ABSTRACT
Pharmacokinetic changes are often seen in patients with severe infections. Administration by continuous infusion has been suggested to optimize antibiotic exposure and pharmacokinetic/pharmacodynamic (PK/PD) target attainment for β-lactams. In an observational study, unbound piperacillin concentrations (n = 196) were assessed in 78 critically ill patients following continuous infusion of piperacillin-tazobactam (ratio 8:1). The initial dose of 8, 12, or 16 g (piperacillin component) was determined by individual creatinine clearance (CRCL). Piperacillin concentrations were compared to the EUCAST clinical breakpoint MIC for Pseudomonas aeruginosa (16 mg/liter), and the following PK/PD targets were evaluated: 100% free time (fT) > 1× MIC and 100% fT > 4× MIC. A population pharmacokinetic model was developed using NONMEM 7.4.3 consisting of a one-compartment disposition model with linear elimination separated into nonrenal and renal (linearly increasing with patient CRCL) clearances. Target attainment was predicted and visualized for all individuals based on the utilized CRCL dosing algorithm. The target of 100% fT > 1× MIC was achieved for all patients based on the administered dose, but few patients achieved the target of 100% fT > 4× MIC. Probability of target attainment for a simulated cohort of patients showed that increasing the daily dose by 4-g increments (piperacillin component) did not result in substantially improved target attainment for the 100% fT > 4× MIC target. To conclude, in patients with high CRCL combined with high-MIC bacterial infections, even a continuous infusion (CI) regimen with a daily dose of 24 g may be insufficient to achieve therapeutic concentrations.
INTRODUCTION
Infections are common in critically ill patients, and early, adequate antimicrobial therapy is essential for reducing mortality (1). Piperacillin-tazobactam is a broad-spectrum and well-tolerated β-lactam–β-lactamase inhibitor combination (2). It is one of the most prescribed antibiotics in critically ill patients with severe infections and is commonly administered in the intensive care unit (ICU) in a ratio of 8:1. The antibacterial activity is time dependent, i.e., bacterial eradication is related to the time for which the unbound drug concentration is maintained above the MIC (fT>MIC) (3). As such, maximizing fT>MIC is assumed to increase drug efficacy and to limit development of antimicrobial resistance (4).
Critical illness is associated with several pathophysiological changes, such as capillary leakage, decrease in plasma protein concentrations, and organ dysfunction with hypoperfusion (5). These changes lead to pharmacokinetic (PK) alterations and increased variability in (unbound) antibiotic plasma concentration, which makes achievement of optimal antibiotic exposure in this patient population a challenge (6).
It has been suggested that administration of β-lactam antibiotics through prolonged or continuous infusion (CI) may be superior to conventional bolus administration with respect to achievement of pharmacokinetic/pharmacodynamic (PK/PD) targets. These modes of drug administration have the advantage of continuous input, which compared to intermittent bolus (IB) dosing results in sustained antibiotic concentrations and may reduce mortality and improve clinical outcome (7–10). Although it is standard practice to administrate piperacillin as intermittent infusions over 30 min, continuous infusion is used as the preferred mode of administration at several ICUs worldwide (11).
The aim of this study was to assess the PK of piperacillin in critically ill patients treated with CI. A population PK model was developed to identify patient covariates and evaluate the probability of target attainment (PTA) for various dosing regimens and renal clearances, to guide dosing recommendations.
RESULTS
Patient population.A total of 78 critically ill patients were included in the study, and 196 samples were available for analysis. Characteristics of the study population are summarized and shown in Table 1, and the observed free concentrations of piperacillin over time at steady state are illustrated in Fig. 1. The average number of piperacillin concentrations determined per patient was 2.5, and the median piperacillin concentration across samples was 58 mg/liter (interquartile range [IQR] 38; 109). The majority of patients were male (68%), and the median age was 64 years (IQR 58; 73). Median plasma creatinine (p-creatinine) across samples was 128 μmol/liter (IQR 73; 206), and the median creatinine clearance (CRCL) was 54 ml/min (IQR 28; 94).
Patient characteristics (n = 78)
Overview of the piperacillin free concentration samples (points, n = 196) collected from included subjects (n = 78). The shaded areas represent the 90% prediction interval (5th to 95th percentiles) of model predicted steady-state concentrations for the individuals in each panel, with the solid line representing the median predicted concentration. The number of patients predicted to achieve the two PK/PD targets is indicated. The short-dashed line indicates a concentration of 16 mg/liter (EUCAST MIC breakpoint for P. aeruginosa), and the long-dashed line indicates a concentration of 64 mg/liter (4× MIC). (a) Patients are split based on the employed dosing regimen (8, 12, or 16 g) according to creatinine clearance (shown in header). (b) Patients in the two middle groups are further divided into narrower intervals to show the predicted concentrations.
Pharmacokinetic modeling and covariates.The final model was a one-compartment model with linear clearance defined according to
Final parameter estimates and variances from the population pharmacokinetic modeling analysis, including uncertainty and shrinkagea,b
Diagnostics of model fit to data. (a) A simulation-based prediction-corrected visual predictive check based on the final model is showing an acceptable fit to the data. The points represent observed piperacillin plasma samples, the dark dashed line represents the median observation per bin, and the faint dashed lines represent the 5th and 95th percentiles of the observed data per bin. The shaded areas represent the 95% confidence intervals of the median, 5th and 95th percentiles from model simulations. (b) Residual diagnostics show an adequate fit of the model to the data. From left to right are conditional weighted residuals (CWRES) versus time, individual predictions (IPRED) versus observations (DV), and individual weighted residuals (IWRES) versus individual predictions (IPRED). A line of identity (light gray) and a smoothing to the points (red) are shown for each subplot.
A model with separation of clearance into a nonrenal and renal component, with the renal component directly scaled to individual CRCL, provided a good fit to the data (objective function value [ΔOFV] = −71.1 compared to the base model) with a reduction in clearance interindividual variability (IIV) (from 71 to 59%CV). Based on the final parameterization of CRCL (which incorporates body weight by use of the Cockcroft-Gault equation), a relation of body weight to clearance did not reach statistical significance (ΔOFV = −3.18).
Simulations.Based on the final model and individual clearances (empirical Bayes estimates), it was assessed how many of the included patients were predicted to achieve the targets for the three CRCL-dependent doses, under steady-state conditions for an MIC of 16 mg/liter. This is illustrated in Fig. 1a by overlaying the observed piperacillin concentrations with the 5th, 50th (median), and 95th percentiles of predicted concentrations. It can be observed that all patients achieved 100% fT > 1× MIC irrespective of CRCL, but steady-state concentrations are lower for higher CRCL. Few patients were predicted to achieve 100% fT > 4× MIC. In Fig. 1b, the two middle groups (30 to 80 and 80 to 130 ml/min) were split and assessed further, showing no clear differences.
For PTA assessment, 10,000 virtual patients, with a CRCL value assigned from a log-normal distribution centered around the observed median value of 54 ml/min and truncated at 5 and 200 ml/min, were simulated. Agreement between observed CRCL in the study population and the sampled values from the virtual patient population was seen by comparing quantiles. The PTA was assessed for the groups of CRCL by assessing the utilized dose as well as doses with increments of 4 g (piperacillin component). For a patient with a CRCL of <130 ml/min, PTA for 100% fT > 1× MIC was above 90% for the daily dosing of 8, 12, and 16 g for an MIC of 16 mg/liter, whereas 20 g was required for the group with CRCL of >130 ml/min (Fig. 3 and Table 3). For the target of 100% fT > 4× MIC, a PTA of ≥90% was observed only in the group with CRCL of <30 ml/min with a high dose of 20 g. None of the other dosing regimens resulted in a PTA of ≥90%.
Probability of target attainment (PTA) for free plasma concentrations at steady state, showing the two PK/PD targets of 100% fT > 1× MIC (top) and 100% fT > 4× MIC (bottom). The predictions show the PTA at six daily dose levels of 8, 12, 16, 20, 24, and 28 g, administered through continuous infusion (CI) preceded by a 4-g loading bolus, with dosing and calculation of PTA split on patients’ sampled creatinine clearance (the number of patients is indicated in each header). The dashed horizontal line indicates that 90% of the simulated population has reached the specified target, and the dashed vertical line represents the EUCAST MIC breakpoints for P. aeruginosa (16 mg/liter).
Overview of PTA for continuous infusion at individual creatinine clearances and an MIC value of 16 mg/litera
DISCUSSION
Appropriate antibiotic dosing in critically ill patients may be a challenge due to the large PK variability seen in this patient population. Administration by CI for β-lactams is more likely to ensure that suggested target exposures are reached compared to conventional bolus dosing (12–14). We determined the population PK of piperacillin following CI in critically ill patients and evaluated the impact of renal function on PTA. Our results indicate that for pathogens with high MICs and in patients with a high CRCL, piperacillin CI 16 g/24 h is insufficient to achieve a suggested target of 100% fT > 4× MIC.
Piperacillin is predominantly eliminated through renal excretion (at CRCL values of >20 ml/min, according to the model), and plasma concentrations are significantly affected by individual renal function (15, 16). As renal impairment is frequently seen in critically ill patients in the ICU, it may be more likely for these patients to achieve therapeutic concentrations compared to patients with augmented renal clearance (ARC) (CRCL > 130 ml/min). Patients with ARC are at risk of suboptimal antibiotic exposure, if dosing is not adjusted based on CRCL (17–19). Both an increase in glomerular filtration rate and an accelerated tubular secretion play a role in ARC, leading to increased elimination of renally cleared antibiotics (20, 21).
As illustrated in Fig. 3, the strong relation between renal function and piperacillin clearance is directly reflected in the probability of PK/PD target attainment. For the lower PK/PD target of 100% fT > 1× MIC, 8 g/24 h results in >90% PTA for patients with a CRCL of <30 ml/min, whereas 16 g/24 h is needed for patients with a CRCL of ≥80/<130 ml/min. For patients with a CRCL of ≥130 ml/min, 20 g/24 h is needed for this target to be achieved. However, 90% PTA for the higher target of 100% fT > 4× MIC was achieved only in patients with a CRCL of <30 ml/min, receiving 16 g/24 h. For patients with a CRCL of ≥30 ml/min, none of the higher dosing regimens simulated (20 g, 24 g, and 28 g/24 h) resulted in PTA of ≥90%. These findings are in line with other studies. Dhaese et al. found that in critically ill patients with a renal clearance above 90 ml/min/1.73 m2, even high-dose piperacillin (24 g daily) given as CI was not enough to achieve adequate exposure against susceptible Pseudomonas aeruginosa isolates, for the target of 100% fT > 4× MIC (19). Furthermore, in an observational study by Aardema et al. (22), renal function played an important role in target attainment. For patients with a CRCL of <50 ml/min, 42.9% maintained a piperacillin concentration of ≥80 mg/liter, whereas only 18.2% of patients with a CRCL of ≥50 ml/min maintained a concentration of ≥80 mg/liter (22). Although the population PK model developed in this study could be used to predict dosing regimens for achieving 90% PTA for 100% fT > 4× MIC, such dosing regimens would have to be examined in a separate study before any treatment recommendations could be made. The exact threshold for antibiotic exposure and clinical effect is still controversial, which makes it difficult to decide which PK/PD target to aim for. However, there seems to be a general belief that critically ill patients benefit from the more strict PK/PD target of 100% fT > 4× MIC (19, 23, 24).
Antimicrobial CI has a PK advantage compared with bolus dosing and results in higher % fT>MIC and a higher PTA for the same daily dose (7, 8, 25, 26). However, data on clinical advantages and mortality due to CI are inconclusive (9, 10). Two relatively large, randomized controlled trials with 140 and 438 patients included, respectively, were recently published, comparing clinical outcome and efficacy of continuous versus intermittent β-lactam antibiotic infusion in septic patients (12, 27). The study by Dulhunty et al. (12) found no difference in clinical outcome whereas the study by Abdul-Aziz et al. (27) found significantly higher cure rates in severe sepsis patients treated with CI. The different findings may partially be explained by the fact that Dulhunty et al. included patients on renal replacement therapy. As previously mentioned, the likelihood of achieving therapeutic concentrations for antimicrobials cleared renally increases in patients with renal impairment. It is important to recognize that patients admitted to the ICU are a heterogeneous population, where patients with high renal clearances may need a higher antibiotic dosing or prolonged infusion to achieve such concentrations, compared to patients with renal impairment. In addition, compared to intermittent infusion, continuous infusion is more likely to result in therapeutic concentrations and improved clinical benefits when the causative pathogen has an elevated MIC (28). As these situations may be difficult to anticipate, CI needs to be considered from the start when critically ill patients are treated with β-lactam antimicrobials.
Furthermore, extended infusion seems to be beneficial in certain ICU populations. A post hoc analysis based on data from the DALI study comparing CI to IB found that in patients with respiratory infections, the 30-day survival was higher in patients treated with CI (23). Identification of such subpopulations is highly relevant, and further research in this area is needed.
While recognizing the crucial importance of sufficient antibiotic dosing in critically ill patients infected with bacteria with MICs near the upper limit of the piperacillin-susceptible range, lowering antibiotic dosing when possible is of equal importance in order to reduce the risk of toxic side effects and antimicrobial overconsumption, as well as the risk of resistance development (29). The massive use of antibiotics in the ICU has ecological side effects, and in order to prevent dissemination of multidrug-resistant bacteria, antimicrobial therapy needs to be targeted and deescalation should be considered whenever possible (30). Results from our simulations (Fig. 3) illustrate that for patients with a CRCL of ≥80/<130 ml/min, piperacillin CI of 16 g results in >90% PTA for bacteria with an MIC of ≤4 mg/liter. Verification of pathogen MIC and application of therapeutic-drug monitoring (TDM) whenever possible may therefore be of help to optimize piperacillin exposure as well as performing deescalation and targeted treatment when possible. Furthermore, initial dosing adjusted to CRCL increases the likelihood of achieving therapeutic concentrations.
Since the collected piperacillin samples were taken under a CI administration regimen, the information with respect to piperacillin PK parameter estimation was limited. The estimated volume of distribution is like one previously reported, likely informed by administration of an initial bolus of piperacillin prior to initiating the CI (19). Six of the 196 samples were taken during the first 24 h from patients with CRCLs between 11.5 and 22.1 ml/min; some information on volume of distribution is present (also considering the low uncertainty of 13%). Moreover, the information on clearance (most important parameter for CI regimens) is likely greater than if data were from a bolus or short infusion. A previous population PK analysis of piperacillin in critically ill patients identified a nonlinear clearance component with an estimated Km of 37.1 mg/liter (total concentrations), though no such nonlinearity could be estimated from the available samples (31). The Km indicates that the nonlinearity may be of relevance at therapeutic concentrations, though the degree of interindividual variability and relation of the nonlinearity to kidney function (CRCL) need further investigation. As such, the here-presented PTA simulations may be viewed as conservative for the doses above the studied 16 g.
The relatively large number of patients and concentration samples included in this study increases the validity of the results. Furthermore, the collected data are obtained from actual CI of piperacillin-tazobactam and not based on simulations from population PK models, built on piperacillin data obtained following intermittent bolus administration (32).
This was a single-center study, and the results may not be representative of other ICUs, thus limiting the external validity of the study. Another limitation with this study is that doses of piperacillin were adjusted according to serum creatinine and not actual CRCL, although this is not a limitation for the population PK modeling. At the time when piperacillin-tazobactam continuous infusion was initiated at the ICU at Aarhus University Hospital, and during the study period, dosing was done according to serum creatinine as described in Materials and Methods. At that time, this was in accordance with the practice of dosing adjustment used in several other countries where piperacillin-tazobactam continuous infusion was applied. However, nowadays, dosing adjustment according to the Cockcroft-Gault formula is more commonly used and is also applied in our ICU setting. It has been shown previously that equations other than Cockcroft-Gault may have higher accuracy and smaller bias (33). Even so, we used the Cockcroft-Gault equation in our study as it is the most commonly used in the literature, which makes it easier to compare our findings and data to similar studies.
In conclusion, our clinical study demonstrates that in critically ill patients, decreasing pathogen susceptibility and increased renal clearance increase the risk of inadequate antibiotic exposure. To prevent subtherapeutic concentrations, initial dosing according to CRCL should be considered. Furthermore, as local susceptibility patterns may vary greatly between countries, determination of pathogen MICs whenever possible, as well as applying TDM, may also help optimize antibiotic exposure in this patient population.
MATERIALS AND METHODS
Study design.This was an observational study conducted in the ICU at Aarhus University Hospital, Denmark, between June 2016 and January 2018. The study was approved by the Danish Data Protection Agency. Given that this study was undertaken parallel to standard-of-care treatment of critically ill patients, the Regional Ethical Committee approved the study without requiring signed informed consent.
Patient population.Critically ill patients treated empirically or targeted with piperacillin-tazobactam, administered as CI, were eligible for study enrollment. Patients were excluded if treated with renal replacement therapy or extracorporeal membrane oxygenation or were under the age of 18 years. The following data were registered for each enrolled patient: age, sex, weight, and serum creatinine. Furthermore, an estimate of individual creatinine clearance (CRCL) was derived from the serum creatinine measurements by use of the Cockcroft-Gault formula (34).
Study drug and blood sample collection.In the ICU at Aarhus University Hospital, piperacillin-tazobactam (in a ratio of 8:1) is routinely administered through 24-h CI. The daily dose depends on the individual serum creatinine levels as follows: 16 g for serum creatinine of <110 μmol/liter, 12 g for serum creatinine between 110 and 300 μmol/liter, and 8 g for serum creatinine of >300 μmol/liter. This corresponds to approximate CRCL cutoffs (as derived by Cockcroft and Gault for a 20-year-old 70-kg male) of >80 ml/min, between 30 and 80 ml/min, and less than 30 ml/min, respectively. At day 1, piperacillin administration through CI is preceded by a 4-g bolus infusion over 3 min, which is not subtracted from the total daily dose on the first day of dosing. Continuous infusion is initiated immediately after the bolus infusion. All patients have their piperacillin plasma concentrations determined three times weekly: Monday, Wednesday, and Friday. All patients included in the study received piperacillin CI for a minimum of 24 h before the first piperacillin plasma concentration was determined. Piperacillin concentrations were determined as described above, for the duration of piperacillin-tazobactam treatment, and the time period ranged from 1 to 22 days. The free concentrations of piperacillin in sera were assessed using ultrahigh-performance liquid chromatography following ultrafiltration. An in-depth description of the method can be found elsewhere (35).
PK/PD targets.The following PK/PD targets were evaluated: 100% fT > 1× MIC (free piperacillin concentration maintained above the MIC throughout the dosing interval) and 100% fT > 4× MIC (free piperacillin concentration maintained at a level 4-fold the MIC throughout the dosing interval). To evaluate the predefined PK/PD targets, the clinical breakpoint MIC for Pseudomonas aeruginosa (16 mg/liter) published by EUCAST was used (36). This pathogen is frequently seen in the ICU setting and is associated with poor outcome. Clinical breakpoint MIC represents a worst-case scenario regarding bacterial susceptibility that needs to be considered when patients are treated empirically.
Statistical analysis.Demographic data were analyzed using Stata version 13 (StataCorp, College Station, TX). Medians, interquartile ranges, percentages, and means were found using the “sum, detail” function in Stata.
Pharmacokinetic modeling and covariate screening.A population pharmacokinetic model was developed based on the collected samples using NONMEM 7.4.3 (Icon Development Solutions, Gaithersburg, MD, USA) assisted by Perl-speaks-NONMEM and Piraña (37, 38). Model parameter estimates were obtained by use of the first-order conditional estimation method with interaction (FOCE-I) on nontransformed data. Selection between two competing (nested) models was done statistically with application of the likelihood-ratio test to the two objective function values (OFV), under the assumption that the OFV follow a χ2 distribution. Consequently, a ΔOFV = −3.84 for one additional parameter would be significant at alpha = 0.05, the significance level applied for structural and stochastic components (39). In addition to statistical assessment, parameter precision, covariance matrix condition number, residual goodness of fit, and prediction-corrected visual predictive checks (VPCs) were used to aid model selection and evaluation (40).
Identification of a suitable structural model included the assessment of one- or two-compartment disposition, with interindividual variability assessed in all relevant parameters (parameterized to result in log-normally distributed individual parameters). After an appropriate structural model was established, initial assessment of parameter-covariate relationships by use of stepwise covariate modeling (SCM) was done for the following covariates: body weight (time-varying), age, sex, serum creatinine (time-varying), and creatinine clearance (derived, time-varying) (41). Statistical significance for incorporation of covariates in the forward search was alpha = 0.05, while a stricter criterion of alpha = 0.01 was applied during backward deletion. Further assessment was done following the initial screening, by separation of the clearance term into nonrenal and renal components as previously shown (42).
Simulation of alternative dosing regimens and probability of target attainment.Based on the developed model, it was of interest to assess the proportion of included patients who achieved the two PK/PD targets (100% fT > 1× MIC and 100% fT > 4× MIC) under steady-state conditions. The individual empirical Bayes estimates from the final model were used to assess target attainment for CI regimens with the specified doses of 16, 12, or 8 g piperacillin per day preceded by a 4-g loading dose, for an MIC value of 16 mg/liter. Separately, the PTA for the same PK/PD targets and dosing regimens was assessed at steady state for a virtual patient population of 10,000 individuals, where the results were stratified on kidney function as given by CRCL of <30 ml/min, ≥30 to <80 ml/min, ≥80 to <130 ml/min, and >130 ml/min. Patients in the two middle groups were further divided into the following, narrower CRCL intervals: ≥30 to <60 ml/min, ≥60 to <80 ml/min, ≥80 to <100 ml/min, and ≥100 to <130 ml/min. Values of identified covariates were sampled from a distribution approximately similar to that observed in the included patient population. PTA assessment was done for the CRCL-dependent doses utilized in the original study design (i.e., 8, 12, and 16 g) and for three higher doses given at 4-g increments (piperacillin component). As such, patients with a CRCL of <30 ml/min were assessed with doses of 8, 12, 16, and 20 g/day (in addition to a 4-g loading dose).
ACKNOWLEDGMENTS
We thank Felix Klastrup for helping with data management and statistical analysis.
No funding was received for this study.
There are no conflicts of interest to report in relation to the contents of this study.
FOOTNOTES
- Received 21 December 2019.
- Returned for modification 30 January 2020.
- Accepted 23 March 2020.
- Accepted manuscript posted online 13 April 2020.
- Copyright © 2020 American Society for Microbiology.
