Potential Benefits and Harms of Gastric Bypass Surgery in Obese Individuals With Type 1 Diabetes: A Nationwide, Matched, Observational Cohort Study

The potential long-term benefits and possible complications of bariatric surgery in patients with type 1 diabetes (T1D) have not been studied extensively. At the same time, the role of bariatric surgery in the treatment of type 2 diabetes as well as its impact on diabetes complications has been thoroughly studied (1–3). In Sweden, over 55% of all patients with T1D were reported overweight, and almost 18% were obese in 2017 (18.5% of female patients and 15.7% of male patients). The treatment of obesity, however, is complicated. Extreme reductions of caloric intake, as with low-calorie diets or initially after bariatric surgery, can make the dosing of insulin very unpredictable and the anabolic effects of insulin can result in difficulties in maintaining reduced weight.

Minor cohort studies and case reports have shown that bariatric surgery leads to weight loss; positive effects on comorbidities, such as hypertension, hyperlipidemia, and albuminuria; and reduction of insulin doses in T1D. However, results with regard to glycemic control postoperatively have not been consistent (4–10). Recent studies have described serious diabetic ketoacidosis (DKA) postoperatively (6,11) as well as hypoglycemia (1).

In the 2018 guidelines, the American Diabetes Association and European Association for the Study of Diabetes called for further research and long-term follow-up on the safety of bariatric surgery for patients with T1D (4). In this study, we therefore compared patients with T1D who had undergone Roux-en-Y gastric bypass (RYGB) surgery to age-, BMI-, sex-, and diabetes duration-matched individuals with T1D who had not undergone bariatric surgery to investigate risks of cardiovascular events, acute and chronic diabetes complications, effects on psychiatric health and substance abuse as well as overall mortality.

The study population originates from the merging of two nationwide Swedish registries: The National Diabetes Register (NDR) (12) and the Scandinavian Obesity Surgery Registry (SOReg) (13). These registries contain information on more than 95% of all patients with diabetes in Sweden and over 98% of all bariatric surgery performed in the country. Health care providers have registered information on treatment and diverse measurements continuously directly to NDR or via electronic patient records from routine clinical practice since 1996. Date of surgery, type of surgical procedure and complications of surgery are reported for up to 10 years postoperatively in SOReg that was started 2007.

Patients with T1D who had undergone RYGB between January 2007 and December 2013 were identified in SOReg and matched with patients with T1D from NDR who had not undergone bariatric surgery. Propensity score matching was performed on the basis of sex, age, BMI, and calendar time. None of the control patients underwent surgical treatment for obesity during the follow up period. We used both an epidemiological and a clinical definition to capture the patients with T1D: either treatment with insulin only and diagnosis at an age of 30 years or younger as in previous publications from NDR (14,15), or as clinically determined and reported by physicians. The epidemiological diagnosis has been validated as accurate in 97% of cases by comparison with the presence of hyperglycemia and autoantibodies (16). Thus, all patients in this study were treated with insulin. The surgical method was performed according to international guidelines and was described by Lönroth et al. (17). For the study, clinical characteristics and information on treatment from NDR was linked to date and registered RYGB surgery from SOReg using personal identification numbers unique to each individual. Information was also gathered from Statistics Sweden (socioeconomic variables), the Swedish Inpatient Registry (hospital admissions, coexisting conditions), Prescribed Drug Register (pharmaceutical treatment), and the Cause of Death Registry (cause and date of death). These registries have all been validated and described previously (18). The ICD-10 diagnosis was used for definition of incident cardiovascular events, hyperglycemia with or without coma (including DKA), hypoglycemia with or without coma, kidney disease, and alcohol or drug abuse as well as psychiatric disorders (Supplementary Material). All events were from registered in-hospital treatment. Each patient in NDR has given informed consent and the Regional Ethical Review Board in Gothenburg, Sweden (DNR 56312) approved the study.

Continuous variables are presented as mean (SD), and categorical variables are presented as means of frequencies and percentages. Standardized mean differences (SMDs) are presented for each variable. Since individuals with T1D who undergo RYGB are not directly comparable to other individuals with T1D, we are mainly interested in the average treatment effect among the treated. Hence, we used time-dependent propensity score matching (19) to find control patients for each treated individual matched on sex, age, BMI, and calendar time. The time-dependent propensity scores were estimated using a Cox proportional hazards model with time-dependent covariates and control patients were selected by matching on the estimated risk score among the people at risk for exposure at the time of each exposure. A distinct advantage of this procedure is that the selected control patients were given the same index date as the corresponding case. Time-to-event variables were subsequently analyzed using Cox proportional hazards regressions models with exposure as the only independent variable.

Because of the explorative nature of the study, there is no adjustment for multiple comparisons, and consequently, individual statistical test should be interpreted with caution.

We identified 387 individuals with T1D who had undergone RYGB in NDR and SOReg, and we matched them with 387 control patients with T1D from NDR. Follow-up for hospital admissions was up to 9 years (mean 4.8 years), and all-cause mortality was for up to 10 years (mean 5.7 years) for both groups. Baseline characteristics as well as an overview of baseline pharmaceutical treatments other than insulin can be seen in Table 1. All patients included in the study were treated with insulin.

There were differences between the groups with regard to sex and levels of physical activity (SMD > 0.2). BMIs were 39.5 and 40.8 kg/m² for the control and RYGB groups, respectively (SMD 0.205). The groups were well matched for age, diabetes duration, HbA_1c levels, blood pressure, LDL-cholesterol levels, cardiovascular and kidney disease, psychiatric disorders, and history of alcohol and drug abuse as well as cancer (SMD < 0.2). The range of diabetes duration in the RYGB group was 0–55 years compared with 0–62 years in the control group. The patients’ previous history of conditions and comorbidities requiring hospitalization before the start of treatment can be seen in Table 2. There were no differences between the groups with regard to comorbidities or pharmaceutical treatment (SMD < 0.2).

During the follow-up period, reported cardiovascular disease (hazard ratio [HR] 0.43; 95% CI 0.2–0.9; P = 0.026) (Fig. 1A) and cardiovascular mortality (HR 0.15; 95% CI 0.03–0.68; P = 0.013) were significantly lower in the RYGB group. The differences were most marked for stroke (HR 0.18; 95% CI 0.04–0.82; P = 0.027) and heart failure (HR 0.32; 95% CI 0.15–0.67; P = 0.003). The differences in incidence of myocardial infarction (HR 0.57; 95% CI 0.24–1.35; P = 0.199) and atrial fibrillation (HR 0.69; 95% CI 0.30–1.62; P = 0.395) were broadly similar. The group that had been treated with surgery had significantly more serious hyperglycemic events (HR 1.99; 95% CI 1.07–3.72; P = 0.030) (Fig. 1B), and there was a numeric but not statistically significant difference in hypoglycemic events that lead to coma (HR 1.57; 95% CI 0.78–3.16; P = 0.205) (Fig. 1C). A difference in the incidence of serious hyperglycemic events, including DKA, was already apparent during the first 2–3 days postoperatively. There was a definite trend in decreased all-cause mortality for the RYGB group, but the difference was not significant statistically (HR 0.57; 95% 0.32–1.02; P = 0.060) (Fig. 1D).

Regarding secondary outcomes, the analysis did not detect a significant difference in risk of kidney disease (HR 0.70; 95% CI 0.39–1.26; P = 0.234) or leg amputation (HR 0.57, 0.17–1.95; P = 0.373), and although there was a trend toward more psychiatric disorders, the difference did not reach significance (HR 1.72; 95% CI 0.96–3.08; P = 0.070) (Fig. 1E). There was a significantly increased risk for alcohol and substance abuse in the RYGB group (HR 3.71; 95% CI 1.03–13.29; P = 0.044) (Fig. 1F). Four patients in the surgical group died because of diabetes-related coma compared with one patient in the control group. Cardiovascular comorbidity or heart failure were the most common causes of death in the control group.

In patients with available data, the mean HbA_1c level was 7.6% (59.6 mmol/mol) and 7.8% (62.1 mmol/mol) at 1 and 2 years after baseline, respectively, in the surgical group compared with 8.3% (67.2 mmol/mol) and 8.3% (67.4 mmol/mol) at 1 and 2 years after baseline, respectively, in the control group. BMI in the surgical group was 30.6 kg/m² and 28.8 kg/m² at 1 and 2 years after surgery, respectively, compared with 37.5 kg/m² and 37.5 kg/m² at 1 and 2 years after surgery, respectively, in the control group. Reported weight was 86.2 and 82.3 kg after 1 and 2 years, respectively, in the surgical group and 106.1 and 105.8 kg after 1 and 2 years, respectively, in the control group. An overview of these changes can be seen in Table 3.

Our observational findings show that although surgical treatment of obesity in patients with T1D is associated with a substantially lower risk for cardiovascular disease and cardiovascular death, stroke, and heart failure, it appears to be associated with a higher risk for serious hyperglycemic events, including DKA and abuse of alcohol and narcotics. Our results also imply that bypass surgery might be associated with a higher risk for serious hypoglycemic events after surgery as well as psychiatric illness. That said, if the results represent true or close-to-true causal effects, the potential net benefits would seem to outweigh potential harms. Such findings may help other centers considering bariatric surgery as a potential treatment in similar patients.

Earlier studies on the positive effects of bariatric surgery in general and in patients with type 2 diabetes have shown similar results with regard to cardiovascular benefits and mortality (20,21). At the same time, studies have also shown higher incidence of serious psychiatric illness, alcohol and substance abuse, and even suicide in patients with type 2 diabetes after bariatric surgery when compared with nonsurgical control patients (1), which is also in line with our results. Using data from register-based cohort studies to interpret whether there was increased psychiatric disease and substance abuse before surgery or if this was a consequence of the surgical operation is of course challenging as these kinds of studies are generally based on diagnoses received only after hospitalization.

This study is by far the largest study of bariatric surgery in patients with T1D. The registers that were used in the study include 98% of patients with T1D and all individuals who have undergone bariatric surgery in Sweden, which makes this study representative for the country. The length of follow-up also strengthens the study. Lack of information about treatment with insulin pumps and usage of continuous blood glucose monitors with alerts to high and low levels of blood glucose can be seen as a limitation as this clearly affects the risk of hyper- and hypoglycemic events. This information was not available from the registries used at the time of the study. Similarly, we lack robust information on the effects on microvascular complications apart from renal events. Both patient groups had access to similar treatment options and any meaningful differences in usage of technology within the groups seem unlikely. Earlier studies have shown a reduction in insulin doses after RYGB (9), but this variable was not included within our study. RYGB was the most commonly used surgical technique at the time of the study and is therefore the only procedure included in this report. As commonly seen in studies on bariatric surgery, as well as in clinical practice, both groups included more women.

Concerning selection bias, there were no substantial differences between the study groups with regard to diseases that could affect the preoperative assessment, but other details of the individual evaluation of the patients cannot be included in this kind of study. By using both the clinical and epidemiological diagnosis of T1D and the fact that this particular group of T1D patients is obese, increases the risk that there are young patients with type 2 diabetes included. However, the epidemiological diagnosis has been validated in earlier studies (16), and it is unlikely that there is a large population of relatively young patients with type 2 diabetes that is only treated with insulin. All of the included individuals were treated with insulin, and there was only one patient in each group with a history of sulfonylurea use, two in the RYGB group that had received dipeptidyl peptidase 4 inhibitors, one patient in the control group with meglitinide treatment, and one patient in the surgical group with reported used of a thiazolidinedione. Metformin and glucagon-like peptide 1 agonists are not exclusively used in type 2 diabetes, especially if the patients are obese, but they are only used off label in T1D.

The groups were matched using a propensity score to attempt to find control patients who were as likely to be exposed to the surgical treatment as the individuals in the surgical group, and thus to reduce effect of possible confounders on the effect of treatment. This approach gives us the average treatment effect among the treated (the effect of RYGB among those who underwent the intervention) as opposed to the average treatment effect in the whole T1D population (the effect of having the complete population with T1D undergo the intervention), which is estimated in a regression model. Individuals with T1D who undergo RYGB differ considerably from other individuals with T1D and RYGB is an unrealistic intervention for most individuals with T1D. Therefore, we are mainly interested in the treatment effect in the treated group, which motivates the choice of propensity score matching as opposed to alternative analysis approaches such as regression modeling. As exposure happens over time, we need to establish an index date for the selected control patients, which is the main reason for using time-dependent propensity scores. Given the limited number of persons in the treated group we were unable to include more than a few variables in the propensity score model.

The reported Cox regressions were unadjusted for additional factors, and because of the size of the groups we believe this is sufficient. A sensitivity analysis after such adjustment for sex, BMI and age showed results in line with the results reported here (data not shown). This was the best matching possible because of the limited availability of obese control patients with T1D. We are aware that this limits the number of events available for analysis but matching 1:2 or higher was not possible. However, control patients were found for each surgically treated patient and no cases were lost or excluded. As NDR contains information on more than 95% of all patients with T1D in the country, the follow-up should not differ between the surgical group and control group as might be seen in the general population. All individuals are included as long as they have not emigrated.

Technological advances have made the dosing of insulin more accurate and prevention of hyper- and hypoglycemia easier, and this is important when assessing the risk involved with bariatric surgery for patients with T1D. However, when assessing a patient for bariatric surgery, one cannot assume that these treatment options are available or of interest to the patients postoperatively. Serious hyperglycemic events directly after surgery indicate insufficient insulin treatment and the need for early contact with the diabetes team. It is important that the surgical team is aware that continuous treatment with insulin is of vital importance for patients with T1D and that treating these patients is not comparable to treating patients with type 2 diabetes with regard to insulin dosing postoperatively. The stress of the surgical operation could also increase the need for insulin. The hyper- and hypoglycemic events included were only those that required hospitalization and are likely an underestimation. Four patients in the surgical group died because of diabetes-related coma, compared with one patient in the control group, which stresses the seriousness of these possible complications.

There are no randomized prospective trials on the effects and complications of bariatric surgery in patients with T1D available, and this sort of study would be ethically questionable. A recent randomized study comparing RYGB and sleeve gastrectomy in type 2 diabetes patients showed better results after RYGB (22), but further research is needed on the choice of surgical method for patients with T1D as well as comparisons between surgical and medical treatment of obesity. It is also of interest to carefully consider the technological aspect of insulin treatment and glucose monitoring for the patients who receive surgical treatment. The usage of these devices might need to be considered preoperatively to reduce risk for serious hypo- and hyperglycemic events after surgery. Finally, it is important to further evaluate whether weight regain is more common in such patients because of the continuous treatment with insulin and what the effects are on overall quality of life.

While our findings stem from an observational (case versus matched control) study, they suggest that bariatric surgery may yield broadly similar benefits on risk for cardiovascular disease and cardiovascular mortality in patients with T1D and obesity as has been reported for patients with type 2 diabetes. However, such surgery may increase the risk for adverse effects, such as hyperglycemia and abuse of alcohol and drugs. We suggest that the preoperative evaluation and postoperative monitoring should be done on an individual basis and with early involvement of a diabetes team.

Acknowledgments. The authors thank Margrét L. Höskuldsdóttir (Reykjavik, Iceland) for assistance with graphics and figures.

Funding. Sveriges Kommuner och Landsting funds the NDR and the Scandinavian Obesity Surgery Register. Västra Götalandsregionen also provides funding for the NDR. Naveed Sattar is supported by the British Heart Foundation Research Excellence Award (RE/18/6/34217).

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

Author Contributions. G.H. drafted the article. G.H., J.E., M.M., V.W., J.O., I.N., S.G., A.-M.S., and B.E. contributed to the conception and design of the study. J.E. performed the statistical analyses. J.E., M.M., J.O., I.N., and A.-M.S. contributed to the acquisition of data. J.E., M.M., V.W., J.O., I.N., S.G., N.S., A.-M.S., and B.E. contributed to critical revision. All authors contributed to the interpretation of data. B.E. 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 55th Annual Meeting of the European Association for the Study of Diabetes, Barcelona, Spain, 16–20 September 2019.