Document Type



A new equation for predicting human body surface area from anthropometric measurements was developed with the aid of Statistical Analysis System (SAS), which allowed optional use of Mallows’ C(p) to select the best subset of regressors. Forty Japanese male students, aged 18 to 26 years served as the subjects. Body surface areas were determined by a direct paper coating technique and 15 anthropometric measurements were taken as regressors. Since the dimension of predictors should be the same as that of body surface area, linear anthropometric measurements were all squared and body weight was converted to 2/3th power in the regression analyses. The procedure RSQUARE in SAS yielded Mallows’ C(p) for all possible combinations of 15 anthropometric variables, thus providing the best subset of regressors having the minimal value of C(p): body weight, body height, head circumference and trunk length. However, we recommend the following 3-regressor model of weight, height and head circumference: Body surface area (cm2) = — 2142.0 + 617.0 X [weight (kg)]2/3 + 0.2453 X [height (cm)]2 + 0.6825 X [head circumference (cm)]2 The reasons are firstly because the reliability of trunk length measurement is very low, secondly because the statistical test proved the above 3-regressor model was statistically the second best, thirdly because the loss of information due to deleting the trunk length amounted only to 0.004 in R2, and lastly because Mallows recommended the model where C(p) first ap­proached the number of predictors p.The predictive reliability of this equation was cross-validated by additional data on 15 Japanese males who were not included in our experiments in an attempt to compare our results with those of representative published formulae. The results showed that the present regressors gave remarkably better predictive values than the representative height and weight models did.