1 Introduction
Candida or
Aspergillus species are the fourth most dominant pathogens causing disease complication in the intensive care unit (ICU) and account for approximately 20% of infections in the ICU [
1,
2]. Micafungin is an antifungal agent belonging to the class of echinocandins that act by inhibiting the synthesis of β-(1,3)-
d-glucan, an important component of the fungal cell wall. It is a semi-synthetic compound, freely soluble in water and it has a molecular weight of 1292.26 [
3,
4]. Micafungin shows in vitro and in vivo activity against
Candida species and it is licensed as a first-line treatment for invasive candidiasis [
3‐
5]. Micafungin, like other echinocandins, achieves adequate clinical responses in about 70% of patients with invasive candidiasis or candidaemia [
2,
6].
The recommended micafungin dose is a daily (QD) intravenous infusions of 100 mg (infusion time approximately 1 h) [
4]. In the case of insufficient response, such as when the clinical condition does not improve or in case of persistent positive cultures, the dose may be increased to 200 mg QD [
4].
Recently, we analysed the pharmacokinetics of micafungin in 20 critically ill patients (ICU patients) by means of a non-compartmental approach [
7]. A lower total exposure [area under the concentration–time curve over 24 h (AUC
0–24h)] was observed in this cohort as compared with healthy volunteers, although the total exposure was comparable to other patient populations [
7].
The pharmacodynamic (PD) index of the echinocandins is best described by the area under the concentration–time curve:minimum inhibitory concentration (AUC:MIC) ratio [
8,
9]. To design optimal dosing regimens for micafungin in critically ill patients, both pharmacokinetic (PK) and PD factors need to be incorporated into a model. A non-compartmental analysis is not sufficient for the purpose of modelling and simulations. Hence, we will deploy an additional PK-PD analysis using non-linear mixed-effect modelling to obtain a further understanding of the pharmacokinetics of micafungin. Defining such a PK model will enable us to simulate different dosing regimens and assess the corresponding exposure (AUC) and PK target attainment, taking into account the susceptibility profiles of the pathogen.
Micafungin clinical breakpoints have been defined in a phase III clinical study among patients with invasive candidiasis. An AUC:MIC between 3000 and 12,000 was associated with 98% success for all
Candida species. A specific target was defined of an AUC:MIC ratio of 5000:12,000 for non
C. parapsilosis species and above 285 for
C. parapsilosis [
10]. Micafungin has an overall favourable safety profile; therapy with 150–200 mg QD is well tolerated [
3].
We aimed to combine our PK data and the clinical breakpoints data to determine the probability of target attainment (PTA) in a population of critically ill patients. This will enable us to make simulations for other regimens to investigate the gain in the PTA in pathogens with altered susceptibility profiles.
4 Discussion
We developed a population PK model for micafungin in ICU patients. The model was successfully used to assess different dosing regimens of micafungin to predict the corresponding PTA, taking into account parameter uncertainty. The simulations revealed that the majority of the population is treated sufficiently with the current licensed dosing regimen, but the PTA for infections with Candida species with a MIC of 0.032 mg/L can be improved by increasing the maintenance dose a priori to 200 mg.
Clearance in our cohort was 1.10 L/h and modestly higher compared with CL in healthy volunteers [
24,
25]. Three population PK models have been published previously. Clearance in our cohort was similar to ICU patients on CVVH, mechanical ventilation or with an intra-abdominal infection [
26‐
29]. Inter-individual variability on CL was 40% CV in our cohort and therefore higher than observed in healthy volunteers [
24] or ICU patients on CVVH (17–20% CV) [
26] but similar to ICU patients with sepsis and mechanical ventilation (34%) [
29]. Volume of distribution was higher than in healthy volunteers (13.3 L for a healthy 70-kg patient) [
24] and the central V was higher than in ICU with severe peritonitis, sepsis or burn injuries [
27‐
29] but very similar to ICU patients on CVVH (22.5 L for a 70-kg patient) [
26]. IIV on
V
1 was 73% CV in our cohort and thereby two- to ten-fold higher compared with other studies (8–38% CV) [
24,
26,
29]. One explanation could be that healthy volunteers and subjects on CVVH are more homogenous populations, in which less variability is observed, while our cohort consisted of ICU patients with and without CVVH. The estimated IIV may also have been inflated by one influential individual (BW 134 kg) who seems to be responsible for the improved model fit when including IIV on
V
1.
V
2 was lower in our analysis but the total V (
V
1 +
V
2) was comparable to other pharmacokinetics. [
26,
29]
BW was a priory added in the model on CL and
V
1 in agreement with physiological plausibility and based on previous work. [
15‐
18,
26,
29]. It remains a matter of debate whether one should estimate or fix the allometric exponents. One group empirically estimated this for micafungin [
29] while another one fixed it [
26]. For illustration purposes, we have estimated the allometric exponent of BW to CL, which was 1.26 and this did not significantly improve the model fit (ΔOFV −2.97); 1.26 is very different from what is physiologically plausible and from the results of another group (0.59) [
29]. Based on our data (
n = 20 patients), the true exponent is not identifiable and using an empirical estimate would not allow for extrapolation beyond our dataset.
Although not present in our cohort, in the case of morbidly obese, critically ill patients, BW may not be the ideal parameter to relate size to pharmacokinetics. Other weight-derived parameters such as fat-free mass could be alternative descriptors to explain the variability in PK parameters in this subpopulation [
30]. As micafungin has a relative low V, only a small fraction is metabolised and kidney function does not play an important role [
3,
26]. BW is assumed to adequately describe the relationship between body size and the PK parameters for the normal weight ICU patients [
3].
We found an important IOV on V, which means that V may change considerably from day to day within one patient. While our data also supported the addition of IOV on CL, we considered IOV on V more physiologically plausible, as we observed important fluctuations in peak concentrations of micafungin. Day-to-day fluctuations in V may be explained by haemodynamic changes including fluid retention or fluid loss as result of a capillary leak, which may occur in ICU patients [
31,
32]. Moreover, including IOV on both CL and
V
1 resulted in an unidentifiable model with parameter correlation and including both was, therefore, not considered. A study among ten ICU patients on CVVH found an IOV on
V
1 and
V
2 of 27–28% CV, which was lower compared with our cohort (37% CV) [
26]. This difference might be explained by the difference in clinical condition: patients on CVVH might be haemodynamically more stable and controlled in terms of fluid retention and diuresis than ICU patients not on CVVH. This is reflected in the lower IOV of
V
1.
As referenced, we have now conducted several population PK studies of echinocandins in ICU patients. We observed some differences between the various echinocandins. Unlike caspofungin with identical CL and V in critically ill patients compared with healthy volunteers, micafungin showed a higher CL (1.10 L/h compared with 0.55 L/h) and a higher V (21.2 L compared with 13.9 L for
V
1 and
V
2 together) in ICU patients [
33]. Anidulafungin showed similar changes in CL when comparing critically ill patients with healthy volunteers, while V was equal in both populations [
34]. It remains unclear why micafungin pharmacokinetics changes but caspofungin pharmacokinetics is less affected by critical illness, as we could not find any covariates, apart from BW, to explain the variability in PK parameters. Covariates might be obscured owing to the overall high variability in pharmacokinetics among ICU patients, caused by a mixture of factors (e.g. systemic inflammatory response, capillary leak, protein-binding capacity) [
35]. Combining datasets and thereby increasing sample size may help to further clarify this.
Our results confirm that the exposure (AUC
0–24h) in the ICU population after 100 mg QD is much lower than the exposure with the same dose in healthy volunteers, which is consistent with our previous analysis [
7]. Increasing the maintenance dose in ICU patients to 150 mg led to equal exposure compared with healthy volunteers. When the labeled 200 mg is chosen, this would lead to higher exposure compared with healthy volunteers. Theoretically, lower exposure risks decreased efficacy and it is known that AUC is inversely linked to disease outcome when exposed to pathogens with increasing MICs [
9].
We have simulated the licensed regimens, as well as alternative regimens. Based on our simulations and the PTA assessment with the 3000 target (all Candida spp.), the majority (83%) of the ICU population is treated adequately with the currently licensed 100-mg maintenance dose. Higher maintenance doses of 200 mg QD are only required in the case of infections with species with decreased susceptibility. However, based on the 5000 target (non-C. parapsilosis), the majority of the population may benefit from a dose increase to 200 mg QD as a result of low target attainment with the 100 mg QD (62% attains the 5000 target with MIC 0.016). Although this can be considered a worst-case scenario, no susceptibility profiles are available upon the start of therapy, for which a 200-mg dose QD may be beneficial until the results of the microbiology are available. If the MIC allows, one can scale down to 100 mg QD. This should be evaluated prospectively. Alternatively, a 200-mg loading dose for all ICU patients can be beneficial, to achieve early adequate exposure (also shown in Fig. S3). Especially given the high IOV of V
1, it is of great importance to adequate saturate all compartments with micafungin.
Our results are comparable to previous work [
26,
29] and confirm that species with attenuated MICs may benefit from higher maintenance doses. It should be noted that PK-PD targets vary considerable among
Candida species and the target greatly influences the outcome: the PTA decreases when the target increases [
8]. Thus, correct selection of the target is crucial for adequate interpretation of the PTA. When starting therapy, the MIC is often unknown. With this in mind, our results encourage the prospective evaluation of a higher initial dose (e.g. 200 mg maintenance), as it may have the potential to decrease the rate of patients not attaining the target. This should be evaluated for superiority in a prospective trial.
In addition, modelling and simulation are always associated with uncertainty, further challenging the interpretation of the estimated PTA. We therefore took parameter uncertainty into account when performing Monte-Carlo simulations of micafungin in ICU patients, subsequently used for the PTA predictions. The variability in exposure (Fig.
1) is a combination of parameter uncertainty and variability in BW. Although this uncertainty is taken into account in the PTA predictions, Fig.
2 does not show any variability as it was based on the proportion of patients attending the PK target. Another approach could be to perform multiple simulations (e.g. 1000 times) using parameter precision information from a bootstrap analysis and calculate confidence intervals around the PTA [
36].
The current work is inevitably associated with some limitations, which are discussed below. The first relates to the high IIV on V
1, which may have been inflated by an influential individual with a BW of 134 kg, the highest BW of the cohort. However, the inclusion of the parameter improved the model fit to the data significantly and it is not surprising to encounter deviating individuals in heterogeneous ICU cohorts. We therefore decided to retain this individual in the cohort and proceed with IIV on V
1, but at the same time underline the uncertainty in the estimated value of this parameter. The second limitation relates to the fact we did not have information on the unbound micafungin concentrations. As micafungin is highly protein bound, fluctuations in albumin (likely to occur among ICU patients) may influence total concentrations and thereby the PTA. This might have led to an underestimation rather than an overestimation of the PTA.