BMI–Mortality Paradox and Fitness in African American and Caucasian Men With Type 2 Diabetes

Participants

Between 1986 and 2010, symptom-limited exercise tolerance tests (ETT) were performed on more than 20,000 veterans at the Veterans Affairs Medical Centers (VAMCs) in Washington, DC, and Palo Alto, California, as part of routine evaluations or to assess exercise-induced ischemia. This information, along with the individual’s medical history, was electronically stored. From this cohort, we identified those with type 2 DM, defined by ICD coding. We excluded women and those with any of the following: 1) history of an implanted pacemaker, 2) left bundle branch block, 3) unable to complete the test, 4) BMI <18.5 kg/m², 5) impaired chronotropic response; and 6) those with AIDS or HIV. After these exclusions, the participant group (n = 4,156) consisted of 2,013 African Americans, mean age (years ± SD) 59.6 ± 10; 2,000 Caucasians, 60.8 ± 10; and 143 unknown races, 58.4 ± 11. The institutional review board at each institution approved the study, and all subjects gave written informed consent before undergoing ETT.

All demographic, clinical, and medication information was obtained from the subject’s computerized medical records just before their ETT and verified by each individual. A standardized scale was used to assess body weight (kg) and height (m), and BMI was calculated as kg/m². Demographic data are included in Table 1.

Dates of death were verified from the VA Beneficiary Identification and Record Locator System File. This system is used to determine benefits to survivors of veterans and has been shown to be 95% complete and accurate (19). Vital status was determined as of 28 December 2010.

Exercise assessments

Exercise capacity was assessed by a standard treadmill test using the Bruce protocol at the VAMC, Washington, DC, and by an individualized ramp protocol, as described elsewhere (20), for men assessed at the VAMC, Palo Alto, California. Peak exercise capacity in metabolic equivalents (METs) was estimated using standardized equations based on peak speed and grade for the ramp protocol (20) and on peak exercise time for the Bruce protocol (21). Subjects were encouraged to exercise until volitional fatigue in the absence of symptoms or other indications for stopping (22). Medications were not altered before testing.

Determination of BMI and fitness categories

Individuals were classified according to four predetermined BMI categories adopted by the American Heart Association as indicators of adiposity (23): BMI 18.5–24.9 (normal weight; n = 668); 25–29.9 (overweight; n = 1,610); 30–34.9 (obese I; n = 1,160) and ≥35 kg/m² (obese II and III; n = 718).

We also formed four fitness categories based on the MET level achieved. Those who achieved a peak MET level ≤5METs (lowest 25th percentile of the cohort) comprised the lowest fitness category (Least-Fit; n = 1,162); those with a peak MET level of 5.1–7.0 (26th to 50th percentile) comprised the moderate fit category (Moderate-Fit; n = 1,163); those with a peak MET level of 7.1–8.7 (51st to 75th percentile) comprised the fit category (Fit; n = 995), and those with a peak MET level >8.7 (>75th percentile), comprised the highest fit category (High-Fit; n = 836).

To further explore the effect of fitness on mortality risk, we formed two groups within each fitness category (a total of eight) based on the presence (fitness category/risk factors) or absence of additional risk factors (fitness category/no additional risk factors).

Statistical analysis

Follow-up time is presented as mean ± SD and median years. Mortality rate was calculated as the ratio of events by the person-years of observation. Continuous variables are presented as mean (SD) values and categoric variables as relative frequencies (%). Baseline associations between categoric variables were tested using χ² analysis. One-way ANOVA was applied to evaluate mean differences between age and BMI among fitness categories. Post hoc procedures (Bonferroni) were performed for multiple comparisons. Equality of variances between groups was tested by Levene’s test.

The hazard ratio (HR) for all-cause mortality was calculated for each BMI and fitness category using Cox proportional hazard models. For BMI categories, individuals with a BMI ≥35 kg/m² (lowest mortality group) comprised the reference group. For fitness categories, individuals with an exercise capacity ≤5 MET were considered the Least-Fit and comprised the reference group. The analyses were adjusted for age, history of cardiovascular disease (CVD), cardiovascular medications (ACE-I, angiotensin receptor blockers [ARBs], β-blockers, calcium channel blockers, diuretics, vasodilators, statins), race, and risk factors (hypertension, dyslipidemia, smoking) and (when appropriate) METs and BMI. Cox proportional hazard models were also used to assess risk among the eight fitness and risk-factor categories. These analyses were adjusted for age, CVD, and cardiovascular medications, as above, and the Least-Fit and no additional risk factors category comprised the reference group. The assumption of proportionality for all Cox proportional hazard analyses was graphically tested and fulfilled the criteria. Values of P < 0.05 using two-sided tests were considered statistically significant. All statistical analyses were performed using SPSS 19.0 software (SPSS Inc., Chicago, IL).

Predictors of all-cause mortality across fitness and BMI categories

Multivariate Cox proportional hazards analysis revealed that age (HR 1.04 [95% CI 1.03–1.05]); hypertension (1.42 [1.2–1.6]), smoking (1.34 [1.1–1.5]), BMI (0.96 [0.95–0.97]), and exercise capacity were strong predictors of mortality. The adjusted mortality was reduced by 12% for every 1-MET increase (0.88 [0.85–0.90]) for the entire cohort. Within each BMI category, the mortality risks per 1-MET increases were 11% lower (0.89 [0.85–0.94]) for those with BMI of 18.5 to 24.9 kg/m², and 14% lower (0.86 [0.83–0.90]) for individuals with a BMI of 25 to 29.9 kg/m². The mortality risks were 18% lower per MET for the two highest BMI groups (BMI 30–34.9 [0.82 (0.76–0.88)] and BMI ≥35 kg/m² [0.82 (0.74–0.91)]).

Mortality risk for the entire cohort was significantly higher (39.7%) among those in the lowest BMI group (BMI <25 kg/m²) and declined progressively to 27.3%, 21.6%, and 16.6% for the next three BMI groups, respectively (P < 0.001 for all comparisons). Similar patterns were observed in African Americans and Caucasians. However, mortality in African Americans in the lowest BMI category (BMI <25 kg/m²) was substantially higher than in Caucasians (45.6% vs. 35.6%, respectively; P = 0.01). The risk for individuals of unknown race was not considered because of the small number of participants (n = 141).

Mortality risk according to BMI categories

The fully adjusted Cox proportional hazards model for the entire cohort revealed a paradoxic association between BMI categories and mortality, with significantly higher risk among those with a BMI between 18.5 and 24.9 (HR 1.70 [95% CI 1.36–2.1]) compared with those with BMI ≥35 kg/m². No significant increase in risk was noted for the remaining BMI categories (Fig. 1). When mortality risk was assessed for African Americans and Caucasians separately, the HR was 1.95 (1.44–2.63) for African Americans in the BMI group 18.5–24.9 kg/m² and 1.53 (1.0–2.1) for Caucasians, with no significant change in mortality risk for the remaining BMI categories (Fig. 1). To partially account for reverse-causation, we excluded those with an exercise capacity ≤5 METs who died within the initial 2 years of follow-up and repeated the analysis. The trend was similar, with a mortality risk of 1.67 (95% CI 1.32–2.1) for the entire cohort, 1.83 (1.34–2.51) for African Americans, and 1.49 (1.0–2.1) for Caucasians in the BMI group 18.5–24.9 kg/m² (data not shown).

Mortality risk according to fitness categories

HRs across fitness categories for the entire cohort and each BMI category are presented in Table 2. The adjusted mortality risks for the four fitness categories are shown, with those in the Least-Fit category (≤5 MET) as the reference group. For the entire cohort in the fully adjusted model, the mortality risk was progressively lower as exercise capacity increased, such that the moderately fit and most fit categories had ∼50% reductions in risk (P < 0.001 for trend). Similarly, the adjusted risks in each BMI category were progressively lower with increased exercise capacity, ranging between 23 and 67% (Table 2). We observed similar trends when we repeated the analyses for African Americans and Caucasians separately, for those with and without CVD, and when we excluded those with low exercise capacity (≤5 METs) who died within the initial 2 years of follow-up (Table 2).

We then assessed possible differences in mortality risk between individuals with additional risk factors compared with those without, for each fitness category. We used the Least-Fit category and no risk factors (Least-Fit/no additional risk factors) as the reference group, adjusting for age, BMI, race, CVD, and cardiac medications. The findings, illustrated in Fig. 2, revealed the following: 1) the mortality risk was significantly higher for those in the Least-Fit category with one or more risk factors (HR 1.50 [95% CI 1.17–1.93]; P = 0.001); 2) there was no difference in risk for individuals in the next fitness category (Moderate-Fit) with risk factors (0.98 [0.75–1.28]; P = 0.88), but was 34% lower in those in this category with no risk factors (0.66 [0.47–0.87]; P = 0.04); and 3) in the two highest fit categories, the risk was similar for those with additional risk factors but substantially lower for individuals in the highest fit category (>8.8 METs) and no additional risk factors (0.37 [0.23–0.60]).

Limitations

The current study was retrospective and, therefore has several limitations. First, only male veterans were included, which limits the ability to generalize the findings to women. There is also the possibility of a beneficial “veteran effect” (6,34). Briefly, veterans must fulfill weight criteria at the time of enlistment and maintain weight during their service, resulting in the exclusion of obese individuals. Therefore, late onset of obesity in our sample may be lessening the duration of exposure to obesity-related risk (35,36). Second, adiposity was determined only by BMI, which does not provide information on body fat distribution. Third, information regarding BMI and fitness levels was available only at baseline, negating evaluation of the effects of changes during the follow-up period in either or both of these factors. Fourth, we did not have enough patients who were extremely lean (BMI <18.5 kg/m²) to draw conclusions on mortality rates in these subjects. Finally, the potential effects of reverse causality cannot be excluded.

Conclusion

Similar to previous studies in Veterans, we observed a paradoxic BMI–mortality risk association in African American and Caucasian patients with diabetes, with heavier subjects exhibiting better survival. This association was more pronounced in African American than in Caucasian individuals. The current study also confirms the strong relationship between exercise capacity and mortality. The exercise capacity–mortality risk association was inverse, independent, and graded for the entire cohort and in all BMI categories regardless of race. The presence of additional cardiac risk factors combined with low exercise capacity increased mortality risk significantly. Conversely, mortality risk declined progressively with increased fitness, regardless of the presence or absence of additional cardiac risk factors. Our findings support the concept that even small improvements in exercise capacity, regardless of BMI level, can counteract the increased mortality risk associated with diabetes.