1 INTRODUCTION

Effective, empirical antimicrobial therapy improves therapeutic outcome and reduces infection-related mortality in children with cancer and febrile neutropenia.1 Piperacillin-tazobactam, a ?-lactam/?-lactamase-inhibitor combination with broad-spectrum activity, is frequently prescribed as empirical therapy.1, 2 Like other ?-lactams, piperacillin-tazobactam exerts time-dependent activity, in which bacterial killing is correlated to the percentage of time where the free drug concentration exceeds the pathogen's minimum inhibitory concentration (MIC) (fT > MIC). By maximizing fT > MIC, bactericidal and clinical effect is improved.3 Based on pre-clinical data, maximum effect is achieved when the free drug concentration remains above MIC during 40–70% of the dosing interval.4, 5 Due to the lack of neutrophils in immunocompromised patients, stricter pharmacokinetic/pharmacodynamic (PK/PD) targets may be necessary for optimal antimicrobial efficacy.3, 6, 7

For less susceptible bacteria, conventional intermittent administration of piperacillin-tazobactam is associated with a significant risk of not attaining the desired targets.8-11 Pathophysiological disturbances, commonly observed in children undergoing aggressive chemotherapy, can widely affect the pharmacokinetics (PK) of antimicrobials, and alterations in volume of distribution and clearance reduce the likelihood of adequate pharmacodynamic (PD) coverage.10, 12, 13 Prolongation of the infusion time, through extended infusion or continuous infusion, form an attractive strategy to maximize fT > MIC without increasing the daily dose. Although a large randomized trial14 found no outcome difference between intermittent administration and continuous infusion of β-lactams in adults, prolonged infusions have demonstrated improved probability of target attainment (PTA) and clinical outcome, including mortality.15-17 Published pediatric studies advocate a PTA benefit with extended infusion and, in particular, continuous infusion of piperacillin-tazobactam. However, available data is mainly based on simulations from PK models built on intermittent administration or extended infusion regimens in small, specific populations.8-11, 18-23 Only a single, small study7 has prospectively evaluated and demonstrated increased PTA with continuous infusion in neutropenic children.

The aim of this study was to evaluate piperacillin PK and PTA in children with cancer receiving continuous infusion of piperacillin-tazobactam, in order to define an optimal dosing regimen and substantiate the use of continuous infusion in a clinical setting.

2 METHODS 2.1 Study design and patient population

This prospective, single-center PK study was conducted at the Department of Pediatrics and Adolescent Medicine, Aarhus University Hospital, Denmark between 1 August 2018 and 30 June 2020. The study was approved by the Central Denmark Region Committee on Health Research Ethics (registration no. 1-10-72-427-17), the Danish Medicines Agency (Clinicaltrialsregister.eu registration no. 2017-004281-10) and the Danish Data Protection Agency (registration no. 1-16-02-924-17). Written informed consent was obtained from both parents or legal guardians. In accordance with Danish law, children aged 15–17 years were allowed to give written informed consent for themselves, in close collaboration with their parents or legal guardians.

Children, aged 1–17 years, with cancer who received empirical piperacillin-tazobactam to treat a suspected or documented infection were eligible for inclusion. Exclusion criteria covered; fully breastfeed infants, history of anaphylaxis to ?-lactams, admission to intensive care unit and body weight <8 kg. Eligible children could be enrolled several times in case of recurrent fever episodes. Fever was defined as a single temperature measurement above 38.5°C and a recurrent fever episode as fever after seven fever-free days. Clinical variables and baseline demographics were registered: sex, age, body weight, height, cancer diagnosis, fever duration, piperacillin-tazobactam bolus dosing prior to initiation of continuous infusion, plasma creatinine, C-reactive protein, absolute neutrophil count and microbiological isolates detected in blood cultures.

2.2 Drug dosing and blood sample collection

As standard care, piperacillin-tazobactam was prescribed at 300 mg/kg/day and administered intermittently every 8 h (qh8). After study enrollment, eligible patients received a 2–5 min piperacillin-tazobactam loading dose of 100 mg/kg (not subtracted from the total daily dose) followed by continuous infusion of 300 mg/kg/day (piperacillin component, maximum of 16 000 mg/day) at a fixed rate over 24 h. A supplemental loading dose of 100 mg/kg (maximum four doses per day) was administered in case of a discontinued infusion for longer than 30 min.

The blood sampling schedule was designed, using optimal design theory, in an attempt to accurately inform the piperacillin concentration-time course with continuous infusion, and each patient was scheduled to contribute three blood samples. Sampling times covered peak concentration immediately after loading dose and prior to continuous infusion initiation (2–30 min), distribution phase (0.5–1.5 h), and steady state concentration (12–24 h). Most study participants had indwelling catheters with two separate infusion lines, and piperacillin samples were drawn from the infusion line that was not occupied by the continuous infusion. In case of a single line-catheter, the samples were obtained after thorough saline flushing of the catheter.

2.3 Ultra-high performance liquid chromatography

The unbound piperacillin concentrations in serum were quantified using validated ultra-high performance liquid chromatography (HPLC) following ultra-filtration (Agilent 1290, Agilent Technologies, USA), as formerly described.24, 25 Intra-run (total) imprecisions (coefficients of variation [%]) were 10.2% (15.3%) at 4.5 mg/L and 4.7% (8.2%) at 15.6 mg/L. The limit of quantification for piperacillin was 0.5 mg/L, and was defined as the lowest concentration with a coefficient of variation <20%. The tazobactam component of piperacillin-tazobactam was not quantified.

2.4 PK/PD targets and MIC profile

PTA was assessed for PK/PD targets of 100% fT > MIC (free piperacillin concentration sustained above MIC throughout the dosing interval) and 50% fT > 4xMIC. Due to the steady state concentration-time profile arising from continuous infusion, 50% fT > 4xMIC and 100% fT > 4xMIC result in similar PTA. The PK/PD targets were evaluated in relation to the susceptibility breakpoint MIC for Pseudomonas aeruginosa (16 mg/L), published by the European Committee on Antimicrobial Susceptibility Testing (EUCAST),26 and MIC50 (2 mg/L) and MIC90 (4 mg/L), calculated from an institutional MIC distribution of bacteria in blood cultures from pediatric cancer patients through 10 years.27

2.5 Pharmacokinetic modeling

Piperacillin samples obtained during continuous infusion were merged with samples previously obtained during intermittent administration in the same patient population9, 20 (482 samples from 43 patients across 89 fever episodes). An extended population PK model was build based on the merged data. The two-compartment model,9, 20 with linear elimination and allometric (fixed exponents) scaling of PK parameters to body weight, was used as starting point. The model structure (two or three compartments) and inter-individual as well as inter-occasion (fever episode) variability were re-evaluated. Previously insignificant structural components, namely kidney maturation28 and non-linear clearance, were reassessed and parameter-covariate relations were tested by stepwise covariate modeling: sex, age, underlying malignancy, bacteraemia, neutropenia, glomerular filtration rate (GFR, estimated by Schwartz equation),29 fever duration and peak temperature. A bootstrap analysis with 2000 samples was performed for the final model to obtain 95% confidence intervals for the model parameters as a measure of uncertainty.

PK modeling was performed using NONMEM version 7.4.4 (ICON Development Solutions, Gaithersburg, MD)30 facilitated by Perl-Speaks-NONMEM.31 Statistical selection of a suitable model structure was performed by the likelihood-ratio test of the objective function values (OFV). Model selection and evaluation was guided by residual goodness-of-fit plots and visual predictive checks of simulated concentration-time profiles. Descriptive statistics were performed using STATA 15.1 software (STATA Corp, College Station, TX), and p < .05 were considered statistically significant.

2.6 Monte Carlo simulations

Based on the updated PK model and individual PK parameter estimates, concentration-time profiles and PTA were predicted for each individual fever episode. Monte Carlo simulations were performed to assess the proportion of children attaining the PK/PD targets with alternative dosing regimens. The simulated dosing regimens covered intermittent administration every six (q6h) and eight hours (q8h), extended infusion (infusion for half of the dosing interval, administered q6h and q8h) and continuous infusion. Daily doses of 300 and 400 mg/kg (maximum 16 000 mg daily) were simulated for all dosing regimens. A population of 10 000 children aged 2–18 years with equal sex distribution was constructed based on the NHANES database (1999–2015)32 with median (2.5th–97.5th percentiles) body weight of 42.7 kg (12.6–94.2 kg). PTA was predicted over MICs from 0.125 to 128 mg/L for body weight groups of <25 kg, ≥25 to <50 kg, ≥50 to <75 kg, and ≥75 kg. To assess the likelihood of treatment success in this population, the cumulative fraction of response (CFR)21, 33 was estimated for each regimen; PTA and the proportion of isolates at each MIC within the institutional MIC distribution27 were multiplied and summed. A dosing regimen was considered successful if PTA or CFR ≥95%.

3 RESULTS 3.1 Study population

A total of 38 children with cancer received continuous infusion of piperacillin-tazobactam across 68 fever episodes (1–4 per child). Median (IQR) for age, body weight and GFR were 6.5 years (4; 15), 27.4 kg (15.1; 54.0) and 175.5 ml/min/1.73 m2 (133.3; 209.3), respectively, and a bacteraemia was detected in six of 68 (9%) fever episodes. This cohort was merged with the intermittent administration cohort that comprised 43 children and 89 fever episodes,9, 20 resulting in PK data from 78 children (three children were enrolled in both cohorts) across 157 fever episodes. Clinical and demographic characteristics of the continuous infusion and intermittent administration cohorts are summarized in Table 1. From a clinical point of view, median age (IQR) differed between the two cohorts (6.5 [4; 15] vs. 12 [7; 14] years), however, the difference was statistically insignificant. The only characteristic that differed significantly was the proportion of fever episodes caused by “local infections” (p = .033).

TABLE 1. Clinical characteristics of children and fever episodes in CI cohort, IA cohort and total cohort (CI + IA) Characteristic CI cohort IA cohort Total population (CI + IA)a p Included patients n = 38 n = 43 n = 78 Gender, n (%) Male 24 (63%) 27 (63%) 49 (63%) 0.995 Female 14 (37%) 16 (37%) 29 (37%) Age (years) 6.5 (4–15)b 12 (7–14) 10.5 (5–14) 0.174 Body weight (kg) 27.4 (15.1–54.0) 39.4 (22.5–50.4) 31.6 (20–51) 0.141 Body surface area (m2) 0.97 (0.64–1.60) 1.31 (0.90–1.58) 1.12 (0.77–1.58) 0.109 Underlying malignancy, n (%) Hematological malignancies 17 (45%)c 18 (42%) 32 (42%) 0.868 Solid tumors 21 (55%)d 25 (58%) 46 (59%) GFR (ml/min/1.73 m2)5 175.5 (133.3–209.3) 171.5 (147.8–208.4) 176.1 (136.6–209.9) 0.923 Fever episodes n = 68 n = 89 n = 157 Positive blood culture, n (%) 6 (9%) 10 (11%) 16 (10%) 1.000 Viridans streptococci 3 (30%) 3 (19%) Gram-positive cocci, unspecified 1 (17%) 1 (6%) Pseudomonas aeruginosa 1 (17%) 1 (6%) Micrococcus 2 (33%) 3 (30%) 5 (31%) CoNS 2 (33%) 2 (20%) 4 (25%) Staphylococcus aureus 1 (10%) 1 (6%) Klebsiella pneumoniae 1 (10%) 1 (6%) Fever of unknown origin 44 (65%) 69 (78%) 113 (72%) 0.100 Local infection 18 (26%) 10 (11%) 28 (18%) 0.033 GFR (ml/min/1.73 m2)e 174.9 (150.4;225.7) 172.4 (139.8–210.8) 174.4 (144.1–220.8) 0.916 Neutropenia (x109/L)f 49 (72%) 77 (87%) 126 (80%) 0.126 Duration of fever (days) 3.5 (3–4) 1 (1–3) 3 (3–4) 0.245 CRP (mg/L) 34.5 (19.7–54.9) - - Maximum CRP (mg/L) 80.9 (44.5–135.9) - - No. of piperacillin-tazobactam bolus doses before start of CIg 5.5 (3–9) - - CI disruption >30 min 9/68 (13%) Note: Continuous variables are presented as medians (IQR) and dichotomous data are presented as a number (%). Abbreviations: ANC, Absolute Neutrophil Count; CI, Continuous infusion; CoNS; Coagulase negative staphylococci; CRP, C-reactive protein; IA, Intermittent administration; IQR, Interquartile range. a Three children were enrolled in both IA and CI cohort, thus n = 78 (total population). b Age distribution: 0–5 years: n = 16; >5–10 years: n = 5; >10–15 years; n = 8; >15 years: n = 9. c Hematological malignancies; acute lymphoblastic leukemia (n = 9), acute myeloid leukemia (n = 3), unspecified leukemia (n = 2), B-cell Non-Hodgkin lymphoma (n = 1), Hodgkins lymphoma (n = 1), myelomatosis (n = 1). d Solid tumors; medulloblastoma (n = 4), neuroblastoma (n = 3), osteosarcoma (n = 2), langerhans histiocytosis (n = 1), other sarcomas (n = 1), Ewing sarcoma (n = 9), Wilms tumor (n = 1). e GFR: glomerular filtration rate calculated from the Schwartz equation28; GFR = kL/Pcr (L: body length in cm, Pcr: plasma creatinine concentration in mg/dl and k: constant of proportionality). f Neutropenia: ANC < 0.5 x 109/L on the day of admission. g Range (1–39). 3.2 Pharmacokinetic modeling and covariates

A total of 192 plasma piperacillin samples were collected during continuous infusion, with a median number (range) of samples per study subject of two (two-three), and the extended PK model was built on 674 piperacillin samples in total. Data were best described by a three-compartment model with a fast distribution phase, which improved the model fit significantly (ΔOFV = 118; Figure S1), and no other structural model improvements were identified. As continuous infusion samples, particularly those at steady state, showed high variability, model fit was significantly improved by adding a separate residual error (residual unexplained variability) for the continuous infusion data (ΔOFV = 141; 65 vs. 27.5% for intermittent administration data). However, residual unexplained variability was reduced to 54% by allowing a separate inter-occasion variability in clearance in the continuous infusion cohort (ΔOFV = 17.4) and inter-individual variability in the residual error (ΔOFV = 16.0), thereby allowing higher residual error in samples from some individuals. For covariates, body weight was significantly associated with the PK parameters, that is, clearance, inter-compartmental clearances, and volumes of distribution. Incorporation of lean body weight rather than total body weight further improved model fit (ΔOFV = 20.8). Still, total body weight was retained as a covariate in the final model, since parameter and variability estimates remained unchanged, and as pediatric dosing typically rely on total body weight. No other statistically significant covariate-parameter associations were identified. Simulation-based, visual predictive checks and residual diagnostics showed adequate description of the concentration-time course during continuous infusion without systematic bias (Figure S1). Compared with the previous model, only minor changes in the PK parameter estimates were identified, and the final PK parameters are summarized in Table 2.

TABLE 2. Pharmacokinetic parameters Parameter Parameter description Estimate 95%CIa CL (L/h) Elimination clearance 14.24b (12.98, 15.27) Vc (L) Central volume of distribution 5.953b (3.468, 7.467) Q1 (L/h) Inter-compartmental clearance (slow) 0.1943b (0.1548, 0.2322) Vp1 (L) Peripheral volume of distribution (slow) 3.537b (1.968, 5.277) Q2 (L/h) Inter-compartmental clearance (rapid) 27.45b (19.58, 37.71) Vp2 (L) Peripheral volume of distribution (rapid) 7.329b (6.140, 8.627) CL IOVIA (%) Inter-individual variability in CL (IA cohort) 18.0 (14.0, 21.8) CL IOVCI (%) Inter-individual variability in CL (CI cohort) 48.1 (30.0, 67.9) ERR IIV (%) Inter-individual variability in residual error 30.7 (19.0, 43.3) ERRIA (%) Residual unexplained variability (IA cohort) 27.5 (23.9, 30.8) ERRCI (%) Residual unexplained variability (CI cohort) 50.0 (38.4, 61.1) a Based on a non-parametric bootstrap of the data set (with 1837/2000 successful samples). b Estimates for 70 kg, scaled to individual body weight according to: CLtypical,individual = CLtypical,70kg·(WTindividual,kg/70kg)0.75 and Vtypical,individual = Vtypical,70kg·(WTindividual,kg/70kg)1.0.

Figure 1 illustrates the different piperacillin concentration-time course with intermittent administration and continuous infusion with 95% prediction interval of individual predictions. A sample was available at steady state in 49 of 68 (72%) fever episodes, and observed median piperacillin concentration (95% confidence interval) was of 47.6 mg/L (17.2; 129.5). The observed steady state concentrations showed high variability, and were spread above and below the model-predicted concentrations, with a few observations placed outside the 95%-prediction interval. Based on observed piperacillin exposure, 89.9 and 22.4% of the children achieved the targets of 100% fT > MIC and 50% (100% for continuous infusion) fT > 4xMIC for the P. aeruginosa breakpoint, respectively.

Piperacillin concentration-time course for intermittent administration every 8 h (circles) and continuous infusion (triangles) of piperacillin-tazobactam (both 300 mg/kg/day), with the body weights indicated (point color). The gray shaded areas represent the 95% prediction interval of individual predictions based on the final model. CI, continuous infusion; IA, intermittent administration; MIC, minimum inhibitory concentration

3.3 Probability of target attainment

PTA for the two PK/PD targets is illustrated in Figure 2 and Table 3. Predictions of median steady state concentrations (95% percentiles) for continuous infusion of 300 mg/kg/day was 47.9 mg/L (18.0; 126) and 56.2 mg/L (20.2; 150) for 400 mg/kg/day (Figure 3A). For the P. aeruginosa breakpoint (16 mg/L), simulations based on the updated PK model confirmed the findings of the previous model-predictions showing that continuous infusion was required to reach the target of 100% fT > MIC (Figure 2A). With continuous infusion at doses of 300 mg/kg/day and 400 mg/kg/day, 98.7 and 99.4% of the children achieved this target, respectively. Intermittent administration and extended infusion regimens achieved MICs of 0.25 and 2 mg/L (MIC50) with the target of 100% fT > MIC, respectively.

Probability of target attainment for the two targets (A) 100% fT > MIC and (B) 50% fT > 4xMIC (effectively 100% fT > 4xMIC for continuous infusion). PTA was simulated for intermittent administration every eight (q8h) and 6 h (q6h), extended infusion q8h and q6h and lasting half of a dosing interval, and continuous infusion. The colors represent different weight groups and MIC50 (2 mg/L), MIC90 (4 mg/L), and the EUCAST MIC breakpoint for P. aeruginosa (16 mg/L) are represented by the dashed vertical lines. The dashed horizontal lines illustrate that 95% of the simulated population have reached the specified fT > MIC target. CI, continuous infusion; EI, extended infusion; IA, intermittent administration; MIC, minimum inhibitory concentration; PTA, probability of target attainment

TABLE 3. Probability of target attainment and cumulative fraction of response for various dosing regimens for PK/PD targets of 100% fT > MIC and. 50% fT > 4xMIC (100% fT > 4xMIC for continuous infusion) Dosing regimen PTA OF MIC50 (2.0 mg/L) PTA OF MIC90 (4.0 mg/L) PTA OF P.A. breakpoint (16.0 mg/L) CFR (%) 100% fT > MIC 50% fT > 4xMIC 100% fT > MIC 50% fT > 4xMIC 100% fT > MIC 50% fT > 4xMIC 100% fT > MIC 50% fT > 4xMIC IA 300 mg/kg/day q8h 2.30% 30.90% 0.10% 5.50% 0% 0% 2.00% 32% q6h 17.90% 67.90% 3.10% 27.10% 0% 0% 15.70% 60.90% IA 400 mg/kg/day q8h 3.10% 37.00% 0.20% 7.40% 0% 0% 2.70% 31.80% q6h 22.00% 76.60% 3.80% 35.00% 0% 0% 19.30% 69.30% EI 300 mg/kg/day q8h 40.60% 100% 12.50% 100% 0% 8.90% 40.10% 96.00% q6h 76.70% 100% 45.60% 100% 0.90% 8.90% 70.30% 96.00% EI 400 mg/kg/day q8h 47.90% 100%