Association of BMI, Fitness, and Mortality in Patients With Diabetes: Evaluating the Obesity Paradox in the Henry Ford Exercise Testing Project (FIT Project) Cohort

Research Design and Methods

This analysis includes 8,528 individuals with diabetes from the Henry Ford Exercise Testing Project (FIT Project) who performed a clinically indicated, physician-referred Bruce protocol exercise treadmill stress test between 1991 and 2009 at the Henry Ford Health System medical centers in metropolitan Detroit, MI, as has previously been described in detail elsewhere (14). Patients were at least 18 years old and had a diagnosis of diabetes, which was defined by patient self-report, use of a blood glucose–lowering medication, or based on electronic medical record (EMR) diagnosis. We required that an EMR diagnosis of diabetes or any other medical condition (e.g., hypertension or hyperlipidemia) be coded on at least three separate encounters in the EMR in order to be included as a diagnosis in our database. Among patients diagnosed with diabetes, 85% had an HbA_1c ≥6.5% and/or were taking a glucose-lowering medication, 11% had an HbA_1c ≥5.7–6.4% (39–46 mmol/mol), and 4% had an HbA_1c <5.7% (39 mmol/mol). We excluded persons with a BMI <18.5 kg/m² (n = 252) and those with prevalent cancer (n = 604).

Total mortality was the primary outcome and was ascertained through a search of the Social Security Death Index with follow-up through the year 2013. A previously described algorithm using a combination of first name, last name, date of birth, and Social Security number was used to perform matching (14). Follow-up was calculated from date of the exercise test to the date of death or through April 2013.

Bruce protocol treadmill stress testing was performed using standard methodology, and the test was stopped if the patient experienced chest pain, dyspnea, or other exercise-limiting symptoms (e.g., chest pain, dyspnea, or dizziness) as determined by the supervising clinician or if the patient requested that the test be stopped. The test could also have been stopped if the patient had an abnormal blood pressure response, significant ST segment changes, or a clinically significant arrhythmia as defined by the American Heart Association/American College of Cardiology guidelines (15,16). Each patient’s maximal exercise capacity (e.g., fitness) was estimated by calculating their METs, which were calculated by the Quinton treadmill controller (Q-Stress; Quinton Instrument Company, Bothell, WA) using their peak exercise workload (treadmill speed and grade) achieved during the stress test based on equations published by the American College of Sports Medicine (17). We categorized fitness as low (<6 METs), moderate (6–9.9 METs), and high (≥10 METs), as consistent with our prior work (18). The stress test indication was categorized into common indications based on the physician referral information, which primarily included chest pain, dyspnea, and preoperative evaluation.

A trained nurse and/or clinical exercise physiologist recorded the patients’ demographics and CVD risk factors along with current medication use and past medical history immediately preceding the treadmill stress test. Patients reported their race, height, and current smoking status. Weight was measured at the time of the treadmill stress test, and the EMR recorded weight was used for any missing values. BMI was categorized as normal (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), or obese (≥30 kg/m²). A diagnosis of hypertension and hyperlipidemia was based on patient self-report, the use of a disease-specific medication, or a database-verified diagnosis. Patients were classified as having a family history of coronary artery disease if they reported a first-degree relative with a history of a clinical coronary artery disease event. Laboratory values for tests performed within 90 days of the stress test were obtained through a retrospective search of the EMR and associated laboratory databases. Hemoglobin A_1c values were only available for 5,786 individuals (68%), and fasting glucose values were not available. For patients who participated in the Henry Ford Health System integrated health plan, a retrospective search of the EMR, administrative databases, and/or pharmacy claims files was performed to obtain additional data on medication use and past medical history.

We calculated age-adjusted mortality rates per 1,000 person-years’ follow-up stratified by BMI and fitness group. We also performed progressively adjusted Cox proportional hazards modeling to examine the association of BMI and total mortality within each fitness group. Using Cox proportional hazards modeling, we also examined the association of fitness and total mortality within each BMI group. Model 1 included age, sex, and ethnicity. Model 2 additionally adjusted for hypertension, current smoking, hypertension medication use, lipid-lowering medication use, oral glucose-lowering medication use, and a history of CVD. Model 3 additionally included insulin use. We also used Cox proportional hazards modeling within each fitness group to examine whether there were differences in the relationship between BMI and mortality for prespecified subgroups of interest.

We performed adjusted cubic spline modeling with BMI as a continuous variable and a reference value of BMI of 22.5 kg/m² (consistent with prior publications) within each fitness group in order to examine the continuous association of BMI with total mortality (19). We also calculated an adjusted cubic spline figure that displays the relative association for each fitness group using a reference value of BMI of 22.5 kg/m² for the lowest fitness group.

We performed multivariable-adjusted logistic regression modeling (model 3) to evaluate the association between 1) per 1 kg/m² increase in BMI and 2) 1 MET increase in fitness. In addition, we performed interaction testing using BMI and fitness as continuous variables. Statistical analyses were conducted using Stata/SE version 15.1 (Stata Corporation, College Station, TX).