Relationship between piperacillin concentrations, clinical factors and piperacillin/tazobactam-associated acute kidney injury

Abstract

Background

Piperacillin/tazobactam, a commonly used antibiotic, is associated with acute kidney injury (AKI). The relationship between piperacillin concentrations and AKI remains unknown.

Objective

Estimate piperacillin exposures in critically ill children and young adults administered piperacillin/tazobactam to identify concentrations and clinical factors associated with piperacillin-associated AKI.

Patients and methods

We assessed piperacillin pharmacokinetics in 107 patients admitted to the paediatric ICU who received at least one dose of piperacillin/tazobactam. Piperacillin AUC, highest peak (C_max) and highest trough (C_min) in the first 24 hours of therapy were estimated. Piperacillin-associated AKI was defined as Kidney Disease: Improving Global Outcomes (KDIGO) Stage 2/3 AKI present >24 hours after initial piperacillin/tazobactam dose. Likelihood of piperacillin-associated AKI was rated using the Naranjo Adverse Drug Reaction Probability Scale. Multivariable logistic regression was performed to identify patient and clinical predictors of piperacillin-associated AKI.

Results

Out of 107 patients, 16 (15%) were rated as possibly or probably having piperacillin-associated AKI. Estimated AUC and highest C_min in the first 24 hours were higher in patients with piperacillin-associated AKI (2042 versus 1445 mg*h/L, P = 0.03; 50.1 versus 10.7 mg/L, P < 0.001). Logistic regression showed predictors of piperacillin-associated AKI included higher C_min (OR: 5.4, 95% CI: 1.7–23) and age (OR: 1.13, 95% CI: 1.05–1.25).

Conclusions

We show a relationship between estimated piperacillin AUC and highest C_min in the first 24 hours of piperacillin/tazobactam therapy and piperacillin-associated AKI, suggesting total piperacillin exposure early in the course is associated with AKI development. These data could serve as the foundation for implementation of model-informed precision dosing to reduce AKI incidence in patients given piperacillin/tazobactam.

Introduction

Critical illness leads to variability in drug exposure due to pathophysiologic alterations in drug pharmacokinetics (PK), resulting in drug inefficacy or toxicity. Model-informed precision dosing can minimize toxicity and maximize efficacy by incorporating data from population-level PK models, individual drug concentrations and patient/clinical factors to generate individual concentration versus time profiles.^1–3 While dosing adjustments may be made on the basis of generated profiles, precision dosing requires knowledge of drug exposure targets to ensure minimum exposures are met to achieve efficacy but toxic exposures are avoided.

Piperacillin/tazobactam, a β-lactam/β-lactamase inhibitor combination broad spectrum antibiotic used commonly in critically ill children with systemic infections,⁴ has efficacy against Gram-positive, Gram-negative and anaerobic bacteria. It is an ideal candidate for precision dosing since high variability in piperacillin concentrations in critically ill children and young adults has been demonstrated.⁵ Low concentrations may increase risk of suboptimal antibacterial efficacy, while high concentrations may place patients at risk of toxicity, specifically acute kidney injury (AKI). Recent meta-analyses from adult studies show increased risk of developing AKI when piperacillin/tazobactam is given combination with other nephrotoxic medications.^6–10 The data in children are less clear.¹¹ Retrospective cohort studies in children show increased risk of AKI development when vancomycin is given with piperacillin/tazobactam compared to vancomycin with another antipseudomonal β-lactam antibiotic.^12–14 However, one study reported that while piperacillin/tazobactam alone increased risk of developing Kidney Disease: Improving Global Outcomes (KDIGO) stage 2 or 3 AKI in critically ill children, there was no association between concomitant use of piperacillin/tazobactam and vancomycin and AKI development.¹⁵ Furthermore, in a rat model where AKI was assessed using urinary biomarkers, histopathology and cellular injury, piperacillin/tazobactam appeared to protect against vancomycin-induced AKI.¹⁶

It is critical to explain the relationship between piperacillin/tazobactam and AKI, since AKI is associated with increased hospital length of stay, cost, morbidity and mortality in children.^17–22 Nephrotoxic medication exposure comprises a common cause of AKI in hospitalized children^23,24 and a modifiable factor to reduce AKI rates. AKI from piperacillin/tazobactam is characterized as a type B adverse drug reaction (i.e. idiosyncratic and cannot be predicted),²⁵ as limited data exist regarding the dose–exposure–response relationship between piperacillin (the active β-lactam component of piperacillin/tazobactam) concentrations and piperacillin-associated AKI. In a single-centre retrospective study in adults, a trough of >452 mg/L was associated with a 50% risk of developing nephrotoxicity.²⁶ We sought to describe the relationship between piperacillin concentration exposures and piperacillin-associated AKI and identify patient and clinical factors that increase the risk of piperacillin-associated AKI in critically ill children and young adults. We hypothesize that patients who develop piperacillin-associated AKI have higher exposures to piperacillin in the first 24 hours of piperacillin/tazobactam therapy, allowing identification of piperacillin exposure thresholds that increase piperacillin-associated AKI risk, which may serve as the foundation for piperacillin/tazobactam precision dosing in the future.

Patients and methods

We conducted a retrospective cohort study utilizing a dataset generated from an ongoing Institutional Review Board (IRB)-approved β-lactam study, for which a waiver of consent was granted (IRB no. 2018-3245). In the parent study, patients admitted to the Cincinnati Children’s Hospital Medical Center (CCHMC) paediatric ICU (PICU) and administered at least one piperacillin/tazobactam dose were eligible. The most common PICU piperacillin/tazobactam dosing regimen is 100 mg/kg every 6 hours. Patients who undergo complex abdominal surgery, including liver transplant, may receive 100 mg/kg piperacillin/tazobactam every 2 hours intra-operatively before transitioning to every 6 hours dosing post-procedure. Piperacillin concentrations were measured from scavenged opportunistic samples, a method we have described previously.^5,27 In this approach, residual blood from clinical samples were requested from the clinical laboratory for measurement of total and free piperacillin concentrations.^5,27 Eligible clinical samples included blood obtained for complete blood counts or metabolic panels drawn within 24 hours after any piperacillin/tazobactam dose in the first 7 days of therapy. Samples drawn during the piperacillin/tazobactam infusion (30 minutes) were not requested. Residual blood was processed as previously described,²⁷ and plasma was stored at −80°C until concentration measurement. Samples obtained through scavenged opportunistic sampling were evenly distributed throughout the dosing interval (data not shown). The distribution confirmed there was no bias in sample collection (i.e. samples were not primarily clustered around the trough or peak). This allowed for robust estimation of the individual-level piperacillin exposures.

One hundred and forty-nine patients were enrolled between October 2018 and June 2021 and had at least one piperacillin concentration measured. Piperacillin concentrations were measured using a validated HPLC assay.⁵

For the current retrospective analysis, we applied the following exclusion criteria: patients who received piperacillin/tazobactam for less than 24 hours or initial piperacillin/tazobactam doses at another institution, patients receiving any modality of kidney replacement therapy or extracorporeal membrane oxygenation, or patients with only total piperacillin concentrations available.

Piperacillin exposure estimates

For each patient, free piperacillin AUC in the first 24 hours (AUC₂₄), the highest maximum concentration (peak) in the first 24 hours (C_max24) and the highest trough in the first 24 hours (C_min24) were estimated. Estimations were based on the free piperacillin concentrations and Bayesian estimation with a piperacillin population PK model in critically ill children²⁸ in MWPharm++ (Mediware, Czech Republic).²⁹ In this method, the selected population PK model represented the patient population of the study well and provided a priori information about the likely distribution of PK parameters. By combining the population PK model with actual measured concentrations from an individual patient, an estimated PK profile with individual PK parameters was generated using Bayesian estimation for each patient in the study. A visual predictive check was performed to ensure that the fit of the concentration versus time profile was appropriate with respect to the concentrations measured. The PK exposure estimates (peak at the end of the infusion, pre-dose trough and AUC) were then determined from the generated PK profile. For the 107 patients included in the study, there were 570 free piperacillin concentrations (average 5.3/patient, range 1–19/patient). These sampling numbers allowed for robust estimation of individual concentration versus time profiles and PK parameters³⁰ (for a previously published example of modelling and simulation, see Tang Girdwood³¹).

Adjudication of piperacillin-associated AKI

We defined piperacillin-associated AKI based on three established criteria for identifying drug-associated AKI²⁵: (i) AKI present for at least 1 day between 24 hours and 7 days after the first piperacillin/tazobactam dose; (ii) AKI that is at least KDIGO³² stage 2 (severe AKI) by creatinine or urine output criteria, with the higher of the criteria used for overall staging and (iii) increased creatinine reaches a concentration above 0.5 mg/dL for patients whose baseline creatinine is less than 0.5 mg/dL. The latter criterion has been used in previous paediatric nephrotoxicity studies to avoid small increases in serum creatinine (e.g. from a baseline of 0.2 to a peak of 0.3 mg/dL leading to an overestimation of AKI incidence).³³ Urine output was analysed in 2-hour increments to determine lowest rate of output in 6-, 12- and 24-hour periods. If a 2-hour period contained urine-stool mix or an unmeasured urine output, it was excluded from urine output calculations.

Four trained physician adjudicators with expertise in AKI first determined the patients’ baseline creatinine values from one of the following: (i) lowest creatinine in the 3 months prior to PICU admission for patients without chronic kidney disease or lowest creatinine in the 3–12 months prior to PICU admission for patients with chronic kidney disease (if available); (ii) creatinine imputed from a presumed baseline creatinine clearance of 120 mL/min/1.73 m² (based on the bedside Schwartz equation) as previously reported in the paediatric literature^34,35 or (iii) lowest creatinine during hospitalization. After setting the baseline creatinine, they determined which piperacillin-associated AKI criteria were met. Since the third criterion would exclude many young children with low baseline criteria or chronically ill patients with low muscle mass, the adjudicators determined whether a patient met all three criteria versus just the first two, for additional analysis. If a patient met the first two or all three criteria, the probability of AKI being caused by piperacillin/tazobactam was evaluated using the Naranjo Adverse Drug Reaction Probability Scale (Table S1, available as Supplementary data atJACOnline).³⁶ Each patient was adjudicated by at least two people; if final decisions on presence or probability of piperacillin-associated AKI differed, a third or fourth adjudicator reviewed the case until consensus was reached. All adjudicators were blinded to the piperacillin concentration results.

Clinical data collection and predictors

Chart review was conducted to extract data regarding demographics, hospitalization characteristics and outcomes, including lengths of stay and mortality. Comorbid condition was defined as receiving medication or requiring subspecialty care for at least one condition. Potential predictors of piperacillin-associated AKI were extracted including age, weight category (underweight, normal/healthy weight, overweight, obese), AKI presence and associated KDIGO stage before first piperacillin/tazobactam dose, baseline creatinine clearance (estimated with bedside Schwartz equation³⁷), cumulative piperacillin dose adjusted for weight in the first 24 h, number of concurrent nephrotoxic medications (vancomycin, systemic aminoglycosides, non-steroidal anti-inflammatory drugs, radiocontrast, tacrolimus and other calcineurin inhibitors, diuretics) in the first 24 h, percentage fluid balance in the first 24 h, lowest albumin within 24 h of the first piperacillin/tazobactam dose and surgical procedure in the 72 h before or within the first 24 h of piperacillin/tazobactam therapy. Clinical data were recorded and stored in REDCap.³⁸

Statistical analysis

The primary outcome was piperacillin-associated AKI. Differences in demographics, hospitalization characteristics and outcomes between patients with piperacillin-associated AKI based on two or three criteria versus those without piperacillin-associated AKI were compared with Wilcoxon rank sum test for continuous variables and Fisher or Chi-squared tests for categorical variables. Comparison of log-transformed AUC₂₄, C_max24 and C_min24 was done with Student’s t-test. Statistical significance was considered as P < 0.05. Multivariable logistic regression was performed to identify clinical predictors of piperacillin-associated AKI. The full logistic regression model initially included all the predictors. Selection of final predictors was done stepwise with backwards direction; the alpha was set at 0.05 for the final model. Statistics were performed in R v.3.6.1 and RStudio v.1.2.1335 (Vienna, Austria).

We also evaluated the sensitivity and specificity of identifying patients with piperacillin-associated AKI using trough thresholds based on the CLSI breakpoints for Pseudomonas aeruginosa for piperacillin/tazobactam, 16 mg/L.³⁹ While it remains under debate what piperacillin targets should be achieved for efficacy in critically ill patients, conservative targets include concentrations remaining above the MIC or four times (4×) the MIC for 100% of the dosing interval.^40–43 Therefore, we evaluated the sensitivity and specificity of trough concentrations at the CLSI breakpoint (16 mg/L) and 4× the breakpoint (64 mg/L).

Results

Demographics and hospitalization characteristics of patients who met piperacillin-associated AKI criteria

The final cohort included 107 of 149 patients in the initial piperacillin concentration cohort (Figure 1, Table 1). The present cohort ranged in age from 1 month to 33 years (median 6.3 years). Most patients (86%) had a comorbid condition.

Figure 1.

CONSORT diagram of reasons for exclusions. CCHMC = Cincinnati Children’s Hospital Medical Center; PICU = Paediatric Intensive Care Unit; PTZ = piperacillin/tazobactam.

Table 1.

Demographics and hospitalization characteristics of overall cohort, patients who met all three piperacillin-associated AKI (PTZ-AKI) criteria and received a rating of possible or probable likelihood of having PTZ-AKI, and patients who did not meet all three criteria or who were rated as unlikely to have PTZ-AKI

	Overall (N = 107)	PTZ-AKI (three criteria) (n = 16)	No PTZ-AKI (three criteria) (n = 91)	P value (PTZ-AKI versus no PTZ-AKI)
Age (years)				<0.001
Median [IQR]	6.3 [2.2, 12.8]	14.8 [11.8, 17.9]	5.4 [1.6, 11.0]
Sex				0.63
Female	51 (47.7%)	9 (56.2%)	42 (46.2%)
Male	56 (52.3%)	7 (43.8%)	49 (53.8%)
Weight (kg)				<0.001
Median [IQR]	21.7 [11.5, 12.8]	44.7 [28.4, 58.3]	17.7 [10.9, 27.3]
BMI (kg/m2)				0.011
Median [IQR]	17.1 [15.3, 19.8]	20.5 [27.3, 21.7]	16.7 [15.1, 19.2]
Self-identified race				0.39
White	68 (63.6%)	12 (75.0%)	56 (61.5%)
Black	25 (23.4%)	3 (18.8%)	22 (24.2%)
Hispanic	7 (6.5%)	0 (0%)	7 (7.7%)
Asian	2 (1.9%)	1 (6.2%)	1 (1.1%)
Native American or American Indian	0 (0%)	0 (0%)	0 (0%)
Hawaiian or Pacific Islander	0 (0%)	0 (0%)	0 (0%)
Other	0 (0%)	0 (0%)	0 (0%)
Unknown	5 (4.7%)	0 (0%)	5 (5.5%)
Presence of comorbid conditions				0.46
No	15 (14.0%)	1 (6.2%)	14 (15.4%)
Yes	92 (86.0%)	15 (93.8%)	77 (84.6%)
PICU length of stay (days)				0.89
Median [IQR]	5.0 [2.0, 11.0]	6.0 [2.8, 9.0]	5.0 [2.0, 11.0]
Hospital length of stay (days)				0.30
Median [IQR]	17.0 [9.3, 40.8]	26.0 [12.0, 49.8]	16.0 [9.0, 39.3]
Missing	1 (0.9%)	0 (0%)	1 (1.1%)
Vasopressor-free days in PICU				0.937
Median [IQR]	5.00 [2.0, 10.0]	6.00 [1.8, 7.5]	4.00 [2.0, 10.0]
Ventilator-free days in PICU				0.29
Median [IQR]	2.00 [1.0, 4.0]	3.00 [1.0, 6.0]	2.00 [1.0, 3.0]
28-day outcome				0.39
Alive	104 (97.2%)	15 (93.8%)	89 (97.8%)
Deceased	3 (2.8%)	1 (6.2%)	2 (2.2%)

Comorbid condition is defined as a condition requiring subspecialty care or medication for treatment. PTZ-AKI = piperacillin-associated AKI. IQR = interquartile range.

In the initial round of adjudications, there was 83% agreement between two adjudicators on the presence of piperacillin-associated AKI. Twelve patients required a third adjudication and six patients required review by a fourth adjudicator. Sixteen (15%) patients met all three criteria for piperacillin-associated AKI and scored as possible or probable on the Naranjo Adverse Drug Reaction Probability Scale (Table 2). Piperacillin-associated AKI patients were older and weighed more than those without piperacillin-associated AKI [14.8 (11.8, 17.9) versus 5.4 (1.6, 11.0) years, P < 0.001; 44.7 (28.4, 58.3) versus 17.7 (10.9, 27.3) kg, P < 0.001, respectively] (Table 1). There were no differences in self-identified race, presence of comorbidities, PICU and hospital lengths of stay, vasopressor- and ventilator-free days or mortality rates between patients adjudicated as having piperacillin-associated AKI versus not. For most patients, the maximum KDIGO stage was first achieved in the second or third 24-hour period after the first dose of piperacillin/tazobactam, regardless of whether they met piperacillin-associated AKI criteria or not (Figure S1).

Table 2.

Piperacillin/tazobactam-associated AKI (PTZ-AKI) adjudication decisions

PTZ-AKI adjudication decision	Number of patients (%) total = 107 patients
No	67 (62.6)
Unlikely	3 (2.8)
Possible	35 (32.7)
ȃMeets first 2 criteria only	20 (18.7)
ȃMeets all 3 criteria	15 (14.0)
Probable	2 (1.8)
ȃMeets first 2 criteria only	1 (0.9)
ȃMeets all 3 criteria	1 (0.9)

Adjudicators also assessed patients who met only the first two criteria for piperacillin-associated AKI, as the rise in creatinine to above 0.5 mg/dL may exclude young patients who have true AKI. Thirty-seven (35%) patients met at least the first two criteria and were rated as possible or probable (Table 2). The piperacillin-associated AKI group meeting the first two criteria were also older and weighed more than the group not meeting two criteria [10.4 (3.0, 15.2) versus 5.6 (2.0, 10.5) years, P = 0.036; 27.4 (15, 46.2) versus 18.2 (11.1, 27.1) kg, P = 0.046, respectively] (Table S2). Demographics, hospitalization characteristics or mortality rates did not differ between those with or without piperacillin-associated AKI.

Differences in estimated piperacillin exposures between patients with and without piperacillin-associated AKI

Patients meeting all three piperacillin-associated AKI criteria had higher estimated free piperacillin AUCs and C_min values in the first 24 h of piperacillin/tazobactam therapy (geometric means ± standard errors of 2042 ± 283 versus 1445 ± 67 mg*h/L, P = 0.032; 50.1 ± 14 versus 10.7 ± 1.8 mg/L., P < 0.001, respectively) (Table 3). C_max did not differ between the group. We observed higher estimated AUCs and C_min but similar estimated C_max when comparing patients who met the first two piperacillin-associated AKI criteria versus those who did not (geometric means ± standard errors of AUC: 1738 ± 131 versus 1445 ± 81 mg*h/L, P = 0.032; C_min: 24.5 ± 5.6 versus 9.91 ± 1.9 mg/L, P = 0.004; C_max: 275 ± 13.3 versus 275 ± 7.0 mg/L, P = 0.77) (Table 3).

Table 3.

Comparison of piperacillin AUC, highest peak (C_max) and highest trough (C_min) in the first 24 hours between those with and without piperacillin-associated AKI (PTZ-AKI). Values are geometric means ± standard error

	Based on meeting all three criteria			Based on meeting at least the first two criteria
	PTZ-AKI	No PTZ-AKI	P value	PTZ-AKI	No PTZ-AKI	P value
Area under the curve (mg*h/L)	2042 ± 283	1445 ± 67	0.032	1738 ± 131	1445 ± 81	0.032
Highest peak (mg/L)	282 ± 30	275 ± 5.8	0.80	275 ± 7.0	275 ± 13.3	0.77
Highest trough (mg/L)	50.1 ± 14	10.7 ± 1.8	<0.001	24.5 ± 5.6	9.91 ± 1.9	0.004

Predictors of piperacillin-associated AKI

Table 4 and Table S3 list all the piperacillin-associated AKI predictors, in addition to estimated piperacillin exposures (AUC, C_max, C_min) and age, that we included in the initial multivariable logistic regression model. Baseline creatinine clearance did not differ significantly between the groups based on three piperacillin-associated AKI criteria (166 versus 199 mL/min/1.73 m², P = 0.091).

Table 4.

Predictors included in the multivariable logistic regression for the overall cohort, patients who met all three piperacillin-associated AKI (PTZ-AKI) criteria and received a rating of possible or probable likelihood of having PTZ-AKI, and patients who did not meet all three criteria or rated as unlikely to have PTZ-AKI. In addition to these predictors, piperacillin exposures and age were tested

	Overall (N = 107)	PTZ-AKI (three criteria) (n = 16)	No PTZ-AKI (three criteria) (n = 91)	P value (PTZ-AKI versus no PTZ-AKI)
Weight category				0.87
Underweight	17 (15.9%)	3 (18.8%)	14 (15.4%)
Normal/healthy weight	59 (55.1%)	10 (62.5%)	49 (53.8%)
Overweight	15 (14.0%)	1 (6.2%)	14 (15.4%)
Obese	16 (15.0%)	2 (12.5%)	14 (15.4%)
Baseline creatinine clearance (calculated by Bedside Schwartz equation, mL/min/1.73 m2)				0.091
Median [IQR]	191 [146, 262]	166 [138, 202]	199 [148, 268]
AKI prior to PTZ dose				1.0
No AKI prior to PTZ	93 (86.9%)	14 (87.5%)	79 (86.8%)
AKI Present prior to PTZ	14 (13.1%)	2 (12.5%)	12 (13.2%)
KDIGO stage prior to PTZ dose				0.48
Cannot be determined	16 (15.0%)	1 (6.2%)	15 (16.5%)
No AKI	77 (72.0%)	13 (81.2%)	64 (70.3%)
KDIGO Stage 1	10 (9.3%)	1 (6.2%)	9 (9.9%)
KDIGO Stage 2	2 (1.9%)	0 (0%)	2 (2.2%)
KDIGO Stage 3	2 (1.9%)	1 (6.2%)	1 (1.1%)
Cumulative piperacillin dose per kg of weight				0.12
Median [IQR]	355 [315, 441]	317 [278, 383]	356 [331, 441]
Number of additional nephrotoxic or nephro-modifying drugs				0.46
0 additional meds	37 (34.6%)	6 (37.5%)	31 (34.1%)
1 additional med	40 (37.4%)	4 (25.0%)	36 (39.6%)
2 + additional meds	30 (28.0%)	6 (37.5%)	24 (26.4%)
Percentage fluid balance				0.72
Median [IQR]	4.7 [1.7, 8.9]	4.6 [2.5, 8.4]	4.7 [1.7, 9.1]
Lowest albumin within 24 h of first PTZ dose				0.82
Median [IQR]	2.6 [2.3, 3.1]	2.6 [2.3, 3.1]	2.6 [2.3, 3.1]
Missing (%)	3 (2.8)	1 (6.2)	2 (2.2)
Surgery in the prior 72 h or following 24 h				0.18
No	51 (47.7%)	5 (31.2%)	46 (50.5%)
Yes	56 (52.3%)	11 (68.8%)	45 (49.5%)

Bivariate logistic regression revealed that the highest estimated C_min and AUC in the first 24 hours to be predictors of piperacillin-associated AKI when using all three criteria (Figure 2). Since pair-wise correlation between estimated AUC, C_max and C_min demonstrated all three exposures were highly correlated with one another, we only included estimated C_min in the final multivariable regression model. Our final model showed that for every 10-fold increase in the highest estimated C_min in the first 24 hours, the odds ratio of developing piperacillin-associated AKI is 5.4 (95% CI: 1.7–23), and for every additional year in age, the OR is 1.13 (95% CI: 1.05–1.25).

Figure 2.

Odds ratio visualization for the bivariate logistic regressions between piperacillin-associated AKI (PTZ-AKI) (three criteria) and individual predictors. Percentage fluid balance, nephrotoxic medications, highest trough (C_min), highest peak (C_max), AUC and cumulative piperacillin dose per weight were all assessed during the first 24 hours of piperacillin/tazobactam therapy. The reference weight for weight is ‘underweight’. KDIGO: Kidney Disease Improving Global Outcomes, PTZ = piperacillin/tazobactam. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

When only using the first two criteria for piperacillin-associated AKI, multivariable logistic regression found that the highest estimated C_min in the first 24 hours and having any stage of AKI prior to the first piperacillin/tazobactam dose to be predictors of piperacillin-associated AKI. For the highest estimated C_min in the first 24 hours, the OR of developing piperacillin-associated AKI is 2.8 (95% CI: 1.4–6.1). Any stage of AKI prior to the first piperacillin/tazobactam dose conferred an increased AKI risk of 5-fold (OR 5.0 95% CI: 1.4–21). For the final selected models from before, we performed Hosmer–Lemeshow goodness-of-fit tests with four categories for both models. Two models show that test P values are 0.319 and 0.250, respectively, indicating that both models are reasonably fit.

We evaluated the sensitivity and specificity of identifying piperacillin-associated AKI based on C_min values of 16 and 64 mg/L. A trough of 16 mg/L would yield a sensitivity of 75% and specificity of 59% for patients who meet all three criteria of piperacillin-associated AKI. A threshold of 64 mg/L would yield a sensitivity of 44% and specificity of 87%.

Discussion

Through our adjudication process, 15% of our cohort met all three criteria for piperacillin-associated AKI and whose likelihood were rated as possible or probable. When we expand our piperacillin-associated AKI definition to include patients who meet only the first two criteria, approximately one-third of patients meet the definition. We found that there were higher estimated AUCs and troughs in the first 24 hours of therapy in patients with piperacillin-associated AKI. Highest estimated trough in the first 24 hours was a strong predictor for piperacillin-associated AKI defined by two or three criteria with ORs between 2.8 and 5.4. We also found that age was a predictor for piperacillin-associated AKI defined by three criteria, while presence of AKI before piperacillin/tazobactam exposure was a predictor for two-criteria piperacillin-associated AKI.

The finding of 15% of patients who received piperacillin/tazobactam developing piperacillin-associated AKI by three criteria is consistent with other studies in children that have shown 11–17% of children who received piperacillin/tazobactam and vancomycin develop antibiotic-associated AKI.^12,15 There were no patients who met either two or three piperacillin-associated AKI criteria and were rated as definite on the probability scale, and only two patients who were rated as probable. This finding was unsurprising for multiple reasons. Since no patients had biopsies to evaluate kidney injury and the toxic concentration threshold for piperacillin has not been defined, a score of definite cannot be achieved (Table S1). Since these patients are critically ill, there are many other reasons for AKI development, which also deducts points on the probability scale. Since this was not an intervention trial, we could not administer a placebo and it was uncommon for piperacillin/tazobactam to be discontinued and restarted, thus further reducing the ability to score highly on the probability scale (Table S1).

We found that patients who met two or three piperacillin-associated AKI criteria and whose likelihood of having piperacillin-associated AKI were rated as possible or probable had higher piperacillin exposures, as evidenced by significantly higher estimated AUCs and troughs in the first 24 hours. When using the stricter definition of meeting all three criteria, the mean estimated highest free trough concentration in the first 24 hours in patients with piperacillin-associated AKI was 50.1 mg/L. We also found that a free trough concentration set four times above the CLSI breakpoint for P. aeruginosa (4× = 64 mg/L) would allow us to identify only 44% of patients with piperacillin-associated AKI but with a high specificity of 87%. Imani had previously reported a trough total concentration of 452 mg/L as the threshold associated with 50% risk of nephrotoxicity in adults.²⁶ Assuming 30% protein binding for piperacillin, this would correspond to a free trough concentration of approximately 300 mg/L, six times higher than the mean trough concentration in our patients who met all three piperacillin-associated AKI criteria. The trough concentration reported in the adult study would be considered very high and would occur infrequently when compared to the prevalence of piperacillin-associated AKI, as population PK models in critically ill children demonstrate that total peak concentrations are usually in the 100–500 mg/L range^28,44 and, thus, trough concentrations would be much lower. The reason for the large discrepancy in concentrations between our study and the adult study could include differences in definition of AKI as a probability scale was not used in the adult study. The adult study also included patients on continuous kidney replacement therapy, and 21% of patients had estimated glomerular filtration rates of less than 50 mL/min/1.73 m², both of which could lead to very high concentrations. Nonetheless, our findings corroborate a relationship between piperacillin concentrations and AKI and identify toxicity thresholds for precision dosing, and specifically for a paediatric-predominant population.

Our finding that age is a significant predictor of piperacillin-associated AKI, when using all three criteria, is of interest. This finding is probably from the third piperacillin-associated AKI criterion that peak creatinine be greater than 0.5 mg/dL, which probably excludes many young patients with low baseline creatinine. We found that patients who met at least the first two criteria for piperacillin-associated AKI were older, but age was not a significant predictor of piperacillin-associated AKI in this cohort. Since most patients included in this cohort had comorbidities, older patients are more likely to have had more exposures to nephrotoxic medications or more incidences of kidney injury in their lifetimes compared to younger patients. These repeated injuries to the kidney can decrease their renal functional reserve and increase their susceptibility to AKI when exposed to a nephrotoxin.⁴⁵ Multivariable logistic regression using only the first two criteria for piperacillin-associated AKI also identified having AKI before even the first dose of piperacillin/tazobactam as a significant predictor of piperacillin-associated AKI. This finding is not surprising as piperacillin probably exacerbates pre-existing injury.

In this cohort, we did not find that the number of nephrotoxic medications was a significant predictor of AKI, despite previous studies showing that exposure to three or more nephrotoxic medication increases AKI risk in children.^17,46 Reasons for not finding a significant association may include a small sample size and identifying nephrotoxic exposures in only the first 24 hours precluded assessment of a cumulative effect of nephrotoxins over days. In addition, presence of nephrotoxic medications led to loss of points on the Naranjo probability scale as they provide alternative reasons for AKI.

We specifically analysed the effect of estimated piperacillin exposures and certain predictors (cumulative piperacillin dose normalized to weight, number of nephrotoxic drugs, percentage fluid balance, lowest albumin) in only the first 24 hours to reduce confounding. Since piperacillin is renally cleared, it is expected that a patient who develops AKI would have high piperacillin concentrations. To assess whether high piperacillin exposures increase the risk of piperacillin-associated AKI development rather than whether AKI leads to high piperacillin exposures, we only looked at the effect of estimated piperacillin exposures in the first 24 hours. While it is plausible that there is a cumulative effect of piperacillin dose and nephrotoxic medications (i.e. the more days a patient is on piperacillin/tazobactam or nephrotoxic medications, the higher likelihood of piperacillin-associated AKI development), we sought to identify patients at risk for piperacillin-associated AKI early in their piperacillin/tazobactam course. We also showed that maximum KDIGO stage was often met within 2–3 days of piperacillin/tazobactam exposure (Figure S1). These patients who are identified early may benefit most from model-informed precision dosing or discontinuation of the drug. We thus limited our analysis of exposures to certain predictors to the first 24 hours.

This study is not without limitations. We excluded patients who received less than 24 hours of piperacillin/tazobactam therapy so we could estimate the AUC and highest C_max and C_min more accurately. These excluded patients represent an important group of patients who may be at risk for developing piperacillin-associated AKI even after a single dose of piperacillin/tazobactam. Further studies into this cohort with limited piperacillin exposures are warranted. We only evaluated piperacillin-associated AKI development in the first 7 days after the first piperacillin/tazobactam dose. In addition, creatinine is an imperfect marker for AKI. SCr concentrations may not rise until 24–36 hours after renal injury.⁴⁷ Multiple processes, other than renal filtration, can affect SCr concentrations including sepsis, liver disease and drug competition with creatinine secretion.⁴⁷ There is also concern that piperacillin may decrease creatinine secretion without causing injury by inhibiting tubular creatinine secretion.^48–50 While we used urine output criteria for KDIGO staging and some patients did meet criteria based on low urine output alone, urine documentation in patients without foley catheters can be inaccurate. To bias ourselves towards the null, we specifically excluded any 6-hour periods with unmeasured urine outputs or urine/stool mixes, which would lead to lower KDIGO staging (i.e. if a patient only had one 6-hour period without unmeasured outputs, the maximum KDIGO stage based on urine output would be 1 even if the unmeasured urine outputs or urine/stool mixes in the other 6-hour periods had minimal urine). Studies relating piperacillin concentrations and more sensitive and specific biomarkers for the detection of early drug-induced nephrotoxicity^51–54 are warranted.

Conclusion

Our single-centre study further defines a relationship between estimated piperacillin concentrations, specifically between trough concentrations in the first 24 hours of piperacillin/tazobactam therapy and the development of piperacillin-associated AKI. These findings support model-informed precision dosing of piperacillin/tazobactam to maximize efficacy, minimize risk and provide a concentration range where we can further investigate to inform potential toxic thresholds. Further studies to understand the mechanism underlying piperacillin-associated AKI and the relationship between piperacillin/tazobactam and non-creatinine AKI biomarkers are warranted.

Funding

This study was supported by the Cincinnati Children’s Hospital Medical Center (CCHMC) Child Health Research Career Development Award 5K12HD028827 (STG) from the National Institute of Child Health and Development and by 5T32DK0079695 (JTC) from the National Institute of Diabetes and Digestive and Kidney Diseases. The original dataset used in the study was created with funding from the National Institute of Child Health and Development Cincinnati Pediatric Clinical Pharmacology Postdoctoral Training Program [5T32HD069054-09], Gerber Foundation Novice Research Award, CCHMC Arnold Strauss Award and Hospital Medicine Fellow Award. REDCap use is supported by the NIH Clinical and Translational Science Award (CTSA) program [2UL1TR001425].

Transparency declarations

None.

Supplementary data

Figure S1 and Tables S1 to S3 are available as Supplementary data at JAC online.