Abstract
OBJECTIVE To characterize the status of cardiometabolic risk factors after late relapse of type 2 diabetes mellitus (T2DM) and to identify factors predicting relapse after initial diabetes remission following bariatric surgery to construct prediction models for clinical practice.
RESEARCH DESIGN AND METHODS Outcomes of 736 patients with T2DM who underwent Roux-en-Y gastric bypass (RYGB) or sleeve gastrectomy (SG) at an academic center (2004–2012) and had ≥5 years’ glycemic follow-up were assessed. Of 736 patients, 425 (58%) experienced diabetes remission (HbA1c <6.5% [48 mmol/mol] with patients off medications) in the 1st year after surgery. These 425 patients were followed for a median of 8 years (range 5–14) to characterize late relapse of diabetes.
RESULTS In 136 (32%) patients who experienced late relapse, a statistically significant improvement in glycemic control, number of diabetes medications including insulin use, blood pressure, and lipid profile was still observed at long-term. Independent baseline predictors of late relapse were preoperative number of diabetes medications, duration of T2DM before surgery, and SG versus RYGB. Furthermore, patients who relapsed lost less weight during the 1st year after surgery and regained more weight afterward. Prediction models were constructed and externally validated.
CONCLUSIONS While late relapse of T2DM is a real phenomenon (one-third of our cohort), it should not be considered a failure, as the trajectory of the disease and its related cardiometabolic risk factors is changed favorably after bariatric surgery. Earlier surgical intervention, RYGB (compared with SG) and more weight loss (less late weight regain) are associated with less diabetes relapse in the long-term.
Introduction
Bariatric surgery has now found a place in the algorithm of management of type 2 diabetes mellitus (T2DM) (1). At the same time, earlier suggestions that bariatric surgery might offer a permanent cure (2) for T2DM have had to be moderated as it became apparent that not all patients with T2DM experience a remission with bariatric surgery (3,4) and that a large number of those who experience a remission in the early period after surgery suffer a relapse on longer-term follow-up even with highly effective procedures such as biliopancreatic diversion (5,6).
Though relapse of T2DM after an initial remission following bariatric surgery is now well recognized (7–10), its exact incidence is difficult to comprehend, as most studies report outcomes on the total cohort and do not focus on only those patients who experienced an early remission. In addition, the severity of T2DM after late relapse is largely unknown. It is further known that bariatric surgery improves metabolic profile including lipid panel and blood pressure (3,4,6,8–12) in patients with diabetes, but the effect on lipid profile and hypertension in patients who experience a relapse of T2DM has not yet been adequately examined. It is therefore unclear whether these patients also experience a gradual reduction in other metabolic benefits, which would potentially diminish the long-term benefits of bariatric surgery in these individuals.
Similarly, there are some data to suggest that those with poor preoperative diabetes control, those with longer duration of T2DM, and those requiring insulin at the time of surgery—the clinical factors indicating diminished pancreatic β-cell reserve—are less likely to experience durable long-term remission (7,8,13). But we do not exactly know whether these factors can predict relapse. The predictors of diabetes relapse have been less well characterized. It is important to be able to predict relapse after an initial remission of T2DM to counsel patients accordingly, encourage closer monitoring, and institute appropriate medical management if and when needed in the long-term. Furthermore, the impact of late weight regain that can be seen several years after bariatric surgery in many patients on relapse of T2DM is not clear.
The purpose of this study was to determine the incidence of relapse of T2DM after initial remission following bariatric surgery at a high-volume academic center, characterize the status of metabolic profile and cardiovascular risk factors in these patients after late relapse, and identify factors predicting relapse of T2DM to construct and validate prediction models for clinical practice.
Research Design and Methods
The institutional review board approved this retrospective study as minimal risk research and permitted us to conduct the study without obtaining the informed consent from patients. All patients with T2DM who underwent primary Roux-en-Y gastric bypass (RYGB) or sleeve gastrectomy (SG) at an academic center (Cleveland Clinic Main Campus, Cleveland, OH) between 2004 and 2012 were identified. Only patients with complete and documented baseline data who also had at least 5 years’ postsurgery follow-up including data on body weight, fasting blood glucose (FBG), HbA1c levels, and status of diabetes medications were included in this study. Patients who had reoperative bariatric surgery for conversion of SG to RYGB, conversion of SG to duodenal switch, or conversion of RYGB to distal gastric bypass with shortening of the common channel were excluded. Patients who died within the initial 5 years after the index procedure and international patients who were not expected to have a long-term follow-up at Cleveland Clinic were also not included in the analysis.
All outcome data included in the analysis were obtained from the electronic medical records from a clinical visit in Cleveland Clinic or from source documentation from the patient’s primary physician. Data on body weight, BMI, diabetes status, blood pressure, lipid panel, and diabetes medications were extracted from the preoperative period until the most recent follow-up.
Diabetes status following bariatric surgery was characterized according to the definitions developed by the American Diabetes Association (14) and the American Society for Metabolic and Bariatric Surgery (15). Diabetes remission was defined as HbA1c <6.5% (48 mmol/mol), FBG <126 mg/dL, and being off diabetes medications (3,4,14,15). Long-term remission was defined as meeting the criteria for remission at the last follow-up time, which was at least 5 years after surgery. According to the American Diabetes Association criteria, HbA1c <7% (53 mmol/mol), irrespective of diabetes medications, was considered to indicate glycemic control. Percent total weight loss (TWL) was calculated as ([operative weight − follow-up weight]/operative weight) × 100. Late weight gain was assessed by ([weight 1st year after surgery − last follow-up weight]/weight 1st year after surgery) × 100. Values >5% were arbitrarily defined as significant late weight regain.
To characterize the subset of patients who developed late diabetes relapse, we first identified a group of patients who met the criteria for diabetes remission within the 1st year of surgery. These patients were then followed up to characterize late relapse of T2DM, which was defined as FBG or HbA1c in the diabetes range (≥126 mg/dL and ≥6.5% [48 mmol/mol], respectively) or need for antidiabetes medication after initial remission.
Statistical Analysis
Categorical variables were presented as frequencies (%). Continuous variables with normal and nonnormal distributions were presented as mean ± SDs and median (interquartile range [IQR]), respectively.
To assess the metabolic status of patients who developed late diabetes relapse, a paired t test and Wilcoxon signed rank test were used to calculate the differences between the last follow-up point and baseline at the time of surgery. Two dependent proportions were compared with Z test.
All tests were two tailed and performed at a significance level of 0.05. Odds ratio (OR) and 95% CI were used as measures of the magnitude of the association. R software, version 3.5.0 (Vienna, Austria), was used for all analyses for this study.
Main Prediction Model
Independent baseline (before surgery) predictive factors of late relapse of T2DM were determined by multivariable logistic regression analysis. Univariate analysis (using χ2, Fisher exact, and two-sample t tests) was performed to compare eight baseline characteristics in patients who developed late diabetes relapse compared with patients who stayed in diabetes remission long-term. Factors considered in univariate analysis included sex, age, known duration of T2DM before surgery, baseline BMI, HbA1c, number of diabetes medications, insulin use, and type of bariatric surgery performed (RYGB vs. SG). Multivariable logistic regression models with stepwise variable selection were built to identify the independent predictors of diabetes relapse. Backward elimination was used to remove insignificant variables. The discriminatory capability of the model was assessed using the c-statistic, which is the same as the area under the receiver operating characteristic curve. The c-statistic with bootstrapping technique (internal validation) was also calculated.
With use of the same enrollment criteria and outcomes definitions, a validation cohort was assembled by pooling the data from three external sources: 1) Hospital Clínic de Barcelona, 2) patients with T2DM who were enrolled in the Sleeve vs Bypass (SLEEVEPASS) randomized clinical trial (16) from Finland, and 3) patients from Geisinger Clinic in Danville, PA. Since the suggested minimum requirement for external validation is having at least 100 patients with the event (i.e., diabetes relapse) (17), data of three cohorts were combined to prevent possible overestimation of model performance by examining smaller cohorts individually.
The regression equation of the prediction model was programmed to construct an online version of the diabetes relapse calculator based on preoperative risk factors.
Second Prediction Model
To account for the effects of weight changes after bariatric surgery on late diabetes relapse, a second prediction model was developed utilizing the statistical methods described above. In the second prediction model, in addition to eight baseline variables (above), the role of percent weight loss within the 1 year after surgery and late weight regain (>5% weight gain from weight recorded at 1 year after surgery) were examined in multivariable logistic regression analysis. The second calculator was constructed, which can be used 1 year after surgery to predict late diabetes relapse utilizing both preoperative and weight change data.
Results
Outcomes of 736 patients with T2DM who underwent primary RYGB or SG between 2004 and 2012 had ≥5 years’ glycemic follow-up and who met other enrollment criteria (representing 80% of patients who would have been available for long-term follow-up) were assessed. Of 736 patients, 425 (58%) experienced diabetes remission in the 1st year after surgery including 62% after RYGB and 41% after SG.
The median postoperative follow-up time of the entire cohort was 8 years (range 5–14). The median postoperative follow-up time after RYGB was 8 years (IQR 6–9) and after SG was 8 years (IQR 5–9). Of those 425 patients who initially achieved diabetes remission in the short-term (Supplementary Table 1), 136 (32%) later relapsed with T2DM.
In patients who experienced late relapse (n = 136), statistically significant improvement in metabolic profile was still observed long-term compared with baseline values before surgery (Table 1). Despite experiencing diabetes relapse, 77% of them maintained adequate glycemic control (HbA1c <7%), with a median HbA1c of 6.4% (IQR 6.0–6.8) (46 mmol/mol [IQR 42–51]). A statistically significant reduction in the number of diabetes medications (median of 2 before surgery vs. 1 long-term, P < 0.001) and percentage of patients on insulin (29% before surgery vs. 12% long-term, P < 0.001) was observed. Similarly, there was a statistically significant improvement in systolic and diastolic blood pressure, LDL, HDL, and triglycerides compared with the baseline values before surgery.
Long-term metabolic profile of post–bariatric surgery patients with late relapse of diabetes following initial remission (n = 136)
Predictors of late diabetes relapse from univariate and multivariate analyses have been summarized in Tables 2 and 3, respectively. Independent baseline predictors of late relapse were preoperative number of diabetes medications (OR 2.04 [95% CI 1.54, 2.71], P < 0.001), known duration of T2DM before surgery (OR 1.08 [95% CI 1.02, 1.14], P = 0.005), and SG versus RYGB (OR 2.21 [95% CI 1.22, 4.01], P = 0.009) (Table 3).
Univariate analyses in patients with and without diabetes relapse following initial remission after bariatric surgery in training and validating cohorts
Independent predictors of late relapse of diabetes after bariatric surgery in multivariable analyses
Characteristics of external cohorts (Spain, 34.7% late diabetes relapse out of n = 196 with initial remission; U.S., 23.2% late relapse out of n = 82 with initial remission; and Finland, 31.7% late relapse out of n = 41 with initial remission) have been detailed in Table 2 and Supplementary Tables 2–4.
The c-statistics for the main model, bootstrapping validation, and external validation were 0.75, 0.74, and 0.73, respectively. The calibration curve on the external validation data suggested great accuracy (Supplementary Fig. 1).
The second model considers the impact of early weight loss and late weight regain on remission and relapse of T2DM. Patients with late relapse of T2DM lost less weight in their 1st year after bariatric surgery compared with patients who had durable diabetes remission (mean difference in weight loss ± SD 3.9 ± 0.8%, P < 0.001). In addition, patients with late relapse of T2DM had significantly higher weight gain long-term (11.6 ± 14.9%) compared with patients without late relapse (6.5 ± 15.5%, P < 0.001) (Supplementary Tables 5 and 6). Late weight gain was observed in 59% of the cohort. In model two, independent predictors of late relapse were preoperative number of diabetes medications, duration of T2DM before surgery, less weight loss in the 1st year after surgery, and greater late weight regain (Table 3). The model had a very good discrimination ability (c-statistic = 0.80 and bootstrap validation c-statistic = 0.79).
On the basis of the regression equations and the parameter estimates, two risk calculators for late relapse were developed. A free user-friendly version of the scoring systems is accessible at https://riskcalc.org/ and as a smartphone application (BariatricCalc, available on the Apple Store and Google Play). When the required values are entered into the calculator, the percent estimate of late relapse of T2DM following initial remission after bariatric surgery is calculated. The first model includes the preoperative factors only. However, in the second model, weight changes after bariatric surgery are among the independent predictors.
Below are two examples of the estimated probability of late relapse of T2DM after initial remission:
Model 1: The estimated risk of relapse in a patient with T2DM for 5 years before bariatric surgery and on one diabetes medication would be 20% after RYGB and 36% after SG.
Model 2: The estimated risk of relapse in a patient with T2DM for 10 years and on two diabetes medications before bariatric surgery who initially lost 30% of weight in 1 year after surgery would be 28% and 59% without and with significant long-term weight regain (>5% weight gain), respectively.
Conclusions
Over the last few years, it has become apparent that a large proportion of patients with T2DM who initially go into remission following a bariatric surgery experience a late relapse. As early as 2010, DiGiorgi et al. (18) and Chikunguwo et al. (19) observed that T2DM recurred or worsened from peak benefits in ∼24% and 43% of their patients, respectively. New-onset T2DM after bariatric surgery has also been reported (20,21) in the scientific literature.
More recently, Arterburn et al. (7), in a large population-based study, examined 2,254 patients who experienced initial remission after RYGB. They found that 35% of their patients relapsed within 5 years of complete remission. The relapse rates of 32% in our primary cohort and 31% in external cohorts are similar to the findings of that study. In the Swedish Obese Subjects (SOS) study, of 342 patients with T2DM before bariatric surgery, 72% experienced diabetes remission at 2 years after surgery but only 36% remained diabetes free at 10-year follow-up (50% relapse rate) (10,22). The relatively higher relapse rate in the SOS study could be explained by the higher number of surgical procedures, which are less effective in improving T2DM. More than 85% of surgical procedures at that time in the SOS study were either vertical banded gastroplasty or gastric banding—the procedures that have fallen out of favor in recent years. In the Surgical Treatment and Medications Potentially Eradicate Diabetes Efficiently (STAMPEDE) trial, out of 32 surgical patients (RYGB and SG) who initially met the study’s end point (HbA1c <6% [42 mmol/mol] without diabetes medications), 47% were not at this target 5 years after surgery (9).
The most frequently identified predictors of relapse in the scientific literature were signs of advanced T2DM at the time of surgery (23,24). Having a longer duration of T2DM before surgery, poorly controlled T2DM, or being on multiple glucose-lowering agents and insulin are all clinical factors that would reflect limited pancreatic β-cell reserve at baseline. In the investigations by Jiménez et al. (25,26), those experiencing a relapse had poorer β-cell function in comparison with those maintaining a durable remission. The independent baseline variables found to be significantly associated with late relapse on multivariate analysis in the current study were the preoperative number of diabetes medications, preoperative duration of T2DM, and type of bariatric surgery (SG vs. RYGB).
We also found higher short-term and long-term weight loss in patients with durable diabetes remission compared with patients with late relapse. Furthermore, we found that patients with late relapse of T2DM had significantly higher weight regain in the long-term compared with those without relapse. These observations mirror findings by Debédat et al. (13), who found that patients who relapsed with T2DM after 5 years exhibited more severe diabetes at baseline, lost significantly less weight during the 1st year after RYGB, and regained more weight afterward. Since postoperative weight trajectories of patients with durable remission and late relapse of T2DM were significantly different, we constructed a second prediction model to highlight the impact of weight loss on long-term diabetes status. Addition of weight change data could improve the c-statistic of the prediction model (from 0.75 to 0.80). Interestingly, with inclusion of weight trajectory data as independent predictive variables in the second model, type of surgical procedures (RYGB and SG) was no longer a significant predictor on multivariable analysis.
The prediction models would assist in shaping patient expectations, informed consent process, decision-making, and evidence-based bariatric procedure selection. Our first model can be used before surgery to estimate the risk of late relapse based on severity of diabetes and can help in procedure selection between RYGB and SG. Procedure selection, however, should not be solely based on diabetes-related outcomes. Diabetes is one outcome—one of many outcomes and conditions that need to be considered in decision making. The surgical risk (generally lower after SG), differential impact of each procedure on body weight and other obesity-related conditions, presence of other medical and mental problems, and patient’s values and goals should be considered in procedure selection (27). The second model can be used at 1st year after surgery in patients with diabetes remission to estimate their risk of late relapse with or without significant weight regain in future. In addition to our model, Debédat et al. (13) recently published their prediction model of diabetes relapse after RYGB (mean follow-up time of 5 years), originated from 94 patients with persistent remission and only 27 patients with late relapse of T2DM. Their model, which is accessible at https://5y-ad-diarem.nutriomics.org/, incorporates both preoperative variables at baseline (duration of T2DM, number of glucose-lowering agents, and HbA1c) and postoperative variables at 1 year (glucose-lowering agents, FBG, and short-term weight loss) to estimate the rate of late relapse after RYGB (c-statistic = 0.90) (13). Our main model, originated from 289 patients with durable remission and 136 patients with late relapse and subsequently validated on an external cohort of 319 patients (100 patients with relapse), considered only baseline variables to predict the risk of relapse in a median 8-year follow-up after RYGB and SG. Our second prediction model, constructed based on data of 425 patients, can inform the patients of their long-term diabetes status with or without weight gain. When the patient’s values are entered into the calculators (accessible at https://riskcalc.org/ and as a smartphone application), the percent estimate of late relapse of T2DM after RYGB and SG is calculated.
This is one of the first studies in the scientific literature to show statistically significant improvement in glycemic control, diabetes medication requirements, and other cardiovascular risk factors in the patients experiencing a relapse of T2DM. In addition, these patients may further benefit from the beneficial effect of the time period spent in diabetes remission on end-organ complications. This beneficial effect of time spent with better control of diabetes on long-term complications, irrespective of diabetes control on long-term follow-up, has variously been described as the “legacy effect” or “metabolic memory” (3,28). In some of the large diabetes trials, continued posttrial follow-up showed that glycemic differences between the groups disappeared early after the trials ended. Nonetheless, treatment effects persisted with respect to the end-organ complications of T2DM. Therefore, it is possible that patients who eventually relapse with T2DM after bariatric surgery will still continue to experience reduced micro- and macrovascular complications in the long-term (3,29,30). In an effort to support the concept of legacy effect after bariatric surgery, Coleman et al. (31) studied 4,683 patients with T2DM who had undergone bariatric surgery in four health care systems in the U.S. Among patients who experienced a late relapse after initial remission, the length of time spent in remission was inversely related to the risk of incident microvascular disease; for each additional year of time spent in remission prior to relapse, the risk of microvascular disease was reduced by 19% compared with that for patients who never achieved remission.
Although some would consider the late relapse of T2DM a treatment failure, these findings must be viewed against the known risks of poorly controlled T2DM in patients who do not undergo bariatric and metabolic surgery. In patients with late relapse, glycemic control and cardiovascular risk are significantly improved compared with baseline and the trajectory of these chronic conditions has been changed by surgery. Furthermore, the possible benefit from the legacy effect of improved glycemic control provides support to look at these long-term results in a positive light (3,32).
This study has several limitations that must be acknowledged. There are obvious limitations of a retrospective design, but the difficulties of prospectively following up any large cohort for a long enough time mean that there is no prospective large study in the scientific literature specifically focusing on the relapse of T2DM after bariatric surgery. For construction and validation of prediction models, a large sample size is needed, which is usually only possible with retrospective studies. This is particularly important for research on diabetes relapse after bariatric surgery, as patients not experiencing remission (∼40% of patients) would necessarily have to be excluded from analysis, leading to reduced sample size. Since this study only includes patients undergoing RYGB or SG, its findings cannot necessarily be extrapolated to patients undergoing other bariatric procedures. But the fact that these two procedures between themselves account for >95% of the primary bariatric procedures worldwide in patients with T2DM (33,34) means that the findings of this study will be relevant to the practice of most of the bariatric surgeons globally. Another weakness of this study is that only 15% of our patients in the training cohort underwent SG. But in the external cohorts, 38% had SG. Limited follow-up is another problem with this study, but a median follow-up of 8 years (with at least 5-years’ follow-up for all patients) on such a large cohort would still be regarded as acceptable. Nonetheless, it is possible that a longer follow-up time may increase the relapse rate and change the study findings. Loss to long-term follow-up of 20% in the base cohort would be associated with selection bias. After RYGB and SG operations, patients may achieve their nadir weight at different time points. Identification of nadir weight depends on the frequency weight measurement. Therefore, we opted to calculate late weight regain from year 1 after surgery and not from the nadir weight. We have not been able to study biochemical parameters like insulin, C-peptide, and incretin hormones at baseline and during follow-up. This means we are not able to consider them for our prediction models and to offer scientific explanations based on these markers for relapse of T2DM in these patients. However, these markers are not available in every clinical practice. Future studies in this area will need to address this. In addition to bootstrap validation, we could pool data of three external cohorts from the U.S. and Europe to assemble a validation cohort with 100 cases with the study event (relapse), which is the requirement for meaningful validation (17). Nonetheless, external validation in other cohorts, especially in different patient populations (e.g., Asian ethnicity or class I obesity), should be considered in future.
In conclusion, consistent with previous studies, our data show that ∼60% of patients with T2DM experience diabetes remission within the 1st year after bariatric surgery. Furthermore, this study, with the largest cohort examining diabetes relapse to date, indicates that while late relapse is a real phenomenon (one-third of our cohort), relapse of T2DM years after bariatric surgery should not be considered a failure, since the trajectories of cardiometabolic risk factors, diabetes severity, and medication requirements are altered favorably after surgery.
Earlier surgical intervention, RYGB (compared with SG), and better weight loss outcomes (less late weight regain) are associated with lower diabetes relapse in the long-term. Understanding pathophysiology of durable remission and late relapse could aid patient and procedure selection. Further research is also needed to study the potential effect bariatric surgery might have on the subsequent incidence of diabetes micro- and macrovascular complications in those who experience a late relapse of T2DM.
Article Information
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. A.A. designed the study and contributed to data acquisition, data analysis and interpretation, literature search, and writing the manuscript. K.M. contributed to literature search and writing the manuscript. J.V., P.S., C.D.S., Z.N.H., G.S., G.C.W., A.I., and A.J. contributed to data acquisition. C.T. analyzed the data. All authors reviewed the manuscript and edited it for intellectual content and gave final approval for this version to be published. A.A. 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.
Prior Presentation. Parts of this study were presented in abstract form at the 35th American Society for Metabolic and Bariatric Surgery Annual Meeting at ObesityWeek, Nashville, TN, 11–15 November 2018.
Footnotes
This article contains Supplementary Data online at https://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc19-1057/-/DC1.
This article is featured in a podcast available at https://www.diabetesjournals.org/content/diabetes-core-update-podcasts.
- Received May 26, 2019.
- Accepted December 21, 2019.
- © 2020 by the American Diabetes Association.
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/content/license.