M Neovius1, F Rasmussen. 1. Department of Public Health Sciences, Karolinska Institutet, Karolinska University Hospital (Norrbacka), Stockholm, Sweden.
Abstract
OBJECTIVE: To determine the impact of commonly used body fat percentage (%BF) references when evaluating the sensitivity, specificity and misclassifications of body mass index (BMI) for obesity screening in adolescence. SUBJECTS/ METHODS: A community-based sample of 3334 adolescents aged 15.2+/-0.6 years was studied. Weight, height and %BF were measured. Sensitivity, specificity, predictive values and type and extent of misclassifications were calculated. True-positive subjects for overweight and obesity were defined by recently published paediatric %BF cutoffs, 85th and 95th percentiles from the present study population, and an older %BF reference. RESULTS: Regardless of %BF reference used, specificities were high (>0.98 for obesity and >0.90 for overweight) in both sexes, with values for overweight consistently higher for girls (P<0.05). Sensitivities were much lower, with boys ranging from 0.52 to 0.74 for overweight, and 0.24 to 0.43 for obesity. Corresponding sensitivities for girls were 0.42-0.66 and 0.22-0.46. Large discrepancies were seen in total number of misclassifications, with published references resulting in twice as many misclassified for overweight (620, 514 vs 362) when compared to the cutoffs from the study population. For obesity, the difference was even greater (367 vs 133). CONCLUSIONS: No matter which %BF reference used, specificity of BMI was high, and sensitivity low for both sexes. Population percentiles resulted in somewhat lower specificity but much higher sensitivity, and many more correctly classified compared to when cutoffs from the literature were used. The choice of %BF reference in evaluations of BMI-based classification systems has a large impact on the outcome. Sensitivity analyses are therefore warranted.
OBJECTIVE: To determine the impact of commonly used body fat percentage (%BF) references when evaluating the sensitivity, specificity and misclassifications of body mass index (BMI) for obesity screening in adolescence. SUBJECTS/ METHODS: A community-based sample of 3334 adolescents aged 15.2+/-0.6 years was studied. Weight, height and %BF were measured. Sensitivity, specificity, predictive values and type and extent of misclassifications were calculated. True-positive subjects for overweight and obesity were defined by recently published paediatric %BF cutoffs, 85th and 95th percentiles from the present study population, and an older %BF reference. RESULTS: Regardless of %BF reference used, specificities were high (>0.98 for obesity and >0.90 for overweight) in both sexes, with values for overweight consistently higher for girls (P<0.05). Sensitivities were much lower, with boys ranging from 0.52 to 0.74 for overweight, and 0.24 to 0.43 for obesity. Corresponding sensitivities for girls were 0.42-0.66 and 0.22-0.46. Large discrepancies were seen in total number of misclassifications, with published references resulting in twice as many misclassified for overweight (620, 514 vs 362) when compared to the cutoffs from the study population. For obesity, the difference was even greater (367 vs 133). CONCLUSIONS: No matter which %BF reference used, specificity of BMI was high, and sensitivity low for both sexes. Population percentiles resulted in somewhat lower specificity but much higher sensitivity, and many more correctly classified compared to when cutoffs from the literature were used. The choice of %BF reference in evaluations of BMI-based classification systems has a large impact on the outcome. Sensitivity analyses are therefore warranted.