Model-Optimized Fluconazole Dose Selection for Critically Ill Patients Improves Early Pharmacodynamic Target Attainment without the Need for Therapeutic Drug Monitoring

Fluconazole has been associated with higher mortality compared with the echinocandins in patients treated for invasive candida infections. Underexposure from current fluconazole dosing regimens may contribute to these worse outcomes, so alternative dosing strategies require study.

ABSTRACT

Fluconazole has been associated with higher mortality compared with the echinocandins in patients treated for invasive candida infections. Underexposure from current fluconazole dosing regimens may contribute to these worse outcomes, so alternative dosing strategies require study. The objective of this study was to evaluate fluconazole drug exposure in critically ill patients comparing a novel model-optimized dose selection method with established approaches over a standard 14-day (336-h) treatment course. Target attainment was evaluated in a representative population of 1,000 critically ill adult patients for (i) guideline dosing (800-mg loading and 400-mg maintenance dosing adjusted to renal function), (ii) guideline dosing followed by therapeutic drug monitoring (TDM)-guided dose adjustment, and (iii) model-optimized dose selection based on patient factors (without TDM). Assuming a MIC of 2 mg/liter, free fluconazole 24-h area under the curve (fAUC₂₄) targets of ≥200 mg · h/liter and <800 mg · h/liter were used for assessment of target attainment. Guideline dosing resulted in underexposure in 21% of patients at 48 h and in 23% of patients at 336 h. The TDM-guided strategy did not influence 0- to 48-h target attainment due to inherent procedural delays but resulted in 37% of patients being underexposed at 336 h. Model-optimized dosing resulted in ≥98% of patients meeting efficacy targets throughout the treatment course, while resulting in less overexposure compared with guideline dosing (7% versus 14%) at 336 h. Model-optimized dose selection enables fluconazole dose individualization in critical illness from the outset of therapy and should enable reevaluation of the comparative effectiveness of this drug in patients with severe fungal infections.

INTRODUCTION

In the critically ill, bloodstream infections with Candida spp. are associated with crude and attributable mortalities of up to 60% and 40%, respectively (1). Initiation of adequate antifungal therapy within 48 h in critically ill patients with candidemia has been demonstrated to reduce mortality, especially when combined with other therapeutic measures, such as the removal of central venous catheters (2). Conversely, inadequate dosing of antifungals has been suggested to contribute to treatment failures and increased mortality (3, 4).

Fluconazole has been considered the antifungal agent of choice for various fungal infections, especially those caused by Candida spp. However, the efficacy of fluconazole was questioned after the publication of a patient-level quantitative review of randomized trials investigating treatment strategies in patients with bloodstream infections (5). Patients with Candida albicans infection treated with the echinocandins, a different class of antifungal agent, had a significantly lower mortality, determined by multivariate analysis, than those treated with other agents to which these organisms were susceptible, including fluconazole. As a consequence, echinocandins are now favored over fluconazole as an antifungal agent in critically ill patients (6–8). However, the use of broader-spectrum antifungal agents is associated with increased treatment costs (9) and resistance pressure (10, 11).

Previous publications have demonstrated that a high proportion of patients with candidemia are critically ill or undergoing renal replacement therapies and that these factors are predictors of death (12–15). Reported fluconazole studies were flawed by failure to utilize loading doses (16, 17) or by use of the same dosing regimen in both critically and noncritically unwell patients (18). Dosing of fluconazole in critically ill patients is complex, as fluconazole pharmacokinetics (PK) are highly variable, primarily due to changes in physiology and the use of extracorporeal procedures such as renal replacement therapy (19–21). It is therefore feasible that fluconazole underexposure from current dosing regimens may contribute to the higher mortality seen in patients treated with this drug compared with that seen with the echinocandins.

Numerous empirical dosing guidelines for fluconazole have been established (7, 8), but few data are available to assist with prescribers’ selection of the most appropriate dosing regimen in critically ill patients. Furthermore, our earlier work established that no single established fluconazole dosing regimen results in optimal early attainment (within 48 h) of pharmacodynamic (PD) targets for organisms with a MIC of 2 mg/liter, particularly in those patients who were obese, those with augmented renal clearance, or those undergoing continuous renal replacement therapy (CRRT) (60). An MIC of 2 mg/liter is deemed to be susceptible to fluconazole by both the European Committee on Antimicrobial Susceptibility Testing (EUCAST) (22) and the Clinical and Laboratory Standards Institute (23), indicating to clinicians a high probability of success when treated with standard drug doses.

Treatment efficacy has been associated with free fluconazole exposure, with the free fluconazole 24-h area under the curve to MIC ratio (fAUC₂₄/MIC) being the PD parameter which best correlates with clinical outcomes. An fAUC₂₄/MIC ratio of ≥100 has been associated with a high clinical cure rate for Candida sp. infections (24). An fAUC₂₄ of ≥800 mg · h/liter represents more than four times the required exposure for the treatment of susceptible Candida infections with an MIC of 2 mg/liter and may represent fluconazole overexposure. Although the fluconazole fAUC₂₄ exposure-toxicity relationship has not been defined, high doses are more commonly associated with central nervous system and liver dysfunction, gastrointestinal side effects, and rash (25). An optimal fluconazole dosing regimen would therefore minimize both under- and overexposure throughout a treatment course.

Given the high pharmacokinetic variability in critically ill patients, therapeutic drug monitoring (TDM) has often been proposed to guide dosing decisions of antimicrobial agents (26). Fluconazole meets previously proposed criteria for TDM in the setting of critical illness, namely, significant PK variability, a relationship between plasma concentrations and clinical effects, and the availability of a cost-effective drug assay (27, 28). Although TDM of fluconazole is not generally recommended because of the predictable linear PK in stable patients (8, 29), it may be beneficial for clinically unstable patients, patients with renal dysfunction, and patients requiring renal replacement therapy (30, 31). Fluconazole TDM has been implemented in some institutions to optimize drug exposure in vulnerable patients (32, 33).

Despite the potential benefits of TDM-guided fluconazole dose selection, initial (i.e., first 48 h) fluconazole dosing decisions are usually made in the absence of TDM data. This is because results that can have a meaningful impact on prescribing decisions are rarely available at the commencement of therapy. In the absence of TDM, prescribers remain reliant on empirical dosing guidelines to guide dosing decisions early in the treatment. Nonetheless, TDM has the potential to aid target attainment later in the treatment course; however, there is little evidence evaluating the performance of fluconazole TDM in this setting.

In light of this, there remains a need for novel strategies to guide clinicians in making informed individualized dosing decisions for critically ill patients to ensure early treatment efficacy in the absence of the measurement of fluconazole concentrations, as well as achievement of optimal exposure over the entire treatment course. The application of population PK modeling methods provides an opportunity to address this unmet medical need (34). Model-optimized dose selection is a novel approach to individualized dosing and can be applied even if TDM is unavailable (35).

This study was conducted to examine predicted attainment of fluconazole therapeutic targets in critically ill patients (i) in the early phase of therapy using established dosing regimens or a novel PK model-optimized dose selection approach and (ii) during the later phase of therapy using established dosing regimens, a TDM-guided strategy, or model-optimized dose selection approach.

(Preliminary results from this study were presented at the Australian Society for Antimicrobials [ASA] 2017 Annual Scientific Meeting [36].)

RESULTS

Optimal loading and maintenance doses.

The median (interquartile range [IQR]) optimal loading fluconazole dose for the model-optimized strategy was 1,450 mg (1,250 to 1,700 mg). The median (IQR) optimal maintenance fluconazole dose for the model-optimized strategy (400 mg [300 to 550 mg]) was higher than that for the TDM-guided approach (250 mg [200 to 350 mg]).

Early target attainment.

Guideline-based dosing of fluconazole was predicted to result in poor early (0- to 48-h) target attainment, with only 70% and 79% of patients achieving an fAUC₂₄ of ≥200 mg · h/liter at 24 h and 48 h, respectively (Fig. 1a). This meant that 30% and 21% of critically ill patients treated with guideline-based dosing were underexposed at 24 and 48 h, respectively. In comparison, model-optimized dose selection was predicted to result in 99% of patients attaining targets at 24 and 48 h.

FIG 1.

Probability of target attainment over time by dosing regimen. (a) Proportion of individuals achieving an fAUC₂₄ of ≥200 mg · h/liter over time by dosing regimen. Guideline-based dosing (open diamond), guideline-based dosing and intensive TDM at 24 to 48 h (open circle), and model-optimized dosing from outset of therapy (filled triangle). (b) Proportion of individuals achieving an fAUC₂₄ of ≥800 mg · h/liter over time by dosing regimen. Guideline-based dosing (open diamond), guideline-based dosing and intensive TDM at 24 to 48 h (open circle), and model-optimized dosing from outset of therapy (filled triangle). Note different y axis. (c) Fluconazole fAUC₂₄ over time across the study population for each dosing regimen. Median (gray line) and graded shaded ribbons for 60, 70, 80, 90, 95, and 99% population intervals for guideline-based dosing, guideline-based dosing and intensive TDM at 24 to 48 h, and model-optimized dosing from outset of therapy from top to bottom. Efficacy and “toxicity” targets of 200 mg · h/liter and 800 mg · h/liter, respectively, are represented with dashed lines.

Overexposure within the first 48 h was predicted to be uncommon with guideline-based dosing, with 0.1% and 0.6% of patients having an fAUC₂₄ of ≥800 mg · h/liter at 24 and 48 h, respectively (Fig. 1b). In contrast, model-optimized dose selection was predicted to result in 1.1% and 2.6% of patients having an fAUC₂₄ of ≥800 mg · h/liter at 24 and 48 h, respectively.

Late target attainment.

Beyond 48 h, guideline-based dosing of fluconazole was predicted to result in the lowest target attainment, with only 77% of patients achieving an fAUC₂₄ of ≥200 mg · h/liter at 72 h (Fig. 1a). Overall target attainment at 168 and 336 h was unchanged, leaving 23% of patients underexposed. Once the TDM-guided doses were implemented at 72 h, this protocol was predicted to lead to lower target attainment at 168 and 336 h compared with guideline-based dosing (65% and 63%, respectively). Of the tested strategies, model-optimized dose selection performed the best, with 99% of patients predicted to attain targets at 72 h, and 98% at 168 and 336 h.

Considering overexposure, guideline-based dosing with or without a loading dose resulted in similar outcomes. At 168 and 336 h, 7% and 14%, respectively, of patients receiving these regimens were predicted to have an fAUC₂₄ of ≥800 mg · h/liter (Fig. 1b). The TDM-guided dosing protocol was predicted to result in the least fluconazole overexposure, with <1% of patients having an fAUC₂₄ of ≥800 mg · h/liter during the entire treatment course.

Relative to guideline-based dosing, model-optimized dose selection was predicted to initially result in slightly higher population overexposure until 96 h, but it ultimately halved the proportion of patients overexposed, with 4% and 7% of patients having an fAUC₂₄ of ≥800 mg · h/liter at 168 and 336 h, respectively.

Interindividual variation in fAUC₂₄.

Examining the exposure across the study population, it was apparent that the TDM-guided dosing protocol was predicted to result in the least interindividual variability in fAUC₂₄ (Fig. 1c). While fluconazole exposure was predicted to be more variable with the model-optimized dose selection approach than with the TDM-guided dosing protocol, it was still less variable than guideline-based dosing.

DISCUSSION

Like many other anti-infective drugs, fluconazole PK in the critically ill have been demonstrated to be highly variable (19, 20, 37, 38, 60). A recent reexamination of current and previously proposed fluconazole dosing regimens revealed that no single regimen achieved optimal exposures during the standard treatment course across a representatively diverse critically ill population (60). Guideline fluconazole dosing resulted in underexposure in patients who were overweight, who had higher renal function, and who were receiving CRRT.

Unlike other triazole agents, TDM is not routinely recommended for fluconazole (39). However, considerable interindividual PK variability has been observed in critically ill patients receiving fixed doses of fluconazole, with 33% of patients not attaining the optimal PK/PD target in a recent multicenter study (38). Fluconazole therefore meets the criteria for undertaking TDM in critically ill patients (27, 28). As there are currently no guidelines available specifically to inform fluconazole TDM processes, it is recommended that the general principles established for other anti-infective drugs be followed (40). In our institution, critically ill patients have several fluconazole concentrations collected in a dosing interval, from which an AUC₂₄ value is calculated using the trapezoidal rule. Proportional dose adjustments are made in response to the AUC₂₄ value. In other institutions, collection of a trough concentration has been used as a surrogate for AUC₂₄ (32, 33, 41), with only limited evaluation of correlation between trough concentrations and AUC₂₄ in their populations (42, 43).

Anti-infective TDM in critically ill patients is a multistep process, of which accurate and timely measurement of the drug concentration is only one part. A structured approach to dose adjustment based on measured drug concentrations is also necessary to limit under- and overexposure of drugs and thus maximize the benefit of TDM to patients (44). In our study, fluconazole TDM did not have an impact on early underexposure due to the inherent procedural delays in the TDM process. Once implemented, it made late underexposure worse for the studied population. This is not surprising, because early TDM is biased by the impact of the loading dose. Additionally, the majority of patients do not achieve steady state until close to the end of therapy, limiting the utility of fluconazole TDM as implemented in our institution.

Previous evaluations of TDM for fluconazole and other azole antifungal agents in critically ill adults have been flawed. They have not examined fAUC₂₄/MIC over the treatment course, focusing instead on surrogate trough concentration measurements collected at steady state (32, 45–49). Studies to date have also been retrospective and uncontrolled in design, failing to standardize the performance of TDM or dose adjustment based on results (32, 45–48). As such, objectively evaluating the impact of TDM in published studies and making comparisons with the current study is difficult. Population PK simulation methods provide a framework by which anti-infective TDM could potentially be systematically studied in the absence of prospective controlled clinical studies. Wojciechowski and colleagues undertook a novel in silico assessment of TDM-based infliximab dosing strategies in inflammatory bowel disease and found that TDM-guided Bayesian estimation utilizing an optimization algorithm was superior at maintaining target concentrations across the study population (50). Their work led to the development of the approach described here. The results of this study highlight that the application of TDM to support fluconazole dosing decisions in critically ill patients is itself not a “one size fits all” solution and needs to be carefully evaluated together with alternative approaches.

Model-optimized dosing utilizing a population PK model and patient covariates alone to select anti-infective drug doses for individuals has had limited evaluation to date, despite its promise as a strategy to tailor dosing in the absence of (or prior to) TDM. Frymoyer and colleagues evaluated a similar approach to tailor empirical vancomycin dosing in neonates based on a population PK model and patient parameters. In their in silico study, 94% of patients were predicted to achieve the PD target with the model-based dosing regimen compared with only 18% to 55% with existing dosing guidelines, while at the same time minimizing extreme trough concentrations of either <5 or >20 mg/liter (35). Consistent with these findings, the model-optimized dose selection strategy in this study was found to be highly effective at both improving early underexposure and limiting late overexposure compared with guideline dosing. Moreover, the model-optimized dose selection strategy performed well across the tested diverse critically ill population. Model-optimized dosing demonstrated that, in general, fluconazole loading doses in critically ill patients need to be much higher than those currently in use (700 to 2,400 mg versus 800 mg). In addition, optimal maintenance doses (150 to 650 mg) for individuals vary substantially. This novel finding highlights the limitations of the current fixed fluconazole dosing in critical care.

Although our model-optimized fluconazole dose selection strategy improves the probability of target attainment (PTA), interfaces which facilitate the application of this approach at the clinical interface will require careful consideration to enable effective translation. The Shiny framework (51) provides a simple avenue to build and host web applications (52). To explore this further, we developed a proof of concept web application which, via a simple interface, delivers the model-optimized dosing protocol recommendations based upon patient covariates input by the user and requires no prior PK knowledge to utilize (see Fig. S3 in the supplemental material). The validity and utility of this application in a clinical setting requires further evaluation.

Several limitations of this study must be considered. First, the results of this study are likely to be affected by the composition of the study population and how well that population is described by the PK model. The original model was developed and validated in a critically ill population that was representative of patients treated in our intensive care unit. External validation of the model found it to compare well with other published models of fluconazole pharmacokinetics in critically ill patients, and therefore the results are likely to be applicable to other intensive care units. Second, the use of an fAUC₂₄ of ≥800 mg · h/liter as a measure of overexposure was somewhat arbitrary, although it did represent four times the exposure required for susceptible organisms as defined by the European Committee on Antimicrobial Susceptibility Testing (EUCAST). Estimation of unnecessary overexposure resulting from the model-optimized dose selection approach compared with established practice to evaluate its performance was useful. Third, the TDM strategy incorporated linear dose adjustment targeting an fAUC₂₄ of 200 h · mg/liter rather than a Bayesian forecasting approach. It is likely that a Bayesian approach would have performed better due to the ability to forecast exposure when patients are not at steady state (53). However, we are not aware of any currently available Bayesian programs for fluconazole, limiting the applicability of this approach at the bedside. The choice to target a higher fAUC₂₄ would likely have resulted in a higher proportion of patients attaining efficacy targets but would also have resulted in a higher proportion of patients being overexposed. The selection of an fAUC₂₄ of 200 h · mg/liter target for fluconazole is reflective of current clinical practice. Finally, the study was an in silico exercise, and prospective clinical validation of the model-optimized dose selection approach in a critically ill population is required. Critically ill patients are also rarely “pharmacokinetically static,” with dynamic changes in drug distribution and clearance occurring over the course of their admissions (26). In this context, TDM may be beneficial, following on from a model optimized initial approach and requires prospective study. In addition, (i) the safety of the higher doses proposed (including the evaluation of drug-drug interactions), (ii) the efficacy of our model-based dosing strategy in terms of both target attainment and clinical outcomes when fluconazole is used as stepdown therapy, and (iii) the potential role for dose-optimized fluconazole as a primary therapy in low-risk settings requires further evaluation. The potential benefits to the health care system in terms of both cost savings and reduced resistance pressure makes conducting these studies important.

In conclusion, we have demonstrated that a model-optimized dose selection approach resulted in optimal fluconazole exposures for almost the entire critically ill study population from the outset of therapy in the absence of TDM. This strategy performed better than standard dosing and an intensive TDM-guided approach. Although the model-optimized dose selection strategy requires clinical validation, it has the potential to allow fluconazole dose individualization in critical illness and may improve patient outcomes in severe fungal infections.

MATERIALS AND METHODS

Study population.

A virtual population of 1,000 individuals was constructed to reflect patient factors representative of an adult critically ill population (60) (weight, 80 ± 20 kg; age, 50 ± 15 years; serum creatinine, 120 ± 40 μmol/liter; gender, 60% male; CRRT, 30%). Distributions of the population patient characteristics are presented in Fig. S1 in the supplemental material.

Fluconazole pharmacokinetics.

A previously evaluated one-compartment population PK model for fluconazole (60) was used to simulate individual PK parameters for the representative patient population in all dosing scenarios (see “Dosing scenarios” below). The model covariates included the influence of weight on volume of distribution and renal function on systemic clearance as described by power models referenced to 70 kg and 120 ml/min, respectively, and the use of CRRT as a dichotomous relationship on clearance (see Fig. S2 and Table S1 in the supplemental material). Population parameter variability was included using an exponential error model. Residual unexplained variability in measured fluconazole concentrations was described by a combined proportional and additive error model. This model was developed in critically ill adult patients and was chosen for its comprehensive description of PK across the entire clinical cohort (19, 20, 37).

Software.

All statistical and graphical analyses were performed using R version 3.4.4 (R Foundation for Statistical Computing) and a number of add-on packages, including mrgsolve, pknca, ggplot2, grid, and plyr (54–56).

Dosing scenarios.

Four dosing regimens were applied to the representative virtual patient population administered once-daily fluconazole therapy over a 14-day period. The three dosing regimens included (i) the guideline dosing, (ii) guideline dosing followed by TDM and proportional dose adjustment, and (iii) model-optimized dose selection. To reflect current clinical practice, all administered doses were rounded to the nearest 50 mg. Doses were administered as an intravenous (i.v.) infusion at a rate of 200 mg/h per the Australian Injectable Drug Handbook (57).

Guideline dosing.

“Guideline dosing” consisted of an 800-mg loading dose followed by renally adjusted maintenance dosing according to the fluconazole product information. (6, 8, 58) In summary, patients with a creatinine clearance (CL_CR) of >50 ml/min and those on CRRT were administered 400 mg once daily, while patients with a CL_CR of ≤50 ml/min were administered 200 mg every 24 h. The maintenance phase (subsequent) dosing commenced at 24 h.

TDM-guided dosing.

TDM-guided dosing of fluconazole consisted of patients initially commenced on label-based dosing with an 800-mg loading dose, with adjustment of subsequent (maintenance) doses based on plasma fluconazole (TDM) concentrations collected on the second day of treatment (per local practice). While not clinically feasible, to represent the “best-case scenario” for TDM, intensive sampling was conducted every 15 min from 24 to 48 h, from which fAUC₂₄ was calculated using noncompartmental methods, utilizing the trapezoidal rule (59). After a 24-h delay for the hypothetical specimens to be processed in the laboratory, proportional dose adjustments targeting an fAUC₂₄ of 200 h · mg/liter were instituted at 72 h. A proportional dose adjustment strategy was selected over a Bayesian forecasting approach due to the lack of currently available Bayesian programs for fluconazole. The TDM-guided dosing procedure is illustrated in Fig. 2.

FIG 2.

TDM-guided dosing to calculate fAUC₂₄. TDM-guided dosing calculation with a worked example. fAUC₂₄, free fluconazole 24-h area under the curve; AUC_target, target AUC (200 mg · h/liter). Unbound fraction = 0.88.

Model-optimized dose selection.

Model-optimized dose selection of fluconazole implemented an optimization algorithm (“optim” in R) that predicted the loading dose with 99% probability of achieving an fAUC₂₄ of 200 mg · h/liter for each individual in the representative population corresponding to their patient data (such as weight, gender, age, serum creatinine, and CRRT status). Once the loading dose was optimized and administered, the same algorithm was applied to the subsequent (maintenance) doses in order to identify the daily dose required to continue achieving the fAUC₂₄ target at 336 h with 99% probability. The model-optimized dose selection code is presented in the Appendix S1 in the supplemental material.

Outcome measures.

Fluconazole pharmacokinetic parameters such as fAUC₂₄ were determined using a population pharmacokinetic approach. The target fAUC₂₄ range of 200 to 800 mg · h/liter was based on an organism with an MIC of 2 mg/liter. Each dosing regimen was assessed according to (i) the proportion of patients who achieve an fAUC₂₄ ≥of 200 mg · h/liter, (ii) the proportion of patients who had an fAUC₂₄ of ≥800 mg · h/liter, and (iii) the interindividual variation in fAUC₂₄ throughout a course of therapy. Outcome measures were considered across the entire treatment course (14 days), specifically for the early (0 to 48 h) and late (48 to 336 h) phases of the treatment course. Early target attainment was not evaluated for TDM-guided dosing, which was not implemented until 72 h.