Background: Body mass index (BMI) is the most widely used measure to diagnose obesity. However, the accuracy of BMI in detecting excess body adiposity in the adult general population is largely unknown.
Methods: A cross-sectional design of 13 601 subjects (age 20-79.9 years; 49% men) from the Third National Health and Nutrition Examination Survey. Bioelectrical impedance analysis was used to estimate body fat percent (BF%). We assessed the diagnostic performance of BMI using the World Health Organization reference standard for obesity of BF%>25% in men and>35% in women. We tested the correlation between BMI and both BF% and lean mass by sex and age groups adjusted for race.
Results: BMI-defined obesity (> or =30 kg m(-2)) was present in 19.1% of men and 24.7% of women, while BF%-defined obesity was present in 43.9% of men and 52.3% of women. A BMI> or =30 had a high specificity (men=95%, 95% confidence interval (CI), 94-96 and women=99%, 95% CI, 98-100), but a poor sensitivity (men=36%, 95% CI, 35-37 and women=49%, 95% CI, 48-50) to detect BF%-defined obesity. The diagnostic performance of BMI diminished as age increased. In men, BMI had a better correlation with lean mass than with BF%, while in women BMI correlated better with BF% than with lean mass. However, in the intermediate range of BMI (25-29.9 kg m(-2)), BMI failed to discriminate between BF% and lean mass in both sexes.
Conclusions: The accuracy of BMI in diagnosing obesity is limited, particularly for individuals in the intermediate BMI ranges, in men and in the elderly. A BMI cutoff of> or =30 kg m(-2) has good specificity but misses more than half of people with excess fat. These results may help to explain the unexpected better survival in overweight/mild obese patients.