International Journal of Antimicrobial Agents 41 (2013) 564– 568

Contents lists available at SciVerse ScienceDirect

International Journal of Antimicrobial Agents

j o ur nal homep age : ht tp : / /www.e lsev ier .com/ locate / i jant imicag

Can population pharmacokinetic modelling guide vancomycin dosingduring continuous renal replacement therapy in critically ill patients?

Andrew A. Udya,∗, Cecilia Covajesb, Fabio Silvio Tacconeb, Frédérique Jacobsc,Jean-Louis Vincentb, Jeffrey Lipmana, Jason A. Robertsa

a Burns, Trauma and Critical Care Research Centre, The University of Queensland, Brisbane, Australiab Department of Intensive Care, Erasme Hospital, Université Libre de Bruxelles, Brussels, Belgiumc Department of Infectious Diseases, Erasme Hospital, Université Libre de Bruxelles, Brussels, Belgium

a r t i c l e i n f o

Article history:Received 18 December 2012Accepted 29 January 2013

Keywords:VancomycinPharmacokineticsContinuous renal replacement therapyCritical illness

a b s t r a c t

Treatment of resistant bacteria such as meticillin-resistant Staphylococcus aureus (MRSA) relies on achiev-ing adequate antibiotic concentrations at the site of infection. Strategies to attain such targets in septiccritically ill patients receiving renal replacement therapy (RRT) are uncommon but could be useful forincreasing the likelihood of therapeutic dosing. The aim of this study was to conduct a populationpharmacokinetic (PK) analysis in septic patients undergoing continuous RRT and to determine whichparameters were associated with inadequate vancomycin concentrations. In total, 81 patients with 199blood samples were included in the study. All patients received vancomycin dosing according to the localprotocol, which included a weight-based loading dose followed by continuous infusion. The vancomycinconcentration–time points were adequately described with a one-compartment model with zero orderinput. The median population PK estimate for vancomycin clearance (CL) was 2.9 L/h [interquartile range(IQR) 2.4–3.4 L/h] and for volume of distribution (Vd) was 0.8 L/kg (IQR 0.6–1.1 L/kg). The goodness-of-fitplots for the model were adequate. When covariates were tested, none were found to adequately explainchanging vancomycin CL or Vd in the population PK model. In particular, the lack of correlation betweenCL and RRT settings was likely due to the multiple confounders known to influence antibiotic prescriptionin this setting. These data provide a cautionary tale of the challenges of describing pharmacokinetics incritically ill patients receiving RRT and highlights the need for a detailed, prospective, multicentre studyto better inform dosing practice.

© 2013 Elsevier B.V. and the International Society of Chemotherapy. All rights reserved.

1. Introduction

Infection in the intensive care unit (ICU), particularly in asso-ciation with severe sepsis and septic shock, continues to manifestin-hospital mortality rates of 35–65% [1]. Current guidelines stressthe importance of early and appropriate antibiotic therapy as a keyintervention linked to improving outcomes [2]. Optimised antibi-otic dosing requires rapid achievement of drug concentrations atthe site of infection, a process often confounded by the physiologi-cal changes encountered in critical illness [3]. In particular, alteredexcretory end-organ function (such as renal impairment) will sig-nificantly affect drug elimination, necessitating the use of modifieddoses in such patients.

∗ Corresponding author. Present address: Burns, Trauma and Critical CareResearch Centre, The University of Queensland, Level 3 Ned Hanlon Building, RoyalBrisbane and Women’s Hospital, Butterfield St., Brisbane, QLD 4029, Australia.Tel.: +61 7 3646 8111; fax: +61 7 3646 3542.

E-mail address: andrew udy@health.qld.gov.au (A.A. Udy).

Acute kidney injury (AKI) complicates the ICU course for manyseptic patients [4], with some of them requiring organ supportwith renal replacement therapy (RRT). Although often associatedwith high illness severity, the need for RRT remains an indepen-dent predictor of mortality [5], whilst infected patients have a 50%higher mortality rate than non-infected patients receiving RRT [6].The mechanisms underlying this observation remain poorly under-stood, although suboptimal antibiotic exposure due to variableextracorporeal drug clearance represents an important possibility[7]. Accordingly, significant interpatient variation in drug con-centrations during RRT has been noted [8], underpinned by thelimited data accurately describing antibiotic pharmacokinetics inthis setting [9]. Pharmacokinetic (PK) models that quantify thedrug exposure effects from variations in RRT settings are thereforeurgently required to better inform dosing schedules.

Complicating this is the increasing prevalence of multidrug-resistant infections in the ICU, including meticillin-resistantStaphylococcus aureus (MRSA), where outcomes are even worse[10,11]. Whilst newer agents with anti-MRSA cover are increas-ingly available, vancomycin remains the most commonly employed

0924-8579/$ – see front matter © 2013 Elsevier B.V. and the International Society of Chemotherapy. All rights reserved.http://dx.doi.org/10.1016/j.ijantimicag.2013.01.018

A.A. Udy et al. / International Journal of Antimicrobial Agents 41 (2013) 564– 568 565

antibiotic therapy in such instances [12]. Vancomycin distributesessentially into extracellular water and is eliminated primarily viaglomerular filtration [13]. For the treatment of MRSA infection,current guidelines recommend trough serum concentrations of15–20 mg/L [12], with recent data highlighting the need for moreaggressive dosing to achieve such targets [14].

Based on its physicochemical properties, vancomycin is likelyto be extensively cleared by haemofiltration [15], although accu-rate dosing schedules in patients receiving continuous RRT remainscarce [16]. Therefore, the aims of this retrospective study were todescribe the population pharmacokinetics of vancomycin in a largecohort of critically ill patients receiving RRT.

2. Methods

2.1. Patients and data collection

A cohort of septic patients treated with continuous infusion ofvancomycin, which has recently been described elsewhere [17],was re-analysed. All patients admitted to the ICU at Erasme Hospi-tal (Brussels, Belgium) between January 2008 and December 2010were screened for inclusion. Patients were included in the study ifthey met the following criteria: (a) aged >18 years; (b) had sepsisaccording to standard criteria [18]; (c) had received vancomycin for≥48 h; (d) were concomitantly treated with continuous RRT; and (e)had daily measurement of vancomycin levels. Patients with previ-ous administration of vancomycin by intermittent infusion (within48 h of the start of the continuous infusion) were excluded, as werethose with residual urine output >500 mL/day, pregnancy, burnsor cystic fibrosis. The study was approved by the Erasme HospitalEthics Committee, which waived the need for informed consent.

2.2. Vancomycin treatment

All patients commenced vancomycin (Vancocin®; Eli Lilly, Saint-Cloud, France) after the initiation of RRT. Treatment was oftenempirical in the setting of presumed or documented Gram-positivehospital- or ICU-acquired infection. Bolus loading dose followedby continuous infusion is the preferred method of administra-tion in the study ICU. Weight-based loading doses (15 mg/kgactual body weight) were recommended in all patients, althoughsubsequent dose selection was at the discretion of the treatingclinician. Daily vancomycin therapeutic drug monitoring (TDM)was then used to adjust the regimen, targeting serum drugconcentrations of 20–30 mg/L. Where the measured serum concen-tration was <20 mg/L, an incremental loading dose of 500–1000 mgwas administered, with an increase in the total daily dose by500–1000 mg. Where the concentration was >30 mg/L, the infusionwas suspended for 4–8 h and the total daily dose was reduced by500–1000 mg.

2.3. Vancomycin assay

Serum concentrations of vancomycin were determinedby particle-enhanced turbidimetric inhibition immunoassay(Dimension® Xpand; Siemens Healthcare Diagnostics, Newark,DE).

The limit of quantification and total imprecision of the assaywere 0.8 �g/mL and <5%, respectively. Blood samples (3 mL) weretaken every day at 08:00 h and were sent immediately to the centrallaboratory. The nursing staff recorded the exact sampling time inthe electronic prescribing system. Samples for PK analysis weredrawn at 24, 48 and 72 h.

2.4. Renal replacement therapy

RRT was provided either as continuous venovenous haemodia-filtration (CVVHDF) or continuous venovenous haemofiltration(CVVH) using a Prisma® or Prismaflex® machine (Gambro Hospal,Bologna, Italy), with polyacrylonitrile (AN69; Hospal, Meyzieu,France) or polysulfone (Gambro Lundia AB, Lund, Sweden)haemofilters. Anticoagulation was achieved using continuousinfusion of either heparin or citrate. Blood flow was typicallyset at 100–200 mL/min, with total effluent rates (ERs) of ca.20–40 mL/kg/h as a combination of ultrafiltration and dialysis. Vas-cular access employs a 20 cm or 25 cm 13.5 Fr silicone double lumenset, inserted under aseptic conditions into a large central vein. Thedecision to initiate and cease therapy was entirely at the discretionof the treating clinician, although it was consistent with estab-lished guidelines [19]. RRT parameters, including the blood flowrate (BFR), ultrafiltration rate (UFR) and dialysis rate (DR), wererecorded prospectively at the time of vancomycin TDM. The inten-sity of RRT was defined by the ER (the sum of the UFR and DR),expressed as mL/kg/h.

2.5. Pharmacokinetic analysis

The concentration–time data for vancomycin in serum wereanalysed by a non-linear mixed-effects modelling approach [20]using NONMEM v.6.1 (GloboMax LLC, Hanover, MD) with doubleprecision with the COMPAQ VISUAL FORTRAN compiler. The NON-MEM runs were executed using Wings for NONMEM (WFN 6.1.3).Data were analysed using the first-order conditional estimationmethod with the Interaction program. Between-subject variabil-ity was calculated using an exponential variability model and wasassumed to follow a log-normal distribution. Residual unexplainedvariability (RUV) was modelled using a combined exponential andadditive random error.

2.6. Model diagnostics

To assess model validity, statistical comparison of nested mod-els was undertaken in NONMEM based on a �2 test of the differencein the objective function. A decrease in the objective function of3.84 units (P < 0.05) was considered significant. Goodness of fit wasevaluated by visual inspection of diagnostic scatter plots, includ-ing observed and predicted concentrations versus time, weightedresidual versus time, and residual versus predicted concentra-tions.

3. Results

In total, 85 patients were identified for inclusion in the study,4 of whom were eventually excluded because timing of samplingwas not available for all vancomycin measurements, leaving a finalcohort of 81 patients. Of the 81 patients, 41 (50.6%) were receivingCVVH at the time of vancomycin administration and 40 (49.4%)were receiving CVVHDF. Table 1 shows the main demographic,physiological, illness severity and dose-related data for the cohort.The mean ± standard deviation (S.D.) vancomycin concentrationachieved after 24 h of therapy was 24.6 ± 9.2 mg/L.

Only 7 patients (8.6%) manifested subtherapeutic concentra-tions (<20 mg/L), whilst 18 (22.2%) had concentrations >30 mg/L.The remainder (56; 69.1%) had values within the range 20–30 mg/Lafter 24 h, suggesting that the empirical weight-based dosing strat-egy was generally quite accurate. By Day 3, only 7% of patients stillhad subtherapeutic vancomycin concentrations, with 67% havingtherapeutic concentrations at this time.

566 A.A. Udy et al. / International Journal of Antimicrobial Agents 41 (2013) 564– 568

Table 1Demographic, physiological, illness severity and dose-related data (n = 81).

Variable

Age (years) (mean ± S.D.) 61.0 ± 15.6Gender (male/female) [n (%)] 53 (65.4)/28 (34.6)ABW (kg) (mean ± S.D.) 83.4 ± 22.1Vasopressor requirement [n (%)] 68 (84.0)Mechanical ventilation [n (%)] 69 (85.2)APACHE II score (mean ± S.D.) 24.5 ± 7.4SOFA score (mean ± S.D.) 10.9 ± 4.0ICU length of stay (days) [median (IQR)] 12 (16–19)Vancomycin loading dose (mg/kg) (mean ± S.D.) 16.4 ± 5.5Infusion dose (Day 1) (mg/kg/24 h) (mean ± S.D.) 23.7 ± 8.1Blood flow rate (mL/min) [median (IQR)] 150 (130–150)Ultrafiltration rate (mL/kg/h) (mean ± S.D.) 21.2 ± 7.2Dialysate rate (n = 40) (mL/kg/h) (mean ± S.D.) 19.4 ± 5.8Effluent rate (mL/kg/h) (mean ± S.D.) 30.8 ± 13.124-h vancomycin concentration (mg/L) (mean ± S.D.) 24.6 ± 9.2Mortality [n (%)] 47 (58.0)

S.D., standard deviation; ABW, actual body weight; APACHE, Acute Physiology andChronic Health Evaluation; SOFA, Sequential Organ Failure Assessment; ICU, inten-sive care unit; IQR, interquartile range.

3.1. Population pharmacokinetic analysis

Each of the patients contributed two to three samples to theanalysis over 3 days of therapy (total of 199 blood samples). Thebest base model, using the model-building criteria, consisted of aone-compartment linear model with zero order input and com-bined proportional and additive RUV. Other models could not besupported as they did not result in an improvement in objec-tive function value or between-subject variability. Between-subjectvariability was included both for clearance (CL) and volume of dis-tribution (Vd). The final objective function for the base model was918.442.

Fig. 1 displays the goodness-of-fit plots for the final model. Allother visual predictive checks were acceptable and confirmed thegoodness of fit of the model. The plots in Fig. 1 show that thefinal PK model describes the measured vancomycin concentrationsadequately. The median population parameter estimate for van-comycin CL was 2.9 L/h [interquartile range (IQR) 2.4–3.4 L/h] andfor Vd was 0.8 L/kg (IQR 0.6–1.1 L/kg). The between-subject vari-ability for CL was 34.7% and for Vd was 49.8%. The RUV was 13.6%.

Surprisingly, we found that no covariate from this data set ade-quately explained changing vancomycin CL or Vd. Specifically, forvancomycin CL, BFR (r2 = 0.02), UFR (r2 = 0.01), DR (r2 = 0.08) and ER(r2 = 0.01) all displayed a poor correlation, whereas for Vd, patientactual body weight did not correlate well either (r2 = 0.01).

4. Discussion

This study has attempted to use a population PK approach toidentify RRT parameters associated with changes in vancomycinpharmacokinetics in critically ill patients receiving continuous RRT.Most notably, despite an adequate base model being described,we were not able to show any significant correlations quantifyingthe effect of RRT on vancomycin pharmacokinetics. This surpris-ing result highlights the complexity of drug handling in criticallyill patients receiving extracorporeal support. In particular, thesepatients can have a number of different RRT settings, including vari-ations in BFR, UFR and DR, all of which may affect the dispositionof a hydrophilic antibiotic such as vancomycin.

Further to this, patient-specific characteristics, including bodyweight, level of sickness severity and degree of residual renalfunction, will also affect vancomycin pharmacokinetics. The con-tribution of each of these factors, and the spectrum of variabilitywithin each factor, means that accurate PK models for describingvancomycin dosing requirements in this setting remain highly chal-lenging. Indeed, this analysis has shown that a limited samplingapproach is unlikely to be able to describe the effect of PK covariatesin this context, particularly when a weight-based dosing protocolis employed.

Current literature regarding the effect of RRT on vancomycinpharmacokinetics in critically ill patients remains sparse. Aprospective study from DelDot et al. examined vancomycin phar-macokinetics in ten critically ill patients receiving CVVHDF [21].Mean (S.D.) total CL of vancomycin was 2.5 (0.7) L/h, whilst thatcleared by CVVHDF was on average 1.8 (0.4) L/h, accounting for76% of total body elimination [21]. In comparison, a recent van-comycin PK study in patients receiving CVVH demonstrated a mean(S.D.) extracorporeal clearance of 0.73 (0.21) L/h, representing ca.50% of total CL [15]. These results contrast the slightly elevatedCL observed in this study (median 2.9 L/h, IQR 2.4–3.4 L/h) andhighlights the variation in pharmacokinetics when using differingmodes of continuous RRT. Importantly, much of the existing dataare from small cohorts, with the current study representing one ofthe largest data sets reported to date.

The Vd reported in this analysis is actually less than that citedin other studies in critically ill patients [14]. This finding, in addi-tion to the poor correlation with body weight, is unexpected. Theevolution of AKI, with impaired water and solute excretion, wouldlikely result in an expanded extracellular fluid compartment and,hypothetically, a larger Vd [3]. However, our analyses do not con-sider the application of RRT prior to the initiation of vancomycintherapy, nor any interventions earlier in the ICU course. As such,changes in fluid management remain an important covariate notincluded in this analysis. Importantly, this highlights the need for

Fig. 1. Diagnostic plots for the final population pharmacokinetic covariate model (n = 199). The left panel describes the observed concentrations versus the populationpredicted concentrations (r2 = 0.08) and the right panel describes the observed concentrations versus the individual predicted concentrations (r2 = 0.89). The grey dotted lineis the line of x = y.

A.A. Udy et al. / International Journal of Antimicrobial Agents 41 (2013) 564– 568 567

extremely detailed investigation when planning future work, asany attempt to accurately model changes in drug pharmacokineticsin this patient group will require well-designed intensive samplecollection and sophisticated data analysis.

The clinical and PK heterogeneity often encountered in thissetting was recently highlighted by the results of the RENALstudy, a large, multicentre, randomised controlled trial of high-versus low-intensity RRT using CVVHDF [22]. In a small subgroupof study participants (n = 10), trough vancomycin concentrationswere obtained, with only 30% of patients achieving the lowertherapeutic target (defined as 15 mg/L in this study) with dosesof 1 g/24 h [8]. In comparing those receiving low- versus high-intensity CVVHDF, median vancomycin trough concentrationswere in fact lower in the low-intensity group, although this wasnot statistically significant [10 mg/L (IQR 9–13.8 mg/L) vs. 14 mg/L(IQR 12–16.8 mg/L); P = 0.13] [8].

Additional findings from the literature highlight factors beyondRRT parameters that may influence vancomycin pharmacokine-tics. The degree of non-renal elimination of vancomycin has beenreported to be between 30% and 50% of total body CL, which willvary depending on the duration of RRT [23]. Importantly, there isa scarcity of data describing changes in non-renal elimination inthe critically ill, which may differ considerably between patients.In addition, many clinicians may commence RRT early in the dis-ease process [24] or for indications other than AKI [19], suggestinga variable degree of native renal function at the time of vancomycintreatment. Finally, whilst referred to as ‘continuous’, issues of filterlifespan mean that patients often receive less actual RRT than is pre-scribed [25], invariably influencing vancomycin exposure. As such,regular measurement of serum vancomycin concentrations, withdose adjustment to ensure optimal levels, remains a key aspectof prescribing practice. Whether a population PK model couldcompletely replace vancomycin TDM in this setting is unlikely.However, such data are still likely to help in initial dose selectionand time to therapeutic concentrations, in addition to minimisingthe reliance on TDM, which may or may not be available locally.

We would like to acknowledge the following limitations of thiswork. The retrospective nature limits the data available for anal-ysis, such that important PK variables (such as sieving coefficientand non-renal CL) cannot be determined. In addition, vancomycindosing was not standardised across the cohort, making PK analysispotentially less precise. Furthermore, no estimates of endogenousrenal function were available, nor was the time spent ‘off’ RRTrecorded. Finally, free or unbound vancomycin concentrations willvary in the critically ill [26], which may potentially influence drugelimination via RRT. In this analysis we did not have access to freevancomycin levels.

In conclusion, we have attempted to explore the relationshipbetween RRT parameters and vancomycin concentrations in a largeretrospective cohort of critically ill patients. No significant relation-ship could be identified between drug CL or Vd and RRT or patientcovariates despite an adequate PK model being described. The lackof correlation between PK parameters and RRT settings is likelydue to the multiple confounders known to influence antibiotic pre-scription in this setting. This study provides a cautionary tale of thechallenges to accurately describing drug pharmacokinetics in crit-ically ill patients receiving RRT and highlights the need for futuredetailed, prospective, multicentre investigations in order to betterinform dosing practice.

Acknowledgments

The authors would like to thank the staff of the Departmentof Intensive Care, Erasme Hospital, Université Libre de Bruxelles(Brussels, Belgium) for their ongoing support.

Funding: AAU was supported, in part, by a Royal Brisbane andWomen’s Hospital Research Scholarship. JAR is supported, in part,by a National Health and Medical Research Council of AustraliaResearch Fellowship (NHMRC APP1048652). The funding sourcehad no role in manuscript preparation.

Competing interests: JL is a consultant to AstraZeneca andJanssen-Cilag and has received honoraria from Astra Zeneca,Janssen-Cilag and Wyeth Australia; JAR has previously consultedfor Janssen-Cilag, AstraZeneca, Pfizer and Gilead, has been involvedin advisory boards for Janssen-Cilag and Astra-Zeneca, and hasreceived unrestricted grants from Janssen-Cilag, AstraZeneca andNovartis; Edwards Lifesciences provide an annual unrestricteddonation to the Burns, Trauma and Critical Care Research Centre(BTCCRC), University of Queensland (Brisbane, QLD, Australia). Allother authors declare no competing interests.

Ethical approval: The Ethics Committee of Erasme Hospital(Brussels, Belgium) approved the present protocol and waived theconsent from patients because of its retrospective nature.

References

[1] Finfer S, Bellomo R, Lipman J, French C, Dobb G, Myburgh J. Adult-populationincidence of severe sepsis in Australian and New Zealand intensive care units.Intensive Care Med 2004;30:589–96.

[2] Dellinger RP, Levy MM, Carlet JM, Bion J, Parker MM, Jaeschke R, et al. SurvivingSepsis Campaign: international guidelines for management of severe sepsis andseptic shock: 2008. Crit Care Med 2008;36:296–327.

[3] Roberts JA, Lipman J. Pharmacokinetic issues for antibiotics in the critically illpatient. Crit Care Med 2009;37:840–51.

[4] Uchino S. The epidemiology of acute renal failure in the world. Curr Opin CritCare 2006;12:538–43.

[5] Metnitz PG, Krenn CG, Steltzer H, Lang T, Ploder J, Lenz K, et al. Effect of acuterenal failure requiring renal replacement therapy on outcome in critically illpatients. Crit Care Med 2002;30:2051–8.

[6] Oppert M, Engel C, Brunkhorst FM, Bogatsch H, Reinhart K, Frei U, et al. Acuterenal failure in patients with severe sepsis and septic shock—a significant inde-pendent risk factor for mortality: results from the German Prevalence Study.Nephrol Dial Transplant 2008;23:904–9.

[7] Kielstein JT, Burkhardt O. Dosing of antibiotics in critically ill patients under-going renal replacement therapy. Curr Pharm Biotechnol 2011;12:2015–9.

[8] Roberts DM, Roberts JA, Roberts MS, Liu X, Nair P, Cole L, et al. Variability ofantibiotic concentrations in critically ill patients receiving continuous renalreplacement therapy: a multicentre pharmacokinetic study. Crit Care Med2012;40:1523–8.

[9] Choi G, Gomersall CD, Tian Q, Joynt GM, Freebairn R, Lipman J. Principles ofantibacterial dosing in continuous renal replacement therapy. Crit Care Med2009;37:2268–82.

[10] Hanberger H, Walther S, Leone M, Barie PS, Rello J, Lipman J, et al. Increasedmortality associated with methicillin-resistant Staphylococcus aureus (MRSA)infection in the intensive care unit: results from the EPIC II study. Int J Antimi-crob Agents 2011;38:331–5.

[11] Ho KM, Robinson JO. Risk factors and outcomes of methicillin-resistant Staphy-lococcus aureus bacteraemia in critically ill patients: a case control study.Anaesth Intensive Care 2009;37:457–63.

[12] Rybak MJ, Lomaestro BM, Rotschafer JC, Moellering RC, Craig WA, Billeter M,et al. Vancomycin therapeutic guidelines: a summary of consensus recommen-dations from the Infectious Diseases Society of America, the American Society ofHealth-System Pharmacists, and the Society of Infectious Diseases Pharmacists.Clin Infect Dis 2009;49:325–7.

[13] Rodvold KA, Blum RA, Fischer JH, Zokufa HZ, Rotschafer JC, Crossley KB, et al.Vancomycin pharmacokinetics in patients with various degrees of renal func-tion. Antimicrob Agents Chemother 1988;32:848–52.

[14] Roberts JA, Taccone FS, Udy AA, Vincent JL, Jacobs F, Lipman J. Vancomycin dos-ing in critically ill patients: robust methods for improved continuous-infusionregimens. Antimicrob Agents Chemother 2011;55:2704–9.

[15] Chaijamorn W, Jitsurong A, Wiwattanawongsa K, Wanakamanee U, DandechaP. Vancomycin clearance during continuous venovenous haemofiltration incritically ill patients. Int J Antimicrob Agents 2011;38:152–6.

[16] Frazee EN, Kuper PJ, Schramm GE, Larson SL, Kashani KB, Osmon DR, et al.Effect of continuous venovenous hemofiltration dose on achievement ofadequate vancomycin trough concentrations. Antimicrob Agents Chemother2012;56:6181–5.

[17] Covajes C, Scolletta S, Penaccini L, Ocampos-Martinez E, Abdelhadii A, BeumierM, et al. Continuous infusion of vancomycin in septic patients on continuousrenal replacement therapy. Int J Antimicrob Agents 2013;41:261–6.

[18] Levy MM, Fink MP, Marshall JC, Abraham E, Angus D, Cook D, et al. 2001SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference. Inten-sive Care Med 2003;29:530–8.

568 A.A. Udy et al. / International Journal of Antimicrobial Agents 41 (2013) 564– 568

[19] Palevsky PM. Indications and timing of renal replacement therapy in acutekidney injury. Crit Care Med 2008;36(4 Suppl.):S224–8.

[20] Beal SL. NONMEM user guides (I–VIII). San Francisco, CA: University ofCalifornia at San Francisco; 1998.

[21] DelDot ME, Lipman J, Tett SE. Vancomycin pharmacokinetics in criticallyill patients receiving continuous venovenous haemodiafiltration. Br J ClinPharmacol 2004;58:259–68.

[22] Bellomo R, Cass A, Cole L, Finfer S, Gallagher M, Lo S, et al. Intensity ofcontinuous renal-replacement therapy in critically ill patients. N Engl J Med2009;361:1627–38.

[23] Macias WL, Mueller BA, Scarim SK. Vancomycin pharmacokinetics in acuterenal failure: preservation of nonrenal clearance. Clin Pharmacol Ther1991;50:688–94.

[24] Karvellas CJ, Farhat MR, Sajjad I, Mogensen SS, Leung AA, Wald R, et al. A com-parison of early versus late initiation of renal replacement therapy in criticallyill patients with acute kidney injury: a systematic review and meta-analysis.Crit Care 2011;15:R72.

[25] Uchino S, Fealy N, Baldwin I, Morimatsu H, Bellomo R. Continuous is notcontinuous: the incidence and impact of circuit ‘down-time’ on uraemiccontrol during continuous veno-venous haemofiltration. Intensive Care Med2003;29:575–8.

[26] Berthoin K, Ampe E, Tulkens PM, Carryn S. Correlation between free and totalvancomycin serum concentrations in patients treated for Gram-positive infec-tions. Int J Antimicrob Agents 2009;34:555–60.