Validity of single-site and multi-site models for estimating body composition of women using near-infrared interactance.

The purpose of this study was to develop a multi-site near-infrared (NIR) model (Model I) and compare its predictive accuracy to single-site models (IIA and IIB). In Model I, the sum of two optical density (OD) measures (Σ2OD), age, body weight, height, and physical activity level were used as potential predictors of body density (D ). In Model IIA, the variables used in the manufacturer's NIR equation (biceps OD and OD , body weight, height, gender, and physical activity level) were the potential predictors. This model was modified by including age as an additional potential predictor in Model IIB. We also examined the test-retest reliability and interrelationships of OD measures taken at 10 anatomical sites, as well as the validity of the manufacturer's NIR equation, for estimating body composition of women. The subjects, 148 women between 20 and 72 years, were hydrostatically weighed to determine criterion D . The Futrex-5000 was used to measure OD and OD at 10 anatomical sites. Only two sites (pectoral OD and biceps OD ) contributed significantly to the variance in D . Thus, the sum of these two ODs (Σ2OD), was used as a potential predictor in the multi-site model. Test-retest reliability was high, with intraclass correlation coefficients ≥0.85 for many of the OD measurements. Intercorrelations of ODs ranged from 0.22 to 0.91. In the multi-site model (I), ΣOD, body weight, age, and height were significant predictors, accounting for 85.7% of the variance in D . The SEE was 0.0076 g/ml or 3.3% BF. In the manufacturer's model (IIA), biceps OD , body weight, and height accounted for 76.3% of the variance in D , and the SEE was 0.0094 g/ml (4.1% BF). When age was included as a predictor (Model IIB), the R increased (86.0%) and the SEE (0.0073 g/ml or 3.1% BF) decreased substantially. Cross-validation of the three equations yielded r s ranging between 0.688 (Model IIA) and 0.748 (Model I) and slightly larger SEEs (0.0094-0.001048 g/ml). There were no significant differences between average criterion D and predicted D for each equation. The manufacturer's equation programmed in the Futrex-5000 yielded a lower r (0.55), higher SEE (5.61% BF), and significantly underestimated criterion % BF by an average of 3% BF. Either the multi-site (model I) or single-site (Model IIB) equations is recommended to estimate body composition of this population. © 1992 Wiley-Liss, Inc.