BMI, Mortality, and Cardiovascular Outcomes in Type 1 Diabetes: Findings Against an Obesity Paradox

Recent studies suggest that insulin resistance in overweight or obese individuals with type 1 diabetes may be associated with an increased risk of vascular complications (1), but few studies have investigated the relationship between BMI in individuals with type 1 diabetes and mortality. Phenomena such as the obesity paradox, which suggests that there is an inverse association between BMI and risk of cardiovascular outcomes, have been debated intensely in the past decade (2). Also, the prevalence of obesity in the general Swedish population has increased steadily over past decades (3), and for patients with type 1 diabetes, weight gain is a potential side effect of intensive insulin therapy, which is the mainstay of modern management (4). Hence, to our knowledge, no study has described the overall trend in BMI for the population with type 1 diabetes.

Among patients with type 1 diabetes, low weight has been associated with an increased mortality risk in African Americans (5), with BMI <20 kg/m² (6) or <18.5 kg/m² (7). Therefore, with new recommendations regarding BMI in registry-based studies (8), we also sought to evaluate the association between BMI and cardiovascular outcomes and mortality within the population with type 1 diabetes to help to inform whether there is a need for altered weight management in this condition. In the current study, we used linked nationwide Swedish registers to investigate these issues.

The Swedish National Diabetes Registry (NDR), initiated in 1996, is a quality assurance tool with nationwide coverage, with informed consent from all participants. The registry includes data on risk factors and medication provided by trained health care professionals (9,10) and continuous work to improve the validity of the data (11). Through the personal identity number assigned to all Swedish residents (12), we linked the following Swedish national registers: the NDR; Inpatient Registry (IPR), which is part of the National Patient Register (13); the Cause of Death Registry; and the Longitudinal Database for Health Insurance and Labor Market. Information regarding coexisting medical conditions for patients and control subjects was retrieved through the IPR using the ICD-9 and ICD-10 from 1987 and onward.

Type 1 diabetes was epidemiologically defined when diagnosed in a patient ≤30 years of age who received only insulin treatment (10). The concordance between the epidemiological classification and diagnosis and that entered by the clinician was 97% (10). However, because the definition used might still include patients with type 2 diabetes, for this study, we also required that the clinical classification was type 1 diabetes.

In the current study, we included patients with at least one registration in the NDR from 1 January 1998 until 31 December 2012, with patients followed until December 2013 or to the occurrence of a cardiovascular event, hospitalization for heart failure (HF), or death from cardiovascular causes or from any cause.

For the first cohort, which included patients with type 1 diabetes (n = 36,872), we excluded 4,329 patients not defined as having type 1 diabetes; 1,846 patients because of previous cardiovascular conditions (8), such as acute myocardial infarction (AMI), stroke, coronary heart disease (CHD), atrial fibrillation (AF), HF, renal dialysis/transplantation (chronic kidney disease), or amputation (for definitions, see Supplementary Table 8); 483 with BMI <18.5 kg/m²; and 4,089 with missing BMI. After exclusions, the final baseline cohort consisted of 26,125 patients (Supplementary Fig. 1). In the follow-up cohort, 3,822 patients with fewer than two visits or follow-up time of <1 year (during the first 5 years of follow-up since baseline) were excluded as were 3,324 patients who died within 5 years or who had a total follow-up time of <5 years from baseline or a heart-related event such as AMI, stroke, AF, CHD, or HF during this period. Hence, the final updated cohort for follow-up consisted of 18,979 patients (Supplementary Fig. 2).

Data on risk factors for patients with type 1 diabetes were obtained by physicians or nurses and entered into the NDR. In Fig. 1, we used all of each patient’s appointments to investigate potential changes in BMI from 1998 to 2012. For all other analyses, we used baseline data and updated baseline data (to account for weight loss/gain the first 5 years from study entry). BMI was calculated as the weight divided by height squared and extracted from medical records by the attending nurse or physician entering data into the registry. Microalbuminuria was defined as two out of three positive urine samples collected within 1 year, with a urinary albumin clearance of 20–200 μg/min (20–300 mg/L) or an albumin-to-creatinine ratio of 3–30 mg/mmol (∼30–300 mg/g). The definition of macroalbuminuria was urinary albumin clearance >200 μg/min (>300 mg/L) or albumin-to-creatinine ratio >30 mg/mmol. The Longitudinal Database for Health Insurance and Labor Market registry provided data on socioeconomic status, such as country of birth, individual income in Swedish kronor, educational level, and marital status. Country of birth was categorized as Swedish or other. Education level was categorized as compulsory only, secondary level, or university level. Marital status was categorized as married/registered partner, never married, widowed, or divorced.

Patients and control subjects were followed from the first registration date in the NDR until occurrence of any of the following: 1) hospitalization for HF; 2) major cardiovascular event, defined as a composite of a first fatal/nonfatal AMI or fatal/nonfatal stroke; 3) cardiovascular mortality, defined as CHD or stroke as the underlying cause of death or these conditions listed as one of the first two contributing causes of death; or 4) death from any cause. Analyses for each outcome were performed separately. For exact definitions of the outcomes, see Supplementary Table 6.

To describe the trends in mean BMI in Swedish patients with type 1 diabetes from 1998 to 2012, we used a mixed model to account for repeated measurements. For this analysis, we used data from all registered appointments with measured weight, which generated 275,111 appointments with a registered BMI of a total of 349,790 appointments for all 32,543 patients. Calendar year, age, sex, and diabetes duration were used as fixed effects, while random effects were set as person and as a trend among subjects. Least square means were used to calculate means with 95% CIs.

We calculated crude incidence rates (with 95% CIs) per 1,000 person-years overall and stratified by baseline BMI. Models among patients with type 1 diabetes were adjusted for age, sex, duration of diabetes, socioeconomic status (14), HbA_1c, and smoking status (8). We performed analyses with Cox regression with CI level set at 95% for all outcomes. BMI was modeled using restricted cubic splines with three knots, where BMI of 25 kg/m² served as the reference. In several steps, we analyzed the potentially increased risk among those with BMI <25 kg/m², where we stepwise excluded smokers and then smokers and patients with HbA_1c ≥60 mmol/mol (<7.6%). In a third step, we additionally set 2002 as the latest year of registration to allow for a follow-up time of ≥10 years. In a sensitivity analysis, we added adjustments for systolic blood pressure and LDL cholesterol. Finally, we included patients in the entire BMI span, including those with BMI <18.5 kg/m², and similarly to the main analyses, we performed a stepwise exclusion of patients who could contribute to reverse causality regarding all-cause mortality and cardiovascular disease (CVD) mortality. To investigate what a change in BMI during the first years of follow-up could mean for the selected outcomes, we used the second cohort (Supplementary Table 2) to assess the change in BMI, which was calculated by taking the absolute change restricted to between 1 and 5 years, dividing the absolute change by the days between measurements, and multiplying the change per day by 365.25 to describe the yearly change in BMI. The new start date for the updated data was set to the date of the last registered BMI within the first 5 years from the day of inclusion. We divided BMI change per year into four groups (<0, 0 to <0.25 [reference], 0.25 to <0.75, and ≥0.75 kg/m²) and performed Cox regression for the outcomes. We experienced a loss of power as a result of exclusion criteria for updated BMI; thus, we did not include the change in BMI in sensitivity analyses with exclusion of patients with poor glycemic control and short follow-up time (<10 years).

We assessed the Cox proportional hazard using Schoenfeld residuals. The proportional hazard assumption was considered to be fulfilled. The analyses were performed using R version 3.4.3 statistical software.

Figure 1 describes the mean BMI from 1998 to 2012 in patients with type 1 diabetes, adjusted for age, sex, and diabetes duration, and shows an increase from a mean of 24.7 to 25.7 kg/m² (see flowchart in Supplementary Fig. 1).

The baseline characteristics of the study population are presented in Table 1. Mean age was 33.3 (SD 13.0) years, mean age at diabetes onset was 14.8 (SD 7.5) years, and mean diabetes duration was 18.5 (SD 13.3) years. All BMI groups had a mean estimated glomerular filtration rate (eGFR) of >95 mL/min/1.73 m². Among the 26,125 patients with type 1 diabetes, the largest group was formed by patients with BMI 18.5 to <25 kg/m² (n = 15,288, 58.5%); 8,677 patients (33.2%) had a BMI between 25 and <30 kg/m², and 2,160 (8.3%) had a BMI ≥30 kg/m². Compared with normal-weight patients, obese patients with type 1 diabetes were more often women; were more frequently on medications; and more frequently had albuminuria, higher mean systolic blood pressure, higher mean HbA_1c, higher LDL cholesterol, and lower eGFR. Baseline characteristics by sex are presented in Supplementary Table 1, where men with obesity had higher mean systolic blood pressure, older age at onset, shorter duration of diabetes, and slightly lower mean HbA_1c than obese women. In baseline data for updated BMI (see flowchart in Supplementary Fig. 2, new updated baseline in Supplementary Table 2, and baseline by sex in Supplementary Table 3), we identified that those who experienced the greatest weight gain had an updated mean BMI of 29.3 kg/m², while those who experienced weight loss had an updated mean BMI of 24.1 kg/m². A comparison between the overall cohort versus patients with missing BMI values is presented in Supplementary Table 4.

With a median follow-up time of 10.9 years, there were 1,031 deaths (342 from cardiovascular causes), 1,460 cases of major CVD events, and 580 cases of HF (Supplementary Table 5). Detailed definitions of preexisting conditions and outcomes by ICD-9 and ICD-10 are presented in Supplementary Table 6. We adjusted for several covariates, including smoking and HbA_1c. In the overall cohort, there was a J-shaped association between BMI and mortality but no significantly increased risk associated with low BMI for any of the other outcomes (Fig. 2 and Supplementary Fig. 3). Crude incidence rates are presented in Table 2. In a stepwise approach, we excluded smokers, patients with HbA_1c ≥60 mmol/mol (≥7.6%), and patients with a follow-up time <10 years (Supplementary Fig. 4) and identified a substantial attenuation in risk for patients with BMI <25 kg/m². With the last model (also presented in Fig. 2), this observation also included those with BMI <18.5 kg/m² (Supplementary Figs. 5 and 6), where there was still a slightly J-shaped curve for all-cause but not for CVD mortality. There was an increased mortality risk among the obesity categories, whereas the association between BMI and death from CVD was essentially flat. However, there was an association between higher BMI and major CVD events as well as with HF. In analyses that comprised annual weight change, there were no significantly increased risks for any of the outcomes compared with the reference level of 0 to <0.25 kg/m² gain in BMI per year (follow-up description in Supplementary Table 7). Additionally, those with the greatest weight gain had an increased risk of mortality and HF compared with their reference. Our sensitivity analyses in Supplementary Fig. 7 also included adjustments for LDL cholesterol and systolic blood pressure, with the results for obese patients similar to those in Fig. 2.

In Supplementary Figs. 8–11, we identified stronger associations between baseline BMI and outcomes for men compared with women. Crude incidence rates are presented in Supplementary Table 8. Men showed an almost linear increase from BMI 25 kg/m² for all outcomes, while associations in women did not show any significant variation by BMI for any outcome. Similarly, weight change had a greater effect on men than on women (Supplementary Figs. 8 and 9). We identified some increased risk for mortality in men who lost weight, but this was nonsignificant in nonsmoking individuals (Supplementary Fig. 12).

This nationwide population-based cohort study of >25,000 patients with type 1 diabetes shows that obesity in type 1 diabetes may contribute to an elevated risk of mortality, major CVD, and HF, but there is negligible support for an increased risk associated with low BMI. In an era of increasing BMI in the population, several chronic conditions show an apparent obesity paradox where a low BMI may signal progressive disease and thus conceal true adverse effects of overweight and obesity. The findings of the current study do not support any association between a healthy BMI, weight loss, and increased risk of overall or CVD mortality in patients with type 1 diabetes. Instead, we observed a positive, albeit modest, association between BMI and the risk of mortality and a positive association between major CVD and HF. This pattern was only apparent in men, whereas associations with BMI were less pronounced and nonsignificant in women, but CIs were wide because of lower event rates. These findings and evidence of worse risk factor profiles with increasing BMI suggest that supporting attempts to maintain a normal weight while maintaining good glycemic control may have long-term benefits.

With respect to type 1 diabetes, there is also long-standing concern that intensive glycemic control may cause weight gain. Over the 15-year observation period, mean BMI at baseline increased by ∼1 kg/m²; hence, individuals with type 1 diabetes in Sweden display a mean increase in BMI similar to the general population (15). Increasing body weight is undesirable for several reasons, and we identified a modest relationship between high BMI and greater mortality. Additionally, there was a clear relationship between increasing BMI and major CVD and HF. In the current study, patients with normal weight, compared with moderately overweight patients, did not show an increased risk of any CVD or HF event, and there was no increased risk for mortality among the leanest patients after adjusting for smoking. Previous studies have shown a substantially higher mortality risk among patients with very low BMI and an apparent protective effect of moderate overweight (BMI 25 to <30 kg/m²) compared with individuals with normal weight (7). Weight loss in type 1 diabetes has been shown to be associated with an increased mortality risk (6). However, reverse causality may be a factor in this, where factors such as short follow-up time, frailness, and prior CVD (all factors where weight loss may be the result of prevalent disease) may not have been accounted for (8,16). In the current study, after exclusion of patients with CVD-related comorbidities at baseline and other factors potentially contributing to reverse causality, the group with normal weight (18.5 to <25 kg/m²) did not display any significantly greater risk compared with those with BMI at the reference level of 25 kg/m². Among patients with severe underweight, the J-shaped curve was attenuated when we successively excluded patients with factors associated with reverse causality, where the relative risk of mortality was a nonsignificant slight increase among those who were the most underweight compared with the nadir point of 25 kg/m². However, we do not rule out the possibility of influence from reverse causality among the very small group of patients with severe underweight. Additionally, and in contrast to previous research (6), we did not find any increased risk for adverse outcomes for those who underwent weight loss. Given the plethora of evidence demonstrating the deleterious effects of poor glycemic control, we believe not only that our results may reassure clinicians about the intrinsic value of pursuing ambitious goals for glycemic control but also that a low weight among otherwise healthy patients with type 1 diabetes does not imply an increased mortality risk. Hence, our results support health care professionals encouraging patients with type 1 diabetes to maintain a lower BMI through lifestyle changes, if they can, to further lessen their risk of vascular complications.

There are few previous studies on BMI and the risk of CVD events, but the Diabetes Control and Complications Trial (DCCT) reported a decrease in CVD events in patients assigned to intensive glycemic therapy, regardless of weight gain in the first years of follow-up. However, despite an initially lower risk of CVD, those with maximal weight gain showed risk level equivalent to that of the conventional group after 13 years of follow-up, suggesting that the weight gain has the potential to harm over the longer term (17). Of note, the obese patients did not improve their glycemic control, suggesting that the weight gain cannot simply be a result of more-intensive insulin therapy but that adverse lifestyle factors must also be involved. Whether glucagon-like peptide 1 receptor agonists could be useful adjuncts for very obese patients with type 1 diabetes should be further explored because some research has identified benefits for such patients on glycemic control (18). Similarly, there is ongoing work investigating the benefits of adjunct sodium–glucose cotransporter 2 inhibitors in type 1 diabetes (19). Interventions such as bariatric surgery may have a positive effect on weight but a more uncertain effect on glycemic control in type 1 diabetes (20).

In contrast to our findings, previous studies using the NDR have shown a very marked increased risk for death (21) and HF (22) among patients with type 1 diabetes with poor glycemic control. There was also a marked increased risk for CVD and HF among patients with poor overall risk factor control (23) or who were very young at onset (24). Accordingly, this could indicate that treating HbA_1c to an optimal level, despite a potential risk of weight gain, is an important aspect in diabetes care (25).

Among the strengths of the current study is a uniquely large sample of patients with type 1 diabetes. We were able to identify a cohort with a type 1 diabetes diagnosis using validated criteria and further required confirmation through a clinical diagnosis by the managing physician, which strengthens the validity of the study. The NDR and the IPR provided detailed data on comorbidities and the opportunity to exclude patients with previous chronic kidney disease, AMI, stroke, CHD, AF, and HF; smokers; and patients with short follow-up. This allowed us to identify a group where the influence of reverse causality would be minimal. Another strength is that this population-based study includes the majority of the Swedish patients with diagnosed type 1 diabetes. With publicly financed health care at low cost to the individual, the present cohort is probably representative of many high-income Western countries where most patients have health insurance provided in some form.

The main limitation of the current study is that no information was available on insulin dosing, which would have been of interest in this context. Neither did we have sufficiently complete data on waist-to-hip ratio. Why women in this study displayed little or no associations with BMI could be due to different fat distribution, and future studies should investigate potential differences regarding the effect of overweight and obesity between men and women with type 1 diabetes.

In conclusion, in this nationwide population-based study of 26,125 patients with type 1 diabetes, a slight positive increase was noted for all outcomes in patients with type 1 diabetes without prior clinical CVD, with no apparent obesity paradox and a modest positive association of BMI with major CVD and HF; this association was only evident in men, at least in this relatively young cohort. In contrast to the effect of glycemic control and other cardiovascular risk factors, the risks conferred by obesity seemed modest but higher than in normal-weight patients. These findings support the pursuit of lifestyle changes to improve weight in patients with type 1 diabetes, although they do not detract from the need to strive for strict glycemic control using, where required, aggressive insulin therapy.

Acknowledgments. The authors thank the regional NDR coordinators as well as the contributing nurses, physicians, and patients. The authors also thank Jodi Smith from Edanz Group (www.edanzediting.com/ac) for editing a draft of this manuscript.

Funding. This work was supported by grants from the Swedish state under an agreement between the Swedish Government and the County Councils Concerning Economic Support of Research and Education of Doctors (ALFGBG-427301), the Swedish Heart and Lung Foundation (grant number 2015-0438), the Swedish Research Council (2013-5187 [SIMSAM], 2013-4236], and the Swedish Council for Health, Working Life and Welfare (FORTE) (2013-0325). The Swedish Diabetes Association and the Swedish Society of Diabetology support the NDR. The Swedish Association of Local Authorities and Regions fund the NDR.

Duality of Interest. M.L. has received lecture fees from AstraZeneca, Eli Lilly, and Novo Nordisk and research grants from AstraZeneca, Dexcom, and Novo Nordisk and has been a consultant for Merck Sharp & Dohme and Novo Nordisk, all outside the submitted work. N.S. reports grants and personal fees from Boehringer Ingelheim; personal fees from Janssen, Novo Nordisk, and Eli Lilly; and grants from AstraZeneca, all of which are unrelated to the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.E. wrote the draft of the manuscript. J.E., A.Ra., and M.A. performed the statistical analyses. J.E., A.Ra., M.A., L.B., M.L., A.-M.S., S.G., N.S., and A.Ro. interpreted data and critically revised the manuscript. J.E., A.Ra., M.A., and A.Ro. developed the study design and concept. J.E. and A.Ro. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.