Population Pharmacokinetic Model and Pharmacokinetic Target Attainment of Micafungin in Intensive Care Unit Patients

This study was conducted in compliance with the Declaration of Helsinki and approved by the local medical ethics committee (Commissie Mensgebonden Onderzoek region Arnhem-Nijmegen number, clinical trials.gov NCT01783379). Informed consent was given by all participants.

All data were collected in a prospective PK study published previously [7]. Patients admitted to the ICU and receiving micafungin for suspected or proven fungal infection were eligible if they met the following inclusion criteria: ≥18 years of age on the day of the first micafungin dose, not receiving micafungin treatment for >2 days before enrolment and having an indwelling central venous or arterial catheter. Exclusion criteria included patients with a history of hypersensitivity to echinocandins or excipients similar to those found in the micafungin preparation, human immunodeficiency virus or hepatitis B/C infection, or abuse of alcohol or drugs. All participants received micafungin 100 mg QD. Micafungin was administered intravenously over approximately 1 h. Micafungin therapy was continued as long as was considered clinically indicated, but the duration of the study was limited to a maximum of 14 days with an additional 3 days after the end of therapy.

Demographic information was gathered and included age, sex, race, weight (available from the medical record previous to ICU admission, estimated or weighted on the ICU), height, relevant co-medication, indication for micafungin use, clinical characteristics, chemistry and haematological parameters. In addition, Acute Physiology and Chronic Health Evaluation II, Sequential Organ Failure Assessment (SOFA), Child-Pugh score and (type of) renal replacement therapy were recorded.

At day 3 (±1) patients were intensively sampled (2 mL) at t = 0 (pre-dose) and 0.5, 1, 2, 4, 6, 8, 12, 18 and 24 h post start infusion during a dosing interval. At day 7 (±1), a second PK curve of six samples was drawn at t = 0, (pre-dose) and 1, 4, 8, 12 and 24 h post start infusion. Additional daily trough concentrations were drawn daily until 3 days after cessation of micafungin therapy. A validated ultra-performance liquid chromatography-fluorescence assay was used to analyse the samples, details have been described previously [7].

Population PK analysis of micafungin was performed with non-linear mixed-effects modelling using the first-order conditional estimation method with interaction between random effects and residual variability as implemented in NONMEM (version 7.3; ICON Development Solutions, Ellicott City, MD, USA) [11]. Pirana interface (PiranaJS, version 1.01) (Pirana Software & Consulting BV 2016, Amsterdam, The Netherlands) [12, 13] was used as an interface for NONMEM, Perl-Speaks-NONMEM (version 4.2.0), Xpose (version 4.5.3) and R (version 3.2.1) [14].

As body weight (BW) is a known confounder for pharmacokinetics, the impact of BW on the pharmacokinetics of micafungin were accounted for by allometric scaling, as proposed previously, with an allometric exponent of 0.75 for all flow parameters and an exponent of 1 for all volume parameters standardised to a 70-kg patient as proposed previously [15‐18]. One-, two- and three-compartment models were considered to describe the drug disposition. The primary PK parameters were clearance (CL) and volume of distribution (V). Inter-individual variability (IIV) and inter-occasion variability (IOV) was estimated using an exponential model. Residual variability was evaluated by proportional, additive and combined additive, and proportional models. Inter-individual random effects were evaluated with a single covariance matrix on the parameters of the first compartment as well as with a full covariance matrix on the parameters of all compartments. IOV was considered on CL and V. Structural model selection was guided by objective function value (OFV) as computed by NONMEM, corresponding to minus twice the log-likelihood (a Δ of −3.84 with 1 degree of freedom, Chi-squared distribution, corresponds to a significance level of p = 0.05) goodness-of-fit plots and physiological plausibility. In addition, the precision of the parameter estimates, ETA shrinkage, and IIV and IOV were assessed, as well as parameter correlation. Candidate models were further evaluated by a prediction-corrected visual predictive check (n = 1000 simulations) and a numerical predictive check based on n = 1000 simulations [19]. With the visual predictive check, the observed data were compared with the model-simulated data to evaluate the internal validity of the model.

After selection of the base model, the following covariates were tested on CL and V based on physiological plausibility: albumin, Continuous Veno-Venous Hemofiltration (CVVH) and SOFA. The effect of albumin was assessed on V ₁ and CL. The effect of CVVH was assessed on V ₁ (dichotomous covariate). SOFA scores were obtained on different days. Missing values of non-sampling days before the observation were replaced by the first value (usually at 48 h) and missing values after observations were carried forward. In the case of completely missing SOFA scores, the median of 7 was used. The SOFA score was divided into a low score (<10) and a high score (>10) based on associated mortality risks [20, 21] and was assessed as a categorical covariate on CL and V.

Covariates were tested in a stepwise fashion (forward inclusion, backward elimination). A covariate was included when it was physiologically plausible and the OFV decreased with 3.84 points (Chi-squared distribution, p = 0.05). The precision of the parameter estimates was assessed using a non-parametric bootstrap method in which resampling of the data to create new datasets with the same size but containing different sets of individuals and yielding new parameter estimates and confidence intervals occurred (n = 1000 replications).

The final model was used to perform a simulation study exploring various dosing regimens. As the ICU cohort used for model building was considered too small for simulation purposes (n = 20), a cohort of 1000 hypothetical individuals with a mean BW of 70 kg with 20% coefficient of variation (CV) in BW was generated for each dosing regimen. This variation in weight (20% CV) was based on the weight distribution of a haematological cohort from our hospital (n = 1706, median 76.4 kg and standard deviation 14.5, resulting in 18.96% CV). Simulations were performed also propagating parameter uncertainty, with the TNPRI function in NONMEM, using priors from the final model.

Five dosing regimens were simulated including licensed regimens and alternative regimens, chosen at the discretion of the researchers. Licensed regimens included (I) 100 mg QD for 14 days and (II) 100 mg QD the first 4 days with 200 mg QD from day 5 (labelled indication for non-responders); alternative regimens included (III) 200-mg loading dose on day 1 followed by 100 mg QD from day 2, (IV) 200-mg loading dose followed by 150 mg QD from day 2 and (V) 200 mg QD. A 200-mg dose is not expected to be associated with toxicity as micafungin doses of 8 mg/kg are well tolerated [4].

The PK-PD target for micafungin in the treatment of invasive candidiasis was determined by the group of Andes et al. [10]. The exposure-response relationship of micafungin has been well established: an AUC/MIC ratio of 3000:12,000 has previously been associated with a 98% mycological response for all Candida infections, while a target of 5000:12,000 has been associated with a 97.8% mycological response for non-C. parapsilosis infections [10]. We used both an AUC/MIC ratio of > 3000 well as 5000 as the PK-PD target for the simulations. Furthermore, the model presented will allow other researchers to use our PK model for simulations on different Candida species whenever needed. However, this is beyond the scope of this article as the focus is on PK variability in a population of critically ill patients.

Target attainment at days 3 and 14 was assessed for a range of clinically relevant MIC values (0.002–1 mg/L) of the Clinical and Laboratory Standards Institute. This range includes both the modal MIC for micafungin for C. albicans (0.015 mg/L) as the epidemiological cut-off MIC (0.03 mg/L, Clinical and Laboratory Standards Institute reference method) [22, 23].

In addition to the determination of the PTA for each individual MIC, the PTA considering the population distribution was evaluated. For each MIC, the fraction of simulated patients who attained the PK target was multiplied by the fraction of the MIC distribution of C. albicans. The cumulative fraction of response was calculated as the sum of fraction products over all MICs. The population distribution of C. albicans was gathered from Pfaller et al., as this species is often involved in candidaemia in ICU patients [22].