Goran M I, Allison D B, Poehlman E T
Department of Nutrition Sciences, University of Alabama, Birmingham 35294, USA.
Int J Obes Relat Metab Disord. 1995 Sep;19(9):638-43.
Body mass index (BMI), % body fat, and the fat:lean ratio are ratios frequently used as obesity indices. Ratios are based on an assumption that the regression between the numerator (e.g. fat mass) and the denominator (e.g. body mass) has a zero-intercept. As shown in the companion paper, non-zero intercepts cause several problems when ratios are used to adjust data and analysis of covariance (ANCOVA) is frequently the preferred statistical tool. The purpose of this paper is to examine whether BMI, % body fat and the fat:lean ratio meet the necessary criteria for suitable obesity indices using gender comparisons as an example.
In 720 healthy men and women, BMI was higher in men (24.9 +/- 3.3 vs 23.4 +/- 3.2 kg/m2, P < 0.001), but fat mass, % fat and the fat:lean ratio were higher in women (14.8 +/- 8.4 vs 19.1 +/- 8.1 kg fat; 18.6 +/- 8.7 vs 29.9 +/- 9.7% body fat; 0.24 +/- 0.14 vs 0.46 +/- 0.20 for the fat:lean ratio; P < 0.001). Body mass (BM) was correlated with height2 in men (r = 0.40) and women (r = 0.36) with equivalent regression slopes (17.1 +/- 1.9 vs 15.6 +/- 2.3 kg per m2 in women), but the intercepts were different from zero (24.1 kg in men, 20.7 kg in women). When BM was adjusted for height2 using ANCOVA, men remained significantly heavier than women (74.4 +/- 11.0 vs 68.8 +/- 11.6 kg; P < 0.001). Fat mass (FM) was significantly correlated with BM in males (r = 0.64) and females (r = 0.78) but the regression slopes were different (0.49 +/- 0.03 vs 0.71 +/- 0.03 kg of fat per kg body mass in females; P < 0.05) and the intercepts were different from zero (-23.2 +/- 2.2 kg in males; -24.8 +/- 2.1 kg in females). FM adjusted for BM was significantly higher in women (11.7 vs 25.6 kg). FM was inversely correlated with fat free mass (FFM) in males (r = -0.17) and females (r = -0.20), with similar regression slopes (-0.16 +/- 0.05 vs -0.26 +/- 0.08 kg of FM per kg of FFM in women) and the intercepts were significantly different from zero (24.8 +/- 3.0 kg in males; 30.7 +/- 3.6 kg in females). When FM was adjusted for FFM, there was no significant difference between men (16.3 kg) and women (17.0 kg).
It is concluded that: (a) the presence of significant intercepts does not support the use of ratios as obesity indices and regression based models should be considered; and, (b) the direction and magnitude of the difference in obesity index between men and women changes with different normalization approaches.
体重指数（BMI）、体脂百分比及脂肪与瘦体重之比是常用的肥胖指标。这些指标基于一个假设，即分子（如脂肪量）与分母（如体重）之间的回归具有零截距。如配套论文所示，当使用这些指标来调整数据时，非零截距会引发若干问题，而协方差分析（ANCOVA）通常是首选的统计工具。本文旨在以性别比较为例，研究BMI、体脂百分比及脂肪与瘦体重之比是否符合作为合适肥胖指标的必要标准。
在720名健康男性和女性中，男性的BMI较高（24.9±3.3 vs 23.4±3.2 kg/m²，P<0.001），但女性的脂肪量、体脂百分比及脂肪与瘦体重之比更高（脂肪量：14.8±8.4 vs 19.1±8.1 kg脂肪；体脂百分比：18.6±8.7 vs 29.9±9.7%；脂肪与瘦体重之比：0.24±0.14 vs 0.46±0.20；P<0.001）。男性和女性的体重（BM）与身高的平方均相关（男性r = 0.40，女性r = 0.36），回归斜率相当（女性为15.6±2.3 kg/m²，男性为17.1±1.9 kg/m²），但截距不为零（男性为24.1 kg，女性为20.7 kg）。当使用ANCOVA对身高平方进行调整后，男性的体重仍显著高于女性（74.4±11.0 vs 68.8±11.6 kg；P<0.001）。脂肪量（FM）在男性（r = 0.64）和女性（r = 0.78）中均与BM显著相关，但回归斜率不同（女性为每千克体重0.71±0.03 kg脂肪，男性为0.49±0.03 kg脂肪；P<0.05），截距也不为零（男性为-23.2±2.2 kg，女性为-24.8±2.1 kg）。经BM调整后的FM在女性中显著更高（11.7 vs 25.6 kg）。FM在男性（r = -0.17）和女性（r = -0.20）中均与去脂体重（FFM）呈负相关，回归斜率相似（女性为每千克FFM -0.26±0.08 kg FM，男性为-0.16±0.05 kg FM），截距也显著不为零（男性为24.8±3.0 kg，女性为30.7±3.6 kg）。当对FFM进行调整后，男性（16.3 kg）和女性（17.0 kg）之间无显著差异。
得出以下结论：（a）显著截距的存在不支持将这些比率用作肥胖指标，应考虑基于回归的模型；（b）男性和女性肥胖指标差异的方向和幅度会因不同的标准化方法而改变。
