Title of article:

Body mass index in screening for adiposity in children and adolescents: systematic evaluation using receiver operating characteristic curves

Authors: Lazarus R, Baur L, Webb K, Blyth F.
Journal: Am J Clin Nutr, Apr 1996;63(4):500-6

Abstract

Body mass index (BMI) has been recommended for use in adolescent screening programs to select subjects with excess body fat for appropriate interventions. No systematic evaluation of MBI in screening for high degrees of adiposity was available when these recommendations were formulated. The purpose of this paper was to evaluated the screening performance of BMI using appropriate epidemiologic methods. Percentage body fat (TBF%) was measured by dual-energy X-ray absorptiometry DXA) in a convenience sample of 230 (119 males, 111 females) health Australian volunteers aged 4-20 y inclusive. Receiver operating characteristic (ROC) curves were prepared for detecting TBF% at or beyond the 85th percentile, using BMI as the screening test. Screening performance was slightly better for girls than for boys, but the differences were not significant. Reasonable true-positive (0.71, 95% CI: 0.53, 0.85) and low false-positive (0.05, 95% CI: 0.02, 0.09) rates were observed at the 85th percentile cutpoint for BMI. At the 95th percentile cutpoint for BMI, both true-positive (0.29, 95% CI: 0.15, 0.47) and false-positive (0.01, 95% CI: 0.00, 0.03) rates were lower. Screening for excess adiposity by using an appropriate percentile cutoff for BMI gives acceptable performance. ROC curves facilitate design of screening programs by allowing an explicit tradeoff between true-positive and false-positive rates. Although large sample sizes are required for precise estimates, the cutoff points that have been recommended appear to offer a reasonable compromise between true-and false-positive rates.

Comments and Key points

Fabulous article. Although it's based in Australia, one of its conclusions is that the CDC overweight criteria for Americans, is also suitable for Australia.

The article supports the CDC criteria for children, using the 85th percentile of BMI for defining "at risk of overweight" and the 95th percentile for defining "obesity". Actually, they did their testing using the NHANES 1 dataset as their 85th and 95th percentile standards.. but as this article shows, the NHANES 1 dataset is extremely close to the current CDC standards.

This fine article provided lots of data tables and charts, showing the ROC curves, the sensitivity and specificity, of choosing various Body-mass index cutoff thresholds. Here is their conclusion:

	Boys		Girls
	Sensitivity	Specificity	Sensitivity	Specificity
85th percentile	72%	95%	85%	93%
95th percentile	33%	98%	46%	100%

The 85th percentile threshold seems to have excellent diagnostic performance. The 95th percentile threshold may be a little too high, and if it were lowered a little, it would gain some sensitivity without losing too much specificity.

I share their beliefs about choosing the correct balance between sensitivity and specificity:

"Being labeled as overweight may have a significant psychological effect on the child concerned, so clinicans involved in screening may prefer to sacrifice some sensitivity to achieve a low false-positive rate. This compromise will result in some overweight individuals being missed by the screening program, but this seems reasonable in the absence of good longitudinal data on which to base any estimates of the long-term health consequences of excess adopisity and inview of the limited options for effective intervention".

I believe the same principle also applies to adults, that specificity should be high (above 90%), even if means lowering sensitivity somewhat.

 

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