Performance of Body Adiposity Index and Relative Fat Mass in Predicting Bioelectric Impedance Analysis-Derived Body Fat Percentage: A Cross-Sectional Study among Patients with Type 2 Diabetes in the Ho Municipality, Ghana

Abstract

Objective: The study sought to determine the diagnostic accuracy of body adiposity index (BAI) and relative fat mass (RFM) to predict BIA-derived BFP among patients with type 2 diabetes in the Ho municipality. Materials and Method. This hospital-based cross-sectional study involved 236 patients with type 2 diabetes. Demographic data, including age and gender were obtained. Height, waist circumference (WC), and hip circumference (HC) were measured using standard methods. BFP was estimated on a bioelectrical impedance analysis (BIA) scale. The validity of BAI and RFM as alternative estimates for BIA-derived BFP was evaluated based on mean absolute percentage error (MAPE), Passing-Bablok regression, Bland-Altman plots, receiver-operating characteristic curve (ROC), and kappa statistics analyses. A p value less than 0.05 was considered statistically significant.

Results: BAI showed systematic bias in estimating BIA-derived BFP in both genders, but this was not evident between RFM and BFP among females (t = -0.62; p = 0.534). While BAI showed "good" predictive accuracy in both genders, RFM exhibited "high" predictive accuracy for BFP (MAPE: 7.13%; 95% CI: 6.27-8.78) among females according to MAPE analysis. From the Bland-Altman plot analysis, the mean difference between RFM and BFP was acceptable among females [0.3 (95% LOA: -10.9 to 11.5)], but both BAI and RFM recorded large limits of agreement and low Lin's concordance correlation coefficient with BFP (Pc < 0.90) in the two gender populations. The optimal cut-off, sensitivity, specificity, and Youden index for RFM were >27.2, 75%, 93.75%, and 0.69, respectively, while those of BAI were >25.65, 80%, 84.37%, and 0.64, respectively, among males. Among females, the values for RFM were >27.26, 92.57%, 72.73%, and 0.65, whereas those of BAI were >29.4, 90.74%, 70.83%, and 0.62, respectively. The accuracy of discriminating between BFP levels was higher among females [BAI (AUC: 0.93) and RFM (AUC: 0.90)] compared to males [BAI (AUC: 0.86) and RFM (AUC: 0.88)].

Conclusion: RFM had a better predictive accuracy of BIA-derived BFP in females. However, both RFM and BAI failed as valid estimates for BFP. Furthermore, gender-specific performance in the discrimination of BFP levels for RFM and BAI was observed.

Figure 1

Relationship of BAI and RFM with BFP (via BIA) stratified by gender presented using Passing-Bablok regression plot. k: intercept, m: slope, r: Pearson correlation coefficient, Pc: Lin's concorrdance correlation coefficient, %BF: percentage body fat using BIA (bioelectrical impedance analysis). Pc < 0.90 rated poor agreement, Pc 0.90 - 0.95 moderate agreement, Pc 0.95-0.99 substantial agreement, and Pc > 0.99 excellent agreement.

Figure 2

Bland-Altman's plot of mean difference between BIA determined BFP and BAI or RFM for males and females.

Figure 3

Receiver-operating characteristic (ROC) curves of BAI and RFM in diagnosing obesity using BIA determined body fat percentage as reference classification. Marked point corresponds to Youden's index.