Population Pharmacokinetic Model and Optimal Sampling Strategies for Micafungin in Critically Ill Patients Diagnosed with Invasive Candidiasis

ABSTRACT

Candida bloodstream infections are associated with high attributable mortality, where early initiation of adequate antifungal therapy is important to increase survival in critically ill patients. The exposure variability of micafungin, a first-line agent used for the treatment of invasive candidiasis, in critically ill patients is significant, potentially resulting in underexposure in a substantial portion of these patients. The objective of this study was to develop a population pharmacokinetic model including appropriate sampling strategies for assessing micafungin drug exposure in critically ill patients to support adequate area under the concentration-time curve (AUC) determination. A two-compartment pharmacokinetic model was developed using data from intensive care unit (ICU) patients (n = 19), with the following parameters: total body clearance (CL), volume of distribution of the central compartment (V1), inter-compartmental clearance (CL12), and volume of distribution of the peripheral compartment (V2). The final model was evaluated with bootstrap analysis and the goodness-of-fit plots for the population and individual predicted micafungin plasma concentrations. Optimal sampling strategies (with sampling every hour, 24 h per day) were developed with 1- and 2-point sampling schemes. Final model parameters (±SD) were: CL = 1.03 (0.37) (L/h/1.85 m²), V1 = 0.17 (0.07) (L/kg LBMc), CL12 = 1.80 (4.07) (L/h/1.85 m²), and V2 = 0.12 (0.06) (L/kg LBMc). Sampling strategies with acceptable accuracy and precision were developed to determine the micafungin AUC. The developed model with optimal sampling procedures provides the opportunity to achieve quick optimization of the micafungin exposure from a single blood sample using Bayesian software and may be helpful in guiding early dose decision-making.

INTRODUCTION

Candida bloodstream infections are associated with an increased length of hospital stay in the intensive care unit (ICU), and the attributable mortality is up to 40% (1). Several studies demonstrated significantly higher survival rates among patients with invasive candidiasis in whom adequate antifungal therapy was promptly started (2–7). Once a bloodstream infection with a Candida species has been diagnosed, guidelines recommend immediate therapy initiation with an echinocandin (8, 9). The efficacy of micafungin is concentration-dependent and mainly related to the ratio of the area under the plasma concentration-time curve over 24 h over the MIC of the microorganism (area under the concentration-time curve [AUC]_0-24h/MIC) (10). An AUC/MIC target of 3000 for micafungin has been suggested for optimal antifungal treatment (11). Despite this target, registered dosages of micafungin are frequently associated with lower drug exposure in ICU patients than in non-ICU patients, resulting in AUC_0-24h/MIC ratios below the target (12–16). In a study comparing efficacy of micafungin in ICU and non-ICU patients, a significantly lower treatment success rate was seen in ICU patients compared with non-ICU patients (17). Since the AUC/MIC is the best predictive parameter for the mycological response (10, 11), therapeutic drug monitoring (TDM) may be a valuable tool to detect drug low exposure potentially requiring dosage adjustment to optimize antifungal treatment (18). Obtaining a full concentration versus time curve to determine the AUC is burdensome for patients and staff, time-consuming and expensive. Therefore, optimal sampling strategies using a limited number of optimally timed samples in concert with a population pharmacokinetic model and Bayesian modeling make AUC guided TDM possible in daily practice (18, 19). The objective of this study was to develop a population pharmacokinetic model including appropriate sampling strategies for assessing micafungin drug exposure in critically ill patients diagnosed with invasive candidiasis to support adequate AUC determination.

RESULTS

In total, 165 samples from 19 patients were used for the development of the population pharmacokinetic model. From each patient, 9 to 11 samples were drawn within 24 h after 4 days (±1 day) of dosing. The demographics of these patients are summarized in Table 1.

TABLE 1.

Patient characteristics and clinical variables^a

Characteristics	Total (N = 19)
Gender (female, no. [%])	9 (47)
Age (yrs, median [IQR])	64 (57–73)
Race (Caucasian, no. [%])	17 (89)
Wt (kg, median [IQR])	85 (65–98)
BMI (median [IQR])	27.5 (22.7–33.9)

^aBMI = Body Mass Index.

Population pharmacokinetic model.

All parameters were fixed in the base model and resulted in an Akaike Information Criterion (AIC) of 100091. After a stepwise approach switching each pharmacokinetic parameter from fixed to Bayesian, the model was obtained with all parameters set on Bayesian and an AIC of 3282, depicted in Fig. 1. Covariates were applied to the model (body weight normalized to 70 kg [with and without an allometric scaling exponent of 0.75]), lean body mass (LBMc), and body surface area (BSA). LBMc applied to V and BSA applied to clearance (CL) statistically improved the model and was included, resulting in a final model with an AIC of 3269.55.

FIG 1.

Obtainment of final model based on the AIC.

Evaluation of the pharmacokinetic model.

The results of the bootstrap analysis of the final population pharmacokinetic model are shown in Table 2. Figures 2A and B show the goodness-of-fit plots for the population and individually predicted micafungin plasma concentrations. The data points in the individually predicted concentrations in Fig. 2B are closer distributed along the line of identity compared to the population predicted concentrations in Fig. 2A.

TABLE 2.

Final population pharmacokinetic parameters and results of bootstrap analysis

	Model (AICa = 3269.55)	Bootstrap (N = 1000)
Parameter	Mean ± SDb	Mean (95% CIc)	SD (95% CI)
CLd (L/h/1.85 m2)	1.03 ± 0.37	0.98 (0.87–1.10)	0.30 (0.19–0.38)
V1e (L/kg LBMc)	0.18 ± 0.07	0.19 (0.16–0.23)	0.09 (0.04–0.12)
CL12f (L/h/1.85 m2)	1.80 ± 4.07	1.71 (0.56–4.01)	3.76 (1.44–6.29)
V2g (L/kg LBMc)	0.12 ± 0.06	0.13 (0.11–0.20)	0.06 (0.04–0.15)

^aAIC = akaike information criterion.

^bSD = standard deviation.

^cCI = confidence interval.

^dCL = total body clearance.

^eV₁ = volume of distribution of the first compartment.

^fCL₁₂ = inter-compartmental clearance.

^gV₂ = volume of distribution of the second compartment.

FIG 2.

(A) Population predicted micafungin plasma concentration based on the final model versus the actually measured micafungin plasma concentration. (B) Individual predicted micafungin plasma concentration based on the final model versus actually measured micafungin plasma concentration. (C) Weighted residuals of the population predicted micafungin plasma concentration versus the predicted micafungin plasma concentration. (D) Weighted residuals of the population predicted micafungin plasma concentration versus the time.

Sampling strategies using the Bayesian approach.

Optimal sampling strategies for micafungin were developed for determination of the AUC in steady state. The results for1 or 2 sampling time points within a 24-h window and an interval of at least 1 h (for 2 sampling points) are shown in Table 3. All 3 best performing sampling strategies for 1 or for 2 time points met the predetermined acceptance criteria for precision (root mean square error [RMSE]) < 15.0% and bias (mean prediction error [MPE]) < 5.0%. A two-time point optimal sampling scheme with sampling at 2 and 13 h after micafungin administration gives an excellent prediction of the micafungin AUC in steady state. A single time point limited sampling scheme with sampling at 8 h after micafungin administration gives a good prediction of the micafungin AUC in steady state. For adequate AUC determination with trough-peak sampling with acceptable precision and bias, at least 2 samples are required. A trough sample with an additional peak sample at 1, 2, or 3 h after start of micafungin infusion also shows good prediction of the AUC in steady state based on the RMSE and MPE as shown in Table 3.

TABLE 3.

Sampling strategies for micafungin

Sampling time point(s) (h)	MPEa (%)	RMSEb (%)	r2c
Optimal sampling strategies
8	4.7	8.8	0.9493
9	4.2	9.0	0.9434
7	3.8	9.1	0.9407
2, 13	0.1	1.8	0.9972
3, 15	−0.1	1.8	0.9972
3, 14	0.3	1.9	0.9966
Conventional sampling strategies (trough - peak sampling)
0	3.7	17.9	0.7639
0, 3	−1.0	5.3	0.9793
0, 2	−1.3	7.2	0.9608
0, 1	−1.5	10.4	0.9195

^aMPE = mean prediction error.

^bRMSE = root mean squared error.

^cr² = coefficient of determination.

DISCUSSION

In this study, we developed a two-compartment population pharmacokinetic model for micafungin in ICU patients and optimal sampling strategies using this pop-PK model and Bayesian approach for micafungin that can be used for optimization of micafungin therapy. The final model predicts the micafungin plasma concentrations in ICU patients adequately based on the goodness-of-fit plot of the individually predicted micafungin concentrations. The bootstrap analysis showed that the final model is robust. This pharmacokinetic model is able to determine the micafungin exposure with acceptable accuracy and precision from a single blood sample drawn 8 h after dosing. However, to increase the precision of the micafungin AUC determination, 2-point sampling could be used. Additionally, trough-peak sampling was tested and sampling at 0 h (pre micafungin administration) and 2 h after micafungin administration gives similar results as for 1-point sampling at 8 h, which could be beneficial in the daily setting for logistical reasons. The proposed sampling strategies can be advantageous in daily clinical practice and in specific clinical situations to determine the micafungin exposure. Critically ill patients often have pathophysiological or iatrogenic conditions resulting in inter individually highly variable pharmacokinetics (20), therefore the per SPC recommended dosing regimen might not achieve the expected drug exposure in every patient. A wide variety of factors with potential impact on micafungin exposure has been proposed, such as increased bodyweight, decreased plasma proteins, higher disease severity score, renal failure and renal replacement therapy, and liver impairment (21, 22). Due to the relatively small number of patients included in this study and the inability to include a wider range of covariates in the MW/Pharm software, some of these factors were not incorporated in the model, what could be considered a limitation. As it is common for ICU patients to have multiple risk factors at the same time which makes predicting drug exposure challenging, as the pharmacokinetic behavior of micafungin may vary in these patients. The developed two-compartment population pharmacokinetic model with optimal sampling strategies provides clinicians in the ICU an easy option for pharmacokinetic optimization of micafungin in patients having the identified risk factors and being prone to underexposure. Since AUC/MIC targets of 3000, 5000, and 285 for micafungin have been proposed for a general Candida population, a non-C. parapsilosis population and a C. parapsilosis population respectively, the AUC determination using the model can be used to provide a good indication of the effectiveness of the antifungal treatment with micafungin.

Even though therapeutic drug monitoring seems an excellent option to optimize antifungal treatment, it has its challenges. Randomized clinical trials may need to be performed to provide the required evidence to determine the added value of TDM in the clinical setting (23). These studies are important to validate PK/PD targets derived from preclinical studies, including the target AUC/MIC values for micafungin for different Candida populations proposed by Andes et al. (11, 18). To obtain the correct AUC/MIC value for micafungin in a specific patient against a specific Candida spp. accurate AUC and MIC determination is important, but challenging. Biological and methodological differences between laboratories could result in significant variability of the MIC of an antibiotic for a specific pathogen (24). For micafungin, the intra- and interlaboratory agreement in assessing its activity against Candida spp. with different test methods has been demonstrated (25). Assays to determine the AUC of different drugs is only available in a limited number of laboratories and most of these laboratories have limited resources to run multiple assays in parallel. To mitigate these resource limitations, researchers have developed methods to analyze multiple drugs on a single system (26, 27). Although well-designed TDM could be an excellent tool to support personalized dosing, the majority of medical centers lack the analytical methods, equipment and knowledge at this moment. For this reason, a simpler and straight-forward approach has been proposed by studies, based on the pharmacodynamic behavior of micafungin in humans. Standard increased (loading) dosages of micafungin for specific patient populations could prevent significant underexposure in these patients (16, 28). Increased micafungin dosages, up to 8 mg/kg daily, have been tested in studies without evidence for significant adverse events (29, 30). Besides lack of significant toxicity due to its good safety profile, micafungin is not associated with drug-drug interactions through CYP450 (31). Although, an increase of the micafungin dose seems safe based on human data, the development of hepatocellular tumors in rats after 3 months of high dose micafungin treatment has been described, and micafungin is associated with inhibition of transporter proteins (32, 33). Multiple studies concluded that ICU patients were significantly underexposed to micafungin, but reported inter-individual variability in AUC up to 50% (34). An increased micafungin dose for all ICU patients may put some of the patients at risk for overexposure and potentially, adverse events due to the substantial inter-individual variability in this patient population.

Although therapeutic drug monitoring has it challenges, it does provide accurate information on the pharmacokinetics for micafungin in an individual, helpful for treatment optimization in every individual patient (35).

Conclusion.

The developed model with an optimal sampling scheme provides the opportunity to achieve quick optimization of the micafungin exposure from a single blood sample using Bayesian software. Using this model, a single blood sample drawn 8 h after dosing is sufficient to have a reliable assessment of micafungin exposure. This data combined with rapid identification and accurate susceptibility testing (MIC) of Candida species provides all information to individualize the dose. The developed model and sampling procedure may be helpful in guiding early dose decision-making for antimicrobial stewardship teams in hospitals.

MATERIALS AND METHODS

This study is a post hoc analysis of data from a prospective study to assess the micafungin exposure in critically ill patients that was conducted at the University Medical Center Groningen (UMCG) from December 2012 to December 2016 (36). The study was approved by the local Institutional Review Board (2012-189), and all patients or their next of kin provided written, informed consent. Micafungin was administered in a dosage of 100 mg once daily by intravenous infusion over 1 h. Micafungin blood samples were obtained on day 4 (±1 day) after treatment initiation, 1, 2, 3, 4, 6, 8, 12, and 24 h after the start of the infusion.

Population pharmacokinetic model.

Plasma concentrations of micafungin were determined using a validated assay where the residual error was equal to the SD of the assay which was calculated as 0.0095 + 0.015 × C, where C is the observed micafungin plasma concentration (mg/L) (37). Pharmacokinetic parameters were assumed to be log-normally distributed. A population pharmacokinetic model was developed using the iterative two-stage Bayesian procedure of the KinPop module of MWPharm (version 3.82; Mediware). Since micafungin behaves according to a 2-compartment model, 2-compartment models were developed based on population pharmacokinetic data from Maseda et al., as the data were derived from an ICU population and a 2-compartment model was used (38). Modeling was performed with the following estimated pharmacokinetic parameters: total body clearance (CL), volume of distribution of the central compartment (V1), inter-compartmental clearance (CL12), volume of distribution of the peripheral compartment (V2). The base pharmacokinetic parameters from the model developed by Maseda et al. were CL of 0.88 (± 0.27) L/h, V1 of 10.86 ( ± 2.84) L, and CL12 of 5.07  (± 0.00) L/h and V2 of 10.94  (± 2.66) L (38). Modeling was initiated without covariates and with all parameters fixed to these literature values where the best model was selected based on the lowest AIC value. In a stepwise approach, each parameter was Bayesian estimated, and each step was evaluated by calculation of the AIC. Each parameter was changed from a fixed value to a Bayesian estimation if the AIC value was significantly lower than the model with the parameter fixed. A reduction of the AIC with at least 3 points was considered a significant improvement of the model (39). If the model did not improve, i.e., obtainment of a significantly lower AIC value, the parameter remained fixed. These steps were repeated in the next cycle using previous population parameters until the lowest AIC was determined with population parameters best fitting the data. Only a limited set of covariates was tested due to the small sample size including bodyweight, lean body mass, fat free mass, and body surface area.

Evaluation of the pharmacokinetic model.

To validate the model, a bootstrap analysis with 1000 replicate sets of the population was performed to determine the robustness of the final model (39). The estimates were tabulated and the lower 2.5% and upper 97.5% value of each parameter were estimated to obtain the 95% confidence interval. The goodness-of-fit of the population predicted and individual micafungin plasma concentrations was determined. The population predicted micafungin plasma concentration based on the final model was compared to the measured micafungin plasma concentration, and the individual predicted micafungin plasma concentration based on the final model was compared to the measured micafungin plasma concentration. The population predicted concentrations were calculated by fixing all parameters to the final model parameters. The individual predicted concentrations were calculated using the KinPop module set to one cycle to prevent the software to change the population parameters. Goodness-of fit plots were created for the population and individual predicted micafungin concentration against the measured micafungin concentrations. In addition, the weighted residuals were plotted against the predicted concentrations and time.

Sampling strategies using the Bayesian approach.

Using Monte Carlo simulation with the limited sampling module of MW\Pharm (version 3.83; Mediware), 1,000 virtual pharmacokinetic profiles were created to represent the pharmacokinetic data used in the development of the optimal sampling strategies. The reference patient for the Monte Carlo simulation was selected based on average patient characteristics in our study, and a 71-year-old male with a bodyweight of 85 kg and a height of 1.70 m was chosen. Optimal sampling time points were assessed for exposure determination during steady state (day 4 of micafungin treatment) (32). The sampling time points were evaluated using acceptance criteria for precision and bias (RMSE < 15.0%, MPE < 5.0%) and for clinical feasibility (19). For sampling it was assumed that 24 h per day (every hour) sampling is feasible on an ICU and a maximum of 2 samples per AUC determination is possible since these critically ill patients were likely to receive an indwelling catheter. Two limited sampling strategies were tested, 1-point sampling and 2-point sampling. For the 2-point sampling, a maximum of 8 h apart was chosen for practical purposes. For comparison, traditional trough and trough-peak sampling was also tested.

All statistical analyses were performed using SPSS version 20.0 (SPSS, Inc.) or as part of the MWPharm population analysis. A P-Value < 0.05 was considered statistically significant.