In multivariate analyses, both fitness and BMI were independently and inversely associated with mortality risk. To ascertain whether low BMI resulted from undetected illness at baseline, we excluded all patients who died during the first 2 years of follow-up and current smokers, but this did not substantially change the primary findings. In joint analyses, elevated BMI generally reduced mortality risk within each fitness category, and higher levels of fitness decreased mortality risk within each BMI category. Highly fit overweight men (n=916) had the lowest mortality risk of any fitness-BMI combination and were 57% less likely to die (HR, 0.43 [95% CI, 0.32-0.59]) as highly fit normal-weight men. Fitness altered the obesity paradox in that overweight and obese men with low fitness were less likely to survive than normal-weight men with high fitness. However, an obesity paradox persisted within fitness group strata.
Our findings that BMI and fitness are inversely associated with all-cause mortality are consistent with earlier results from the VETS27
and extend them to joint analyses of fitness and BMI in a larger cohort of middle-aged men. Explanations for better survival with higher BMI in the current study include the following: (1) reverse causation in clinically referred patients,33
(2) increased coronary artery size,34
(3) the “veteran effect,”27
(4) healthy obesity,35
and (5) the survival effect.36
Several previous studies have reported an obesity paradox in specific patient populations.4-10
First, in patient populations similar to ours, Galal et al33
and Johnson et al37
found significantly lower mortality in overweight and obese patients with known or suspected coronary artery disease compared with patients with normal BMI; this finding was consistent with our results.
Second, greater coronary artery size among patients with higher BMI has been proposed as a possible mechanism for the obesity paradox.34
This may also be a factor in the favorable survival outcomes we observed among overweight and obese men in the current study of clinically referred patients, one third of whom had documented CVD.
Third, a related issue that might further explain our counterintuitive findings is the veteran effect, which we have previously noted.27
Veterans differ from other populations in that they all must meet selection criteria at the time of enlistment. These criteria include minimum height requirements and maximum weight requirements, which must be maintained for the duration of military service. Specifically, maximum allowable weights for different branches of the service correspond to BMI of 25.9 to 29.9 for men.38
Hence, obesity, when present in our population, must have developed after discharge from the service in later life. Some investigators have suggested that adult-onset obesity is less hazardous than obesity developing in childhood or adolescence.39
In addition, men qualifying for military service may have greater than average muscle mass. This might explain the inverse associations between BMI and mortality found in other self-selected populations with physical attributes similar to veterans, such as longshoremen.11
Fourth, in the larger population of obese adults, some experience good health. This is illustrated in some of the findings from the Framingham Heart Study.1
Participants who had a normal BMI at 40 years of age but developed obesity during 20 years of follow-up had no increased mortality risk. Moreover, in healthy obese mice, preferential storage of triglycerides in adipose tissue and reduced levels in the liver40
may result in improved insulin sensitivity, preventing diabetes and heart disease in such animals. A similar mechanism has been proposed for obese humans,41
and a metabolically benign form of obesity has been recently identified.35
Fifth, our findings may be explained in part by the well-known survival effect,36
which has particular relevance in epidemiological studies of older adults. Although our study patients were middle-aged, those in the upper range of this age group were possibly less susceptible to the negative effects of overweight.
Because objective measures of fitness (maximal exercise testing on a treadmill) are often unavailable, data on the joint effects of BMI and fitness on mortality are sparse. We know of 9 published studies that have specifically assessed the joint effects of fitness (as measured from standard exercise testing) and BMI on mortality.18-26
examined this issue by using data from the Lipid Research Clinics Study. All 7 of the remaining published studies used data from the Aerobics Center Longitudinal Study.20-26
Collectively, these studies demonstrated that fitness was a stronger predictor of mortality than BMI and that higher fitness eliminated the mortality risk of elevated BMI (the fat-but-fit hypothesis). Our results differ from these studies in that both high fitness and higher BMI independently reduced mortality risk. The effect of higher BMI on fitness in our cohort was that it generally reduced mortality risk across fitness categories. When highly fit patients were compared by BMI category, those who were overweight and obese experienced dramatic reductions in mortality risk (HR [95% CI], 0.43 [0.32-0.59] and 0.52 [0.34-0.82], respectively). However, only 916 overweight and 390 obese men registered high fitness at baseline, which represents only 7% and 3% of our cohort, respectively. Further study of individuals having this rare phenotype is needed to confirm our findings.
Our study has several strengths. First, all participants underwent an extensive physical examination, and detailed information on medication use was obtained, providing thorough information on the presence or absence of baseline disease. Second, fitness level was determined objectively by maximal exercise testing. Third, the study had a large sample size of more than 12,000 men and an average follow-up of nearly 8 years. Fourth, when smokers and patients who died during the first 2 years of follow-up were excluded, the results were not meaningfully altered.
Our study also has limitations. First, because waist circumference measures were not obtained, we were unable to evaluate body fat distribution characteristics. Second, we included only men who had prior military service and were referred for exercise testing for clinical reasons. Any effort to predict mortality by using fitness, BMI, or clinical or demographic data should be considered population-specific. Although most men were free of CVD, all patients were referred for exercise testing for clinical reasons. Third, fitness is a single measure that is influenced by many factors, including age, heredity, and recent and lifelong activity patterns.32,42
The extent of fitness improvement in adults, or the influence this may have on mortality, cannot be determined from the current investigation. Fourth, we had insufficient information about diet or physical activity patterns to study these factors. Fifth, because we have only baseline data on weight, exercise capacity, and other exposures, we do not know if changes in any of these variables occurred during follow-up and how this might have influenced the results.