Statistical identification of compartmental models with application to plasma protein kinetics

A numerical method for fitting linear compartmental models to data is presented which is similar to the method proposed by Jennrich and Bright (R. I. Jennrich and P. B. Bright, Technometrics 18, 385 (1976] and Feldman (H. A. Feldman. Amer. J. Physiol. 233, R1 (1977] but avoids some of the numerical difficulties. The method is direct and does not use numerical integration thereby avoiding time consuming and expensive calculations. Emphasis is on statistical procedures for model selection. In addition to the usual F tests for comparing two models, Akaike's Information Criterion (AIC) is introduced for choosing from among several models. The combined use AIC with an F test, chi-square test, or confidence intervals on estimated parameters gives a practical method to obtain a balance between underfitting and overfitting. Application to models of plasma protein kinetics illustrates the method.