The results of our study, involving a large sample from the US population, demonstrates that BMI has a limited diagnostic performance to correctly identify individuals with excess in body fatness, particularly for those with BMI between 25 to 30 kg/m2, for men and for the elderly. Body mass index has good general correlation with BF %, but it fails to discriminate between BF % and lean mass. In addition, the sensitivity of BMI ≥ 30 kg/m2 to diagnose obesity is relatively low, missing more than half of people with BF %defined obesity, while the specificity and positive predictive value are good. Furthermore, for a given BMI value there is significant inter-subject variability in BF %.
Previous studies have also shown a good overall correlation between BMI and BF % and significant variability at individual level. Studies testing the diagnostic performance of BMI 26,27
, including our report in a group of patients with coronary artery disease17
, have shown that BMI has a good specificity but a low sensitivity to diagnose obesity. This limitation of BMI has been also reported in pediatric populations. Studies in adolescents have shown that BMI and body fat content have a good correlation only in the highest percentiles of BMI, while in lower percentiles the correlation can be considered limited 28, 29
. Regretfully, these previous studies have been performed in a small sample of subjects from selected populations and are limited to specific age groups. Our present study is the first report to describe the diagnostic performance of BMI and its correlation with BF % and lean mass for men and women and across different age groups tested in a large multiethnic sample of the US population.
From our findings it is apparent that the diagnostic performance of BMI in intermediate ranges of body weight is limited mainly because of the inability of BMI to discriminate between BF % and lean mass, understandable since the majority of human body weight (numerator of the BMI) comes from lean mass. Indeed, our analyses found that BMI correlated in similar fashion with lean mass as it did with body fat. In fact, in men BMI correlated significantly better with lean mass than with body fat. In contrast, in women BMI appears to perform better than men, which may explain why BMI-defined overweight in women has been more consistently related to increased mortality than in men in previous studies 30–31
Based on the overwhelming evidence of the deleterious effects of adipose tissue on body systems and organs, it would be expected that the association between body weight (indexed to height) and outcomes would be linear. To the contrary, most studies testing the effects of body weight on survival have generally shown a U or J-shape survival curve, or at best, they have shown a horizontal survival line for BMI values in the overweight BMI ranges (25–27 kg/m2
) followed by an upward trend in risk11–13
at higher levels of BMI. In fact, the U-shape association between BMI and mortality has been previously reported in the NHANES III population 13
. In our results, people with normal and mildly elevated BMI clearly had a mixture of different combinations of adipose tissue and lean mass which are likely to explain the inconsistent relationship between BMI and adverse events. Moreover, in the elderly, where most of the mortality occurs in survival studies, BMI had its worst diagnostic performance. This poor accuracy of BMI in the elderly can also explain the inconsistency in the association between BMI and survival.
Because BMI is calculated using total body mass, it contains two factors that have opposite biological effects, namely adipose tissue and lean mass. While adipose tissue has been associated with deleterious health outcomes, preserved lean mass is positively associated with physical fitness, higher caloric expenditure and exercise capacity, all of which are associated with a better survival32–34
. A scenario to exemplify this would be a person with a BMI of 25 with preserved lean mass and mildly increased fat content, compared to another person with the same BMI of 25 with limited lean mass and a high body fat content, both representing completely different levels of exposure to the deleterious effects of adipose tissue, a fact that limits the BMI ability to predict long-term health outcomes.
Our findings also suggest that the magnitude of the obesity epidemic may be greatly underestimated by the use of BMI as the marker of obesity 35
. In our results, BMI showed an unacceptable low sensitivity for detecting body fatness, with more than half of obese subjects (by body fat measurement) being labeled as normal or overweight by BMI. The true prevalence of obesity might be strikingly higher than that estimated by BMI. Unfortunately, the adjustment of BMI cut-offs for obesity does not overcome the limitations of using BMI as a marker of obesity. Decreasing the BMI cut-off for obesity to ≥ 25 kg/m2
for instance, will still result in misclassifying as obese 38% of men and 16% of women.
The implications of mislabeling patients are not trivial. By using BMI as a marker of obesity, we misclassify ≥ 50% of patients with excess body fat as being normal or just overweight and we miss the opportunity to intervene and reduce health risk in such individuals. Conversely, BMI may lead to misclassification of persons with normal levels of fat as being overweight, a fact that could cause unnecessary distress and prompt to unnecessary and costly interventions. In addition, such mislabeling has a deleterious effect on public trust for healthcare providers, particularly from fit patients with evident preserved muscle mass.
While our study of BMI illustrates the significant limitations in using BMI for the diagnosis of obesity, it is important to point out that the use of BMI is not without value. A BMI ≥ 30 kg/m2 has an excellent specificity and positive predictive value for diagnosing obesity in both sexes. This particular finding could explain why the risk for total and cardiovascular mortality usually peaks up when BMI is ≥ 30 kg/m2 and suggests that suboptimal diagnostic performance of BMI to detect excess fat could be limited only to intermediate ranges of BMI. Body mass index should continue to be used in clinical practice to identify those at the two extremes of the body weight spectrum, those with a BMI ≥ 30 kg/m2 who most likely have an excess in body fat, and those with a BMI < 20 kg/m2. Furthermore, BMI or plain body weight might still be the best way to evaluate changes in body fatness over time because increments on body weight or BMI most likely represents fat gain, with the exception of body builders, athletes or patients with conditions that increase the volume of third space such as heart failure, ascitis or renal failure. However, we do challenge the use of BMI to detect excess in body fat for those individuals with intermediate levels of BMI, where it fails to distinguish between excess in body fat or preserved lean mass.
The main limitation of our study is the use of bioelectrical impedance as the method of assessing body fatness. Other more accurate methods to estimated BF %, such as hydrostatic weighting, energy-dual X-ray absorptiometry and air displacement plethysmography would be preferred 36
. Nevertheless, the accuracy of bioelectrical impedance is acceptable, and its ease use, lack of radiation and relatively low cost suggest it is a feasible alternative for measuring body fatness, particularly in large populations.
In conclusion, despite the good correlation between BMI and BF % in a large sample of adults in the US population, the diagnostic accuracy of BMI to diagnose obesity is limited, particularly for individuals in the intermediate BMI ranges. Direct but simple measures of body fatness and measures of body fat distribution may be helpful in such individuals to further stratify them according to their level of body fatness. Future studies are necessary to determine if body composition measurements predict obesity-related risk better than does BMI, waist circumference, waist-to-hip ratio or other measures of body fat distribution.