ABSTRACT
A retrospective study was conducted in hospitalized patients receiving intravenous polymyxin B who underwent therapeutic drug monitoring during treatment. The aim of this study was to assess the population pharmacokinetics (PK) of intravenous polymyxin B in patients with variable total body weights and create a population model for clinical use. Nonlinear mixed-effects modeling analyses were performed. A total of 43 patients were included, and 70% of these patients were male. The median age was 58 years, and the median weight was 78 kg. The median polymyxin B dose was 180 mg/day or 2.8 mg/kg/day. A one-compartment model described the polymyxin B PK well with conditional mean parameter estimates of a clearance (CL) of 2.37 liters/h and a volume of distribution of 34.4 liters and can be employed for clinical population modeling. Total body weight was not significantly associated with CL (Akaike information criterion, 361.6 for the weight-based model versus 359.5 for the non-weight-based model). These data suggest that dosing according to patient body weight requires further exploration. Greater study is needed to assess the relationships between polymyxin B exposures and efficacy and toxicity.
INTRODUCTION
Over the past decade, the polymyxins have emerged as useful agents in the treatment of infections caused by carbapenem-resistant Gram-negative pathogens (1). The polymyxins, however, have not been subject to the scrutiny of modern-day pharmacokinetic (PK) and pharmacodynamic (PD) studies. In addition, most of the data over the past 5 years have focused on colistin (i.e., polymyxin E). Colistin's pharmacokinetics have recently been defined and may be more complex than those of polymyxin B. Further, colistin pharmacodynamics may also pose problems, as colistin has been associated with more nephrotoxicity than polymyxin B (2–7). Thus, a better understanding of polymyxin B PK/PD is needed.
The current understanding of polymyxin B pharmacokinetics is based on four pharmacokinetic studies in a total of 60 patients (7–10). Collectively, these studies suggest that polymyxin B, unlike colistin, is not eliminated by the kidney and should not be adjusted for renal dysfunction. These studies are limited in the number of patients studied at extremes of body weight and in the loading doses given. Only one study was performed after pharmacokinetic data suggested that 3 mg/kg of body weight/day divided as a 1-h infusion every 12 h with no adjustment for renal dysfunction was optimal to achieve an area under the concentration-time curve for the free, unbound fraction of drug/MIC ratio of approximately 20 (7). As such, previous studies evaluated doses of <2 mg/kg/day with various infusion times ranging from 1 to 6 h for each dose (8, 9). The aim of this study was to develop a population model for common clinical use. The companion paper (11) assesses the clinical covariates regularly considered in polymyxin B dosing and explores the simulated exposures expected with various dosing schemes.
RESULTS
Patient characteristics.A total of 43 patients accounting for 134 polymyxin B plasma assays were included in the study. It should be noted that all 43 patients used for this study formed the 80% of the patient population used to develop the base model in the study described in the companion paper (11). The demographic and clinical information for the patients is summarized in Table 1. Most patients were male (70%), and the median age was 58 years. The median weight of the population was 78 kg, with the range being 40 kg to 126 kg and with 8 (19%) patients weighing greater than 100 kg and 12 (28%) weighing less than 60 kg. A polymyxin B loading dose was administered to 24 (56%) patients, and the median daily dose was 2.8 mg/kg/day. Polymyxin B was administered at the following dosing frequencies: twice daily, 32 (74%) patients; once daily, 8 (19%) patients; and every other day, 3 (7%) patients. Many patients (37%) had a creatinine clearance of <50 ml/min, with the creatinine clearance in 6 (14%) patients being <30 ml/min. The median duration of polymyxin B dosing was 13 days.
Characteristics of polymyxin B-treated patientsa
Population pharmacokinetic model.A one-compartment model described the polymyxin B pharmacokinetics better than a two-compartment model (Akaike information criterion [AIC], 356.7 and 361.2, respectively). In the one-compartment model, a combined error model minimized the residual variability compared to that achieved with a constant error model (AIC, 359.7 and 424.2, respectively). Final population pharmacokinetic parameters are shown in Table 2. The population estimates of the parameters clearance (CL) and volume of distribution (V) demonstrated acceptable precision, with the relative standard error (RSE) values being ≤10% and shrinkage being −1% and 9%, respectively.
Estimation of population PK parameters of polymyxin B in 43 adult patientsa
When the covariate model assessing total body weight (TBW) was fit, the model was not improved (AIC, 361.6 and 359.5, respectively). Model diagnostics of population weighted residuals, individual weighted residuals (IWRES), and quantile-quantile plots are displayed in Fig. 1. All diagnostic plots displayed adequate fits for the final model. Notably, normalized prediction distribution error values were distributed randomly with a normal distribution, and the predictions on the IWRES and population weighted residuals plots were generally centered on zero with constant variance. Figure 2 demonstrates the simulated polymyxin B concentrations (in milligrams per liter) and the associated probability distribution for the mean loading dose and the mean maintenance dose that the patients in this study received (i.e., 202.7 and 104.5 mg, respectively).
Diagnostic population weighted residuals (PWRES), individual weight residuals (IWRES), and normalized prediction distribution error (NPDE) plots as a function of time and population prediction (pred. y) plots, population density function (pdf) plots, and quantile-quantile plots (qqplot). cond. mean, conditional mean. Pink line, reference line for zero; black line (in third row of graphs), theoretic density; green line (in third row of graphs), empirical density.
Simulated polymyxin B concentrations (mg/liter) and associated probability distribution (proba) for the mean loading dose (202.7 mg) (A) and maintenance dose (104.5 mg) (B) in the regimens received by patients in the clinical cohort. Concentrations were generated every 30 min for 0 to 48 h.
DISCUSSION
This report describes polymyxin B pharmacokinetics in the largest patient population studied to date. Unlike previous studies, however, no relationship between total body weight and clearance was observed, implying that optimal dosing in obese patients should not be based on total body weight, as previously suggested (7). This is similar to the findings described in our companion paper (11), where we demonstrated through simulation that TBW should not be used for clinical dose selection, especially in repeated maintenance dosing.
Our study is the first to include a significant number of patients with extremes of body weight and obesity. A previous study included the findings for one patient weighing 41 kg and one patient weighing 250 kg, with the more obese patient also being confounded by renal replacement therapy (7). In our study, 8 patients weighed ≥100 kg and 5 patients weighed <50 kg. In addition, 12 patients had a total body weight greater than 30% above their ideal body weight. We found no impact of TBW on polymyxin B CL. Dosing in obese patients using total body weight may result in greater exposure to polymyxin B. Excess exposures have the potential to result in more acute kidney injury, as suggested by previous studies in which total daily doses exceeding 150 to 250 mg were associated with more nephrotoxicity (12–14).
This study has several limitations. First, one patient contributed concentrations that were outliers in the model. This was likely due to an incorrect documentation of the sampling time. The sample was documented as a trough concentration but appeared numerically to be a peak concentration relative to the other concentrations for this patient. We chose to include these data, even though their inclusion resulted in worse model fits. Second, only pharmacokinetic monitoring was performed, and clinical outcomes were not rigorously evaluated. Urinary polymyxin B concentrations were not measured, but it has already been established that the rate of urinary recovery of polymyxin B is low (7). Third, we excluded patients on renal replacement therapy and/or extracorporeal membrane oxygenation and patients with cystic fibrosis to maintain the homogeneity of our population. Generalizing our findings to these populations is not possible. Lastly, the relatively sparse sampling from this study could have resulted in biased estimates of clearance. Additional studies are needed.
In conclusion, we have presented a reasonably large population pharmacokinetic study of polymyxin B in patients with variable body weights. Weight did not appear to affect clearance. This relationship is further explained in our companion paper (11). Our population model is translated into a freely available simulator that clinicians can use to estimate first polymyxin B exposures (see supplemental material). Further detailed studies linking pharmacokinetics to pharmacodynamic outcomes (i.e., efficacy and toxicity) are still needed.
MATERIALS AND METHODS
Study design.A retrospective study of adult patients (≥18 years of age) admitted to New York-Presbyterian Hospital between January 2009 and December 2015 who underwent therapeutic drug monitoring of intravenous polymyxin B was conducted. The study was approved by the Columbia University Irving Medical Center Institutional Review Board with a waiver for informed consent and by Midwestern University for deidentified data analysis.
Inclusion required receipt of polymyxin B for treatment of a suspected or documented Gram-negative bacterial infection. Polymyxin B treatment, dose, and duration of use were ultimately at the discretion of the primary medical team; however, institutional dosing guidelines were provided to practitioners. Prior to 2014, institutional guidelines for polymyxin B dosing recommended dose adjustment for renal dysfunction, specifically, for an estimated creatinine clearance of less than 80 ml/min. Starting in 2014, dosing guidelines no longer recommended dose adjustments for renal dysfunction (1.5 mg/kg/dose every 12 h). Across all study years, loading doses of 2.5 to 3 mg/kg of total body weight were recommended for dosing of polymyxin B, with no specific maximum dose being recommended. The standard infusion time was 1 h but could be lengthened as needed to increase patient tolerability.
The following demographic and clinical data were collected from the electronic medical record: basic demographics; the type of infection; the bacterial organism targeted; the serum creatinine concentration prior to and during polymyxin B therapy; details of intravenous polymyxin B therapy, including the timing and concentration of every dose; the clinical resolution of infection; and any toxicity, including acute kidney injury or neurotoxicity. Only patients with at least two plasma polymyxin B assays were included. Cystic fibrosis patients and those receiving renal replacement therapy or treatment via some other extracorporeal device (e.g., extracorporeal membrane oxygenation) at the start of polymyxin B treatment or at the time of sampling were excluded from the analysis.
Blood samples for polymyxin B therapeutic drug monitoring were obtained on or after day 3 of therapy and prepared as plasma (15). Troughs were typically obtained within 60 min of a dose, and peaks were typically obtained 15 to 60 min after the end of the infusion. All samples included in the analysis for an individual patient were collected during the same dosing interval. Exact polymyxin B dosing times and blood collection times were documented for each patient. Polymyxin B concentrations were analyzed using a validated high-performance liquid chromatography-mass spectrometry method, as previously described (15). In brief, the lower limit of quantitation was 100 ng/ml.
Population pharmacokinetic modeling.A population pharmacokinetic model was developed and fit to the polymyxin B concentration-time data using a nonlinear mixed-effects modeling approach with Monolix 2016R1 software (Lixoft, Orsay, France). Model parameters were estimated using the stochastic approximation expectation-maximization algorithm with a Markov chain Monte Carlo procedure to compute the maximum likelihood estimates for the final population pharmacokinetic parameters (16, 17). All parameters were assumed to be log normally distributed. One- and two-compartment models were tested on the basis of previous pharmacokinetic studies (7, 10, 18, 19). The log-likelihood ratio test was utilized to compare competing models. A P value of <0.05 was required for increased complexity with agreement from the Akaike information criterion (AIC). The two-compartment model was fit to the polymyxin B data, parameterized as CL (in liters per hour), V (in liters), the volume of distribution of the peripheral compartment (in liters), and intercompartmental clearance (in liters per hour).
Because polymyxin B is conventionally dosed on the basis of weight, the effect of total body weight on clearance was evaluated as follows: log(CLi) = log(CLpop) + beta · log[TBW(i)/70], where CLi and CLpop are the clearance of the individual and the population, respectively, and TBW(i) is the total body weight of the individual.
Retention of total body weight (TBW) as a covariate required an improved AIC. Constant and combined error models were explored to describe the residual unexplained variability of the finalized model. The final model was selected on the basis of minimization of AIC, visual assessment of goodness-of-fit plots (including quantile-quantile plots), the distribution of the population and individual weighted residuals, and the intersubject variability for the estimated pharmacokinetic parameters. Model-based values for CL and V were generated for each patient on the basis of the conditional mean estimate using the final pharmacokinetic model (20). Predicted polymyxin B plasma concentrations for the regimens with mean loading and nonloading doses within the clinical cohort were simulated using the R package Simulx (21). For each scenario, the mean population parameters and intersubject variabilities (i.e., omega values) were used to generate 2,000 PK exposure profiles using the mlxR package. See Supplemental Text S1 in the supplemental material for the mlxR code (22).
ACKNOWLEDGMENT
We declare that we have no conflicts of interest related to this work.
FOOTNOTES
- Received 15 August 2017.
- Returned for modification 12 September 2017.
- Accepted 11 December 2017.
- Accepted manuscript posted online 8 January 2018.
For a companion article on this topic, see https://doi.org/10.1128/AAC.01475-17.
Supplemental material for this article may be found at https://doi.org/10.1128/AAC.01493-17.
- Copyright © 2018 American Society for Microbiology.