On establishing of predictive model of total body water. (II) / 体力科学

ABSTRACT

In this paper, a predictive equation for TBW (Total Body Water) from various anthropometric measurements was established. Fifty-seven healthy adult males, aged 19 to 54 years old, volunteered as subjects in this experiment.<BR>Ten anthropometric measurements were taken for each subject such as standing height (HT), body weight (WT), breadth of humerus and femurs (B1, B2), girth of upper arm and calf (G1, G2), and skinfold thickness of triceps, subscaplar, suprailliac and abdomen (S1, S2, S3, S4) along with the amount of ingested deutrium oxide (D<SUB>2</SUB>O) .<BR>Total body water was quantitied by the analysis of dilution of orally ingested D<SUB>2</SUB>O in urine. The method of forward stepwise regression analysis was adopted to establish the predictive equation. The stopping rule to select the variables was <I>F</I> statistics (<I>F</I>=2.0) . Furthermore, some criteria such as <I>AIC</I> (Akaike's an information criterion), Mallows' <I>Cp</I>, <I>Schwarz's</I> criterion, <I>R</I>* (adjusted for degree of freedom <I>R</I>) were derived as the regression equation was constructed at each step. These criteria contributed to selecting the best and most valid equation of all equations possible. The results obatained was summarized as follows.<BR>1) Firstly, descriptive statistics were derived for all subjects. Mean (±S. D, ) of TBW was 34.85 (±5.38) <I>l. Skewness</I> and <I>kurtosis</I> were not significant. Multico-linearlity was suggested by correlation matrix (10×10) of all independent variables.<BR>2) Six variables were entered into the equation such as sequences WT, S4, HT, G2, S1, B1. The multiple correlation coefficient (<I>R</I>) and standard error of estimates (<I>SEE</I>) of this equation were 0.961 and 1.568, respectively. It was derived as follows<BR><I>Y</I>=-34.56+0.170×HT+0.231×WT+0.567×G2+1.37×B1-0.167×S1-0.086×S4<BR>3) Since the analysis of residuals suggested that abnormal values were contained in this sample, the next regression analysis was adopted after deleted the results of 7 subjects whose standardized residuals were over 1.5. Consequently, the regression equation composited from WT, S4, S2, HT, S1 was evaluated as the best equation according to <I>Cp</I> and <I>Schwarz</I> criterion. <I>AIC</I> selected the equation which added S3 as the 6 th variables. The multiple regression equation established at this stage was described as follows, and <I>R</I> and <I>SEE</I> were 0.9857, 0.940, respectively.<BR><I>Y</I>=8.10+0.4573×WT-0.0839×S4-0.0951×S2+0.1089×HT-0.1368×S1<BR>4) The specific problem was not obtained from statistics of residuals. However, the co-ordinates of <I>eis</I> (standardized residuals) and predicted value suggested a specific changing pattern. The low possibility of existing multico-linearlity was determined from the eigen value of correlation matrix between independent variables.