a 3-h OGTT and an IV glucose challenge at pre-RYGB and various times after surgery. While the protocol for OGTT was identical for each study, the glucose amount differed, 50 g for cohort 1 and 2 and 75 g for cohort 3. The IV glucose challenges were a glucose clamp isoglycemic to the glycemic level of the corresponding OGTT or a graded glucose infusion (GGI). When applicable, presurgery, GLP-1 analogs, dipeptidyl-peptidase 4 inhibitors, and thiazolidinediones were discontinued at least 2 months before the first study visits and replaced as needed by sulfonylureas or insulin. All oral antidiabetes medications were withheld for at least 3 days before each study visit, and the last injection of insulin was given at least 24 h before the morning of the test. Only preoperative data are used for β-cell function and body composition assessment.

Participants underwent a 3-h oral OGTT, with blood samples collected at 0, 15, 30, 45, 60, 90, 120, and 180 min using an antecubital IV catheter from an arterialized arm vein before RYGB and at various times after surgery for up to 7 years. For cohort 1, the study visits were conducted presurgery and at 1 month, 6 months, and yearly up to 7 years postsurgery. For cohort 2, study visits were conducted presurgery and 1 month and 1 and 2 years postsurgery. For cohort 3, study visits were conducted presurgery and 3 months and 1 and 2 years postsurgery.

An IV isoglycemic glucose (IV-isoG) clamp was performed at each time point in participants from cohort 1 and 2 to calculate the incretin effect, as previously described (8). Briefly, during the glucose clamp, glucose (20% dextrose solution) was infused using a Gemini pump over 180 min to match the plasma glucose concentration profiles achieved for each subject during the OGTT. Blood glucose was monitored using contralateral antecubital IV access every 5 min, and the glucose infusion rate was adjusted accordingly. Blood samples were collected over 3 h, as previously described (4–7)

During the GGI, antecubital IV catheters were placed in the right and left arms. If fasting glucose was >8.33 mmol/L, a bolus of IV regular insulin (0.6 units/kg), followed by a continuous intravenous infusion of 0.01 units/kg/h was initiated until the blood glucose decreased to ≤8.33 mmol/L. At that point, the insulin infusion was stopped, and the GGI started 60 min later (at time 0) with an infusion of 20% dextrose at 2 mg/kg body weight/min. Every 30 min, the infusion rate was increased to 4, 6, 8, and 12 mg/kg body weight/min. The blood was drawn every 10 min for the entire experiment.

Fat mass and waist circumference were measured by a three-dimensional photonic scanner (3DPS) in all three cohorts. The 3-DPS scans (model #C9036-02 Body Line Scanner; Hamamatsu Photonics KK, Shizuoka, Japan) were performed at the New York Nutrition Obesity Research Center Body Composition Unit at St. Luke’s-Roosevelt Hospital until 2014, and thereafter at Columbia University. The 3DPS scanner system and procedures have been previously described in detail (26–28). During scan acquisition, participants wore a tight-fitting cap to minimize air spaces between the hair/skull and well-fitting underwear that clung to skin surfaces. Arms were positioned with hands holding the handlebars, and participants stood on footprints that were equidistant from the center point. With this standardized position, participants’ arms were abducted from the trunk, and there was no contact between the legs. Participants respired normally and remained motionless during three repeated scans.

Plasma glucose was determined at the bedside by the glucose oxidase method with an Analox glucose analyzer (Lunenburg, MA). Plasma insulin, C-peptide, and GLP-1 were measured by radioimmunoassay (EMD Millipore, Burlington, MA) at the Columbia University Diabetes Research Center Translational Biomarkers Analytical Core. For cohort 3 only, GLP-1 was measured by ELISA (EMD Millipore). Intra- and interassay coefficients of variance for all assays ranged from 3.4 to 7.4%. Blood HDL, LDL, total cholesterol, and triglycerides were derived from patients’ medical records.

Total (tAUC) and incremental (iAUC) areas under the curves for glucose and hormones during the OGTT and GGI were calculated by using the trapezoidal method (12). HOMA of insulin resistance (HOMA-IR) was calculated as fasting insulin (μU/mL) × fasting glucose (mmol/L)/22.5 (13).

Insulin secretion rates (ISR) were calculated by C-peptide deconvolution using a two-compartment model (16). β-Cell glucose sensitivity (BCGS) was calculated as the slope of the relationship between the ISR and the change of glucose from fasting to peak glucose. HOMA of β-cell function (HOMA-B) was calculated as (360 × insulin_0min)/(glucose_0min − 63) (17). Insulin resistance was estimated as the HOMA-IR: (fasting insulin_µU/mL × fasting glucose_mg/dL)/405 (29). The insulin sensitivity index (ISI) was derived from OGTT values using the Matsuda index (30). The insulinogenic index (IGI) was calculated as (insulin_30min − insulin_0min)/(glucose_30min − glucose_0min). The disposition index (DI) was calculated as BCGS × (1/HOMA-IR). All variables derived from OGTT are labeled with an O- (e.g., O-BCGS for BCGS calculated during the OGTT), and variables calculated during the IV challenges (IV-isoG and GGI) are labeled with the IV prefix.

Two diabetes remission scores (DiaRem and Ad-DiaRem) were calculated for all participants, as previously described (15,16). Each score is a weighted scoring system based on baseline presurgery factors and was created to predict diabetes remission postbariatric surgery.

Statistical analysis was performed using the statistical package R 4.0.1 for Mac. Data are presented as percentages for categorical values and mean and SD for continuous variables. The Shapiro-Wilk test was used to evaluate the distribution of continuous variables, and nonnormally distributed variables were log-transformed where applicable. Correlation coefficients between O-BCGS, IV-BCGS, and independent variables were obtained via the Pearson correlation coefficient test. The Fisher exact test analyzed categorical variables between postoperative remission groups, while differences in mean continuous predictors were analyzed using both the Kruskal-Wallis test and ANOVA, followed by the post hoc Tukey test. All tests were two-tailed, and P values < 0.05 were considered significant. Variables with P < 0.05 in the univariate analysis were included in the multivariate multinomial logistic regression model of diabetes remission. Multivariate multinomial logistic regression models, including validated diabetes remission scores, were compared using the likelihood ratio test.

Presurgery participants’ characteristics are presented in Table 1. Participants (N = 73) were predominantly female (80.8%) and Hispanic (60.3%). Nearly 12% of participants were treated with insulin preoperatively, and the number of noninsulin medications ranged from 0 to 5. Compared with participants in cohorts 1 and 2, participants in cohort 3 had more advanced diabetes, as indicated by longer known diabetes duration, higher rates of insulin use, higher number of diabetes medications, higher HbA_1c %, and higher DiaRem/Ad-DiaRem scores (Table 1). Notably, there was no difference in postsurgery weight loss between cohorts.

As hypothesized, O-BCGS on average was 1.6-fold greater than IV-BCGS (Table 2), and the difference was more marked for F-REM. O-BCGS and IV-BCGS were strongly positively correlated with each other (R² = 0.67, P < 0.001). Both were strongly negatively correlated with known diabetes duration, preoperative HbA_1c %, insulin use, and the DiaRem and Ad-DiaRem scores (Table 3). Except for insulin use, O-BCGS was more strongly correlated to clinical characteristics compared with IV-BCGS. However, neither anthropometric measures nor ISI, HOMA-IR, lipids, and GLP-1 response during OGTT were associated with O-BCGS or IV-BCGS (data not shown).

Data on diabetes remission phenotype were obtained on all participants (N = 73) at 1 year post-RYGB. Postoperative F-REM was achieved in 43.8% (n = 32), P-REM in 39.7% (n = 29), and N-REM in 16.4% (n = 12) of participants. Age, racial and ethnic distributions, body composition, ISI, and HOMA-IR did not differ between remission groups; however, N-REM had more men compared with the P-REM and F-REM (Supplementary Table 1). As expected, participants in the N-REM group were more likely to have longer diabetes duration, a greater number of noninsulin medications, a higher HbA_1c at presurgery, and to use insulin (Supplementary Table 1). There was no difference in percentage of total weight loss at the latest time point between the three remission groups. DiaRem and Ad-DiaRem differed significantly between P-REM and N-REM and between F-REM and N-REM; however, neither score differed between P-REM and F-REM. Data on diabetes remission phenotype at 2 years post-RYGB and at the latest time point (up to 7 years) showed a similar pattern of results (Supplementary Table 7).

Preintervention glucose and insulin response after the OGTT and parameters of β-cell function (both oral and IV) differed according to postoperative diabetes remission outcomes (Supplementary Table 2). Insulin and C-peptide iAUC, BCGS, and DI were ∼2- to 5-times lower in N-REM than in F-REM. GLP-1 response during the OGTT trended (P = 0.50) lower in F-REM compared with both P-REM and N-REM. Unlike BCGS and DI, HOMA-B and IGI values did not differ between groups. The difference between F-REM and N-REM was of lesser magnitude for parameters measured during the IV glucose challenge (Supplementary Table 2).

To identify preintervention predictors of post-RYGB diabetes remission status, a multinomial logistic regression analysis was used (Table 4). A univariate multinomial logistic regression analysis showed, as expected, that the baseline DiaRem scores predicted the 1-year postoperative diabetes remission status. The models revealed that for every one additional point increase in the DiaRem or Ad-DiaRem score, participants were ∼1.2- to 1.5-times more likely to be P-REM versus F-REM or N-REM versus F-REM (Table 4). Preintervention biomarkers of β-cell function, with the exception of IV-DI, HOMA-B, ISI, HOMA-IR, and IGI, predicted diabetes remission after surgery (Table 4). In univariate multinomial logistic regression analysis, DiaRem, O-BCGS, IV-BCGS, and O-DI predicted outcome between F-REM and N-REM but were unable to differentiate between F-REM and P-REM, while only Ad-DiaRem could predict both (Table 4). However, when data were pooled and analyzed at the latest available time points in the three cohorts (3 months to up to 7 years), Ad-DiaRem and DiaRem were both able to differentiate between F-REM and N-REM as well as between F-REM and P-REM (Supplementary Table 7).

To assess whether the addition of baseline biomarkers of β-cell function and/or insulin resistance improve the prediction model with validated DiaRem scores alone, BCGS, DI, HOMA-B, IGI, ISI, and HOMA-IR were added to the analysis and held constant (Table 4). Although the addition of HOMA-B and IGI into a multivariate analysis containing one of the diabetes remission scores did increase the odds ratio, the difference between the multivariate analysis and diabetes remission scores alone was not statistically significant when compared via the likelihood ratio test (data not shown) or when assessing accuracy and among other diagnostic measures (Supplementary Table 9). Contrary to our hypothesis, neither O-BCGS nor IV-BCGS added significantly to a clinical model of prediction. Interestingly, weight loss at 1 year post-RYGB neither differed between the three remission groups nor did it improve the prediction when added to a model consisting of baseline clinical or metabolic characteristics.

We evaluated the association of baseline clinical variables and measured β-cell function and studied preoperative predictors of postoperative diabetes remission 1 year after RYGB in individuals with severe obesity and type 2 diabetes. Our key findings are 1) DiaRem and Ad-DiaRem scores were both positively and strongly correlated with β-cell function as measured by O-BCGS and IV-BCGS, with O-BCGS demonstrating the strongest correlation; 2) DiaRem and Ad-DiaRem scores and β-cell function independently predicted remission (P-REM or F-REM) versus N-REM and differentiated between P-REM and F-REM depending on analysis; and 3) contrary to our hypothesis, the addition of measured β-cell function to defined clinical models of diabetes remission, with or without DiaRem and Ad-DiaRem scores, did not improve the probability to predict diabetes remission.

We show that most preoperative factors that significantly associated with O-BCGS were also significantly associated with IV-BCGS; however, the associations between individual clinical variables, DiaRem scores, and β-cell function were always strongest for O-BCGS compared with IV-BCGS. This trend was also observed in most of our analyses. The difference between the two measurement methods was not unexpected, because IV-based glucose stimulation tests have limitations due to their inability to reflect the normal physiological gut stimulation and the incretin effect (31,32).

Our data confirm that higher preoperative β-cell function is associated with a greater likelihood of undergoing diabetes remission after RYGB (21,33). Clinical markers of less advanced diabetes, including shorter known diabetes duration, better glycemic control, fewer medications, and no insulin use, are associated with preoperative β-cell function, specifically O-BCGS and IV-BCGS. This finding coincides with the progressive decline of β-cell function over time in the natural course of diabetes (34). HbA_1c, insulin use, and diabetes duration show the strongest correlation with O-BCGS.

A more accurate estimate of β-cell function may combine multiple clinical factors in order to minimize the inherent flaws in each isolated measurement. Not surprisingly, we demonstrate that the diabetes remission composite scores DiaRem and Ad-DiaRem, compared with each individual clinical parameter, correlate more strongly to β-cell function as measured by BCGS. The use of these scoring systems suggests that the known duration of diabetes, age, HbA_1c and preoperative insulin use can be used clinically as a good surrogate marker for β-cell functional reserve.

However, none of the clinical variables showed very strong correlation with diabetes remission status (all R² ≤ 0.42), and they do not explain the entire relationship between preoperative factors and postoperative remission. This lack of tight relationship may be explained by their relative temporality. Diabetes duration spans years, HbA_1c a few months, and the measurement of β-cell function is done over a few hours in a single day. Moreover, known diabetes duration is not an accurate reflection of disease length, as delay in diagnosis is not atypical, with the initial onset of hyperglycemia often occurring 4–7 years before the clinical diagnosis is made (35). HbA_1c only represents 2–3 months of glycemic control, can be falsely elevated/lowered due to other underlying conditions, and may not fully capture disease progression, the extent of β-cell failure, and long-term glycemic control (36). Genetic factors (37) are also important determinants of β-cell function. The cohorts were ethnically diverse, and ethnicity has been shown to be an important determinant of β-cell loss (38).

The strongest individual predictors of postoperative remission were diabetes duration, preoperative insulin use, number of noninsulin diabetes medications, and total number of diabetes medications. These results are also consistent with previous literature (11,39). Preoperative BMI was found to be an inconsistent predictor (39); here we failed to find a correlation between preoperative BMI and remission status. However, participants in the F-REM group were more likely to experience greater absolute weight loss compared with both the P-REM and N-REM groups.

Several scoring systems exist for predicting diabetes remission postsurgery, including Ad-DiaRem and DiaRem (15,17). The latter was externally validated in independent populations (19) and shown to be highly predictive of postoperative diabetes remission. Our data confirm the performance of both scores to differentiate between both F-REM and N-REM as well as F-REM and P-REM via logistic regression analysis; however, mean scores were unable to differentiate between F-REM and P-REM remission. This inability to differentiate is likely a consequence of the original development of these scoring systems, which used higher HbA_1c % cutoffs for the definition of both P-REM (6.0–6.5%) and F-REM (<6.0%) compared with our definition, which was based on the latest American Diabetes Association guidelines (25). The Ad-DiaRem was initially created to improve the prediction ability of DiaRem by adding two variables—known diabetes duration and number of glucose-lowering agents (16); however, previous study (15,17) concluded that it failed to solve the misclassification of midrange-scored patients and suggested the additional biological factors may help to resolve this problem.

Our data demonstrate that glucose and various biomarkers of insulin secretion (insulin and C-peptide iAUC) were able to differentiate between N-REM and F-REM, but not between other diabetes remission statuses. As expected, participants with higher baseline levels of certain indices of insulin, C-peptide, and ISR were more likely to be in the P-REM or complete remission group, and those with lower levels of glucose and GLP-1 AUC were more likely in the F-REM group. C-peptide indices derived from OGTT had a stronger predictive value compared with insulin. C-peptide has a longer half-life, which allows for a more stable testing window (40) and therefore provides a more accurate assessment of pancreatic β-cell function reserve. In addition, surrogate markers of β-cell function (BCGS and DI), both of which are derivatives of C-peptide indices, also demonstrate an ability to predict the differences between F-REM and N-REM groups. This result was also expected as previous studies have demonstrated that patients who do not remit have worse β-cell function at baseline compared with their remitter peers (21). Contrary to our hypothesis, the addition of measured baseline β-cell function (both oral or IV derived) to define clinical models of diabetes remission, with or without DiaRem and Ad-DiaRem scores, did not improve the probability to predict diabetes remission after RYGB. Given the lack of improvement with the addition of preoperative β-cell function, there may be great value in finding other unique and independent preoperative biomarkers reflecting genetic or epigenetic diversity that may better predict postoperative diabetes remission. In all, the best prediction in our cohort was the Ad-DiaRem score alone.

Our study has many strengths, including limited attrition at the 1-year follow-up, single research group, standardized measurement of β-cell function, and comparison of both oral and IV glucose challenges. We also demonstrate similar pattern of results and level of significance when analyzing data in a subset of participants at longer time points, 2 years and up to 7 years after surgery (Supplementary Tables 4–8). However, limitations include some heterogeneity in inclusion criteria, predominance of women and Hispanic ethnicity, and differing glucose challenge protocols. Finally, we lack a control group with normal glucose tolerance and a group with vertical sleeve gastrectomy.

In conclusion, the addition of surrogate measures of presurgery β-cell function or of insulin resistance does not improve the predictive value of previously validated diabetes remission scores. Further studies are needed to identify preoperative biomarkers and postoperative mechanisms responsible for diabetes remission after RYGB and other types of bariatric surgery.

Funding. The data presented were collected in part with grants from the American Diabetes Association (7-05-CR-18 and 1-09-CR-34) and from the National Institutes of Health National Institute of Diabetes and Digestive and Kidney Diseases (R01-DK-067561, R01-DK-098056, P30-DK-063608, and P30-DK-26687) and National Center for Advancing Translational Sciences (UL1TR001873 and past grants UL1TR000040 and UL1TR024156), and KL2TR003018 (to A.S.).

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. C.L. performed all of the data analyses and wrote the manuscript. A.S. and M.L. and edited the manuscript. B.L. designed each individual study, reviewed the data analyses, and edited the manuscript. B.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.