Figure 7. Comparison of predictive performance measures obtained for model versions A, A', D, and D' in 53 patients. The individual scatters of the patients are plotted versus their individual biases for the four versions. The triangles and the thick lines on the abscissas and the ordinates represent the group biases (plus/minus SD) and scatters (plus/minus SD), respectively. The group biases indicate that all versions suffer from overprediction, with A taking the worst position. Ideally, all observations on one version would coincide with the star representing the point of ideal performance (root mean squared error (rmse) = 0%). The semicircles are the iso-rmse-lines from 10% to 50%. The distance between the star and a dot is the rmse representing the typical error size for an individual patient. The left upper plot visualizes the mathematical relationship between the rmse, bias, and scatter (Appendix II, Equation 5A). The rmse is the hypotenuse of a right-angled triangle of which the scatter and the bias are the opposite and the adjacent, respectively. The adoption of a nonlinear nonpulmonary elimination (D versus A) reduces both the biases and the scatters. The incorporation of age-adjusted partition coefficients (A' versus A and D' versus D) reduces the rmses. The plot for D' shows that 91% of the patients are found in the semicircular area where the rmse is < 30%.