Pharm Res (2020) 37:96

https://doi.org/10.1007/s11095-020-02832-w

RESEARCH PAPER

An Integrated Dialysis Pharmacometric (IDP) Model to Evaluate

the Pharmacokinetics in Patients Undergoing Renal Replacement
Therapy
Astrid Broeker 1 & Matthias G. Vossen 2 & Florian Thalhammer 2 & Steven C. Wallis 3 &
Jeffrey Lipman 3,4 & Jason A. Roberts 5,6,7 & Sebastian G. Wicha 1

Received: 21 February 2020 / Accepted: 22 April 2020

# The Author(s) 2020

ABSTRACT Conclusion The IDP model allows accurate, precise and sen-
Purpose Clearance via renal replacement therapy (RRT) can sitive characterization of clearanceRRT, adsorption and deg-
significantly alter the pharmacokinetic profile of drugs. The aim radation. Successful quantification of all aspects of
of this study was (i) to improve the use of clinical trial data and (ii) clearanceRRT in clinical data demonstrated the benefit of the
to provide a model that allows quantification of all aspects of IDP model as compared to conventional approaches.
drug elimination via RRT including adsorption to dialysis mem-
branes and/or degradation of the drug in the dialysate. KEY WORDS adsorption . doripenem . pharmacokinetics .
Methods An integrated dialysis pharmacometric (IDP) mod- renal replacement therapy . teicoplanin
el was developed to simultaneously incorporate all available
RRT information. The sensitivity, accuracy and precision of
the IDP model was compared to conventional approaches in
clinical trial simulations and applied to clinical datasets of ABBREVIATIONS
teicoplanin and doripenem. CVVHD Continuous veno-venous hemodialysis
Results The IDP model was more accurate, precise and sensi- CVVH Continuous veno-venous hemofiltration
tive than conventional plasma-concentration-based approaches CVVHDF continuous veno-venous hemodiafiltration
when estimating the clearanceRRT (relative bias <1%). In con- dOFV Drop in objective function value
trast to conventional approaches, adsorption and degradation Eq. Equation
were quantifiable using the IDP model (relative bias: −1.1% IDP model Integrated dialysis pharmacometric model
and − 1.9%, respectively). Applied to clinical data, LLP-SIR log-likelihood-profiling based sampling-
clearanceRRT, drug degradation (effluent-half-lifedoripenem: importance-resampling
13.5 h −1 ) and adsorption (polysulphone adsorption rBias relative Bias
capacityteicoplanin: 31.2 mg) were assessed.

Astrid Broeker and Matthias G. Vossen contributed equally to this work.

Electronic supplementary material The online version of this article
(https://doi.org/10.1007/s11095-020-02832-w) contains supplementary
material, which is available to authorized users.

* Sebastian G. Wicha 4
Department of Intensive Care Medicine, Royal Brisbane and Women’s
[email protected] Hospital, Brisbane, Australia
5
1
University of Queensland Centre for Clinical Research, Faculty of
Department of Clinical Pharmacy, Institute of Pharmacy, University of Medicine & Centre for Translational Anti-infective Pharmacodynamics,
Hamburg, Bundesstraße 45, 20146 Hamburg, Germany School of Pharmacy, The University of Queensland, Brisbane, Australia
2
Division of Infectious Diseases and Tropical Medicine, Department of 6
Internal Medicine I, Medical University of Vienna, Vienna, Austria Departments of Pharmacy and Intensive Care Medicine, Royal Brisbane
and Women’s Hospital, Brisbane, Australia
3
University of Queensland Centre of Clinical Research, Faculty of
7
Medicine, The University of Queensland, Brisbane, Queensland, Division of Anaesthesiology Critical Care Emergency and Pain Medicine,
Australia Nîmes University Hospital, University of Montpellier, Nîmes, France
 96 Page 2 of 15 Pharm Res (2020) 37:96

rRMSE Relative root mean squared error Even when multiple RRT specimens were measured, these
RRT Renal replacement therapy data are commonly analyzed separately and compared after-
SSE Stochastic simulation and estimation wards [4, 6, 8, 9].
Using the above conventional approaches to estimate RRT
clearances is associated with obvious as well as hidden restric-
tions. The post-filter approach considers blood flow settings
INTRODUCTION only and ignores effluent settings on a mechanistical level, and
vice versa for the effluent approach. This leads to difficulties
Renal replacement therapy (RRT) can alter the pharmacoki- when the respective other setting influences the RRT clear-
netic profile of drugs and potentially lead to therapeutic fail- ance. The simplification to sieving- or saturation coefficients
ure or increased toxicity. In order to provide reliable guidance that are linearly correlated to the resulting RRT clearance is
on dose adaptation, all effects of RRT need to be quantified accordingly restricted by possible non-transferability between
sensitively, accurately and precisely. The impact of RRT on different settings, modes or membranes. The assumption of
the pharmacokinetic profile is usually investigated in studies linearity in RRT clearance to flow rates, is seldom confirmed
containing only a small number of patients. This underlines in clinical practice [10]. The correct implementation of he-
the importance of using the clinical data from the limited matocrit, blood flow rate and filtration effect holds some chal-
patient sample undergoing RRT in the most optimal way. lenges [11]. When the blood flow rate is used for calculation, a
Estimation of RRT clearance is often simplified in clinical correction factor for hematocrit and drug concentration in the
practice. As a rule of thumb, drug characteristics like the pro- red blood cells is required as well as a correction for potential
tein binding, the renally-cleared fraction and lipophilicity are dilution and concentration effects associated with filtration
considered to guide an educated guess of the RRT clearance. processes in RRT.
If RRT specimens, e.g. post-filter plasma samples or dialysate
samples, are available, they are not used in an integrated Drug Degradation in the Cumulated Effluent
analysis approach, but analyzed separately. In addition, im-
portant details on RRT mode or the calculated or estimated Drug degradation in the cumulated effluent can poten-
RRT clearance are often not reported when the pharmacoki- tially influence or bias the results for estimated RRT
netics of patients undergoing RRT is investigated [1, 2]. Since clearance when the cumulated effluent approach is used.
the patients receiving RRT are often heterogenous and suffer The cumulated effluent is often measured only once per
from comorbidities, distinguishing between RRT clearance dosing interval and the stability of drugs in the effluent
and biliary clearance or remaining renal clearance to RRT is typically not investigated. This might be problematic
clearance can be challenging. These points emphasize the for drugs that are known to be instable, such as some
necessity to investigate dialysis processes in more detail. beta-lactam antibiotics [12].

Conventional Approaches to Investigate RRT Drug Adsorption to the Dialysis Membrane

Clearance
Adsorption of the drug to the dialysis membrane is not
Beyond educated guess on the clearance of drugs via RRT, we considered in conventional approaches. The use of the
identified four conventional approaches to estimate RRT effluent based approaches might underestimate the RRT
clearance: First, the plasma concentrations in patients under- clearance in case of adsorption leading to wrong
going RRT are analyzed and compared to periods, when the assumptions. Highly protein bound drugs are usually
patients are not undergoing RRT, or to other patients not considered to be not removed via RRT [8], but adsorp-
undergoing RRT [3]. Secondly, the post-filter plasma concen- tion to the dialysis membrane might still significantly
trations are analyzed and used to calculate dialysis clearance contribute to elimination of these drugs [13, 14].
based on the difference between pre- and post-filter plasma Moreover, non-linear adsorption processes that are sat-
concentrations and the blood flow rate [4]. Thirdly, the efflu- urable or reversible, i.e. leading to time-dependent
ent concentrations are analyzed and used to calculate dialysis RRT clearance cannot be quantified with the conven-
clearance based on effluent flow rates [5]. Fourthly, the cu- tional models. Adsorption to the dialysis membrane was
mulated effluent concentrations and the volume of the cumu- observed and described over decades, recognized as po-
lated effluent, i.e. the collected effluent in the waste bag over a tentially influencing or even requiring dose adjustments
specified time period, are analyzed to calculate the dialysis [14, 15]. But to the authors knowledge no quantitative
clearance. The total amount of drug removed over time is modelling transformation from in vitro to in vivo or suc-
derived and the principle of mass balance is used [6] as sug- cessful quantification using clinically available data of
gested by the FDA [7]. adsorption was performed so far.
Pharm Res (2020) 37:96 Page 3 of 15 96

C pl ðpreÞ −C pl ðpost Þ
Purpose CL RRT ¼ Q blood adj:  ð2Þ
C pl ðpreÞ
Sections 1.1 to 1.3 emphasize that no one-dimensional con-
where Hct represents the hematocrit used to calculate the
sideration of RRT is mechanistically accurate, and RRT
adjusted blood flow rate (Qblood adj.) based on blood flow rate
mode (dialysis, filtration or dia-filtration), flow rate settings,
(Qblood), Cpl(pre) and Cpl(post) represent the plasma concentration of
pre- or post-filter dilution, membrane material and surface
drug pre- and post-filter, respectively. A correction factor for
area and patient characteristics can influence a drug’s RRT
blood cell concentration (CRBC) was required depending on
clearance. In this work, the focus was set on CVVHD, CVVH
the drug specific blood to plasma ratio [4]. For filtration pro-
and CVVHDF (continuous veno-venous hemodialysis, −fil-
cesses, a correction term for post-filter measurements is re-
tration and dia-filtration, respectively), but the principles and
quired when a fluid removal rate is used and in CVVH or
issues apply to all forms of RRT.
CVVHDF when sampling is done before adding the post-
The aim of this study was (i) to improve the use of data
filter replacement fluid as described previously [9].
derived in clinical trials on RRT and (ii) to propose solutions  
to published and hidden problems, such as drug adsorption Q blood ad j: − Q R F post þ Q F RR
and degradation when estimating RRT clearance. C pl ðpost Þ corr: ¼ C pl ðpost Þ meas:  ð3Þ
Q blood ad j:

Here, QRF post represents the flow rate of the replacement

MATERIALS AND METHODS
fluid added post-filter and QFRR the fluid removal rate. A more
robust way to correct for post-filter measurements is the nor-
Population Pharmacokinetic Modelling
malization to the hematocrit, when pre- and post-filter hemat-
ocrit measurements are available.
For all population pharmacokinetic modelling work in this
study, NONMEM® (ICON development service,
Effluent
Gaithersburg, MD, version 7.4) was used. The inter-
individual variability of the pharmacokinetic parameters was
For the effluent specimen, the RRT clearance was calculated
assumed to be log-normally distributed. For the intra-individ-
as follows:
ual, residual variability an additive, proportional or combined
error model was used [16]. In the clinical data examples, C effl:
CL RRT ¼ Q effl:  ð4Þ
models were evaluated by graphical and numerical criteria C pl ðpreÞ
(goodness of fit plots, visual predictive checks and drop in
Q effl: ¼ Q dial þ Q RF pre þ Q RF post þ Q FRR ð5Þ
objective function value (dOFV)). The LLP-SIR (log-likeli-
hood-profiling based sampling-importance-resampling) meth-
where Qeffl. represents the total effluent flow rate, Qdial, QRF pre
od was employed for the assessment of parameter uncertainty,
and QRF post represent the dialysate, the pre- and post-
since a small number of subjects or cases was investigated [17].
filter replacement fluid flow rate, respectively, ceffl. represents
the concentration of drug in the effluent and Cpl(pre) represents
RRT Data
the pre-filter plasma concentration.
An integrated pharmacometric model was developed to in-
Cumulated Effluent
clude all RRT specimens in one model. The RRT clearance
(CLRRT) was implemented as an additional elimination route
For the cumulated effluent specimen, RRT clearance was
to body clearance of the patient. The RRT clearance repre-
calculated using the amount of drug in the cumulated effluent
sented the sum of all clearance processes mediated via RRT,
i.e. dialysis, filtration and adsorption to the hemofilter. amt cum: effl: ¼ ∫tt 21 CL RRT  cpl ðpreÞ dt ð6Þ

Post-Filter The amount in the effluent compartment (amtcum. effl.) was

modelled as an output compartment in a similar fashion as
The post-filter and dialysate specimens were included based suggested for urine measurements [16].
on the mass balance equations [4]. For the post-filter speci-
men, the RRT clearance was parameterized as follows: Simulation Study

  For the stochastic simulation and estimation (SSE) study, the au-
C RBC
Q blood adj: ¼ Q blood  1−Hct þ Hct  ð1Þ tomated SSE tool by PsN [18] was used with n = 1000 simula-
C pl ðpreÞ tions. Clinical trials with different extents of RRT clearance,
 96 Page 4 of 15 Pharm Res (2020) 37:96

ranging from 0.001% to 100% of the total body clearance, were conventional cumulated effluent approach as compared to
simulated. The developed IDP model (Eqs. 7–10) was employed using the IDP model was investigated. Only with the IDP
for the clinical trial simulations including all concentration-time model it was feasible to account for drug degradation in the
data (pre- and post-filter plasma, effluent and cumulated effluent). cumulated effluent, being informed by time-dependent differ-
The model parameters were (re-)estimated from the ences in the estimated clearance between effluent and cumu-
concentration-time data of the simulated trials using the reduced lated effluent. The dosing interval and the sampling interval of
model (only pre-filter plasma data), the post-filter model, the ef- effluent were set to 8 or 12 h. The RRT clearance was simu-
fluent model, the cumulated effluent model (pre-filter plasma data lated to be as high as the body clearance (3 L/h).
and one of the RRT specimens, respectively) or the IDP model
(using all data simultaneously). This simulation and (re-)estimation Drug Adsorption to the Dialysis Membrane
allowed the assessment of model performance in simulated
(virtual) clinical trials in order to draw conclusions on the model’s The example drug colistin was used to investigate accuracy
performance on real clinical data. Furthermore, the simulated and precision of the conventional effluent and cumulated ef-
truth is known, e.g. the true dialysis clearance and further phar- fluent approaches as compared to the IDP model, when ad-
macokinetic parameters, and allows the calculation of perfor- sorption to the dialysis membrane was simulated. The con-
mance metrices. All five approaches were compared regarding ventional approaches were not accounting for adsorption,
their accuracy (relative bias, rBias), precision (relative root mean while the IDP model was estimating the adsorption processes.
squared error, rRMSE) and power to detect RRT clearance. The RRT clearance was assumed to be as high as the body
clearance in the simulations (2 L/h). When non-reversible
Study Design adsorption was simulated, the fraction of the RRT clearance
mediated by adsorption was set to 65% of the RRT clearance
A dataset including 10 dialysis and 10 non-dialysis patients without saturation. When reversible adsorption was simulat-
was used for the SSEs. Four pharmacokinetically different ed, the fraction of the RRT clearance mediated by adsorption
antibiotics (piperacillin, tigecycline, colistin and linezolid) was set to 95% of the RRT clearance and the reversible ad-
served as example drugs and typical dosing schemes were sorption rate constant to 0.07 h−1 resulting in lower, non-
simulated. In all simulation examples the blood cell concen- linear adsorption clearance. The simulated fraction of RRT
tration was assumed to be zero. clearance mediated via adsorption was considered plausible
A typically used rich sampling schedule for the first and for with respect to the fraction of RRT clearance mediated via
the fifth dose in pre- and post-filter plasma and effluent (t = adsorption of teicoplanin (see results section).
0.5 h, 1 h, 2 h, 4 h, 6 h, 8 h; t = 1 h, 2 h, 4 h, 6 h, 8 h, 12 h for
8 h and 12 h dose interval, respectively) was chosen [5]. Case Studies with Pre-Clinical and Clinical Datasets
Cumulated effluent concentration and volume were collected
over the dosing interval and determined once at the end of a Doripenem
dosing interval. For the non-dialysis patients, RRT clearance
was set to zero and only pre-filter plasma concentrations with A previously published pharmacokinetic study of 12 patients
the same sampling schedule were included. undergoing CVVHDF receiving doripenem was investigated
Residual variability was set conservatively to avoid overes- as clinical data example for the application of the IDP model
timation of power or precision to 25% for the concentration [6]. 195 pre-filter plasma concentration, 194 post-filter plas-
measurements (pre- and post-filter plasma, (cum.) effluent) ma concentration, 71 effluent concentration, 40 cumulated
and to 10% for the volume measurements. Further details effluent concentration and 31 volume of the effluent measure-
on the drug examples used for the simulations are provided ments were included in the analysis. No samples below the
in the supplementary material (Table S1). quantification limit were reported. The blood flow rate was
CVVHD was chosen for the SSEs with settings of Qblood= 200 mL/min, the dialysate flow rate 1000 mL/h and the
10 L/h (167 mL/min), Hct = 0.3 and QDial= 3.0–3.3 L/h, replacement fluid rate 1000 or 2000 mL/h with variable fluid
reflecting typical settings in CVVHD [1]. removal rate. A polyacrylonitrile filter (AN69 Nephral ST
200, Gambro Lundia AB, Lund, Sweden) was used. Further
Drug Degradation in the Cumulated Effluent details of the study can be found in the publication by Roberts
et al. [6]. The general model structure was resumed as de-
Degradation of drug in the cumulated effluent was simulated scribed by Roberts et al. [6], where pre-filter plasma concen-
using the piperacillin example. A degradation rate constant of trations and amounts in cumulated effluent had been used.
0.03 h−1 corresponding to a half-life of approximately 24 h For the dialysis data, the IDP model was used to describe all
was simulated to mimic instable beta-lactams [19]. Accuracy data simultaneously. For the post-filter plasma samples, a cor-
and precision of the estimated parameters using the rection factor was used, since the post-filter plasma samples
Pharm Res (2020) 37:96 Page 5 of 15 96

were obtained pre-addition of replacement fluid (Eq. 3). concentrations and volume of the cumulated effluent simulta-
Possible adsorption to the dialysis membrane with and with- neously within a single model (Fig. 1). These measurements
out a capacity limitation term was investigated, being param- are included as dependent variables, while flow rates, hemat-
eterized by a potentially time-dependent clearance estimated ocrit, and blood to plasma ratios are usually independent var-
from the pre-filter-post-fitler clearance and pre-filter-effluent iables, but the integrated structure allows the estimation of
clearance. The hematocrit was set to 0.30 for all patients [20]. particular variables, if required, e.g. the dialysis flow rate
Moreover, the IDP model was used to quantify potential based on the volume measurements. The integration of all
degradation of doripenem in the cumulated effluent, being measurements allowed to quantify adsorption to the dialysis
parameterized by the difference in the estimated clearance membrane and degradation in the cumulated effluent. The
using the pre-filter-effluent clearance and the cumulated- total clearance mediated via RRT, i.e. dialysis, filtration and
effluent-based clearance. adsorption processes, is given as CLRRT. An example for a
NONMEM control stream and a NONMEM dataset speci-
Teicoplanin fication is provided in the supplementary material (Appendix
model 1, simulation_pip.csv). Implemented into an ordinary
An in vitro study investigating RRT of teicoplanin in bovine one compartment pharmacokinetic model, the IDP model is
blood was evaluated using the IDP model. 51 pre-filter plasma given as follows:
samples and 45 post-filter plasma samples were included. 45
dA1 CL RRT CL body
effluent samples were collected which were all below the lower ¼−  A1 −  A1 ð7Þ
limit of quantification (5 mg/L), which was considered in the dt V cent V cent
 
analyses by comparison to simulations after estimation. A dA2 CL RRT A2
¼  A1  F Ads  1− −k ads rev
blood concentration of 25 mg/L using a volume of 2.2– dt V cent Adsmax
2.5 L was studied. The blood flow rate was set to 250 mL/
 A2 ð8Þ
min and the dialysate flow rate to 500 mL/min without filtra-
  
tion. In five experiments, a polysulphone membrane (n = 3: dA3 CL RRT A2
¼  A1  1− F Ads  1−
FX80, Fresenius Medical Company, Austria, n = 2: F60 S, dt V cent Adsmax
Fresenius Medical Company, Austria), and in one experiment
a triacetate membrane (Ni 21 e, Surflux-21E, Nipro þ k ads rev  A2 −k deg  A3 ð9Þ
Corporation, Japan) was used. The IDP model was used to dA4
describe the RRT clearance and adsorption to the dialysis ¼ Q effl: ð10Þ
dt
membranes was evaluated.
An in vivo case study with two patients receiving teicoplanin
undergoing CVVHF [14] was evaluated using the IDP model. The drug amounts in the central compartment (A1), ad-
One patient received 1000 mg every 24 h and samples were sorption compartment representing the binding capacity of
taken in the first dosing interval. The other patient received the dialysis membrane (A2) and effluent compartment (A3)
400 mg teicoplanin every 24 h and samples were taken in the were included in the IDP model as described in Equation 7,
second dosing interval. The replacement fluid rate was 75 and Equation 8 and Equation 9, respectively, where CLbody repre-
60 mL/h; blood flow rate was 200 and 150 mL/h, respectively. sents the clearance mediated by the body; A1 − 3 the drug
A polyamide membrane (Gambro FH66D, Gambro, Austria) amounts in the compartments 1–3; FAds, Adsmax and kads rev
was used. 11 pre-filter plasma samples, 11 post-filter plasma the fraction of RRT clearance mediated via adsorption, max-
samples and 2 effluent samples were included in the analyses. imal amount adsorbed and reversible rate of adsorption, re-
No samples below the lower limit of quantification were spectively; kdeg the degradation rate in the cumulated effluent.
reported. The IDP model was used to describe the RRT clear- The volume of the cumulated effluent (A4) was described using
ance. It was tested for adsorption to the dialysis membranes. Qeffl. as shown in Eq. 10. The adsorption compartment (A2)
was empty and opened with the first dose of drug and reset
with every recorded filter change in a patient. The effluent
RESULTS compartment (A3) and the volume of the effluent compart-
ment (A4) were set to zero at the start of effluent collection
The Integrated Dialysis Pharmacometric (IDP) Model and reset with every change of effluent bag.

Model Structure Drug Adsorption to the Dialysis Membrane

The IDP model allowed to include measured pre- and post- Possible adsorption to the dialysis membrane was included in
filter plasma concentrations, effluent and cumulated effluent the model by adding an adsorption compartment
 96 Page 6 of 15 Pharm Res (2020) 37:96

Fig. 1 Schematic overview of the IDP model with reversible adsorption. Bold arrows indicate measurements, thin arrows mass transfer, dashed arrows
mechanistic flows of the dialyzer. cpl(pre), cpl(post), ceffl.: concentration of pre- and post-filter plasma, of effluent and cumulated effluent; CLbody: Clearance
mediated by the human body; CLRRT: total RRTclearance; kdeg: degradation rate; corr: corrected; cum.: cumulated FAds, Adsmax and kads rev: fraction, maximal
and reversible rate of adsorption; Qeffl., Qbood adj.: effluent and adjusted blood flow rate.

representing the binding capacity of the dialysis membrane

dA1 CL RRT
(A2). It described the adsorbed amount of the drug (A2) as a ¼−  A1
fraction (Fads) of the amount of drug cleared by dialysis. dt V cent
   
This expression can be expanded by a term describing a A2 CL body
 ð1−F Ads Þ þ F Ads  1− −
maximal amount adsorbed to the membrane (Aadsmax) Adsmax V cent
and/or by a term that describes the reversibility of the
adsorption by the rate kads rev. Only the amount that is  A1
removed by dialysis and not adsorbed to the membrane ð11Þ
is collected in the effluent compartment. The IDP mod-
el enables flexible estimation of the adsorption process The adaptions to the other equations for this scenario can
allowing inclusion of covariates, e.g. membrane material be found in the Appendix model 1.
on the maximal amount adsorbed to the membrane,
and adaption in the number of estimated parameters Drug Degradation in the Cumulated Effluent
based on quantifiability and plausibility.
Notably, the RRT clearance is split in a fraction In order to estimate possible degradation of the drug, a deg-
mediated by dialysis and/or filtration, and a fraction radation rate constant kdeg was included in the cumulated ef-
mediated by adsorption. The notation shown in Eqs. fluent compartment (Eq. 9).
7–10 represents a stable total RRT clearance due to
an increase of dialysis and filtration efficacy, when the Observations
adsorption maximum was reached. Another parameteri-
zation represents a stable fraction of RRT clearance The predictions of the concentration measurements for
mediated by dialysis and filtration not depending on the RRT specimens pre- and post-filter plasma, effluent
the amount of drug bound to the membrane. In such and cumulated effluent and for the volume of the cu-
a scenario, the total RRT clearance affecting the central mulated effluent were obtained as shown in Eq. 12–16.
compartment was accordingly not stable over time, but Notably, no assumption on the volume of post-filter
dependent on the amount of drug bound to the mem- plasma and effluent was required to incorporate these
brane (A2): measurements in the IDP model and accordingly no
Pharm Res (2020) 37:96 Page 7 of 15 96

assumptions on the volume and no ‘virtual’ compart- filter model, using pre- and post-filter plasma concentrations, an
ments were required. RRT clearance of >20% of the body clearance was needed to be
detected with a power >80% for piperacillin. The effluent based
approaches, using the effluent and the cumulated effluent, re-
A1
cplasma pre−filter ¼ ð12Þ spectively, were much more sensitive and detected an RRT
V cent clearance >0.03% and 0.06% of the body clearance with a pow-
A1 A1 CL RRT er >80% in the piperacillin example, respectively. The IDP
cplasma post− f ilter ¼ − 
V cent V cent Q blood ad j: model, using all RRT specimens at a time, was the most sensitive
approach, where already an RRT clearance of only 0.02% of the
Q blood ad j:
   ð13Þ body clearance resulted in >80% power for piperacillin. The
Q blood ad j: − Q R F post þ Q F RR same pattern was observed in the other drug examples (tigecy-
cline, colistin and linezolid).
A1 CL RRT dA2 1
ceffluent ¼  −  ð14Þ
V cent Q effl: dt Q effl: Accuracy and Precision
A3
ccumulated effluent ¼ ð15Þ The accuracy and precision of RRT clearance assuming no deg-
V effl:
radation or adsorption determined via rBias and rRMSE using
cvolume cumulated effluent ¼ A4 ð16Þ all five approaches is presented in Fig. 3. The RRT clearance
was estimated least precisely by the reduced model and by the
The post-filter plasma concentration was calculated based
post-filter model (rBias: 13.4% and − 1.6%, rRMSE: 82.3% and
on the mass balance equations. For post-filter plasma sampled
32.7% rRMSE for an RRT clearance 20% of the body clear-
before addition of the replacement fluid, a correction as de-
ance with the reduced and post-filter model, respectively).
scribed in Eq. 3 was employed. However, especially in CVVH
Generally, with increasing RRT clearance, the estimation of
and CVVHDF dilution effects are complex. In order to ensure
the RRT clearance was more precise by the plasma-based
no omitted dilution effects, the determination of the hemato-
approaches. The (cumulated) effluent based models gave accu-
crit in pre- and post-filter samples might be useful.
rate and precise estimates of the RRT clearance over all investi-
The measured effluent concentration was influenced by the
gated drug examples (rBias <1%, rRMSE<7%). The IDP mod-
drug adsorbed to the dialysis membrane (Fig. 1). Therefore,
el was slightly more precise as compared to the other approaches
the change of drug amount per time in the membrane com-
and accurate over all tested scenarios.
partment (dAdt2 , Eq. 8) per dialysate flow was included in Eq. 14.
In contrast to calculating amounts from measured cumulated Drug Degradation in the Cumulated Effluent
effluent concentration and volume before performing the analy-
sis, cumulated effluent concentration and volume are integrated Drug degradation in the cumulated effluent was simulated and
in the analysis. The amount in the effluent (Eq.9) was linked to its influence on accuracy and precision of the estimated RRT
volume and concentration of the cumulated effluent, which are clearance was evaluated for the cumulative effluent approach
directly measurable in a dialysis study. and for the IDP model. A simulated degradation half-life of
The volume of the cumulated effluent was expressed as 24 h in a dosing interval of 8 and 12 h lead to a rBias of
dialysate flow rate over time (Eq.10) and linked to the obser- −12.1% and − 19.7% and an rRMSE of 13.4% and 20.3%,
vations as described in Eq. 16. The drug concentration in the respectively, when RRT clearance was estimated with the cumu-
cumulated effluent was described based on the prediction for lated effluent approach. The IDP model resulted in accurate and
the volume and amount (Eq. 15). precise estimates of the RRT clearance (rBias: −0.07% and −
0.07%, rRMSE: 3.9% and 3.9%, for an 8 and 12 h dosing
The Simulation and Estimation Study interval). The degradation rate constant kdeg was quantifiable with
an rBias of −1.9% and − 1.1% and an rRMSE of 42.0% and
Power to Detect RRT Clearance 26.0% for an 8 and 12 h dosing interval, respectively.

The power to detect RRT clearance with the reduced, the post- Drug Adsorption to the Dialysis Membrane
filter, the effluent, the cumulated effluent and the IDP model is
presented in Fig. 2. The reduced model, using only pre-filter Adsorption to the dialysis membrane was simulated and its influ-
plasma data of dialysis and non-dialysis patients, had the lowest ence on accuracy and precision of the estimated RRT clearance
power to detect RRT clearance. For instance for piperacillin, an was evaluated for the effluent and cumulated effluent approaches
RRT clearance as high as 60% of the body clearance (1.8 L/h vs. as well as for the IDP model. The irreversible adsorption led to
3 L/h) would be detectable with a power >80%. For the post- imprecise and biased estimates of the RRT clearance using the
 96 Page 8 of 15 Pharm Res (2020) 37:96

Fig. 2 Power to detect RRT

clearance by reduced, post-filter,
effluent, cumulated effluent and IDP
approach in four drug examples.
Horizontal line indicates 80%
power.

effluent approaches (rBias: −65.0% and −64.8%, rRMSE: resulted in biased estimates for Vcent (rBias: 15.2%) as well and for
65.0% and 64.8% for the effluent and cumulated effluent ap- CLbody (rBias: 16.3%), but the structural model was not affected
proach, respectively). The reversible adsorption caused imprecise using the cumulated effluent approach except for body clearance
and biased estimates of the RRT clearance for the effluent ap- (rBias 22.3%). The IDP model gave accurate and precise esti-
proach (rBias −42.2%, rRMSE 42.3) as well as for the cumulated mates of the RRT clearance for the irreversible and reversible
effluent approach (rBias −35.9%, rRMSE 36.3%). For irrevers- case (0.49% and 0.71% rBias, 18.7% and 6.05% rRMSE, re-
ible adsorption, the structural model parameter estimates (i.e. spectively). No bias occurred in the structural model (rBias<5%)
Vcent, CLbody) were unbiased except for body clearance with both and the adsorption process was quantifiable with the IDP model.
effluent approaches. For reversible adsorption, the effluent model The reversible and irreversible fraction of adsorption were

Fig. 3 Accuracy and precision of

estimated RRTclearance by
reduced, post-filter, effluent,
cumulated effluent and IDP
approach. Horizontal line indicates
20% relative bias (rBias) and 20%
relative root mean squared error
(rRMSE).
Pharm Res (2020) 37:96 Page 9 of 15 96

estimated with an rBias of 0.07% and − 1.9% and an rRMSE of characterized with a fraction of the RRT clearance mediated
0.65% and 12.5%, respectively. via adsorption of 0.891 and a capacity limitation term. The
drug removed via dialysis processes was stable over time and
Pre-Clinical and Clinical Dataset Examples not dependent on the drug adsorbed to the membrane. The
maximal adsorption capacity was dependent on the material
Doripenem of the membrane and was 8.6 mg (5.1 mg - 11.8 mg, 95%
confidence interval) and 31.2 mg (26.5 mg - 36.7 mg) for
The IDP model was successfully applied to the doripenem triacetate and polysulphone membranes, respectively.
clinical dataset (Supplementary Table S2). Pre- and post- When the IDP model was applied to the data of the teico-
filter plasma, effluent, cumulated effluent and volume of the planin in vivo case study patients, a similar pattern as in the
effluent data were all described simultaneously within one in vitro study was observed (Supplementary Table S2, Fig. 5c).
model as visualized in the visual predictive checks (Fig. 4). The RRT clearance was in a similar range (4.91 L/h in vivo as
The RRT clearance was estimated to 2.46 L/h and the total compared to 4.59 L/h in vitro) and also mainly mediated via
effluent flow rate, here the sum of dialysis rate, replacement adsorption (fraction of the RRT clearance mediated via ad-
rate and fluid removal rate, was used as a proportional cova- sorption of 0.896 and 0.891 in vivo and in vitro, respectively).
riate relationship on effluent flow rate normalized to 3 L/h. A The drug removed via dialysis and filtration processes was not
degradation half-life of 13.5 h for doripenem in the cumulated dependent on the drug adsorbed to the membrane, but stable
effluent was estimated and improved the model fit significantly over time. Using the capacity limitation term, a maximal ca-
(dOFV: −9.16, p = 0.0025). A capacity limited binding of pacity of 40.3 mg was estimated for the polyamide membrane.
164 mg and a fraction adsorbed of 24.1% were estimated
and resulted in a dOFV of −14.65. The total RRT clearance
including adsorption mediated clearance did not decrease DISCUSSION
when the capacity limit was reached.
When comparing the results obtained from the IDP model The Integrated Dialysis Pharmacometric (IDP) Model
to the conventional models, similar estimates for RRT clear-
ance were provided by the post filter model, where adsorption The IDP model unites all RRT samples, i.e. pre- and post-
processes are included but not differentiated from the RRT filter plasma, effluent and cumulate effluent concentrations
clearance mediated via dialysis and filtration processes and volume measurements of the effluent within a unique
(2.46 L/h). The effluent approach, where only RRT clearance model. All settings regarding RRT type (continuous/inter-
mediated via dialysis is captured, provided lower results mitted, dialysis, filtration, dia-filtration), drug properties
(2.07 L/h). The cumulated effluent model provided an even (protein binding, red blood cell concentration), corrections
lower RRT clearance of 1.91 L/h, deviating by 8.4% from for pre- and post-filter replacement fluid or fluid removal
the effluent approach and by 28.8% from the IDP approach, rates and membrane types are considered with the IDP
since degradation of doripenem is omitted leading to under- model.
estimation of the RRT clearance. For the IDP model, more data as compared to the conven-
The inclusion of the effluent flow rate as covariate on RRT tional approaches are needed. As shown in the clinical exam-
clearance resulted in a dOFV of −25.09 (p < 0.1−5) for the ples, also reduced datasets can be evaluated with the IDP
IDP model and of −15.5, −17.7 and − 18.4 in the conven- model, but dependent on which RRT specimen is missing at
tional approaches, post-filter, effluent and cumulated effluent, the cost of accuracy and precision or not describing adsorption
respectively, and reduced IIV CLRRT in all approaches (e.g. or degradation (Table 1). The costs of collecting additional
from 25.4% to 6.7% for the IDP model). data to the pre-filter plasma samples can be deemed accept-
able, given that these trials are usually performed in small
Teicoplanin patient collectives and increase of the patient sample size is
not an option. Hence, as much information as possible should
The IDP model was successfully applied to the teicoplanin be derived from these trials and even samples from compart-
in vitro data (Supplementary Table S3) and the graphical eval- ments that are commonly considered waste (effluent, cumu-
uation of the individual fits showed good alignment of the lated effluent) can be highly informative.
model prediction with the observed data (Fig. 5a). An RRT The IDP model can be beneficial to guide therapy by pro-
clearance of 4.59 L/h was estimated, which was changing viding a better understanding and quantification of RRT pro-
over time. The RRT clearance was mainly mediated via ad- cesses potentially allowing improved or precision dosing. Its
sorption and only a minor part of the drug was removed benefit can potentially be expanded in clinical applications,
constantly via the dialysis processes. This was reflected in the e.g. in TDM processes. Modern software supporting TDM
mass balance analyses (Fig. 5b). The adsorption was could potentially allow parallel evaluation of plasma and
 96 Page 10 of 15 Pharm Res (2020) 37:96

Fig. 4 Visual-predictive checks on doripenem concentrations in pre- and post-filter plasma, effluent and cumulated effluent and on volume of the cumulated
effluent in the 5th, 50th and 95th percentile with the shaded area describing the 90% confidence interval.

effluent measures alongside information on the RRT like the underlying mechanisms of RRT. The pharmacokinetic
membrane type or flow rates. model used by Leuppi-Taegtmeyer et al. was incorporating
Some inconsistency of the usage of the mass balance de- post-filter plasma measurements and effluent measurements
rived conventional approaches to calculate RRT clearance within one model and described RRT clearance based on the
were discussed by Atkinson [11], who emphasized using the mass balance. However, the model required assumptions on
adjusted blood flow rate corrected for hematocrit and red the fictive volume of filter and cartridge, while the IDP model
blood cell concentration in the post-filter approach. Notably, incorporates all measurements based on the actual flow rates
no adjustment to the effluent was required in case of pre-filter independent of the hemofilter “distribution” volume and is
dilution or post-dilution. A correction term for pre-filter dilu- therefore more in line with the true mechanisms of RRT.
tion as described by Pea et al. [2], characterized the translation The characterization and quantification of adsorption pro-
of efficacy of a post-dilution system to a pre-dilution system. cesses was not discussed by Leuppi-Taegtmeyer, even though
Pre-filter dilution is less efficient due to a dilution of the blood colistin has been described to display relevant adsorption to
before it enters the dialysis membrane. Notably, when con- the hemofilter [22, 23]. The IDP model allowed to describe
centrations were determined in a pre-dilution system, this cor- and quantify irreversible, reversible and capacity limited
rection factor was not needed and would even violate the mass adsorption.
balance equation. The measured concentrations in the efflu-
ent in a pre-filter dilution system will be lower as compared to The Simulation and Estimation Study
the concentration in a post-filter dilution system and accord-
ingly the differences are addressed in the mass balance equa- Power, Accuracy and Precision
tion without correction factor.
Some steps towards simultaneous analyses were taken by The IDP model was superior regarding precision, accuracy
Leuppi-Taegtmeyer et al. [21], but were not fully in line with and power especially to the reduced model, where only pre-
Pharm Res (2020) 37:96 Page 11 of 15 96

Fig. 5 (a) individual fits teicoplanin

in vitro, points: observation, solid
lines: individual model fit, black: pre-
filter plasma, red: post-filter plasma,
green: effluent, dashed line: lower
limit of quantification. (b) mass
balance teicoplanin in vitro, solid
lines: amount adsorbed to the
dialysis membrane, dashed line:
amount of drug in plasma, red:
triacetate membrane, blue:
polysulphone membrane. (c)
individual fits teicoplanin in vivo,
points: observation, solid lines:
individual model fit, black: pre- filter
plasma, red: post-filter plasma,
green: effluent.

filter plasma concentrations were included. The reduced Drug Degradation in the Cumulated Effluent and Drug Adsorption
model provided a biased estimate for the RRT clearance. to the Dialysis Membrane
This can be explained by the distribution of the esti-
mates for RRT clearance (Supplementary Fig. S4), When degradation or adsorption was included, the IDP mod-
where the lower boundary was 0 but no upper limit el was superior regarding accuracy and precision of the esti-
was given, which resulted in a skewed distribution of mated RRT clearance as compared to the effluent based
estimates (mean: 0.43 L/h, median: 0.36 L/h, true: approaches. For the cumulated effluent model, the impact of
0.3 L/h). The post-filter model was less sensitive and degradation on the RRT clearance depended on the collec-
precise as compared to the approaches using tion interval and the stability of drug in the effluent. However,
(cumulated) effluent data. When the differences between using the IDP model, RRT clearance was estimated accurate-
pre- and post-filter plasma are small, these might be ly and precisely over all scenarios and the degradation half-life
hidden due to the residual error of both specimens. was directly estimable within the model.
For the (cumulated) effluent approaches, RRT clearance For the (cumulated) effluent approaches, adsorption lead to
is detectable more reliably since drug concentrations ex- biased RRT clearance estimates. Using the conventional ef-
ceeding the range of the additive error strongly support fluent approach, even the estimated structural model param-
the identification of an RRT clearance. This indicates eters (i.e. Vcent, CLbody) were affected as well in case of reversible
that detectability of drug and hence the lower limit of binding processes to the dialysis membrane. The underlying
quantification of the bioanalytical method influences the assumption in the effluent approach, that pre-filter plasma
power to detect RRT processes. Adsorption or degrada- concentrations and effluent concentrations are in a constant
tion processes potentially reduce the power to detect ratio not changing over time (Equation 4) lead to over-
dialysis clearance since the less sensitive plasma-based estimation of volume of distribution and under-estimation of
approaches are of higher relevance in such scenarios. total clearance (Supplementary Fig. S5). Instead, the IDP
Overall, regarding accuracy, precision and power, an model accounted for the changing ratio of pre-filter plasma
approach using (cumulated) effluent data as in the IDP to measured effluent concentration (Eq. 14) and provided un-
model is beneficial. biased estimation of the structural model parameters, the
 96 Page 12 of 15 Pharm Res (2020) 37:96

Hct postfilter
(frequently)
RRT clearance and the adsorption processes in all tested

X
scenarios.

Hct pre-filter Pre-Clinical and Clinical Dataset Examples

Doripenem
(once)

X
X
The IDP model described pre- and post-filter plasma, effluent
and cumulated effluent concentrations and the volume of the
Volume cumulated

cumulated effluent simultaneously and successfully character-

Study Design Recommendations for Application of the IDP Model Approach and Contribution of Specimen to Derive Quantitative Information. Hct: Hematocrit

ized degradation in the cumulated effluent. A capacity limited

adsorption to the dialysis membrane was estimated. With re-
effluent

spect to the surface area, similar adsorbed amounts of the

X

X
X
same material, polyacrylonitrile filters, was found for ticarcil-
lin in vitro (up to 85 mg, 0.6 m2 surface area) [24], while
Cumulated

167 mg adsorbed doripenem was estimated for 1.05 m2 sur-

effluent

face area with the IDP model in the present study. The RRT
X

X
X

clearance determined with the IDP model was more in line

with the RRT clearance determined with the post-filter ap-
Effluent

proach while effluent based approaches would lead to smaller

X

X
X

RRT clearances not capturing adsorption processes. The es-

timated degradation rate constant for doripenem (half-life of
Post-filter
Specimen measured

Plasma

13.5 h) was in line with degradation observed in destabilizing

conditions in the literature such as elevated temperature, ox-
X

X
X

idative stress or exposure to UV radiation (half-life of 12 h and

Pre-filter

27 h vs. >100 h in destabilizing vs. stabilizing conditions, re-

Plasma

spectively) [12, 25, 26]. Hence, the estimated degradation was

X

X
X

in the range of previously determined half-lives and suggested

destabilizing conditions in the effluent. The dOFV and the
Degradation determined

reduction of IIV in RRT clearance due to covariate inclusion

No filtration processes
No filtration processes

Adsorption excluded,
Adsorption excluded

of effluent flow rate as a covariate in all approaches underlined

(only CVVHD)

(only CVVHD)

a dependency of the actual RRT clearance to the dialysis

Requirements

settings following the mass balance equations. However, it

remains unclear if RRT clearance is in all cases linearly chang-
ing with effluent and/or blood flow rates. Saturable dialysis
and filtration processes are possible scenarios as well as non-
linearity through adsorption processes or protein binding [2].
No quantification of degradation in the cumulated effluent possible
No quantification or detection of adsorption processes possible
No quantification or detection of adsorption processes possible
Quantification of drug degradation in the cumulated effluent

Employing the IDP model in a systematic investigation across

different dialysis settings could elucidate such correlations. For
Quantification of adsorption to the dialysis membrane

doripenem, this was not observed and the linearity between

Accurate and precise determination of RRTclearance

Verify and correct for dilution processes occurring

Quantitative information gained from IDP approach

RRT clearance and effluent flow rate was considered

in CVVH(D)F Estimation of effluent flow rate

plausible.
No correction for dilution processes possible
No correction for dilution processes possible

No estimation of effluent flow rate possible

No estimation of effluent flow rate possible

The integrated analysis of concentration in the cumulated

effluent and volume of the cumulated effluent instead of pre-
viously calculating the amount of drug in the cumulated efflu-
ent allowed the use of all concentration measurements (n =
40), even when no volume measurement (n = 31) was
recorded.

Teicoplanin
Table 1

The in vitro study of teicoplanin revealed a capacity limited

binding of teicoplanin to the membrane. The maximal
Pharm Res (2020) 37:96 Page 13 of 15 96

amount adsorbed to the membrane was depending on the The effluent specimen allowed accurate, precise and sensi-
membrane type, being in line with findings of Shiraishi et al. tive estimation of the RRT clearance when no adsorption of
[13]. However, only the IDP model allowed for quantification the drug to the dialysis membrane occurred. However, the
of the fraction and maximal amount of drug adsorbed to the effluent-bases approach is highly sensitive to the effluent flow
membrane while adsorption or capacity limitation could be rate and therefore undocumented or unmeasured discrepan-
explored only graphically with the conventional approaches. cies would lead to erroneous results in the estimated RRT
Hence, wrong conclusion would be drawn from the calcula- clearance.
tions based on the conventional approaches not accounting The cumulated effluent approach was providing accurate,
for adsorption. The fraction of teicoplanin removed via precise and sensitive estimation of RRT clearance at a low
dialysis was very small, which was in line with the pre- sample number. However, adsorption to the dialysis mem-
vious studies [27]. brane and undetected drug degradation in the cumulated
The adsorption and filtration behavior of teicoplanin was effluent can lead to biased results. Also, time-and concentra-
in line with the in vivo results, suggesting that the behavior of a tion-dependencies in the RRT clearance cannot be detected.
drug in vitro described with the IDP model is a sufficient base to The cumulated effluent approach is independent to flow rates
transfer adsorption and dialysis behavior of a drug to clinic. when the volume is determined and the volume information
Protein binding plays an important role when the RRT can even support flow rate estimation for the effluent
clearance is only estimated based on drug characteristics, since approach.
it is assumed that only unbound drug can be dialyzed or fil- When degradation of the drug in the cumulated effluent
trated. The integrated model allows direct assessment of the was not excluded in previous in vitro studies or quantified in
RRT clearance without assumptions on the efficacy in dialysis stability tests, the additional collection of effluent samples and
and filtration processes based on the unbound fraction. the application of the IDP model is recommended.
Moreover, elimination processes, that are independent of di- When adsorption to the dialysis membrane was not exclud-
alysis and filtration, e.g. adsorption to the dialysis membrane, ed for the respective membrane type, the collection and the
can be quantified. RRT clearance of highly protein bound use of all RRT specimens is highly recommended. The post-
drugs, e.g. teicoplanin presenting a protein binding of >90% filter approach including hematocrit measurements is re-
in patients with normal serum albumin concentrations [28], quired to quantify the extent of total RRT clearance, while
can be quantified overcoming the assumption of low RRT the effluent specimen allows time dependent insight on the
clearance due to the high protein binding acknowledging binding process and the cumulated effluent provides robust,
adsorption. flow independent mass balance information.
When degradation and adsorption are excluded in previ-
Study Design Recommendations ous or in vitro studies, dialysis based approaches are recom-
mended and the sampling size can be reduced by using the
For future studies, we recommend to consider which RRT cumulated effluent approach.
specimen might provide useful input to the analysis. An over-
view of implications, requirements and the respective RRT Limitations
specimens to obtain is provided in Table 1.
The post-filter RRT specimen can provide valuable infor- This study was focusing on continuous RRT and no proof of
mation and is itself not influenced by drug degradation. concept for intermitted RRT in patients was provided. This
Moreover, drug adsorption to the dialysis membrane is intrin- was considered reasonable, since patients undergoing contin-
sically included or detectable when the post-filter approach is uous RRT are often more vulnerable, the dosing after RRT is
used to calculate an RRT clearance. However, the post-filter not possible like for intermitted patients and in-patient com-
approach was less sensitive and less precise as compared to the parisons of pharmacokinetic with and without RRT is often
effluent approaches and is highly dependent on precise knowl- hard to get. However, the underlying theory of RRT is similar
edge of the blood flow rate and the hematocrit, where unde- and thus, our findings might be transferable.
tected variabilities or missing information would lead to erro- Since the main focus was on RRT clearance, body clear-
neously calculated RRT clearances. In contrast to the effluent ance was estimated as one parameter and it was not distin-
flow rate, which can be verified by the cumulated volume in guished between non-renal and renal clearance.
the effluent bag, no direct verification is possible for However, the IDP model can easily be extended to
blood flow. In addition, for methods using filtration or estimate non-renal and renal clearance separately as
replacement fluids (CVVH and CVVHDF), a determi- demonstrated by Roberts et al. [6].
nation of the hematocrit in each sample, i.e. in the pre- Due to a lack of hematocrit measurements in the doripe-
and post-filter sample, is recommended to verify the nem dataset, a hematocrit of 0.3 in all patients was assumed,
occurring dilution processes. which matches the hematocrit in patients undergoing RRT.
 96 Page 14 of 15 Pharm Res (2020) 37:96

Yet, as the hematocrit influences the calculated post-filter exceeds the permitted use, you will need to obtain permission
clearance, use of a different hematocrit will lead to a different directly from the copyright holder. To view a copy of this
adsorption profile and hence the estimated amount of doripe- licence, visit http://creativecommons.org/licenses/by/4.0/.
nem bound the dialysis membrane needs to be interpreted
with caution. Moreover, undetected dilution effects cannot
be excluded. We therefore want to underline the importance REFERENCES
of measuring the hematocrit both in pre- and post-filter sam-
ples to parameterize the pre-post-filter clearance correctly and 1. Li AMMY, Gomersall CD, Choi G, Tian Q, Joynt GM, Lipman J.
directly account for dilution effects in the measured post-filter A systematic review of antibiotic dosing regimens for septic patients
receiving continuous renal replacement therapy: do current studies
samples.
supply sufficient data? J Antimicrob Chemother [Internet]. 2009
For the teicoplanin in vivo example only two patients with [cited 2017 Apr 25];64:929–37. Available from: https://academic.
full pharmacokinetic data were available and it can therefore oup.com/jac/article-lookup/doi/10.1093/jac/dkp302
only be seen as an indicator for transferability of the IDP 2. Pea F, Viale P, Pavan F, Furlanut M. Pharmacokinetic considera-
approach from in vitro to in vivo and not as a base for reliable tions for antimicrobial therapy in patients receiving renal replace-
ment therapy. Clin Pharmacokinet [Internet]. 2007;46:997–1038
clinical conclusions. Available from: http://link.springer.com/10.2165/00003088-
200746120-00003.
3. Caltabiano S, Cizman B, Burns O, Mahar KM, Johnson BM,
CONCLUSION Ramanjineyulu B, et al. Effect of renal function and dialysis modal-
ity on daprodustat and predominant metabolite exposure. Clin
Kidney J [Internet]. 2019;12:693–701 Available from: https://
To conclude, the IDP model allowed the accurate, precise and academic.oup.com/ckj/article/12/5/693/5333182.
sensitive determination of RRT clearance and was superior to 4. Lee CS, Marbury TC, Benet LZ. Clearance calculations in hemo-
conventional approaches. Adsorption and degradation pro- dialysis: application to blood, plasma, and dialysate measurements
cesses can be quantified with the IDP model, which was not for ethambutol. J Pharmacokinet Biopharm [Internet]. 1980 [cited
2017 Jun 13];8:69–81. Available from: http://www.ncbi.nlm.nih.
possible with the conventional approaches. The IDP model gov/pubmed/7381692.
was successfully applied to in vitro and in vivo clinical data and 5. Broeker A, Wicha SG, Dorn C, Kratzer A, Schleibinger M, Kees F,
degradation of the drug in the cumulated effluent and adsorp- Heininger A, Kees MG, Häberle H Tigecycline in critically ill
tion to the dialysis membrane was quantifiable. Accordingly, patients on continuous renal replacement therapy: a population
pharmacokinetic study. Crit Care [Internet]. Crit Care; 2018;22:
the IDP model is a promising approach to better make use of 341. Available from: https://ccforum.biomedcentral.com/articles/
clinical trial data and will provide quantitative insights in re- 10.1186/s13054-018-2278-4
moval of drug via RRT. 6. Roberts JA, Udy AA, Bulitta JB, Stuart J, Jarrett P, Starr T, et al.
Doripenem population pharmacokinetics and dosing requirements
for critically ill patients receiving continuous venovenous haemo-
ACKNOWLEDGMENTS AND DISCLOSURES diafiltration. J Antimicrob Chemother [Internet]. 2014;69:2508–16
Available from: https://academic.oup.com/jac/article-lookup/
J.A. Roberts would like to acknowledge funding from the doi/10.1093/jac/dku177.
Australian National Health and Medical Research Council 7. FDA. Guidance for Industry Pharmacokinetics in Patients with
Impaired Renal Function — Study Design , Data Analysis , and
for a Centre of Research Excellence (APP1099452) and a
Impact on Dosing ans Labeling. 2010; Available from: http://
Practitioner Fellowship (APP1117065). w w w . f d a . g o v / d o w n l o a d s / D r u g s /
GuidanceComplianceRegulatoryInformation/Guidances/
FUNDING INFORMATION UCM204959.pdf
8. Meyer B, Ahmed el Gendy S, Delle Karth G, Locker GJ, Heinz G,
Jaeger W, et al. How to Calculate Clearance of Highly Protein-
Open Access funding provided by Projekt DEAL. Bound Drugs during Continuous Venovenous Hemofiltration
Demonstrated with Flucloxacillin. Kidney Blood Press Res
Open Access This article is licensed under a Creative [Internet]. 2003 [cited 2017 Jul 10];26:135–40. Available from:
Commons Attribution 4.0 International License, which per- www.karger.com
mits use, sharing, adaptation, distribution and reproduction in 9. Weiler S, Seger C, Pfisterer H, Stienecke E, Stippler F, Welte R,
et al. Pharmacokinetics of Caspofungin in critically ill patients on
any medium or format, as long as you give appropriate credit continuous renal replacement therapy. Antimicrob Agents
to the original author(s) and the source, provide a link to the Chemother [Internet]. 2013;57:4053–7 Available from: http://
Creative Commons licence, and indicate if changes were aac.asm.org/lookup/doi/10.1128/AAC.00335-13.
made. The images or other third party material in this article 10. Liesenfeld K-H, Staab A, Härtter S, Formella S, Clemens A, Lehr
T. Pharmacometric Characterization of Dabigatran Hemodialysis.
are included in the article's Creative Commons licence, unless
Clin Pharmacokinet [Internet]. Springer International Publishing
indicated otherwise in a credit line to the material. If material AG; 2013 [cited 2017 Apr 13];52:453–62. Available from: http://
is not included in the article's Creative Commons licence and link.springer.com/10.1007/s40262-013-0049-6
your intended use is not permitted by statutory regulation or
Pharm Res (2020) 37:96 Page 15 of 15 96

11. Atkinson AJ. Pitfalls in the calculation of hemodialysis clearance 21. Leuppi-Taegtmeyer AB, Decosterd L, Osthoff M, Mueller NJ,
and in the assessment of Dialysis efficacy. Transl Clin Pharmacol. Buclin T, Corti N. Multicenter Population Pharmacokinetic
2016;24:153–6. Study of Colistimethate Sodium and Colistin Dosed as in Normal
12. Berthoin K, Le Duff CS, Marchand-Brynaert J, Carryn S, Tulkens Renal Function in Patients on Continuous Renal Replacement
PM. Stability of meropenem and doripenem solutions for adminis- Therapy. Antimicrob Agents Chemother [Internet]. 2018;63:
tration by continuous infusion. J Antimicrob Chemother [Internet]. e01957–18. Available from: http://aac.asm.org/lookup/doi/10.
Oxford University Press; 2010 [cited 2017 Jul 12];65:1073–5. 1128/AAC.01957-18
Available from: https://academic.oup.com/jac/article-lookup/ 22. Markou N, Fousteri M, Markantonis SL, Zidianakis B, Hroni D,
doi/10.1093/jac/dkq044 Boutzouka E, et al. Colistin pharmacokinetics in intensive care unit
13. Shiraishi Y, Okajima M, Sai Y, Miyamoto K, Inaba H. Elimination patients on continuous venovenous haemodiafiltration: an observa-
of teicoplanin by adsorption to the filter membrane during haemo- tional study. J Antimicrob Chemother [Internet]. Oxford
diafiltration: screening experiments for linezolid, teicoplanin and University Press; 2012 [cited 2017 Jul 13];67:2459–62. Available
vancomycin followed by in vitro haemodiafiltration models for tei- from: https://academic.oup.com/jac/article-lookup/doi/10.
coplanin. Anaesth intensive care [internet]. 2012;40:442–9. 1093/jac/dks257
Available from. http://www.ncbi.nlm.nih.gov/pubmed/ 23. Honore P, Jacobs R, Lochy S, De Waele, De Regt J, Van Gorp V,
22577909. et al. Acute respiratory muscle weakness and apnea in a critically ill
14. Meyer B, Traunmüller F, Hamwi A, Hollenstein UM, Locker GJ, patient induced by colistin neurotoxicity: key potential role of
Sperr WR, et al. Pharmacokinetics of teicoplanin during continu- hemoadsorption elimination during continuous venovenous hemo-
ous hemofiltration with a new and a 24-h used highly permeable filtration. Int J Nephrol Renovasc Dis [Internet]. Dove Press; 2013
membrane: rationale for therapeutic drug monitoring-guided dos- [cited 2017 Jul 13];6:107. Available from: http://www.ncbi.nlm.
age. Int J Clin Pharmacol Ther [internet]. 2004;42:556–60. nih.gov/pubmed/23776390.
Available from. http://www.ncbi.nlm.nih.gov/pubmed/ 24. T.P. Wan W, Guerra Valero Y, Choi GYS, Ordóñez Mejia JL,
15516025. Wallis SC, Joynt GM, et al. In-vitro adsorption and sieving coeffi-
15. Ulldemolins M, Martín-Loeches I, Llauradó-Serra M, Fernández J, cient of ticarcillin-clavulanate during continuous haemofiltration.
Vaquer S, Rodríguez A, et al. Piperacillin population pharmacoki- Int J Antimicrob Agents [Internet]. Elsevier B.V.; 2019;54:261–4.
netics in critically ill patients with multiple organ dysfunction syn- Available from: https://doi.org/10.1016/j.ijantimicag.2019.03.
drome receiving continuous venovenous haemodiafiltration: effect 018
of type of dialysis membrane on dosing requirements. J Antimicrob 25. de Souza Barbosa F, Cassol JPE, Batista LA das C, Cordeiro EWF,
Chemother [Internet]. Oxford University Press; 2016 [cited 2017 Santos MC, Pohlmann AR, et al. Stability of doripenem in recon-
Apr 13];71:1651–9. Available from: https://academic.oup.com/ stituted solution - thermal and oxidative decomposition kinetics and
jac/article-lookup/doi/10.1093/jac/dkv503 degradation products by LC-MS. Biomed Chromatogr [Internet].
16. Boeckmann AJ, Sheiner LB, Beal SL, Bauer RJ, editors. 2017;31:e3940. Available from: http://doi.wiley.com/10.1002/
NONMEM Users Guide. Ellicott City, Maryland: NONMEM bmc.3940
Project Group University of California at San Francisco; 2011.
26. Fuhr F, Paliosa PK, Costa MCN, Garcia CV, Schapoval EES,
17. Broeker A, Wicha SG. Assessing parameter uncertainty in small-n
Steppe M, et al. Bioassay Applied to Quantitative Determination
pharmacometric analyses: value of the log-likelihood profiling-
of Doripenem in Powder for Injection – Method Validation and
based sampling importance resampling (LLP-SIR) technique. J
Degradation Kinetics Study. Curr Pharm Anal [Internet]. 2013;9:
Pharmacokinet Pharmacodyn [Internet]. Springer US; 2020;
244–51. Available from: http://www.eurekaselect.com/openurl/
Available from: https://doi.org/10.1007/s10928-020-09682-4.
content.php?genre=article&issn=1573-4129&volume=9&issue=
18. Lindbom L, Pihlgren P, Jonsson N. PsN-Toolkit—A collection of
3&spage=244
computer intensive statistical methods for non-linear mixed effect
modeling using NONMEM. Comput Methods Programs Biomed 27. Hillaire-Buys D, Peyriere H, Lobjoie E, Bres J, Ossart M, Despaux
[Internet]. 2005;79:241–57. Available from: https://linkinghub. E. Influence of arterio-venous haemofiltration on teicoplanin elim-
elsevier.com/retrieve/pii/S016926070500101X ination. Br J Clin Pharmacol [Internet]. 1995;40:95–7 Available
19. Viaene E, Chanteux H, Servais H, Mingeot-Leclercq M-P, Tulkens from: http://doi.wiley.com/10.1111/j.1365-2125.1995.tb04543.
PM. Comparative stability studies of antipseudomonal beta- x.
lactams for potential administration through portable elastomeric 28. Byrne CJ, Parton T, McWhinney B, Fennell JP, O’Byrne P, Deasy
pumps (home therapy for cystic fibrosis patients) and motor- E, et al. Population pharmacokinetics of total and unbound teico-
operated syringes (intensive care units). Antimicrob Agents planin concentrations and dosing simulations in patients with hae-
Chemother [Internet]. American Society for Microbiology matological malignancy. J Antimicrob Chemother [Internet].
(ASM); 2002 [cited 2017 Jul 12];46:2327–32. Available from: 2018;73:995–1003 Available from: https://academic.oup.com/
http://www.ncbi.nlm.nih.gov/pubmed/12121900. jac/article/73/4/995/4759196.
20. Guenter SG, Iven H, Boos C, Bruch H-P, Muhl E.
Pharmacokinetics of levofloxacin during continuous Venovenous
Hemodiafiltration and continuous Venovenous hemofiltration in
critically ill patients. Pharmacotherapy [Internet]. 2002;22:175–83 Publisher’s Note Springer Nature remains neutral with regard to juris-
Available from: http://doi.wiley.com/10.1592/phco.22.3.175. dictional claims in published maps and institutional affiliations.
33546.