Pharmacokinetic (PK) models of propofol have been used in target controlled infusion (TCI) for the rapid achievement and maintenance of a stable plasma concentration (C_p) within a clinically acceptable range of performance errors [1-3]. However, each PK model has different parameter values, even though the parent drug is the same. Recent reports using a non-linear mixed effect model present complex parameter sets that are expressed with complex equations of covariates. Commercially available TCI devices and PK software programs have various PK models for one drug. Therefore, we have faced some difficulties in choosing the appropriate PK model for routine clinical settings. Furthermore, time courses of predicted C_p and effect-site concentration (C_eff) of propofol were different when the manual bolus and continuous infusion were simulated using different PK models. Likewise, the amounts of propofol administered for maintenance of a stable C_p were also different between the PK models [4]. Therefore, we compared the PK models of propofol to investigate different infusion patterns using the reproducing simulation of TCI data of one PK model using another PK model.

With approval of the Institutional Review Board and signed informed consent, a total of 60 female patients, American Society of Anesthesiologists physical status I, age 20-50 years, and scheduled for gynecologic surgery under general anesthesia were enrolled in this study. Exclusion criteria included body weight of over 150% of ideal body weight, medications for the central nervous system, patients with chronic pain, or neurological and endocrine disorders. Patients received glycopyrrolate 0.2 mg intravenously for premedication; however, no other sedatives were administrated before and during the period of this study. Electrocardiograph, pulse oximeter, noninvasive blood pressure monitor, and capnogram were applied. Heart rate and blood pressure were monitored and recorded every 30 s during the study and thereafter every 3 min. After pre-oxygenation, lidocaine 30 mg was given and propofol TCI was started.

Patients were randomfly assigned to one of two groups. The three compartment mammillary pharmacokinetic model of Marsh PK [5] and 1.21/min of K_e0 [6] were used for Group M, and the multiple covariate-adjusted model of Schnider et al. [7,8] was used for Group S. The PK parameters and their equations with covariates and the K_e0 are illustrated in Table 1. Propofol (1% Fresofol®, Fresenius Kabi, Graz, Austria) was administrated using a syringe pump (Graseby 3500, Sims Graseby Ltd., Herts, England) targeting at 6.0 µg/ml of C_p for induction of anesthesia. The syringe pump was controlled using pharmacokinetic/pharmacodynamic (PK/PD) software (STANPUMP©, written by Steven L. Shafer, Palo Alto, CA, USA) through RS232 serial communication. The propofol-filled syringe was connected to the 3-channel extension tube with a unidirectional valve preventing the back flow to the intravenous bag. In the case of any alarm triggered from the TCI device during the study, the patient was also excluded from the study.

There were no significant differences in the demographic data and hemodynamic variables between groups (Table 2). Group M showed shorter time to LOR compared to Group S (P < 0.001). However, the amount of propofol infused until LOR did not differ significantly between the groups (Table 3 and Fig. 1). C_eff at LOR did not differ between groups, but after cross-simulation using the other PK/PD model, C_eff at LOR were significantly different between groups (P = 0.009) (Table 3).

During TCI of propofol, the infusion data files that were supposed to maintain a stable C_p did not produce identical results when played back to a different PK model. The differences were greater during the early period of infusion, whereas later a relatively stable range of differences was maintained, but with a continual increase during long-term infusion. These discrepancies highly correlated with the body weight of the subject.

Using simulations, we attempted to perform a straightforward comparison of the integrated difference of the two PK models. Apparently, the volume of the central compartment of Marsh PK was nearly triple that of Schnider PK. Therefore, Marsh PK required a larger amount of propofol to fill up the central compartment when there was no drug in the body, and, as a result, showed a more rapid loss of responsiveness than Schnider PK. However, the clearance of the central compartment of Marsh PK was smaller than that of Schnider PK. Therefore; we could deduce that Marsh TCI requires less propofol than Schnider TCI. These differences appeared to be distinct when we simply compared the cumulative amounts of propofol and the PK parameters (3 volumes and 3 clearances). However, the degree of discrepancy during TCI and time course of the predicted concentration could not be determined.

Therefore, we performed total and separate cross-simulation. During the early phase of infusion, the difference of C_p was significant. Thereafter, overall differences were shown to be roughly 10% (at 30 min) and 15% (at 2 hr) under- or overprediction of C_p, and the differences showed a gradual increase with time. However, total cross-simulations could not explain the impact of each parameter on the overall difference of prediction. Therefore, we performed separate cross-simulations. Young and Shafer [9] also demonstrated the changes of C_p and decrement time of three opioids by independently changing the volumes and clearance of the three compartments. They used simulations that were similar to our method, while predicting the change in C_p of opioid after a bolus if each pharmacokinetic parameter were independently increased by 5%. Our separate cross-simulation showed that V_C mainly contributed to the initial differences, and Cl₁ and V₃ to the difference of long-term infusion. The difference of Cl₃ also seemed to be an important factor during the initial period, but contributed less during long term infusion.

Even though Marsh TCI showed a more rapid loss of responsiveness than Schnider TCI, the amount of propofol infused until LOR and C_eff at LOR did not differ between the groups (Table 3). The k_e0 used for Marsh PK was 1.21/min [6], whereas 0.456/min was used for Schnider PK. The volume of distribution of Schnider PK was small and time to LOR was long during Schnider TCI, but the predicted C_eff slowly increased because of its smaller k_e0 than Marsh PK. Therefore, C_eff at LOR appeared to be insignificant between groups. However, the reproduced C_eff at LOR after simulation was significantly different between groups. During TCI, C_eff is calculated using the k_e0 and C_p. However, the value of k_e0 is highly influenced by the PK model and it was reported that it is unwise to use the k_e0 from one PK model with another PK model from a different study [6,10,11].

Body weight and LBM correlated well with the discrepancies of prediction. LBM is calculated using body weight and height. In Marsh PK, body weight influences V_C, V₂, V₃, Cl₁, Cl₂, and Cl₃. As a result, decreasing body weight leads to a smaller volume and clearance of each compartment. In addition, the relationship between Cl₁ and height-body weight showed a flat mesh (Fig. 7, left graph, gray mesh). In Schnider PK, only Cl₁ is influenced by body weight and height. Decreasing body weight and height leads to smaller Cl₁, and increasing height with lower body weight or decreasing height with higher body weight leads to a higher Cl₁ (Fig. 7, left graph, transparent mesh). Therefore, a taller patient with a lower body weight or a shorter obese patient showed a larger difference of clearance between the two models (Fig. 7, Right graph). Accordingly, there could be a large difference in the level of hypnosis between models during the TCI to these types of patients.

In our study, the mean Cl₁ of Schnider PK was 8.8% higher than that of Marsh PK. Wietasch et al. reported that they found a smaller V_C (3.55 L) than Marsh PK, and Cl₁ was reduced to 1.31 L/min in comparison with that of Marsh PK (2.04 L/min) and Schnider PK (1.89 L/min) [3]. In this study, contrary to our study, Cl₁ of Marsh PK was greater than Schnider PK, and mean body weights of subjects were around 75 kg, which was the heaviest body weight in our study. Kim et al. [12] reported that the V_C of lipid emulsion of propofol was 3.9 L and Cl₁ was 1.53 L/min. Jung et al. [13] reported that V_C of lipid emulsion of propofol was 6.78 L and Cl₁ was 1.46 L/min. The populations of these two reports were similar overall to those of our study. Marsh PK and Schnider PK are known to under-predict C_p [3,14]. White et al. [15] also suggested the use of the parameters of Marsh PK adjusted with the covariates, such as age and gender. Therefore, the possibility of overdose of propofol using Schnider PK could increase during long term infusion, especially for patients with low body weight.

There are some caveats that need to be taken into consideration. First, our study evaluated the PD difference between the models, only for the period of induction of anesthesia. Further study on the PD differences during maintenance of anesthesia and recovery period will be helpful in differentiation of the characteristics between the models. Second, the population of this study was limited to females. TCI using Marsh PK administers propofol irrelevant to gender. However, Schnider PK estimates different Cl₁ according to gender. The covariate, LBM, used for calculation of Cl₁, is differently calculated according to gender; men, LBM = 1.1 × weight - 128 × (weight / height)²; women, LBM = 1.07 × weight - 148 × (weight / height)² [16]. Cl₁ for women is calculated higher than that of men when other covariates are equal. Therefore, the results of this study could be different if investigated for both genders. Third, it took nearly 4 min to induce LOR using Schnider PK. The longest time to LOR was 310 s in Group S, which is considered to be inadequate for the routine induction of anesthesia. Therefore, other modalities, such as higher target of C_p or targeting C_eff, are indicated.

The primary goal of this study was to increase our understanding between the different PK models and the influence of PK parameters on the infusion patterns during TCI. This comparison method between PK models would be helpful in selection of an appropriate PK model of a certain drug in clinical settings and research fields.