Beruflich Dokumente
Kultur Dokumente
1
30 ABSTRACT
31 To ensure safe and effective dosing of gentamicin in children therapeutic drug monitoring
37 was developed from retrospective data. Data were collected from 423 patients for model
38 building and a further 52 patients for external evaluation. A two-compartment model with
39 first-order elimination, best described gentamicin disposition. The final model included renal
40 function (described by fat-free mass and post-menstrual age) and serum creatinine as
41 covariates influencing gentamicin clearance (CL). Final parameter estimates were as follow
42 CL: 5.77 L/h/70 kg; central volume of distribution: 21.6 L/70 kg, peripheral volume of
44 evaluation suggested that current models developed in other paediatrics cohorts may not be
46 observations in this population. The final model developed in this study displayed good
47 predictive performance during external evaluation (root mean squared error: 46.0% and mean
48 relative prediction error: -3.40 %) and may therefore be useful to personalise gentamicin
49 dosing in this cohort. Further investigations should focus on evaluating the clinical
51
2
52 INTRODUCTION
54 oncology patients. Neutropenic patients are more susceptible to developing infections (1).
55 Sepsis is the primary cause of mortality and morbidity in oncology children with febrile
59 managing infectious complications in these individuals and are used as second line therapies
60 when treating Gram-negative infections and when resistance develops to first line agents (4).
62 variability. High and prolonged exposure to gentamicin has been associated with nephro- and
63 ototoxicity (5, 6). Paediatric oncology patients often receive long courses of gentamicin on
64 multiple occasions alongside chemotherapy, which further increases the risk of toxicity.
67 alter patients body composition and the pharmacokinetics of gentamicin, increasing the
68 difficulty of identifying the optimal dose (7-9). Increased clearance (CL) and volume of
69 distribution (Vd) for gentamicin has been seen in adults (7) and paediatrics with cancer (8,
70 10). Augmented renal clearance, possibly due to systemic inflammation, also has been
72 normalized by body surface area, it decreased in those patients receiving prior nephrotoxic
73 chemotherapy compared to those who did not receive nephrotoxic chemotherapy (8).
74 The Australian Therapeutic Guidelines (12) recommend that for safety and efficacy
76 such as Bayesian forecasting. While the log-linear regression method is more commonly used
3
77 (13) it may be more invasive, tedious and reactive in nature compared to Bayesian
78 forecasting. The former requires at least 2 blood sample measurements for predictions (14)
79 in contrast to Bayesian forecasting, which may require only 1 sample; and it is not able to
80 give dosage predictions until a dose has first been administered. In adults with febrile
82 achieving desired target exposure to gentamicin compared to using the log-linear regression
83 method (15).
87 and influential covariate factors associated with the drug of interest in the cohort receiving
88 such therapy. Few population pharmacokinetic models of gentamicin have been developed in
89 children between 2 and 18 years of age (16-18), and to our knowledge no models currently
96 population and externally evaluate the predictive performance of this model as well as
98 METHODS
100 Data for model building and model evaluation were collected retrospectively from the
101 medical records of paediatric oncology patients at the Lady Cilento Childrens Hospital,
4
102 Brisbane, Australia (formally known as the Royal Childrens Hospital). These patients
103 received gentamicin and had drug concentrations measured during routine therapeutic drug
104 monitoring (TDM) between 2008 and 2013 (model-building data set) and 2014 and 2015
105 (model-evaluation data set). Ethics approval for this study was obtained from the University
107 Childrens Health Queensland Hospital and Health Service Human Research Ethics
110 gentamicin serum drug concentrations and time of sample collection for drug measurement
111 along with patient age, sex, total body weight, height, serum creatinine concentration (Scr)
112 and body temperature readings were recorded. In addition, information on the date,
113 administration time and dose of specific nephrotoxic drugs (cisplatin and carboplatin)
114 patients received within 6 months preceding gentamicin usage were collected.
116 Paediatric oncology patients with a neutrophil count <1.0 cells/mm3 who developed a
117 fever (temperature 38.5 C for at least 1 hour or 38.0 C for at least 2 hours) and who were
118 considered critically ill or at high risk received gentamicin in addition to empiric broad
119 spectrum penicillin based therapy. Gentamicin was also added as directed treatment if Gram-
120 negative positive cultures were confirmed. Gentamicin was given as a 30 minute infusion
121 once daily, initiated at a dose of 7.5 mg/kg for patients under 10 years of age and 6 mg/kg for
122 patients over 10 years according to local hospital guidelines. Blood samples for measurement
123 of gentamicin serum concentrations during TDM were generally taken at 2 hours and 6-12
124 hours post-infusion. In addition, based on clinical judgment, samples before a subsequent
5
126 Gentamicin Assay
128 polarization immunoassay method at the Pathology Queensland Central laboratory, Brisbane,
129 Australia. Different immunoassays were applied over time. From 2008 to August 2011, a
131 used. From August 2011 to November 2011, an Abbott CMIA Immunoassay run on the
132 Abbot Architect i1000 Immunoassay Analyser was used. From November 2011 to December
133 2012, a Beckman Coulter CEDIA Immunoassay run on a Beckman Coulter DXC800
134 Analyser was used. From December 2012 onwards a Beckman Coulter PETINIA
135 Immunoassay run on a Beckman Coulter DXC800 Analyser was used. Assay lower limit of
136 quantification (LLOQ) varied across immunoassay method from 0.3 to 0.6 mg/L. Within- and
137 between-assay coefficients of variation were below 10% for all immunoassays employed.
139 Software
140 Population modelling was performed using NONMEM software v.7.3 (Icon
141 Development Solutions, Ellicott City, MD, USA) (20) with an Intel FORTRAN compiler
142 and PsN (Perl-speaks-NONMEM) version 3.5.3 (21). Typical population pharmacokinetic
144 (BOV) and residual unexplained variability (RUV) were estimated via the first-order
145 conditional estimation method with interaction (FOCE+I). Xpose (version 4.4.0;
149 Firstly, a base model was developed by testing one-, two- and three- compartment
150 structural disposition models with first-order elimination. Pharmacokinetic parameter values
6
151 were assumed to be log-normally distributed. BSV was modelled using an exponential model
152 (equation 1), and initially applied to all pharmacokinetic parameters and removed if not
153 supported. BOV was modelled using an additional random effects parameter (equation 2) and
156 P = TVP e( , , )
(2)
157 Here, TVP is the typical population parameter estimate and is the parameter estimate for
158 individual i, i,P is a symmetrically distributed random variable with a mean of zero and
159 variance of 2BSV_P, where the square root of the variance represents the reported BSV, and
160 is the parameter value for individual i on the jth occasion, ki,P is the within patient random
161 effect, which is independent and normally distributed with a mean of zero and variance of
162 2BOV_P. Proportional, additive and combined models and different error models for
163 concentration measured using different assays were compared during model building to
164 describe RUV. Gentamicin drug concentrations reported as below LLOQ were substituted
167 included patients renal function (described by body size and age), Scr measurements, sex,
168 use of nephrotoxic drugs and body temperature. After a structural base model was
169 established, potential covariates were screened initially via graphical exploration, considering
170 shrinkage (23). Covariate model building was in the first instant based on physiological
172 of glomerular filtration rate, and standardization for age-expected serum creatinine
173 concentration.
7
174 The impact of changes in renal function due to maturation on gentamicin clearance
175 was included using the glomerular filtration rate prediction model developed by Rhodin et al.
176 (GFRmat; mL/min, see equation 3) (24). Three descriptors of body size were tested in the
177 GFRmat equation: total body weight, normal fat mass (NFM) (25) and fat-free mass (FFM)
179 fraction of fat mass being estimated. Allometric scaling (27, 28) using each of these three
180 descriptors of body size (total body weight, NFM and FFM) was applied to all other
( ) . ( ) .
182 GFR = . 112 (3)
. ( ) .
( )
183 P = TVP (4)
184 Here, P is the parameter of interest, TVP is the typical population parameter value, the
185 size descriptori represents either body weight, FFM or NFM (kg) for an individual i, which
186 was standardised to 70 kg, to ease comparison to adult values. PMA is post-menstrual age
187 (weeks), calculated for each child by addition of 40 weeks, presenting a typical gestation
189 Afterwards, Scr and body temperature were subsequently tested using a linear,
190 exponential and/or power model. An inverse relationship between Scr and CL was examined,
191 with Scr standardised to the mean expected creatinine concentration for a person of the
192 relevant age and sex (29). For patients who had a Scr measurement below the lowest
193 reportable limit (< 30 mol/L), this value was fixed to the mean expected physiological
194 creatinine concentration for a person of the relevant age and sex (29). The categorical
195 covariates, sex and use of cisplatin and/or carboplatin within 6 months prior to the gentamicin
196 observation, were tested by estimating the typical parameter value for the most common
8
197 category and then obtaining the fractional difference in the parameter value for the second
201 for two models approximates the Chi-square (X2) statistic which can be tested for
202 significance (X21, 0.05 = 3.84). The Akaikes information criterion (AIC) was used for non-
203 nested models. Further support for model selection was obtained by evaluating the difference
204 in the magnitude of the change in explained parameter variability (EPV) after covariate
205 inclusion compared to the base model (30). Basic goodness-of-fit plots for observations
206 versus population predictions and conditional weighted residuals versus time after dose were
207 examined for internal model evaluation. Model predictive performance was assessed by
208 generating visual predictive check (VPCs) plots based on 500 simulations. Plot of normalized
209 prediction distribution errors (npde) versus time after dose based on 1000 simulations were
210 also generated and evaluated. Precision of model parameters was obtained from the relative
211 standard errors and confirmed through non-parametric bootstrapping of the final model with
213 External model evaluation of the final model and models published in the literature
214 The model-evaluation data set was used to externally evaluate the predictive
215 performance of the final model as well as any previous population models of gentamicin
216 published in the literature that were based on paediatric patients. A literature search was
217 conducted in September 2016 using PubMed and EMBASE databases with a restriction on
219 models. The following keywords were used during the search: gentamicin, paediatric,
221 developed in a paediatric population with a post-natal age 5 months using a compartmental
9
222 approach and were excluded if they did not provide enough information to be fully
223 reproducible.
225 NONMEM software, as stated above. Gentamicin concentrations were predicted using
227 measurement and relevant patient covariate values recorded for the evaluation data set, with
228 pharmacokinetic parameter values and covariate relationships set as determined in the final
229 model and in the previous publications. The predictive performance of each selected
230 population pharmacokinetic model was assessed visually (predicted concentration (Cpred) vs.
231 observed concentration (Cobs) plots and VPC plots) and numerically. Bias and precision were
232 calculated using the relative root mean square error (RMSE %) (equation 5) and mean
236 RESULTS
238 Data from a total of 423 paediatric oncology patients (219 males (52%)) with 2422
239 gentamicin concentration-time data points were used for model development. 88% of these
240 patients had febrile neutropenia and 12% had fever-free neutropenia. Patients median (range)
241 post-natal age was 5.2 years (0.2 18.2 years), weight was 19.4 kg (4.8 102.8 kg) and FFM
242 was 15.7 kg (3.4 72.6 kg), respectively. Seventy-three patients (17%) were younger than 2
243 years, 247 patients (59%) were 2 to 10 years old and 103 patients (24%) were older than 10
244 years. Gentamicin concentration measurements ranged from 0.3 to 62.5 mg/L with sampling
245 times after the end of the infusion ranging from 0.5 to 36.0 hours. A total of 372 (15%) of
10
246 gentamicin measurements were below the LLOQ at an average of 15 hours after dose. For
247 model evaluation, data from 52 paediatric oncology patients (22 males (42%)) with 174
248 gentamicin concentration-time points were used. Patients median (range) post-natal age was
249 7.9 years (0.7 16.8 years), weight was 22.2 kg (6.2 121.0 kg) and FFM was 20.5 kg (5.3
251 were 2 to 10 years old and 11 patients (21%) were older than 10 years. Gentamicin
252 concentration measurements ranged from 0.5 to 30.8 mg/L with samples taken from 0.75 to
253 34.3 hours after the end of the infusion. A total of 16 (9%) gentamicin measurements were
254 below the LLOQ. Key patient demographic characteristics, laboratory values and gentamicin
257 A two-compartmental model with first-order elimination best described the data
259 L/h/70 kg and typical central volume of distribution (V1) was 17.4 L/70 kg. BSV was
260 estimated for CL, V1 and peripheral volume of distribution (V2), including a correlation
262 Inclusion of GFRmat on CL and either FFM, body weight or NFM on V1,
263 intercompartmental clearance (Q) and V2 improved the fit and increased explained
264 variability. Using any body size descriptor (body weight, NFM or FFM) in the GFRmat
265 equation and for V1, Q and V2 resulted in a significant drop in the OFV (OFV=-406.6,
266 OFV=-403.1 and OFV=-408.3, respectively) (Supplemental material Table S1). However,
267 according to the AIC, FFM fit the data best compared to body weight and NFM. Addition of
269 (EPV=12.67 (L)) (Supplemental material Table S1). In addition to GFRmat, inclusion of an
270 association between Scr measurement and gentamicin CL using a power model (OFV=-
11
271 324.2) further significantly improved the fit. Inclusion of Scr increased EPV on CL compared
272 to the previous model that included only GFRmat (Supplemental material Table S1). In
273 subsequent steps, inclusion of use of nephrotoxic drugs and body temperature on gentamicin
277 VPC plots associated with the final model indicated good agreement between the
278 observed and predicted data (Figure 1). VPC, npde and goodness of fit plots associated with
280 Results from the non-parametric bootstrap were in agreement with the population
281 pharmacokinetic parameters in the final model with narrow 95% confidence intervals (see
282 Table 2). As eta-shrinkage on CL, V1 and V2 was relatively high in the final model (43%,
283 58% and 66%, respectively), diagnostic use of goodness-of-fit plots was limited (23) and
285 External model evaluation of the final model and models published in the literature
286 From the literature search only three population pharmacokinetic models of
287 gentamicin developed in paediatric populations were identified (Table 3) (16, 17, 32). These
288 models together with the final model from this study were tested regarding their predictive
289 performance with the external dataset. Observed versus population predicted concentration-
290 time plots and VPC plots obtained from all four models generated during external evaluation
291 are displayed in Supplemental material Figure S1 and Figure 2 respectively. Of the four
292 models evaluated the final population model developed in this study was the least biased with
293 a MRPE (%) of -3.40 and the most precise with a RMSE (%) of 46.0 compared to the other
12
295 DISCUSSION
296 This study developed and externally evaluated a population pharmacokinetic model of
297 gentamicin in paediatric oncology patients that could be used to improve dosage adjustment
298 of this drug in this cohort in the future. The final model included an effect of renal function
300 and carboplatin, was not identified as having a significant influence on the CL of gentamicin,
301 after taking into consideration maturation changes in renal function and potential function
302 impairment observed given patients Scr concentration. FFM was identified as the best
303 descriptor to model the influence of body size on gentamicin CL, V1, Q and V2. The final
304 model displayed good predictive performance, when applied to a new dataset not used for
306 Selection of a two-compartment model with first order elimination to describe the
307 disposition of gentamicin in this study is in agreement with a number of studies performed in
308 non-oncology children, neonates and adults included in a recent review (33). Only one
309 population pharmacokinetic model of gentamicin has been developed previously in children
310 who were of a similar age to those included in this study (18). This previous study involved
311 non-oncology children (median age of 2.4 years) who received gentamicin intramuscularly
312 (18). Typical gentamicin apparent CL (CL/F) in these children (7.21 L/h/70kg for a patient
313 with an average Scr of 33.0 mol/L), was higher than that estimated in this study based on
314 intravenous administration (5.77 L/h/70kg for a patient with an average Scr of 37.4 mol/L).
315 Two studies have been performed in paediatric oncology patients, but using non-
316 compartmental analysis methods (8, 10). In the first study (8), typical CL of 9.33 L/h/70kg
317 and 10.3 L/h/70kg was reported in paediatric patients receiving and not receiving
318 chemotherapy, respectively. In the second study (10), typical CL of 8.10 L/h/70kg was
319 reported. Comparison of CL of gentamicin obtained in this study with values published in the
13
320 literature is difficult because most previous studies are primarily based on neonates or infants
321 and involved different analysis techniques. Typical V1 estimated in this study (21.6 L/70kg)
322 was similar to other reports, however lower than that reported previously in non-oncology
323 children (24.5 L/70kg) and pre-term neonates (31.1 L/70kg) and higher than that reported in
325 The final model included an effect of body size in terms of FFM on all
326 pharmacokinetic parameters. When the final model was applied in an external dataset, it
327 predicted gentamicin concentrations more accurately and precisely than a model using NFM
329 gentamicin found total body weight was the most common covariate included as a body size
330 descriptor on CL and V1 (33). Considering that before 2015 no FFM equation for paediatrics
331 was available (26) the inclusion of FFM in models for paediatric populations was limited.
332 This study found FFM explained more variability between subjects than other body size
333 descriptors, possibly due to the fact that gentamicin is highly hydrophilic and mainly
334 distributes into muscle mass or lean body weight rather than fat. One study of gentamicin in
335 elderly patients has included FFM on V1 (34) and another has reported that FFM improved
337 An equation describing changes in renal function over time due to organ maturation
338 (GFRmat) was utilised here. This equation was originally developed from a large study
339 involving 923 patients, from premature neonates to young adults, of which most had normal
340 renal function and only 5% were known oncology patients (24). We found that additional
341 inclusion of Scr on CL after consideration of renal function maturation through GFRmat
342 improved model fit further. An accurate bedside estimate of renal function does not exist for
343 young children and Scr is not generally considered to be a good marker to describe renal
344 function maturation in paediatrics, especially in oncology patients due to the lack of muscle
14
345 mass (35). Though, in this study, for patients with a Scr > 30 mol/L, Scr appears to be a
346 marker of renal dysfunction, which is not being accounted for in the GFR renal maturation
347 equation. According to our final model a two-fold increase in Scr in patients with Scr levels >
348 30 mol/L led to a 32% decrease in gentamicin CL. Inclusion of Scr in the model was
350 Scr in this study was that 52% of Scr measurements were < 30 mol/L; thus the model does
351 not account for renal dysfunction in patients with a Scr < 30 mol/L. Inclusion of covariates
352 resulted in a large decrease in BSV in CL and V1 between the base and the final model,
353 namely 28% and 29%, respectively (Table 2). Both BSV and BOV were below 20% for CL,
354 however BOV was slighter higher than BSV which may be due to underlying changes in the
356 Usage of cisplatin and carboplatin, was not found to have a significant effect on
357 gentamicin CL. It should be noted however that only 24% of patients received these drugs in
358 the 6 month period prior to gentamicin administration, the window considered during
359 modelling. This window may be too wide or narrow for examination (36, 37) and cumulative
360 lifetime use was not assessed. Moreover, inclusion of patient Scr values within the model
361 may already account for any potential nephrotoxicity caused by these agents. Further
362 attempts were made to assess the influence of cumulative doses of cisplatin and carboplatin
363 and the timing of their co-administration on gentamicin CL, but no significant relationships
365 One previous study reported that body temperature can influence gentamicin CL in
366 paediatric patients (18). Here, the influence of body temperature on gentamicin CL did not
367 show significance. Body temperature may not have been a suitable marker for critically ill
368 patients here, possible due to the potential use of antipyretics, resulting in a median body
15
369 temperature of 36.6 C during gentamicin therapy. Information on antipyretic usage was not
370 collected in this study, however is part of the standard treatment for febrile neutropenia.
372 suggested gentamicin starting doses between 7.5 and 10.8 mg/kg for paediatric oncology
374 guidelines typically recommend an area under the concentration-time curve (AUC24) to
375 bacteria minimum inhibitory concentration (MIC) ratio (AUC24/MIC) between 80 to 100
376 and/or a maximum concentration (Cmax) to MIC (Cmax//MIC) ratio greater than 10 to be
377 targeted (39, 40). These targets are obtained from clinical studies performed in adults and
378 consider several different microorganisms (41, 42). Given these targets and the estimated
379 population CL in this study an initial dose of 8.0 mg/kg would be required for a typical
380 patient in this study to achieve this target, which is in agreement with previous dose
382 External model evaluation showed that previously published paediatric models over-
383 predicted exposure compared to our model. This is not unexpected, as previous models, even
384 though only models including paediatric patients were considered, were based on different
385 patient cohorts in terms of age, disease state and medication usage. In addition, two out of
386 three previous models evaluated in this study (16, 32) did not scale CL using an allometric
387 weight with an exponent of 0.75. This may partly explain the poor performance of these
388 models in predicting gentamicin concentrations in our paediatric oncology patients (28). This
389 reinforces that a population pharmacokinetic model should not be applied too far beyond the
391 Based on a visual and numerical assessment the final model presented in this study
392 displayed good predictive performance and should be clinically useful in the paediatric
16
393 oncology setting. In addition, all PK parameters were estimated with relative standard errors
395 This study had limitations in terms of the sparse number of samples per patient per
396 dose collected from TDM data available. This is common in special population such as
398 diagnostic plots of individual parameter estimates and covariates (23). Furthermore, this
399 study involved retrospective data collection, which relies on the accuracy of written medical
400 charts.
401 After implementation of the model in a Bayesian forecasting program, future studies
402 should focus on evaluating its clinical applicability to personalise gentamicin dosing
403 regimens, ability to reduce blood sampling during TDM and capacity to improve target
405 CONCLUSIONS
406 To the best of our knowledge this study presents the first population pharmacokinetic
407 model for gentamicin in paediatric oncology patients with febrile neutropenia. This model
409 principles of allometric scaling, maturation of glomerular filtration rate, and standardization
410 for age-expected serum creatinine concentration. Serum creatinine measurement in this
411 model accounted for renal dysfunction not explained by maturation. The model developed
412 displayed superior predictive performance compared to a very limited number of models
413 currently describing gentamicin pharmacokinetics in paediatric patients. This model is now
415 [http://www.tdmx.eu/] (44) for initial dose individualisation and adaptive dose adjustment of
17
418 ACKNOWLEDGEMENTS
419 The authors would like to acknowledge the Australian Centre of Pharmacometrics for
420 the hardware and NONMEM software licences. CCLP was supported by Becas Chile
421 (CONICYT) through a scholarship and the University of Queensland Graduate School
425 CCLP collected and analysed the data and wrote the manuscript. CES, RL and SH applied
426 for the ethical approval, supported data collection and critical review of the manuscript. CES
427 and SH supported and reviewed the data analysis and contributed to the writing of the
428 manuscript.
429
18
430 REFERENCES
431 1. Pizzo PA, Poplack DG, Ovid Technologies I. 2011. Principles and practice of
432 pediatric oncology, vol 6th;. Wolters Kluwer/Lippincott Williams & Wilkins Health,
433 Philadelphia.
437 outcome from febrile neutropenic episodes in children compared with adults: results
438 from four EORTC studies. International Antimicrobial Therapy Cooperative Group
439 (IATCG) of the European Organization for Research and Treatment of Cancer
441 4. Freifeld AG, Bow EJ, Sepkowitz KA, Boeckh MJ, Ito JI, Mullen CA, Raad, II,
442 Rolston KV, Young JA, Wingard JR. 2011. Clinical practice guideline for the use
443 of antimicrobial agents in neutropenic patients with cancer: 2010 update by the
445 5. Dahlgren JG, Anderson ET, Hewitt WL. 1975. Gentamicin blood levels: a guide to
447 6. Paterson DL, Robson JM, Wagener MM. 1998. Risk factors for toxicity in elderly
448 patients given aminoglycosides once daily. J Gen Intern Med 13:735-739.
451 8. Ho KK, Bryson SM, Thiessen JJ, Greenberg ML, Einarson TR, Leson CL. 1995.
452 The effects of age and chemotherapy on gentamicin pharmacokinetics and dosing in
19
454 9. Rosario MC, Thomson AH, Jodrell DI, Sharp CA, Elliott HL. 1998. Population
456 236.
457 10. Bialkowski S, Staatz CE, Clark J, Lawson R, Hennig S. 2016. Gentamicin
460 11. Udy AA, Roberts JA, Lipman J. 2011. Implications of augmented renal clearance in
462 12. Antibiotic Expert Group. 2014. Therapeutic guidelines: antibiotic. Version
464 13. Paviour S, Hennig S, Staatz CE. 2016. Usage and monitoring of intravenous
465 tobramycin in cystic fibrosis in Australia and the UK. J Pharm Pract Res 46:15-21.
466 14. Barras MA, Serisier D, Hennig S, Jess K, Norris RL. 2016. Bayesian estimation of
468 doi:10.1128/aac.01131-16.
469 15. Avent ML, Teoh J, Lees J, Eckert KA, Kirkpatrick CM. 2011. Comparing 3
472 16. Lares-Asseff I, Perz-Guille MG, Camacho Vieyra GA, Perez AG, Peregrina NB,
475 17. Lopez SA, Mulla H, Durward A, Tibby SM. 2010. Extended-interval gentamicin:
476 population pharmacokinetics in pediatric critical illness. Pediatr Crit Care Med
477 11:267-274.
20
478 18. Seaton C, Ignas J, Muchohi S, Kokwaro G, Maitland K, Thomson AH. 2007.
481 19. Fernandez de Gatta MM, Garcia MJ, Lanao JM, Dominguez-Gil A. 1996.
483 20. Beal S, Sheiner LB, Boeckmann A, Bauer RJ. 2009. NONMEM User's Guides
485 21. Lindbom L, Pihlgren P, Jonsson EN. 2005. PsN-Toolkit--a collection of computer
486 intensive statistical methods for non-linear mixed effect modeling using NONMEM.
488 22. Beal SL. 2001. Ways to fit a PK model with some data below the quantification limit.
490 23. Savic RM, Karlsson MO. 2009. Importance of shrinkage in empirical bayes
492 24. Rhodin MM, Anderson BJ, Peters AM, Coulthard MG, Wilkins B, Cole M,
493 Chatelut E, Grubb A, Veal GJ, Keir MJ, Holford NH. 2009. Human renal function
494 maturation: a quantitative description using weight and postmenstrual age. Pediatr
496 25. Anderson BJ, Holford NH. 2008. Mechanism-based concepts of size and maturity in
498 26. Al-Sallami HS, Goulding A, Grant A, Taylor R, Holford N, Duffull SB. 2015.
500 27. Holford NH. 1996. A size standard for pharmacokinetics. Clin Pharmacokinet
501 30:329-332.
21
502 28. Germovsek E, Barker CI, Sharland M, Standing JF. 2017. Scaling clearance in
503 paediatric pharmacokinetics: All models are wrong, which are useful? Br J Clin
505 29. Ceriotti F, Boyd JC, Klein G, Henny J, Queralto J, Kairisto V, Panteghini M.
508 30. Hennig S, Karlsson MO. 2014. Concordance between criteria for covariate model
510 31. Sheiner LB, Beal SL. 1981. Some suggestions for measuring predictive performance.
515 33. Llanos-Paez CC., Hennig S., Staatz CE. 2017. Population pharmacokinetic
518 doi:10.1093/jac/dkw461.
519 34. Johnston C, Hilmer SN, McLachlan AJ, Matthews ST, Carroll PR, Kirkpatrick
520 CM. 2014. The impact of frailty on pharmacokinetics in older people: using
523 35. Skinner R, Cole M, Pearson AD, Keir MJ, Price L, Wyllie RA, Coulthard MG,
524 Craft AW. 1994. Inaccuracy of glomerular filtration rate estimation from
22
526 36. English MW, Skinner R, Pearson AD, Price L, Wyllie R, Craft AW. 1999. Dose-
528 37. Skinner R, Parry A, Price L, Cole M, Craft AW, Pearson AD. 2009. Persistent
531 38. Inparajah M, Wong C, Sibbald C, Boodhan S, Atenafu EG, Naqvi A, Dupuis LL.
532 2010. Once-daily gentamicin dosing in children with febrile neutropenia resulting
537 40. The children's hospital at Westmead. 2015. Aminoglycoside Dosing and
540 41. Moore RD, Lietman PS, Smith CR. 1987. Clinical response to aminoglycoside
543 42. Kashuba AD, Nafziger AN, Drusano GL, Bertino JS, Jr. 1999. Optimizing
546 43. Jones CD, Sun H, Ette EI. 1996. Designing Cross-Sectional Population
547 Pharmacokinetic Studies: Implications for Pediatric and Animal Studies. Clin Res and
23
549 44. Wicha SG, Kees MG, Solms A, Minichmayr IK, Kratzer A, Kloft C. 2015.
550 TDMx: a novel web-based open-access support tool for optimising antimicrobial
552 45. Schwartz GJ, Munoz A, Schneider MF, Mak RH, Kaskel F, Warady BA, Furth
554 20:629-637.
555
556
557
24
558 FIGURE LEGENDS
559
560 FIG 1 (a) Visual predictive check plot based on 500 simulations associated with final
561 population pharmacokinetic model. Observed data (grey dots), median, 95th and 5th
563 (black dashed line) and 95% confidence interval of simulated data (grey area); (b) normalized
564 prediction distribution errors based on 1000 simulations associated with the final population
565 pharmacokinetic model versus time after dose; (c) observed versus population predicted
566 gentamicin serum concentrations and (d) conditional weighted residuals versus time after
568
569 FIG 2 Visual predictive check plot based on 500 simulations associated with this studies final
570 population pharmacokinetic model and another three published population pharmacokinetic
571 models of gentamicin in children generated during external evaluation. Median, 95th and 5th
572 percentiles of observed data (red line), median, 95th and 5th percentiles of simulated data
573 (black dashed line) and 95% confidence interval of simulated data (grey area). The predictive
574 performance for each model is presented numerically with the relative root mean squared
575 error (RMSE %) and the mean relative prediction error (MRPE %).
576
577
25
Table 1. Demographic and clinical information of all patients included in model building and
evaluation analysis
Values
Patient characteristics Model building data Model evaluation
Values are represented as median (range) or number (%); a: serum creatinine concentration
below the lowest reportable limit set to an expected creatinine concentration physiological
mean (28); b: serum creatinine concentration above the lowest reportable limit; c: creatinine
clearance calculated using the Schwartz formula (44) with serum concentration below the
lowest reportable limit set to an expected creatinine concentration physiological mean (28).
Downloaded from http://aac.asm.org/ on May 22, 2017 by CORNELL UNIVERSITY
Table 2. Population pharmacokinetic parameter estimates of the base and the final model and bootstrap results from the final model
Parameter Base model Final model Bootstrap (n=1000) [mean (95 %CI)]
RSE BSV RSE BOV RSE Estimate RSE BSV RSE BOV RSE
Estimate Estimate BSV CV% BOV CV%
% CV% % CV% % % CV% % CV% %
Number of
26 27 27 - -
parameters
CL (L/h) 2.08 3.02 43.8 5.53 24.3 3.93 CL (L/h/70kg) 5.77 1.78 16.0 9.91 20.7 2.15 5.76 (5.59 5.93) 16.0 (12.9 18.5) 20.8 (19.2 22.1)
V1 (L) 5.16 4.26 50.3 6.55 - - V1 (L/70kg) 21.6 2.86 21.5 17.8 - - 21.5 (20.3 22.6) 21.3 (13.7 27.1) -
V2 (L) 3.34 15.7 73.5 15.4 - - V2 (L/70kg) 13.8 10.5 62.4 7.34 - - 13.5 (10.8 16.7) 60.9 (44.1 76.3) -
Covariate
model
SCR - - - - - - 0.55 8.84 - - - - 0.55 (0.47 0.63) - -
Correlations
(%)
CL~V1 93.6 6.55 - - - - 69.2 17.6 - - - - 69.1 (53.9 75.5) - -
Residual error
model (%)
Proportional 26.8 5.86 - - - - 27.5 3.47 - - - - 27.2 (24.8 30.1) - -
Additive
0.05 11.5 - - - - 0.04 15.1 - - - - 0.05 (0.03 0.06) - -
(mg/L)
BOV: between-occasion variability; BSV: between-subject variability; CL: clearance; CV: coefficient of variation; OFV: objective function
value; RSE: relative standard error; Q: intercompartmental clearance; Scr: serum creatinine; V1: apparent volume of distribution of the central
compartment; V2: apparent volume of distribution of the peripheral compartment.
Downloaded from http://aac.asm.org/ on May 22, 2017 by CORNELL UNIVERSITY
Final model:
.
GFR
CL L/h = .
Scr
FFM
V L = .
.
FFM
Q L/h = .
FFM
V L = .
.
FFM PMA .
GFR mL/min = . .
. + PMA
Where,
GFRmat: maturation of glomerular filtration rate; Scrmean: mean serum creatinine concentration calculated using a formula described by Ceriotti et
al. (28); Scri: individual patient serum creatinine concentration; FFM (kg): fat-free mass; PMA (weeks): post-menstrual age considering 40
weeks of gestation.
Downloaded from http://aac.asm.org/ on May 22, 2017 by CORNELL UNIVERSITY
Table 3. Summary of published population pharmacokinetic models of gentamicin identified for external evaluation
Add: additive error; BOV: between-occasion variability; CL: clearance of gentamicin; CLCR: creatinine clearance; CMT: compartmental model;
GFRmat: maturation of glomerular filtration rate using the Rhodin formula (24); FFM (kg): fat-free mass; NR: not reported; PNA (years): post-
Downloaded from http://aac.asm.org/ on May 22, 2017 by CORNELL UNIVERSITY
natal age; Prop: proportional error; Q: intercompartmental clearance; Scr: serum creatinine concentration; Scrmean: Scr standardised to an
expected creatinine concentration physiological mean (28) V1: central volume of distribution; V2: peripheral volume of distribution; WT (kg):
body weight.
Downloaded from http://aac.asm.org/ on May 22, 2017 by CORNELL UNIVERSITY
Downloaded from http://aac.asm.org/ on May 22, 2017 by CORNELL UNIVERSITY