**DOI:**10.1128/AAC.00054-09

## ABSTRACT

Cefuroxime axetil is widely used to treat respiratory tract infections. We are not aware of a population pharmacokinetic (PK) model for cefuroxime axetil. Our objectives were to develop a semiphysiological population PK model and evaluate the pharmacodynamic profile for cefuroxime axetil. Twenty-four healthy volunteers received 250 mg oral cefuroxime as a suspension after a standardized breakfast. Liquid chromatography-tandem mass spectrometry was used for drug analysis, NONMEM and S-ADAPT (results reported) were used for parametric population PK modeling, and NPAG was used for nonparametric population PK modeling. Monte Carlo simulations were used to predict the duration for which the non-protein-bound-plasma concentration was above the MIC (*f*T_{>MIC}). A model with one disposition compartment, a saturable and time-dependent drug release from the stomach, and fast drug absorption from the intestine yielded precise (*r* > 0.992) and unbiased curve fits and an excellent predictive performance. The apparent clearance was 21.7 liters/h (19.8% coefficient of variation [CV]) and the volume of distribution 38.7 liters (18.3% CV). Robust (≥90%) probabilities of target attainment (PTAs) were achieved by 250 mg cefuroxime given every 12 h (q12h) or q8h for MICs of ≤0.375 mg/liter or ≤0.5 mg/liter, respectively, for the bacteriostasis target *f*T_{>MIC} of ≥40% and for MICs of ≤0.094 mg/liter or ≤0.375 mg/liter, respectively, for the near-maximal-killing target *f*T_{>MIC} of ≥65%. For the ≥40% *f*T_{>MIC} target, the PTAs for 250 mg cefuroxime q12h were ≥97.8% for *S**treptococcus pyogenes* and penicillin-susceptible *S**treptococcus pneumoniae*. Cefuroxime at 250 mg q12h or q8h achieved PTAs below 73% or 92%, respectively, for *H**aemophilus influenzae*, *M**oraxella catarrhalis*, and penicillin-intermediate *S. pneumoniae* for susceptibility data from various countries. Depending on the MIC distribution, 250 mg oral cefuroxime q8h instead of q12h should be considered, especially for more-severe infections that require near-maximal killing by cefuroxime.

Cefuroxime axetil is the acetoxyethyl-ester prodrug of cefuroxime. Cefuroxime axetil is reliably absorbed and can be taken with or without a meal, although its extent of bioavailability is enhanced under the influence of food (20, 54). Cefuroxime has been successfully used in the treatment of upper and lower respiratory tract infections as well as genitourinary tract infections (45) and is active against *H**aemophilus influenzae*, *M**oraxella catarrhalis*, *S**treptococcus pyogenes*, *K**lebsiella pneumoniae*, *N**eisseria gonorrhoeae*, penicillin-susceptible *S**treptococcus pneumoniae*, and some isolates of penicillin-intermediate *S. pneumoniae* (6, 7, 25-27, 34, 35, 37, 38, 41, 55).

A susceptibility breakpoint of ≤1 mg/liter has been determined for cefuroxime by national organizations in Britain (BSAC) (8) and Germany (DIN) (17). The susceptibility breakpoints from the Clinical and Laboratory Standards Institute (CLSI) (11) are ≤1 mg/liter for *S. pneumoniae* and ≤4 mg/liter for *Haemophilus* spp., *Enterobacteriaceae*, and *Staphylococcus* spp.

Several authors (31, 37, 40) determined the pharmacokinetic-pharmacodynamic (PK-PD) MIC breakpoint for cefuroxime axetil on the basis of the average plasma concentration profiles but did not incorporate between-subject variability (BSV) in their analysis. Ambrose et al. (2) determined the PK-PD MIC breakpoint for intravenous cefuroxime via Monte Carlo simulation (MCS) based on literature data, and Viberg et al. (51, 52) developed a population PK model for intravenous cefuroxime. We are not aware of a population PK model or MCS for cefuroxime axetil.

Population PK and the MCS methodology account for the BSV in PK parameters and for the variability in the bacterial susceptibility. A PK-PD target is used as a surrogate measure to predict successful microbiological or clinical outcome (13, 18, 22, 28, 29, 48). For beta-lactams, the duration for which the non-protein-bound-plasma concentration exceeds the MIC (*f*T_{>MIC}) best predicts these outcomes (3, 14, 18). For cephalosporins, data from animal infection models showed that a target *f*T_{>MIC} of 40% correlates with bacteriostasis at 24 h and that an *f*T_{>MIC} of 60 to 70% is required for near-maximal bactericidal activity at 24 h (3, 12, 14, 18). On the basis of these PK-PD targets, MCS can predict the probability of target attainment (PTA) at various MICs. If the PTA-versus-MIC profile is combined with the expected MIC distribution of the pathogen(s) of interest in a local hospital, the probability of successful microbiological or clinical outcome can be predicted.

In addition to increasing the extent of bioavailability (20, 54), administration after a meal may cause a lower rate of cefuroxime absorption due to a prolonged gastric transit time. The rate of gastric emptying after a high-fat meal is likely to be variable and may change over time. Parameters describing the absorption phase may also not be normally or log-normally distributed. As MCS based on parametric population PK models uses parametric distributions to describe the variability in PK parameters, we additionally applied nonparametric population PK modeling. The latter offers the advantage that it does not assume any shape for the multivariate distribution of PK parameters. However, sample sizes larger than 24 subjects may be required to adequately describe the shape of the multivariate distribution by a nonparametric variability model.

Our first objective was to develop a semiphysiological population PK model for cefuroxime axetil by using parametric and nonparametric population PK methods. Second, we sought to determine the PTA-versus-MIC profiles and the PTAs for specific MIC distributions of various pathogens for dosage regimens with oral cefuroxime given every 12 h (q12h) or q8h.

(This work was presented in part at the 46th Interscience Conference on Antimicrobial Agents and Chemotherapy, 2006 [8a].)

## MATERIALS AND METHODS

Subjects.Twenty-four healthy male Caucasian volunteers participated in the study after they had given their written informed consent. Their demographic data were as follows (means ± standard deviations [SD]): age, 24.5 ± 3.3 years (range, 18 to 31 years); weight, 73.8 ± 9.2 kg (range, 58.2 to 93.6 kg); and height, 179 ± 8.0 cm (range, 166 to 193 cm). The subjects' health statuses were assessed by physical examination, electrocardiography, and laboratory tests, including urinalysis and screening for drugs of abuse. Intake of food and fluid was strictly standardized during the study days. Consumption of tobacco, methylxanthines, and alcohol in any form was prohibited from 12 h before administration of the study drug until acquisition of the last sample. The volunteers were closely observed by physicians for the occurrence of adverse events on the study days. The study protocol had been approved by the local ethics committee, and the study was conducted by following the revised version of the Declaration of Helsinki.

Study design and drug administration.The study was a single-dose, single-center study. All subjects received an oral suspension of 300.72 mg cefuroxime axetil (equivalent to 250 mg cefuroxime) with 240 ml low-carbonated, calcium-poor mineral water at room temperature. The study drug was administered directly after intake of a standardized breakfast with a significant amount of fat. This breakfast contained four slices of crisp bread (50 g), 20 g margarine, 2 slices of cheese (40 g; 30% fat content), 25 g jam, 100 ml fruit tea, and 100 ml milk (3.5% fat content).

Blood sampling.All blood samples were drawn in heparinated tubes from a forearm vein via an intravenous catheter. Blood samples were drawn immediately before administration and at 30, 60, and 90 min and at 2, 2.33, 2.67, 3, 3.33, 3.67, 4, 4.5, 5, 6, 8, 10, and 12 h after administration of the study drug. Samples were immediately centrifuged and immediately frozen and stored at −70°C until analysis.

Drug analysis.Samples were analyzed by means of a liquid chromatography-tandem mass spectrometry method validated for 0.1-ml samples of human plasma. Plasma samples (0.1 ml) were diluted with buffer containing the internal standard and deproteinized by addition of 400 μl of acetonitrile. After thorough mixing, the samples were centrifuged for 5 min at 3,600 rpm at approximately +4°C, and acetonitrile was removed by extraction with 1 ml dichloromethane. The mixture was centrifuged again, and 30 μl of the aqueous phase of each sample was then chromatographed on a reversed-phase column (Waters Symmetry C_{8}), eluted with an isocratic solvent system consisting of ammonium acetate buffer and acetonitrile (70%-30%, vol/vol), and monitored by liquid chromatography-tandem mass spectrometry with a multiple-reaction-monitoring method as follows: precursor → product ion for cefuroxime (*m/z* 423 → *m/z* 207) and an internal standard (*m/z* 426 → *m/z* 156). Both analyses were in negative mode. Under these conditions, cefuroxime and the internal standard were eluted after approximately 1.4 and 1.5 min. The Mac Quan software program (version 1.5; PE Sciex, Thornhill, Ontario, Canada [1991 to 1997]) was used for evaluation of chromatograms.

The linearity of the cefuroxime calibration curve was shown from 0.00900 mg/liter to 10.2 mg/liter. The coefficient of correlation for all measured sequences of cefuroxime was at least 0.999. The lowest calibration standard of 0.00900 mg/liter was set as the lower limit of quantification of the assay for cefuroxime in human plasma. There was no observation below this quantification limit. For the spiked quality control standards of cefuroxime, the interday precision ranged from 3.2 to 5.0%, with interday accuracy (relative error) between −4.3 and 2.1%. The intraday precision and relative error of the cefuroxime assay ranged from 0.7 to 4.0% and from −0.1 to 3.4%, respectively.

Population PK analysis. (i) Computation.We applied the first-order conditional estimation method with the interaction estimation option in NONMEM, version VI, release 1.2 (NONMEM Project Group, University of California, San Francisco) (5). Initial models were developed in NONMEM V. Model development was primarily performed in NONMEM. The final population PK model was additionally estimated in S-ADAPT (version 1.55; parallelized on a computer cluster), using the importance-sampling parametric Monte Carlo expectation maximization method (with a pmethod value of 8 in S-ADAPT) (4) and by the nonparametric adaptive grid (NPAG) algorithm implemented in the USC*PACK program (version 12.00) (33). WinNonlin Professional (version 4.0.1; Pharsight Corp., Mountain View, CA) was used for noncompartmental analysis and statistics.

Parameter uncertainty was assessed by standard asymptotic formulas in S-ADAPT (4). As NONMEM could not compute asymptotic standard errors in the $COV step for the final model, nonparametric bootstrap methods with 100 replicates were used to calculate standard errors in NONMEM as described previously (9).

(ii) Structural model.We considered one- and two-compartment disposition models with first-order, zero-order, or mixed-order (Michaelis-Menten) absorption with or without a lag time. Additionally, a semiphysiological model with a time-dependent release from the stomach to the intestine and subsequent absorption into the central compartment was developed (Fig. 1). The differential equations for this model are as follows:
$$mathtex$$\[\frac{dA_{1}}{dt}{=}{-}\frac{V_{\mathrm{max}}\ {\cdot}\ A_{1}}{K_{m}{+}A_{1}}\]$$mathtex$$(1)
$$mathtex$$\[\frac{dA_{2}}{dt}{=}\frac{V_{\mathrm{max}}\ {\cdot}\ A_{1}}{K_{m}{+}A_{1}}{-}k_{\mathrm{abs}}\ {\cdot}\ A_{2}\]$$mathtex$$(2)
$$mathtex$$\[\frac{dA_{3}}{dt}{=}k_{\mathrm{abs}}\ {\cdot}\ A_{2}{-}\frac{\mathrm{CL}}{V}\ {\cdot}\ A_{3}\]$$mathtex$$(3) where *A*_{1} is the amount of drug in the stomach, *A*_{2} is the amount of drug in the intestine, and *A*_{3} is the amount in the central compartment (see Table 2 for parameter explanations). All initial conditions for all three compartments were zero. The stomach compartment (*A*_{1}) received a bolus dose of 250 mg cefuroxime at zero hour. The model is simplified, as it did not contain a specific compartment for the prodrug cefuroxime axetil. It was assumed that cefuroxime axetil is converted to cefuroxime before the ester prodrug reaches the peripheral sampling site. This assumption is considered justifiable for an ester prodrug. The maximum rate of release (*V*_{max}) of drug from the stomach compartment was described by a time-dependent function:
$$mathtex$$\[V_{\mathrm{max}}(\mathrm{TPM}){=}V_{\mathrm{max}\ 0}\ {\cdot}\ \left(1{+}\frac{E_{\mathrm{max}}\ {\cdot}\ \mathrm{TPM}^{{\gamma}}}{\mathrm{TC}_{50}^{{\gamma}}{+}\mathrm{TPM}^{{\gamma}}}\right)\]$$mathtex$$(4) Time is denoted as time past meal (TPM). Starting from a *V*_{max} at time zero (*V*_{max 0}), the Hill coefficient (γ) modifies the *V*_{max} over time, with TC_{50} denoting the time past the meal at which *V*_{max} changed by 50% and *E*_{max} the extent of the change in *V*_{max}. For TPM values of ≫TC_{50}, the *V*_{max} approaches *V*_{max} _{0}·(1 + *E*_{max}). Therefore, an *E*_{max} of −1 represents complete inhibition of gastric release, an *E*_{max} of 0 an unchanged maximum rate of gastric release, and an *E*_{max} of 1 a maximum rate of gastric release twice as fast.

Competing models were discriminated by their predictive performance assessed via visual predictive checks (VPCs), their objective function (or log likelihood), and standard diagnostic plots.

(iii) Parameter variability and observation model.For parametric population PK modeling in NONMEM and S-ADAPT, we estimated the BSV of PK parameters by assuming a log-normal distribution. The maximum extent of inhibition or stimulation of gastric release for the *i*th subject (*E*_{max} _{i}) was constrained to a lower bound of −1 by the following logit transformation:
$$mathtex$$\[E_{\mathrm{max}\ _{i}}{=}{-}1{+}\ \left(10\ {\cdot}\ \frac{\mathrm{exp}(\mathrm{Lg}_E_{\mathrm{max}}{+}\mathrm{BSV}\ E_{\mathrm{max}\ _{i}})}{1{+}\mathrm{exp}(\mathrm{Lg}_E_{\mathrm{max}}{+}\mathrm{BSV}\ E_{\mathrm{max}\ _{i}})}\right)\]$$mathtex$$(5) Lg_*E*_{max} is the estimated population mean (arithmetic mean) on the transformed scale and BSV *E*_{max} _{i} the random deviate for the *i*th subject on transformed scale. This transformation ensures that all *E*_{max} _{i} values range from −1 to 9. A sensitivity analysis showed that the upper limit of 9 did not affect the curve fits or predictive performance of this model.

Plasma concentration time profiles were simulated for at least 4,800 subjects for each competing model to calculate the median and nonparametric 80% prediction interval (10th to 90th percentile) for the predicted concentrations. The same percentiles were calculated for the observations to visually assess whether the simulated percentiles closely matched the percentiles of the observations. For nonparametric population PK models in NPAG, this VPC was performed either based on the nonparametric distribution of PK parameters characterized by the support point matrix or based on a parametric, multivariate log-normal distribution of PK parameters. In all three programs full variance-covariance matrices were estimated and used for MCS.

The residual unidentified variability was described by a combined additive and proportional error model. We used the adaptive gamma option in NPAG to estimate the residual error described by the assay error polynomial.

(iv) MCS.A target *f*T_{>MIC} of 60 to 70% has been identified for near-maximal bactericidal activity of cephalosporins, and a target of 40% is required for bacteriostasis (14, 18). Therefore, we used PK-PD target *f*T_{>MIC}s of 65% for near-maximal bactericidal activity and 40% for bacteriostasis. A range of MICs from 0.031 to 64 mg/liter was considered. As the protein binding levels for cefuroxime have been reported to range between 33 and 50% (21, 23, 45), we assumed an average protein binding of 42% for cefuroxime.

We compared dosage regimens of 250 mg and 500 mg oral cefuroxime given q12h and q8h at steady state. For the final population PK models in NONMEM, S-ADAPT, and NPAG, we simulated 10,000 subjects for each dosage regimen in the absence of residual error. NONMEM was used to simulate the full concentration time profiles at steady state with very frequent sampling based on the final population PK model and the estimated full variance-covariance matrix. The *f*T_{>MIC} values were calculated by linear interpolation between simulated time points as previously described (9).

The PTA was estimated by calculating the fraction of subjects who attained the PK-PD target at each MIC. The highest MIC with a PTA of at least 90% was used as the PK-PD MIC breakpoint.

To put these PTAs into clinical perspective, we calculated the PTA expectation value (39) for successful treatment against pathogens from specific MIC distributions as described previously (9). The PTA expectation value is the PTA for treatment of infections caused by bacteria from a specific MIC distribution (ideally the MIC distribution of each local hospital).

The PTA expectation value was calculated based on published MIC distributions. We used susceptibility data from the United Kingdom (38) collected in 2002 and 2003 for *H. influenzae* (*n* = 581), *M. catarrhalis* (*n* = 269), and *S. pneumoniae* (*n* = 519); susceptibility data from Canada (55) collected in 2001 and 2002 for *H. influenzae* (*n* = 1,350); and susceptibility data from Germany (7) collected in 2002 for *H. influenzae* (*n* = 300), *M. catarrhalis* (*n* = 308), *S. pneumoniae* (*n* = 331), and *S. pyogenes* (*n* = 340). Additionally, we used susceptibility data from a global surveillance study (6) collected between 1997 and 2000 for penicillin-susceptible *S. pneumoniae* (*n* = 2,102) and penicillin-intermediate *S. pneumoniae* (*n* = 1,024); susceptibility data from a European surveillance study (26) collected between 1997 and 1999 for *S. pneumoniae* (*n* = 2,018) and *S. pyogenes* (*n* = 662); and susceptibility data from North America (25) collected between 1997 and 1999 for *S. pyogenes* (*n* = 119), *S. pneumoniae* (*n* = 417), *H. influenzae* (*n* = 300), and *M. catarrhalis* (*n* = 231). The PTA expectation values were also calculated for the MIC distributions for *H. influenzae* (*n* = 66,947), *K. pneumoniae* (*n* = 34,629), *M. catarrhalis* (*n* = 14,308), *S**taphylococcus aureus* (*n* = 10,620), *N. gonorrhoeae* (*n* = 655), and *N**eisseria meningitidis* (*n* = 257) on the basis of the multinational database of the European Committee on Antimicrobial Susceptibility Testing (EUCAST) (http://www.eucast.org/mic_distributions_of_wild_type_microorganisms/).

## RESULTS

The PK parameters from noncompartmental analysis (Table 1) were in good agreement with the literature (41, 45). We found (average ± SD) a terminal half-life of 1.34 ± 0.13 h and a peak concentration of 2.64 ± 0.64 mg/liter between 2 and 5 h postdose. The variability in terminal half-life (9.4% coefficient of variation) was lower than the variabilities in apparent clearance (20%), peak concentration (24%), and time of peak (24%).

A biphasic absorption pattern was found for 5 of 24 subjects and a “plateau-like” peak for 8 of 24 subjects (Fig. 2). These shapes could not be described by standard first-order or zero-order absorption models that included a lag time. Compared to the objective function value for the final semiphysiological model, the objective function differences in NONMEM were 721 points for the first-order absorption model with a lag time, 359 points for the zero-order absorption model with a lag time, and 189 points for the Michaelis-Menten absorption model with a lag time (likelihood ratio test: *P* < 0.0001 for all comparisons). The semiphysiological absorption model with a two-compartment disposition model had a 25-point-lower objective function than the same absorption model with one disposition compartment. As the latter model showed precise curve fits and an excellent predictive performance, we chose the simpler model as the final model.

The individual curve fits for the semiphysiological model (Fig. 1) were excellent in all three programs (Fig. 2). The model was flexible enough to fit profiles with “sharp” peaks, “plateau-like” peaks, and dual peaks. No estimation algorithm (program) provided the best fit for every subject. The linear regression plot of individual fitted versus observed concentrations had a slope of 1.007 and an intercept of −0.014 mg/liter in NONMEM (*r* = 0.9941), a slope of 1.011 and an intercept of −0.011 mg/liter in S-ADAPT (*r* = 0.9928), and a slope of 1.003 and an intercept of +0.011 mg/liter in NPAG (*r* = 0.9935).

The final estimates (Table 2 and Table 3) were precise. The relative standard errors were 31% or less for all population means (except for *K _{m}*, 70% in NONMEM and 39% in S-ADAPT) and 41% or less for all BSV estimates. The estimates for apparent clearance and volume of distribution were similar for all three programs. For the absorption parameters, the differences were more apparent. The low

*K*/dose value (mean, 0.433%) from NONMEM indicated that the release from the stomach was estimated to be essentially a zero-order process and that

_{m}*V*

_{max}was inhibited in some subjects and stimulated in others, as indicated by the negative or positive individual

*E*

_{max}value.

In S-ADAPT, the release from the stomach to the intestine had partial first-order and zero-order properties, as indicated by the estimate of 42.6% for *K _{m}*/dose. This rate of release was more notably stimulated, since the median (90th percentile) of individual

*E*

_{max}values was 1.82 (5.55). NPAG estimated the release from the stomach to the intestine primarily as a first-order process (

*K*/dose, 343%), and stimulation of gastric emptying was more pronounced than for S-ADAPT. The mean TC

_{m}_{50}values for the rate of gastric emptying after a standardized breakfast were 1.61 h in NONMEM and S-ADAPT and 2.08 h in NPAG. Individual TC

_{50}estimates were variable (Table 2).

The VPCs (Fig. 3) indicated that the nonparametric simulation based on the support points from NPAG yielded the closest match between predicted and observed concentrations. This was expected, since this simulation is based on the nonparametric distribution of PK parameters that yielded precise and unbiased fits for all 24 concentration-time profiles. The parametric simulations based on the estimates from S-ADAPT and NPAG matched the median and 10th to 90th percentiles for the observations more closely between approximately 1 and 4 h than those based on NONMEM (Fig. 3). For the three parametric simulations, S-ADAPT yielded the best representation of the observations during the terminal phase, followed by NONMEM. The predicted variability was slightly too wide during the terminal phase for the parametric simulation (Fig. 3) using the variance-covariance matrix derived from the support point matrix in NPAG.

As the VPCs showed that the nonparametric simulation based on NPAG and the parametric simulation based on S-ADAPT had the best predictive performance, breakpoints for MCS are reported for these two models. The breakpoints for the other two simulation models were within one 1.5-fold dilution. The PTA-versus-MIC profiles were similar for these two models (Fig. 4). For the bacteriostasis target *f*T_{>MIC} of ≥40%, the PK-PD MIC breakpoints in NPAG and S-ADAPT were 0.5 and 0.375 mg/liter, respectively, for 250 mg q12h and 0.5 and 0.5 mg/liter, respectively, for 250 mg q8h. As these PK models are linear with dose, 500 mg q12h achieved breakpoints of 1 and 0.75 mg/liter and 500 mg q8h achieved breakpoints of 1 and 1 mg/liter in NPAG and S-ADAPT, respectively, for the ≥40% *f*T_{>MIC} target. For the near-maximal-killing target *f*T_{>MIC} of ≥65%, the PK-PD MIC breakpoints for S-ADAPT and NPAG were identical (0.094 mg/liter for 250 mg q12h, 0.375 mg/liter for 250 mg q8h, 0.188 mg/liter for 500 mg q12h, and 0.75 mg/liter for 500 mg q8h). Additional simulations with a hypothetical higher rate of absorption showed that the PK-PD MIC breakpoints were lower if the rate of absorption was faster. The decrease in breakpoints was most pronounced for the 65% *f*T_{>MIC} target and q12h dosing.

High PTA expectation values (≥97.8% for the 40% *f*T_{>MIC} target) were achieved by all three dosage regimens shown in Table 4 for *S. pyogenes*, penicillin-susceptible *S. pneumoniae*, *N. gonorrhoeae*, and *N. meningitidis* (results not shown for the latter two pathogens). High (>90%) PTA expectation values were achieved for some but not all MIC distributions for *S. pneumoniae*. The PTA expectation values were notably lower for penicillin-intermediate *S. pneumoniae*, *H. influenzae*, *M. catarrhalis*, *S. aureus*, and *K. pneumoniae* (results not shown for the latter two pathogens).

## DISCUSSION

Cefuroxime axetil has been used widely for treatment of community-acquired upper and lower respiratory tract infections (including community-acquired pneumonia) (45) that are often caused by *S. pneumoniae*, *H. influenzae*, *M. catarrhalis*, and *S. pyogenes*. The reported MIC_{90}s of cefuroxime against *H. influenzae* and *M. catarrhalis* are often 2 or 4 mg/liter (41, 45). Whereas most isolates would be deemed susceptible by the CLSI breakpoint of ≤4 mg/liter for those pathogens, a significant number of isolates would be considered resistant according to the BSAC and DIN susceptibility breakpoint of ≤1 mg/liter. To evaluate which breakpoint is in better agreement with the predicted PK-PD MIC breakpoint, we applied parametric and nonparametric population PK modeling and MCS for cefuroxime axetil.

The *f*T_{>MIC} best predicts the clinical and microbiological success for cephalosporins. As the prolonged absorption phase of cefuroxime axetil has a notable influence on the *f*T_{>MIC} values, it was critical to develop a population PK model that adequately captures the rates of absorption and BSVs that were observed in our studies and in the literature. We intensively qualified the predictive performance of our population PK model to ensure that the model-predicted PK-PD MIC breakpoints are sound. An 8-fold increase (from 125 to 1,000 mg) in the oral cefuroxime axetil dose after a meal causes a 7.5-fold increase in the area under the curve and a 6.5-fold increase in peak concentrations (20) and causes no systematically altered time of peak concentration (*T*_{max}) (20). Data from rats suggest a saturable component for the rate of absorption (42-44).

We found a range of complex absorption patterns in healthy volunteers (Fig. 2). Models with first-order or zero-order absorption with or without a lag time can describe the dose proportionality in area under the curve and peak concentration (20) but cannot describe a mixed-order rate of absorption and the complex absorption profiles observed in our study. A mixed-order absorption model with a lag time could describe the plasma concentration time profiles of cefuroxime axetil at one dose level (results not shown). However, such a mixed-order absorption model would predict a notable increase in *T*_{max} with cefuroxime doses. This is in disagreement with the data reported by Finn et al. (20). A mixed-order absorption model cannot describe profiles with a dual peak.

Food increases the extent of bioavailability of cefuroxime axetil from 36% in fasting subjects to 52% after a meal (20). Similar results were found by Williams and Harding (54). In both studies (20, 54), *T*_{max} is prolonged by approximately 0.6 to 0.7 h for administration with food. Potential reasons for the increased bioavailability and slightly longer *T*_{max} observed under fed conditions include a more complete dissolution of cefuroxime axetil due to a longer residence time in the stomach and due to bile acid secretion stimulated by the presence of lipids in the intestine (36, 50).

The proposed absorption model (Fig. 1) is in agreement with the observations of literature studies for various dose levels of cefuroxime axetil (20, 54). One limitation of our study is that we had data only for 250 mg oral cefuroxime. Therefore, our simulation results for 500 mg oral cefuroxime q12h should be interpreted conservatively. The saturable rate of absorption is described by a mixed-order release of drug from the stomach to the intestine that is primarily saturated due to the presence of food and not due to the cefuroxime axetil dose. In our model, *V*_{max} and *K _{m}* are expressed as fractions of dose, and this causes

*T*

_{max}to be independent of dose. The second peak in some profiles was described by an increase in the rate of gastric release over time.

This semiphysiological model proved to be robust (Table 2) and to yield excellent individual curve fits (Fig. 2) for all three population PK algorithms and programs. This absorption model was able to capture relevant features of complex oral absorption profiles (24, 32, 53) which showed that the rate of gastric emptying is important for the absorption of amoxicillin (amoxicilline) and clavulanic acid (53).

Estimation of a full variance-covariance matrix and its use during simulations yielded the best predictive performance, caused no instability during estimation, and did not prolong run times in S-ADAPT and NPAG. This saves modeling time, since there are fewer decisions about the choice of the parameter variability model in S-ADAPT and NPAG than in NONMEM. NPAG does not directly estimate the variance-covariance matrix but always derives this full matrix from the estimated support points. Estimating a full variance-covariance matrix in NONMEM tended to cause model instability (i.e., unsuccessful termination messages and the inability of NONMEM to obtain asymptotic standard errors) and notably increased estimation times in NONMEM.

The most important difference between the parametric and nonparametric approaches is that the former describes BSV by a parametric, multivariate distribution (often a multivariate log-normal distribution). In contrast, nonparametric methods use a discrete set of support points to exactly store the BSV and correlation structure of all estimated PK parameters in the studied patient population. In the simplest case, each support point essentially represents a complete set of PK parameters for one patient and has a probability of 1 divided by the number of subjects.

As the variability of individual PK parameter estimates in the studied subject population is “exactly” represented by the set of support points, it was expected that the nonparametric VPC had the best predictive performance (Fig. 3). The parametric VPC using estimates from S-ADAPT had the best predictive performance among the parametric VPCs. We reported the results from a parametric MCS in S-ADAPT with 10,000 virtual patients and from a nonparametric MCS in NPAG. As every support point had the same probability for this study with frequent sampling, the latter MCS was identical to simulation from the individual PK parameter estimates of the 24 subjects.

The PTA-versus-MIC profiles from S-ADAPT and NPAG were similar (Fig. 4). For 250 mg oral cefuroxime q12h or q8h, the PK-PD MIC breakpoints fell between 0.375 and 0.5 mg/liter for the bacteriostasis target *f*T_{>MIC} of 40% but were approximately fourfold lower (0.094 mg/liter) for dosing of 250 mg q12h and the near-maximal-killing target *f*T_{>MIC} of 65%. Dosing of 250 mg or 500 mg q8h increased the latter breakpoints to 0.375 mg/liter or 0.75 mg/liter, respectively. Koeth et al. (31) determined a PK-PD MIC breakpoint of ≤1 mg/liter for susceptibility for standard cefuroxime dosage regimens based on an *f*T_{>MIC} of ≥40 to 50%. This higher breakpoint is expected, since Koeth et al. (31) used average PK parameters for simulation and did not include BSV.

We did not manually increase the BSV in clearance and volume of distribution to mirror the higher variability in critically ill patients, as the relatively low PK-PD MIC breakpoints for oral cefuroxime do not support treatment of critically ill patients. A higher variability in PK parameters and lower average extent of bioavailability after intake of cefuroxime in the fasting state (20, 54) are expected to result in lower PK-PD MIC breakpoints than those reported here. As the sample size of 24 subjects probably did not allow us to obtain precise estimates of the BSV in the whole patient population, the results of our MCS should be interpreted conservatively. The probability of clinical success of the simulated cefuroxime axetil dosage regimens will ultimately depend on the MIC distribution of the pathogen(s) of interest in the local hospital. Dosing at 250 mg cefuroxime q8h instead of q12h had only a small benefit for *S. pyogenes* and penicillin-susceptible *S. pneumoniae*, since 250 mg q12h achieved high PTA expectation values, especially for the bacteriostasis target (Table 4).

Administering cefuroxime axetil q8h yielded notably higher PTA expectation values for some but not all MIC distributions of *H. influenzae* and *M. catarrhalis*. Although 500 mg oral cefuroxime q8h is above the typically recommended oral cefuroxime dose, parenteral doses of up to 6,000 mg split into four daily doses are recommended for severe infections.

To put the results of our MCS into a clinical perspective, we compared our PTA expectation values to the microbiological and clinical outcomes in clinical studies. Clinical data for children with pneumococcal acute otitis media suggest a breakpoint of about 0.5 mg/liter for oral cefuroxime (15). Shah et al. (47) studied hospitalized patients and outpatients in 14 countries in Europe, Africa, and South America with acute exacerbation of chronic bronchitis (AECB) and found a 60% bacteriological overall cure rate for 250 mg oral cefuroxime given twice daily (BID). This cure rate is comparable to the placebo response rate for AECB (30, 46), depending on the severity of disease. Interestingly, 56% (22 of 39) of the *H. influenzae* isolates were eradicated. Using per-protocol and intention-to-treat analyses, the authors reported clinical cure rates of 66% and 61%, respectively, at the clinical endpoint (5 to 14 days posttreatment) and of 53% and 39%, respectively, at the follow-up 3 to 4 weeks posttreatment. Although the authors did not report the MICs in those patients, the failures for the treatment of *H. influenzae* show a suboptimal effectiveness of oral cefuroxime against this pathogen.

Chodosh et al. (10) found a significantly lower microbiological eradication rate for 500 mg oral cefuroxime BID (82%) versus 500 mg oral ciprofloxacin BID (96%) in an outpatient trial with AECB patients. Cefuroxime eradicated *S. pneumoniae* in 100% of the cases (13/13) but had eradication rates of only 76% (19/25) for *M. catarrhalis* and 86% (19/22) for *H. influenzae*. In an outpatient trial with AECB patients, de Abate et al. (16) found a clinical cure rate of 77% for 250 mg oral cefuroxime BID, which was significantly lower than the clinical cure rate of 89% for 400 mg gatifloxacin once daily. The microbiological eradication rates were 77% for cefuroxime and 90% for gatifloxacin.

A trial using patients with community-acquired pneumonia (19) showed a significantly lower rate of microbiological eradication of *H. influenzae* for combinations of intravenous ceftriaxone (1 to 2 g once daily or BID) and/or oral cefuroxime axetil (500 mg BID) than for intravenous and/or oral levofloxacin (500 mg once daily). The former regimen had an eradication rate of 79% and the latter 100%. Upchurch et al. (49) found a clinical cure rate of 74.5% for treatment of acute bacterial sinusitis with 250 mg oral cefuroxime for 10 days but did not document the bacterial etiology. Alvarez-Sala et al. (1) found a clinical success rate of approximately 82% for patients with *H. influenzae* and *S. pneumoniae* for 250 mg oral cefuroxime q12h.

In conclusion, we developed a semiphysiological population PK model for oral cefuroxime which provided precise and unbiased individual curve fits for complex absorption profiles and had an excellent predictive performance. The nonparametric VPC based on NPAG showed a better predictive performance than the best parametric VPC in S-ADAPT. The PK-PD MIC breakpoints were 0.375 to 0.5 mg/liter for 250 mg oral cefuroxime q12h and 0.5 mg/liter for 250 mg oral cefuroxime q8h for the bacteriostasis target *f*T_{>MIC} of ≥40%. Dosing at 250 mg cefuroxime q8h instead of q12h increased the breakpoint for the near-maximal-killing target *f*T_{>MIC} of 65% from 0.094 mg/liter to 0.375 mg/liter. These breakpoints were (slightly) lower than the susceptibility breakpoint of ≤1 mg/liter provided by the BSAC and DIN, whereas the CLSI breakpoint of ≤4 mg/liter for most pathogens is higher than the PK-PD MIC breakpoints predicted by MCS in this study. Oral cefuroxime (250 mg q12h or q8h) achieved high PTA expectation values for *S. pyogenes* (≥96.7%) and penicillin-susceptible *S. pneumoniae* (≥91.4%) but notably lower PTA expectation values for *M. catarrhalis*, penicillin-intermediate *S. pneumoniae*, and *H. influenzae* for most studied MIC distributions for various countries. Administering 250 mg oral cefuroxime q8h instead of q12h was most beneficial for the near-maximal-killing target for MICs between 0.094 and 0.375 mg/liter. Future clinical studies that assess the MIC of the causative pathogen are warranted to validate these predictions for the clinical and microbiological success for q12h and q8h cefuroxime axetil dosage regimens.

## ACKNOWLEDGMENTS

We thank George Drusano for fruitful discussions about this project.

## FOOTNOTES

- Received 14 January 2009.
- Returned for modification 26 March 2009.
- Accepted 3 June 2009.
↵▿ Published ahead of print on 15 June 2009.

↵† This article is dedicated to Ulrich Stephan, the cofounder of IBMP, who passed away on 6 February 2009. Without his inspiration and support, IBMP would not exist, and neither would the present research have been performed. We keep him in our hearts.

- American Society for Microbiology