The performance of compartmental and physiologically based recirculatory pharmacokinetic models for propofol: a comparison using bolus, continuous, and target-controlled infusion data.

BACKGROUND: With the growing use of pharmacokinetic (PK)-driven drug delivery and/or drug advisory displays, identifying the PK model that best characterizes propofol plasma concentration (Cp) across a variety of dosing conditions would be useful. We tested the accuracy of 3 compartmental models and 1 physiologically based recirculatory PK model for propofol to predict the time course of propofol Cp using concentration-time data originated from studies that used different infusion schemes.
METHODS: Three compartmental PK models for propofol, called the "Marsh," the "Schnider," and the "Schüttler" models, and 1 physiologically based recirculatory model called the "Upton" model, were used to simulate the time course of propofol Cp. To test the accuracy of the models, we used published measured plasma concentration data that originated from studies of manual (bolus and short infusion) and computer-controlled (target-controlled infusion [TCI] and long infusion) propofol dosing schemes. Measured/predicted (M/P) propofol Cp plots were constructed for each dataset. Bias and inaccuracy of each model were assessed by median prediction error (MDPE) and median absolute prediction error (MDAPE), respectively.
RESULTS: The M/P propofol Cp in the 4 PK models revealed bias in all 3 compartmental models during the bolus and short infusion regimens. In the long infusion, a worse M/P propofol Cp at higher concentration was seen for the Marsh and the Schüttler models than for the 2 other models. Less biased M/P propofol Cp was found for all models during TCI. In the bolus group, after 1 min, a clear overprediction was seen for all 3 compartmental models for the entire 5 min; however, this initial error resolved after 4 min in the Schnider model. The Upton model did not predict propofol Cp accurately (major overprediction) during the first minute. During the bolus and short infusion, the Marsh model demonstrated worse MDPE and MDAPE compared with the 3 other models. During short infusion, MDAPE for the Schnider and Schüttler models was better than the Upton and the Marsh models. All models showed similar MDPE and MDAPE during TCI simulations. During long infusion, the Marsh and the Schüttler models underestimated the higher plasma concentrations.
CONCLUSION: When combining the performance during various infusion regimens, it seems that the Schnider model, although still not perfect, is the recommended model to be used for TCI and advisory displays.