Contrasting Associations of Body Mass Index and Hemoglobin A1c on the Excess Risk of Acute Myocardial Infarction and Heart Failure in Type 2 Diabetes Mellitus

Patients and comparators were registered from January 1, 1998, until December 31, 2012 and followed until December 31, 2013, the event of interest or death. Patients and comparators were linked to the Swedish Inpatient and Cause of Death Registers through their personal identification number^{13, 14} to obtain information about coexisting conditions such as stroke, AMI, CHD, hospitalization for heart failure, atrial fibrillation (AF), renal dialysis/transplantation (chronic kidney disease), cancer and dementia. Codes from the International Classification of Diseases, Ninth and Tenth Revisions (ICD‐9 and ICD‐10), were used from 1987 and onwards (Table S1).

For the outcomes of AMI and heart failure (Table S1), we similarly used the Swedish Inpatient and Cause of Death Registers, where AMI was defined as the principal or contributory diagnosis at the first identified case of either hospitalization or death in AMI (I21). The outcome of heart failure was defined as either the principal or contributory diagnosis at the first identified case of a hospitalization for heart failure (I50). The Longitudinal Database for Health Insurance and Labor Market studies provided information about socioeconomic variables, marital status (divided into single, married, divorced, and widowed), education level (divided into compulsory education or lower, intermediate, ie, upper secondary, and high, ie, university) and country of birth (Swedish/other).

BMI was calculated from data on height and weight measured by the reporting unit (primary care units or hospital outpatient diabetes mellitus clinics) as weight (kg)/height (meters).² HbA1c was initially measured as percent (mono‐s) and converted into mmol/mole per mole (IFCC [International Federation of Clinical Chemistry]) (10). Microalbuminuria was defined as 2 positive tests from 3 samples taken within 1 year, with an albumin/creatinine ratio of 3–30 mg/mmol (≈30–300 mg/g) or U‐albumin of 20 to 200 μg/min (20–300 mg/L), and macroalbuminuria as albumin/creatinine ratio >30 mg/mmol (≈ >300 mg/g) or U‐albumin >200 μg/min (>300 mg/L). BMI was measured as kg/m² and imputed by using first observation carried backward if missing. We only imputed BMI if values were available within 365 days and provided that no intervening major cardiovascular event occurred (stroke, AMI, CHD, AF, or chronic kidney disease). Before imputation, 112 848 (24.7%) patients had missing BMI, which decreased to 81 721 (17.9%) after imputation.

See flowchart in Figure S1. The original cohort consisted of 457 473 patients and 2 287 365 comparators from the general population. Patients and comparators were excluded if they had inconsistent survival data, which are usually explained by reuse of the unique Swedish personal identification number. We additionally excluded patients and comparators with survival time of 0; after these exclusions 457 453 individuals with type 2 diabetes mellitus and 2 260 994 matched‐population comparators remained. We excluded the complete matched set (1 patient along with 5 comparators) if any of the following conditions were fulfilled: a patient or control had a previous diagnosis of AMI, stroke, CHD, or heart failure (patients and comparators left after exclusion, n=216 183 and n=1 077 471, respectively); if the patient had BMI <18.5 (patients and comparators left after exclusion, n=215 590 and n=1 074 521, respectively) or if the patient had missing BMI after imputation (13.4%; patients and comparators left after exclusion, n=181 045 and n=902 302, respectively).

Patients were divided into 5 BMI categories; 18.5 to <25, 25 to <30, 30 to <35, 35 to <40 and ≥40 kg/m². Each BMI group was further divided into HbA1c categories; <53 mmol/mole, 53 to 70 mmol/mole, and ≥71 mmol/mole. Age‐adjusted incidence rates were calculated as events per 1000 person‐years with 95% exact (Poisson) CIs. To present the excess risk for patients with type 2 diabetes mellitus versus the average age‐ and sex‐matched comparators, we performed Cox regression analyses, adjusted for age, sex, duration of diabetes mellitus, socioeconomic status, and comorbidities at baseline (AF and chronic kidney disease) additionally performed with the outcome of heart failure as the principal diagnosis. Comparators served as the reference group for each Cox regression performed separately by each presented HbA1c group (<53, 53–70, and ≥71 mmol/mole), to demonstrate the effect from BMI and HbA1c compared with the general population comparators with a presumed normal HbA1c and population mean BMI with a similar design as previously published research from NDR.^{1, 2, 4} Duration of diabetes mellitus was centered around the grand mean, while the duration for comparators was set to 0 days. Thus, the excess risk for patients represents the excess risk after a diabetes mellitus duration of 4.3 years.

To determine the risk association within the group of patients with type 2 diabetes mellitus for AMI and heart failure by BMI and HbA1c, respectively, we performed analyses without comparators from the general population to present the differences within the group of type 2 diabetes mellitus, adjusted for variables not available for comparators. These analyses were adjusted for age, sex, diabetes mellitus duration, education, marital status, income, immigrant status, and additionally risk factors; low‐density lipoprotein (LDL) cholesterol, systolic blood pressure, and smoking. BMI and risk factors were modeled using restricted cubic splines with 4 equally distanced knots. We performed analyses separately for each outcome and risk factor, and additionally modeled an interaction term with HbA1c groups and BMI. We noted that there was no significant interaction between HbA1c and BMI. Since AMI is an established mediator for heart failure, we performed a time updated Cox regression with the outcome of heart failure, where we adjusted for AMI during follow‐up, BMI with an added interaction term between AMI during follow‐up and BMI. Of the 8622 patients who were diagnosed with heart failure, 2210 (25.6%) patients were diagnosed with AMI earlier during follow‐up, or at the same day as the diagnosis of heart failure. In the cases where incident AMI and heart failure occurred at the same day, we added 1 day to the follow‐up time of heart failure. The proportional hazards assumption was checked with Schoenfeldt residuals and there were no significant deviations from the assumption. The analyses were 2‐tailed where a value of 0.05 was considered statistically significant. We used R (ver. 3.2.1; R Foundation for Statistical Programming).

For this study, we identified 181 045 patients with type 2 diabetes mellitus and 902 302 age‐ and sex‐matched population comparators. Mean age for both cases and comparators were 58.3 years, and 50.0% were women. Fewer patients (18.8%) than population comparators (28.8%) had a college/university degree. In terms of coexisting conditions at baseline, patients with type 2 diabetes mellitus had more frequently AF. Among patients with type 2 diabetes mellitus increasing body weight was associated with successively younger age, a higher proportion of women, and a lower proportion with college or university education. Similarly, there were associations between increasing body weight and lower age at onset of diabetes mellitus, shorter diabetes mellitus duration, more albuminuria, higher eGFR, and more treatment with statins and antihypertensives. (Table). Baseline characteristics for the patients with missing BMI at study entry after imputation, are presented in Table S2.

Over a median follow‐up time of 5.7 years, there were a total number of 33 060 cases of AMI and 28 855 hospitalizations for heart failure (Table S3, with ICD‐code presented in Table S1). Age‐adjusted incidence rates, shown in Figure 1 (exact crude‐ and age‐adjusted rates presented in Tables S4 and S5), were higher among patients with type 2 diabetes mellitus for all outcomes and poor glycemic control was associated with increased incidence rates. With respect to AMI, comparators displayed similar age‐adjusted incidence rate of 4.0 to 4.6 cases per 1000 person‐years. Since comparators worked as reference to Cox regression analyses stratified separately by each HbA1c group, the similar incidence rates between the matched comparators strengthened our matching process. Among individuals with diabetes mellitus, those with higher HbA1c had higher incidence rates, but there was no clear association between BMI and risk of AMI. For heart failure, we observed a strong association for both HbA1c and BMI, with an incidence of heart failure for individuals who were both poorly controlled and severely obese of 21.4 (17.9–25.5) per 1000 person‐years, compared with 4.6 (4.2–5.1) per 1000 person‐years among patients with optimal glycemic control and BMI 18.5 to <25 kg/m² and ≈3.0 to 3.8 cases of heart failure per 1000 person‐years among matched‐population comparators.

Patterns for hazard ratios for AMI and heart failure were in line with those noted for incidence rates (Figure 2). Glycemic control beyond target level of 52 mmol/mole was associated with increased risks of both outcomes. There were no clear associations between higher BMI and risk of AMI at any level of glycemic control. In contrast, there was a distinct almost linear increase in excess risk of heart failure with increasing BMI at all levels of glycemic control which was substantial among the very obese compared with comparators. In sensitivity analyses using heart failure registered as the principal diagnosis (about half of all heart failure cases; Figure S2) findings with respect to excess risk for BMI and glycemic control among patients with type 2 diabetes mellitus were similar to those for heart failure in any diagnostic position. Separate analyses in men and women displayed similar associations (Figure S3 and S4) as the main analyses. Women, compared with men, had somewhat higher excess risks for AMI, however, with respect to heart failure men and women displayed more or less similar excess risks.

Among analyses restricted to patients with type 2 diabetes mellitus (Figure 3), additionally adjusted for LDL cholesterol, systolic blood pressure, smoking status, and BMI/HbA1c (depending on the exposure), which was not possible to adjust for in analyses comparing with the population comparators, BMI was a stronger independent risk factor for heart failure than HbA1c, whereas for AMI, the risk increased linearly by HbA1c, but associations between BMI and risk were essentially flat when compared with the reference level of 25 kg/m². The analyses modeled as an interaction between HbA1c and BMI (Figure S5) showed similar results as analyses versus comparators, with higher estimates among the poorly controlled patients and a strong link of heart failure to increasing BMI. Our sensitivity analysis (Figure S6) suggested largely the same risk trajectory for the risk of heart failure as in the main analysis (Figure 3), regardless of incident AMI.