Figure 4. Prediction accuracy of three-compartment models for alfentanil disposition in men treated with troleandomycin. Data were fit to (A) conventional [k10] and (B) time-dependent inhibition [k10(t)] models. Each line represents data from a single participant. A perfect model fit would result in a straight line at 1. There was systematic bias and inaccuracy using the conventional model. Data fitting was improved by the k10(t) model, but did deviate at times after the second dose of troleandomycin (arrow).