Sensitivity and specificity of the body mass index to assess low percent body fat in African women

The objective of this study was to evaluate the ability of the body mass index (BMI, kg/m(2)) to reflect low percent body fat (%BF) in a population with a rather mild but widespread prevalence of low BMI. A sample of 586 women was studied in the Plateau Koukouya, a rural area of the Republic of Congo, Central Africa. Percent BF was estimated from bioelectrical impedance (BIA). BIA parameters were assumed to reflect lean body mass. The correlation between %BF and BMI was high (r = 0.84; P < 0.001). Low %BF or low BIA parameters were defined as the first quartile of the distribution. Sensitivity, specificity, positive and negative predictive value of BMI <18.5, an accepted international cutoff for thinness, in relation to %BF was 58.5%, 93.6%, 75.4%, and 87.1%, respectively. A continuous sensitivity/specificity analysis (receiver operator characteristic [ROC] curves) for characterizing low %BF or low BIA parameters was done for a large range of BMI values. ROC curve analysis for %BF suggested that an acceptable trade-off between sensitivity (89.8%) and specificity (77.9%) occurred at a BMI of 19.7 kg/m(2). However, the positive predictive value was low (57.6%). For the prediction of low BIA parameters, results were similar, showing moderate sensitivity and high specificity for BMI <18.5, a cutoff point of BMI = 19.6, and low positive predictive values (<48%). The data suggest that BMI was not a good predictor of low %BF. This is consistent with the assumption of a decrease in both fat and fat free body mass in cases of low BMI. (C) 2000 Wiley-Liss, Inc.