Title of article:
Relationships between the Body Mass Index and body composition.
Authors: Wellens RI, Roche AF, Khamis HJ, Jackson AS,
Pollock ML, Siervogel RM.
Journal: Obes Res, Jan 1996;4(1):35-44
Abstract
The aims of this study were to evaluate the Body Mass Index (BMI) (weight/stature2) as a proxy for percent body fat (%BF) and to determine its association with fat-free mass (FFM). Multivariate analysis of variance and partial correlations were used to examine relationships between BMI and %BF and FFM from densitometry for 504 men and 511 women, aged 20 to 45 years. Sensitivity/specificity analyses used cut offs of 28 kg/m2 in men and 26 kg/m2 in women for BMI, and 25% in men and 33% in women for %BF. Significantly higher associations existed in each gender between BMI and %BF in the upper BMI tertile than in the lower BMI tertiles. In the lower BMI tertiles, correlations between BMI and FFM were approximately twice as large as those between BMI and %BF. The BMI correctly identified about 44% of obese men, and 52% of obese women when obesity was determined from %BF. BMI is an uncertain diagnostic index of obesity. Results of Receiver Operator Characteristic (ROC) analyses using %BF and total body fat, both provided a BMI of 25 kg/m2 in men and 23 kg/m2 in women as diagnostic screening cut offs for obesity.
Comments and Key points
The authors made an assumption that I disagree with. They set their BMI cut-off values to maximize sensitivity for obesity, by saying:
"sensitivity is given more importance than specificity since a false positive is not considered as serious as a false negative."
Paraphrased, this means they think it is more acceptable to call some "normal" people as "overweight", rather than risk under-calling "overweight" people as "normal". This is opposite to how most people think. Most overweight people don't mind being called normal, but normal people are angered if wrongly labelled as overweight.
They concluded that a BMI cutoff of 25.0 for men and 23.0 for women is optimal, to define "obesity" as a body fat percentage over 25% in men or over 33% in women. (Note, these body fat percentages are commonly used arbitrary values.)
I think their terminology is confusing because most people would say these body fat percentages are "overweight" (not "obese").
This study's data sample is weighted towards young adults, with average age of approximately 31 years.
I have gleaned from this article's ROC curves, the following possible cut-off values. In my opinion, a Body Mass Index threshold should have about 90% specificity and at least 60% sensitivity, to gain popular acceptance.
| Cutoff BMI |
Men | Women | ||
|---|---|---|---|---|
| Specificity | Sensitivity | Specificity | Sensitivity | |
| 23 | 84.4% | 81.5% | ||
| 24 | 89.5% | 77% | ||
| 25 | 68.8% | 78.6% | 92% | 72% |
| 26 | 81% | 68% | 97.8% | 51.8% |
| 27 | 88% | 57% | ||
| 28 | 93.5% | 43.6% | ||
This criteria is satisfied for women at a BMI of 25.0. For men, there is no perfect answer, but I would choose 27.0 for men.
Review & comments by Steven B. Halls, MD, Last edited 23-June, 2008, Copyright.
Return to Body Mass Index articles index.
