Population pharmacokinetics of rifampicin in adult patients with osteoarticular infections: interaction with fusidic acid

Abstract

Aims

Rifampicin represents the key antibiotic for the management of osteoarticular infections. An important pharmacokinetic variability has already been described, particularly for absorption and metabolism. All previous pharmacokinetic studies have been focused only on patients treated for tuberculosis. The objective of the present study was to describe a population pharmacokinetic model of rifampicin in patients with staphylococcal osteoarticular infections, which has not been investigated to date.

Method

Rifampicin concentrations were collected retrospectively from 62 patients treated with oral rifampicin 300 mg three times daily. Plasma concentration–time data were analysed using NONMEM to estimate population pharmacokinetic parameters. Demographic data, infection characteristics and antibiotics taken in addition to rifampicin antibiotics were investigated as covariates.

Results

A one‐compartment model, coupled to a transit absorption model, best described the rifampicin data. Fusidic acid coadministration was identified as a covariate in rifampicin pharmacokinetic parameters. The apparent clearance and apparent central volume of distribution mean values [95% confidence interval (CI)] were 5.1 1 h^–1 (1.2, 8.2 1 h^–1)/23.8 l (8.9, 38.7 l) and 13.7 1 h^–1 (10.6, 18.0 1 h^–1)/61.1 1 (40.8, 129.0 1) for patients with and without administration of fusidic acid, respectively. Interindividual variability (95% CI) in the apparent clearance and apparent central volume of distribution were 72.9% (49.5, 86.0%) and 59.1% (5.5, 105.4%), respectively. Residual variability was 2.3 mg l^–1 (1.6, 2.6 mg l^–1).

Conclusion

We developed the first population pharmacokinetic model of rifampicin in patients with osteoarticular infections. Our model demonstrated that fusidic acid affects rifampicin pharmacokinetics, leading to potential high drug exposure. This finding suggests that fusidic acid dosing regimens should be reconsidered.

What is Already Known about this Subject

• Rifampicin represents the key antibiotic for the management of staphylococcal osteoarticular infections.

• An important pharmacokinetic variability has already been described.

• All previous pharmacokinetic modelling studies focused only on patients treated for tuberculosis.

What this Study Adds

• We developed the first population pharmacokinetic model of rifampicin in patients with osteoarticular infections.

• Rifampicin pharmacokinetic parameters were in line with those previously reported in patients treated for tuberculosis and presented a wide interindividual variability.

• In association with fusidic acid, we observed a decrease in the apparent clearance and apparent volume of distribution of rifampicin, correlated with potentially supratherapeutic concentrations of rifampicin.

Tables of Links

TARGETS
Enzymes
CYP 3A4

LIGANDS
Rifampicin

These Tables list key protein targets and ligands in this article that are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY 1, and are permanently archived in the Concise Guide to PHARMACOLOGY 2015/16 2.

Introduction

The management of osteoarticular infections (OAIs) remains a difficult challenge, notably due to a lack of standardization of antibiotic choice, dosage schedule and treatment duration 3, 4, 5. According to the French Infectious Diseases Society (SPILF) and the Infectious Diseases Society of America recommendations, rifampicin in combination therapy represents a major antibiotic in OAI treatment, especially for infections associated with the presence of prosthetic materials 6, 7. Indeed, rifampicin is characterized by having bactericidal activity against susceptible strains such as Staphylococcus aureus and coagulase‐negative staphylococci 8, a very good bioavailability and a good diffusion in bone tissue and biofilms 9, 10. However, rifampicin pharmacokinetics is also characterized by an important interindividual variability, which has been well described and modelled in tuberculosis patients 11, 12, 13, 14, 15. Absorption is known to be fairly variable and the rifampicin peak concentration might be affected by food 11, 14. Hepatic metabolism is the primary elimination pathway for rifampicin, with autoinduction of its own metabolism 16, 17. The drug is metabolized mainly by deacetylation and hydrolysis 18, and rifampicin excretion occurs mainly through the biliary route. Moreover, the pharmacokinetics of rifampicin is known to be nonlinear when the dose is increased beyond that recommended currently 19. This observed pharmacokinetic variability may affect drug disposition and explain sub‐ or supratherapeutic concentrations, leading to an incomplete treatment response, an increased risk of emergence of drug resistance or an exacerbation of toxicity. Thus, it seems essential better to characterize rifampicin pharmacokinetic parameters and variability, to improve infection management 20. Nonlinear mixed‐effect model methodology was adapted to determine the interindividual variability in pharmacokinetics, accounting for both fixed and random effects. Fixed effects are population parameters considered as constant, while random effects refer to unexplained variation. In addition, a nonlinear mixed‐effect model allows the utilization of both intensive and sparse data.

The aim of the present study was to develop a rifampicin population pharmacokinetic model using sparse data in adult patients with OAI, in order to characterize rifampicin pharmacokinetic parameters and to identify any contributing covariate factors. The results from this study could be applied to optimize the current OAI therapy.

Materials and method

Data sources

Plasma rifampicin concentrations were collected retrospectively from adult outpatients followed up in the Infectious and Tropical Diseases Unit at Conception Hospital (Marseille, (France) between August 2012 and August 2015. Patients with OAI were treated with rifampicin for 3 months or 6 months (for infections associated without/with the presence of prosthetic materials, respectively), at a dose of 300 mg three times a day, administered by the oral route. Determination of the plasma rifampicin concentration was requested by clinicians to assess treatment adherence, to detect absorption impairment or in the case of suspected toxicity. The following therapeutic drug monitoring data were collected from the database of the Clinical Pharmacological Unit of Timone University Hospital (Marseille, France): patient identification, plasma rifampicin concentrations, along with associated information such as the treatment regimen and sampling times. Only plasma rifampicin concentrations measured at steady‐state (i.e. at less than 4 weeks after the start of rifampicin treatment), were analysed. Clinical and biological data were recovered from patient medical records, including age, gender, weight, height, type of infection, type of prosthetic materials, suspected bacteria and antibiotics taken in addition to rifampicin antibiotics. Information concerning whether or not food was ingested during drug intake was not available for all patients.

Analytical method

Rifampicin was quantified in the plasma by a validated method, using high‐performance liquid chromatography with ultraviolet detection. The extraction procedure was carried out using acetonitrile, with diazepam as the internal standard. Chromatography was performed on a C18 column. Elution was achieved isocratically with potassium buffer/acetonitrile at a flow rate of 1.0 ml min^–1. The method was linear in the 0.5–20 μg ml^–1 range. Accuracy and precision were assessed at four concentrations – the lower limit of quantification (LLOQ), low‐quality control (LQC), mid‐quality control (MQC) and high‐quality control (HQC) – and the method presented provided adequate results according to European Medicines Agency guidelines 21. Intraday accuracy and precision ranged from 101.6% to 111.0% and from 1.7% to 3.6%, respectively. Interday accuracy and precision ranged from 103.9% to 108.0% and from 0.8% to 3.1%, respectively. Stability studies were assessed at three different concentrations (LQC, MQC and HQC), in duplicate. The bench‐top stability of the samples at room temperature over the course of 5 h (the maximum time for samples to reach the laboratory following collection) showed no relevant changes in the signal intensity of target compounds, with concentration differences ranging from −7.7% to 2.2%. After collection, samples were centrifuged immediately and then stored at −20°C until quantification, which was performed within 8 days. Rifampicin concentrations after 1 month at −20°C showed no significant deviations from the nominal concentrations (ranging from 5.3% to 9.4%).

Population pharmacokinetic analysis

Population pharmacokinetic analysis was performed using a nonlinear mixed‐effects model, with NONMEM version 7.3 software (ICON Development Solutions, ICON plc South County Business Park Leopardstown Dublin 18 Ireland) 22. Data were analysed using a first‐order conditional estimation method. The R version 3.1.2 software (www.r‐project.org) was used for goodness‐of‐fit diagnostics and graphical displays. An initial analysis was performed to identify the base model that best described the data. One‐ and two‐compartment models with linear and nonlinear absorption and elimination were tested. Models incorporating lag times or transit compartments were fitted to the data during the initial stage of model building. The absorption model was subsequently modified as described later. Owing to the auto‐induction phenomenon observed in rifampicin pharmacokinetics, the well‐stirred liver model described by Chirehwa et al. 23 was tested. Elimination was assumed to take place from the central plasma compartment in all models tested. Linear and nonlinear elimination (Michaelis–Menten) were tested. Models were parameterized by clearance [apparent clearance (CL)/F, in l h^–1, where F is unknown bioavailability] and apparent volume of distribution (V/F, in l). Interindividual variability was tested for each pharmacokinetic parameter, to determine any covariance. An exponential variance model was used to describe the pharmacokinetic variability across individuals, in the form: P_i = θ_k*exp(η_ki), in which P_i is the estimated parameter value for the individual subject i; θ_k is the typical population value of parameter k; and η_ki are the interindividual random effects for individual i and parameter k. Various estimation methods were compared. Different error models were tested to describe the residual unexplained variability of the model. Additive, proportional and combined (additive and proportional) residual error models were considered during the model‐building process. Ultimately, an additive error model was used to model random residual variability according to: C_ij = C_Pij + ε_ij, in which C_ij is the observed concentration j in individual i; C_Pij is the individual predicted concentration; and ε_ij is the additive residual random error. Subsequently, the developed model was used to investigate if covariates could improve the fit. The influences of the following covariates on the pharmacokinetic parameters were then tested: demographic data (age, weight, height and gender), pathology (suspected bacteria, type of infection, type of prosthetic materials) and antibiotics taken in addition to rifampicin antibiotics (ofloxacin, fusidic acid, clindamycin, teicoplanin, ciprofloxacin, vancomycin, amoxicillin, cotrimoxazole, ceftazidime, cloxacillin). Covariates were selected and retained in the population model if: (i) their effect was biologically plausible; (ii) they produced a minimum reduction of 4 units in the objective function value (OFV); and (iii) they produced a reduction in the variability of the pharmacokinetic parameter, assessed by the associated intersubject variability (ISV). All covariates identified as potentially relevant were included in a full model. The final model was then obtained by dropping all nonsignificant covariates –that is, producing a reduction of less than 8 units in the OFV. Graphical plots were used to assess the quality of fits.

Absorption, which a graphical inspection of the data revealed to be highly variable, was modelled using transit compartments. Transit compartments were used to mimic a delay in absorption onset followed by a gradual increase in absorption rate, in a more physiologically plausible manner than that offered by the use of lag times. Drug transfer from the final transit compartment to the central compartment occurred through an absorption compartment, according to the first‐order rate constant ka.

The internal model evaluation included the inspection of goodness‐of‐fit plots (plotting observed vs. predicted concentration and individual predicted concentration). Estimated parameters were checked for plausibility and precision (using the standard error) and simulation‐based evaluations were performed using a bootstrap analysis, a normalized prediction distribution error (NPDE) and a prediction‐corrected visual predictive check (VPC). Bootstrap procedures were performed using Wings for NONMEM (www.WFN.sourceforge.net) to evaluate the 95% confidence interval nonparametrically. Bootstrapping is a method of resampling with a replacement, and has the advantage of using the entire dataset. One thousand bootstrapped data sets were generated by resampling subjects from the original data set with replacements. These data sets were analysed using the final model described previously. Finally, the 2.5th and 97.5th percentiles of the parameter estimates were obtained, to construct the confidence intervals. NPDE was performed using the add‐on software package that was run in R 24. The NPDE takes into account the full predictive distribution of each individual observation and handle multiple observations within subjects. Under the null hypothesis that the model under scrutiny describes the validation dataset, the NPDE should follow the standard normal distribution. Five hundred model‐predicted concentrations were generated for each observation with the parameter values that were obtained for the final model. The observed concentrations were subsequently compared with these 500 predicted concentrations. Bayesian forecasting was performed to estimate individual rifampicin pharmacokinetic parameters and peak concentrations (2 h after administration, as recommended for rifampicin therapeutic drug monitoring) 11.

Ethical issues

Ethical approval was granted by the local ethics committee at Aix Marseille Université (n° 2016–06–07‐06).

Results

Sixty‐two patients were included in the study, corresponding to 103 rifampicin concentration observations. The clinical characteristics of the patients are presented in Table 1. Patients were mainly being treated for infections associated with the presence of prosthetic materials infections (n = 47, 75.8%). The main suspected bacteria were Staphyloccoccus aureus (n = 39, 62.9%), followed by coagulase‐negative staphylococci (n = 22, 35.5%). The main antibiotics taken in addition to rifampicin antibiotics were ofloxacin (oral route, 200 mg three times daily) and fusidic acid (oral route, 500 mg three times daily), prescribed for 32 (51%) and 10 (16.1%) patients, respectively.

Table 1.

Patient characteristics used for covariate model development

Variable	n (range or %)
Demographic data
No. of patients	62
Men/women	46/16
Age (years)	57.4 (20–89)
Weight (kg)	72.3 (46–119)
No. of samples	103
Type of prosthetic materials
Prosthesis	32 (51.6)
Osteosynthesis	16 (25.8)
No material	13 (21.0)
Not informed	1 (1.6)
Suspected bacteria
Staphylococcus aureus	40 (64.5)
Coagulase‐negative staphylococci	22 (35.5)
Streptococcus spp.	4 (6.5)
Other bacteria	6 (9.7)
Antibiotic coadministration
Ofloxacin	32 (51.6)
Fusidic acid	10 (16.1)
Clindamycin	6 (9.7)
Teicoplanin	6 (9.7)
Ciprofloxacin	5 (8.1)
Vancomycin	5 (8.1)
Amoxicillin	5 (8.1)
Cotrimoxazole	2 (3.2)
Ceftazidime	2 (3.2)
Cloxacillin	1 (1.6)

In the final model, a one‐compartment model with first‐order elimination best described the plasma rifampicin concentration–time data (Figure 1). Two‐compartment models provided no advantage in terms of improvement in diagnostic plots or change in OFV, and were removed early in the model‐building process. A model with Michaelis–Menten elimination and a well‐stirred liver model as described by Chirehwa et al. 23 provided no improvement in our studied population. Rifampicin oral absorption described by one transit compartment proved superior to the other oral absorption models (zero‐ and/or first‐order absorption models or a lag‐time model). The structural model was parameterized in terms of absorption rate constant (ka), CL/F of rifampicin and the V/F of the central compartment of rifampicin. The absorption rate was fixed at 1.15 h⁻¹ in our study. The model included interindividual variability in CL/F and V/F. The residual unexplained variability of rifampicin, as selected by goodness‐of‐fit plots, weighted residuals (WRES) vs. time and decrease in OFV, was best described by an additive model on natural log‐transformed data. The potential effects of demographic and clinical covariates were evaluated systemically in a univariate analysis. Only one covariate altered the OFV significantly. Fusidic acid coadministration was considered as a significant covariate and was added in the clearance and volume estimation equation (Table 2). The final pharmacokinetic parameters (CL/F and V/F) were similar to those in the base model except for patients with fusidic acid coadministration (Table 2). The goodness‐of‐fit plots (Figure 2), NPDE (Figure 3) and VPC (Figure 4) for the final model show that model predictions were in reasonable agreement with the observed plasma concentrations of rifampicin. The final model was also validated internally by means of 1000 bootstrap runs (Table 2), which were successful in 98% of the runs and confirmed the parameter values. Figures 5 and 6 show estimated individual pharmacokinetic parameters and estimated concentrations using Bayesian forecasting for rifampicin administration alone and concomitantly with fusidic acid, respectively.

Figure 1.

Representation of the final model for rifampicin pharmacokinetics. CL, apparent clearance; ka, absorption constant; V, volume of distribution

Table 2.

Final population pharmacokinetic parameter estimates

	Estimate		Bootstrap
Parameter	Mean	RSE%	Mean	95% CI
CL/F (l h –1 )
With fusidic acid	5.1	48.6	4.6	1.2, 8.2
Without fusidic acid	13.7	26.3	13.8	10.6, 18.0
V/F (l)
With fusidic acid	23.8	78.2	40.0	8.9, 38.7
Without fusidic acid	61.1	56.5	71.2	40.8, 129.0
Interindividual variability
ωCL/F	0.531	38.3	0.517	0.245, 0.740
ωV/F	0.349	238.1	0.237	0.003, 1.110
Residual variability
σadd (mg l–1)	2.256	12.4	2.174	1.613, 2.649

Add, additive; CI, confidence interval; CL/F, apparent clearance of rifampicin; RSE, relative standard error; V/F, apparent volume of distribution of the central compartment of rifampicin; ω, variance of ηki; σ, variance of εij

Figure 2.

Diagnostic plots of the final rifampicin population pharmacokinetic model. Observed rifampicin concentration (DV) vs. population predictions (PRED) and vs. individual predictions (IPRED). Solid line: identity line; dashed line: trend of fit; dotted line: regression line

Figure 3.

Normalized prediction distribution error (NPDE) vs. time after dose (TAD) and predicted concentration (PRED)

Figure 4.

Prediction‐corrected visual predictive check. Solid lines: the 5th, 50th and 95th percentiles of the distribution of the observations; dashed lines: confidence bands around the percentiles for simulated predictions; points: observed data

Figure 5.

Box plots of rifampicin clearance (A) and volume of distribution (B) when this drug is administrated alone or concomitantly with fusidic acid in the final model. Fusidic acid (0 = absence, 1 = presence)

Figure 6.

Bayesian forecasting of individual concentrations 2 h after administration. Fusidic acid (0 = absence, 1 = presence)

Discussion

We developed the first population pharmacokinetic model of rifampicin in patients with OAI. A one‐compartment model coupled to a transit absorption model best fitted the rifampicin data well.

The oral absorption of rifampicin was best described using a transit compartment model with one transit compartment. As in the study by Wilkins et al. 14, our attempt to fit a transit compartment absorption model with an estimated number of transit compartments led to model overparameterization and resulted in a high relative standard error percentage in the estimation of parameters. A two‐transit‐compartment model was also tested, as in the study by Seng et al. 25. However, in our study the addition of a second transit compartment did not lead to a better estimation of parameters, probably due to insufficient blood sampling prior to the attainment of the maximum postdose concentration. A model with one transit compartment resulted in a structural model presenting a lower OFV compared with a model with an absorption lag‐time parameter, as reported in a previous study 26. Indeed, the lag‐time parameter, which reproduces an instantaneous shift from no absorption to maximal absorption, is physiologically implausible and associated with computational difficulties. As partial derivatives are defined for the predicted concentration from the entire absorption profile, the transit compartment model offers advantages from a numerical point of view 27. The autoinduction phenomenon was not incorporated into our model. This difference with the model of Chirehwa et al. 23 is probably explained by the collection of plasma rifampicin concentrations only in steady‐state conditions (i.e. at less than 4 weeks after the start of treatment) in the present study. Moreover, nonlinear elimination was not included in our model, probably due to the low doses of rifampicin used in the study (median: 12.4 mg kg^–1). Indeed, nonlinear pharmacokinetics for rifampicin has been observed with doses from 20 mg kg^–1 23.

Rifampicin CL/F and V/F estimated parameters were in line with those previously reported in patients treated for tuberculosis 14, 25, 26, 28. Moreover, these parameters were highly variable and confirmed the wide interindividual pharmacokinetic variability of rifampicin 14, 15. Rifampicin CL/F and V/F were correlated with the coadministration of fusidic acid (Figure 4). Patients cotreated with fusidic acid presented an important decrease in the CL and V/F of rifampicin: 5.1 l h^–1 vs. 13.7 l h^–1 and 23.8 l vs. 61.1 l, respectively (Table 2). Fusidic acid is orally bioavailable, extensively bound to albumin (about 97%) 29, 30, 31 and extensively metabolized, with metabolites predominantly eliminated by biliary excretion. Moreover, fusidic acid exhibits complex and nonlinear pharmacokinetics. Previous studies have demonstrated a decrease in apparent total clearance after multiple dosing regimens at intermediate and high doses, related to fusidic acid accumulation 32, 33, 34. The rifampicin–fusidic acid pharmacokinetic interaction has not been described previously. However, previous studies have reported drug–drug interactions with fusidic acid, suggesting an inhibition of the cytochrome P450 3A4 (CYP3A4) enzyme. Khaliq et al. 35 observed elevated concentrations of ritonavir and saquinavir after the addition of fusidic acid, and further studies reported rhabdomyolysis in a patient treated with atorvastatin or simvastatin, after fusidic acid coadministration 36, 37. The inhibition of CYP enzymes is an important source of drug–drug interactions and can result in pronounced changes in drug pharmacokinetics. Although the hepatic metabolism of rifampicin is principally due to deacetylation and hydrolysis 18, an inhibition of metabolism induced by fusidic acid could potentially explain the observed modifications in the CL of rifampicin. Moreover, Prakash et al. 38 observed an increase in rifampicin oral bioavailability, according to a demonstrated by a decreased expression of CYP3A4, suggesting a role for CYP3A4 in the presystemic biotransformation of rifampicin. However, Reimann et al. 39 found a contradictory result, showing a time‐dependent activating effect of fusidic acid on the CYP450 enzyme system. To date, the actual explanation for the effect of fusidic acid on rifampicin exposure remains unclear.

As described in Figure 5, an association between rifampicin and fusidic acid could lead to potentially supratherapeutic plasma rifampicin concentrations. These results should be taken into account in clinical practice, and clinical monitoring should be strengthened. Moreover, the therapeutic drug monitoring of rifampicin, currently recommended in tuberculosis patients when there is a suspicion of a drug–drug interaction 11, might be helpful to ensure an effective concentration and detect high drug exposure in patients with fusidic acid coadministration.

The other covariables tested on the model, including gender, age, height and body weight, did not improve the pharmacokinetic model. These results were in accordance with those described previously in other populations 14, 25, 28. Indeed, in tuberculosis patients, only gender was retained as a covariable in only one study 15. Our data did not highlight any influence of gender on the pharmacokinetic parameters of rifampicin, similarly to the study by Wilkins et al. 14.

The present study had several limitations. Firstly, a limited number of rifampicin plasma concentration–time data points were used to build the model. The difficulty in estimating the absorption rate and the interindividual variability of the volume of distribution can probably be explained by the timing of sample collection. In the presented model, the absorption rate constant was fixed at 1.15 h⁻¹ to improve the fit, as presented in previous studies 14, 28. Interindividual variability was not estimated reliably for the volume of distribution (relative standard error = 238.1%). However, this variability was still retained in the final model to improve parameter estimation. The study was based on routine therapeutic drug monitoring samples collected retrospectively, with insufficient data adequately to describe the absorption and distribution phases. Consequently, further prospective studies, based on full‐profile or sparse‐sampling designs, are necessary adequately to elucidate rifampicin pharmacokinetics in this population.

The second limitation of the study was the lack of clinical response data. Unfortunately, data concerning clinical response (efficacy, tolerability) were not available in the study. It would be interesting to assess the difference between patients with higher vs. lower rifampicin exposures, in the presence of fusidic acid coadministration.

Conclusion

We developed a structural pharmacokinetic model that adequately described the disposition of rifampicin in adult patients treated for OAI. Our analysis highlighted a drug–drug interaction between rifampicin and fusidic acid. This concomitant administration induced a significant increase in rifampicin CL/F and V/F, leading to potentially supratherapeutic rifampicin concentrations. Thus, therapeutic drug monitoring of rifampicin could be recommended in cases where there is a suspicion of a drug–drug interaction. Although the interindividual variability in rifampicin pharmacokinetic parameters was important, our model provided predictability of drug exposure and allowed dosage regimens to be tailored to the needs of individual adult patients with OAI. Nevertheless, additional studies will be crucial better to characterize the variability of pharmacokinetic parameters of rifampicin in patients treated for OAI and identify precisely the impacts on clinical outcome in this indication.

Competing Interests

There are no competing interests to declare.

The authors would like to thank Laurence Delaunay and Laurent Allanioux for their excellent technical assistance.

Contributors

A.M., J.D. and R.G. analysed the data. A.M performed population pharmacokinetic analysis. A.M, J.D., C.M. and R.G. wrote the reply to reviewers and the revised manuscript. A.M. recruited the patients. A.M., J.D., A.M. and R.G. wrote the manuscript. O.B. validated the manuscript. R.G. conceived and designed the study and supervised the work. All authors read and approved the final manuscript.

Marsot, A. , Ménard, A. , Dupouey, J. , Muziotti, C. , Guilhaumou, R. , and Blin, O. (2017) Population pharmacokinetics of rifampicin in adult patients with osteoarticular infections: interaction with fusidic acid. Br J Clin Pharmacol, 83: 1039–1047. doi: 10.1111/bcp.13178.