Title of article:

Body mass index compared to dual-energy x-ray absorptiometry: evidence for a spectrum bias

Authors: Curtin F, Morabia A, Pichard C, Slosman DO.
Journal: J Clin Epidemiol, Jul 1997;50(7):837-43

Abstract

The objective of this study was to evaluate the accuracy of body mass index (BMI) in the diagnosis of obesity. The relationship of this weight-for-height index to body composition was determined using dual-energy x-ray absorptiometry (DEXA) in 226 Caucasian subjects. BMI-obesity was defined as a BMI greater than 27.8 kg/m2 in males and greater than 27.3 kg/m2 in females. BMI-obesity was weakly sensitive (males: 12.5%, females 13.6%) compared with diagnosis by DEXA, defined as percent body fat mass (%BF) greater than 20% in males and greater than 25% in females. Conversely, the specificity of BMI-obesity was high (males and females: 100%). When analyzing subgroups of subjects according to weight, sensitivity was higher among heavier subjects than among lighter ones. In both sexes and in all subgroups, the specificity was 100%. The clinical implication of this spectrum bias is that, in men or women weighing less than 80 kg, measures of obesity other than BMI, such as bioelectrical impedance, should be preferred.

Comments and Key points

A study of Switzerland hospital employees, of mean age 38.1, had male mean BMI of 21.7 and female mean BMI of 21.6. That's an extraordinarily weird , too low, non-representative sample. You should ignore their abstract.

They defined obesity, based on DEXA-measured body fat percentages, as 20% for men and 25% for women. That's ridiculous, lower thresholds than anyone else uses.

They showed a graph of their ROC curve, so we can look at the ROC curve (below) and choose better thresholds ourselves.

The ROC curve shows the trade-offs between Sensitivity and Specificity. This article's Authors believed that a BMI of 20.5 was the optimum threshold to define obesity, with a Sensitivity of 84% and Specificity of 60%. Can you believe it? A BMI of 20.5 to define obesity? What were they thinking?

That's nonsense, not only because their body fat percentage thresholds are too low, but also because choosing higher sensitivity over specificity, is not wise... because it will upset too many normal people who are mis-labelled as obese.

The authors did mention, that if they had used higher body fat percentage cutoffs, 25% in men and 35% in women, then their optimum choice Body Mass Index cutoff would be 24 kg/m². Unfortunately, the article didn't show the actual data or ROC curve for these better body fat thresholds. That's more like it. But still, the problem of their bias still exists. They presumably chose higher sensitivity over specificity again, so don't use 24 kg/m².

To remedy this, look at the graph above. There is a 3 kg/m² BMI difference between the pink and blue thresholds. Therefore, I think you should add 3 onto their value of 24 kg/m², to give my recommendation of 27kg/m² as the suitable BMI cutoff to define overweightness, for 38 year olds, adding 1 for men, subtracting 1 for women.

 

Review & comments by Steven B. Halls, MD, Last modified 23-June, 2008, Copyright
Back to scientific BMI articles index.