Population Pharmacokinetics of Intravenous Phenobarbital in Neonates after Congenital Heart Surgery

Abstract

Objectives

To develop a population pharmacokinetic (PK) model for intravenous phenobarbital in neonates following cardiac surgery and perform simulations to identify optimal dosing regimens.

Design

Retrospective single-center PK study.

Setting

Cardiac Intensive Care Unit at Children’s Hospital of Philadelphia.

Patients

Consecutive neonates who received ≥1 dose of intravenous phenobarbital and had ≥1 phenobarbital concentration drawn per standard of care from June 15, 2012 to October 15, 2018.

Interventions

None.

Measurements and Main Results

A population PK model was developed using nonlinear mixed-effects modeling. Simulations were performed using the final model parameters. Optimal phenobarbital loading doses were determined based on attainment of peak and maintenance concentrations between 20–40 mg/L. A total of 37 neonates contributed 159 PK samples. The median (range) weight, postmenstrual age, and postnatal age were 3.2 kg (1.3, 3.8), 39 2/7 weeks (28 2/7, 42 6/7), and 5 days (0, 26), respectively. Twelve (32%) patients were on extracorporeal membrane oxygenation (ECMO). A one-compartment model best described the data. The final population PK model included (1) weight and postnatal age for clearance, and (2) weight, ECMO and albumin for volume of distribution. In neonates not on ECMO, loading doses of 30 mg/kg and 20 mg/kg reached goal concentration with albumin values ≤ 3 mg/dL and 3.5 mg/dL, respectively. Loading doses of 30 mg/kg reached goal concentration on ECMO regardless of albumin values. Maintenance doses of 4 to 5 mg/kg/day reached goal concentration in all neonates.

Conclusions

In neonates following cardiac surgery, phenobarbital clearance increased with postnatal age. Volume of distribution increased with ECMO and lower albumin values. Loading doses of 30 mg/kg on ECMO and 20–30 mg/kg without ECMO were needed to reach goal concentration based on simulations.

Introduction

Seizures are common in neonates following cardiac surgery with cardiopulmonary bypass (CPB) and have been associated with worse long term neurodevelopmental outcomes compared to neonates who did not have seizures (1–5). Phenobarbital (PHB) is widely recognized as a first-line antiseizure medication in this population (1, 6). PHB is a long-acting barbiturate that is metabolized by hepatic cytochromes with approximately 25% of the drug excreted unchanged by the kidneys (7). Previously described neonatal pharmacokinetics (PK) include clearance (CL) values ranging from 3.5 to 9.6 mL/h/kg and volume of distribution (V) values ranging from 0.49 to 1.56 L/kg (8–18). The variability between studies suggests that disease states and physiological parameters may affect PHB PK.

Despite the multitude of factors that may modify drug disposition in critically ill neonates after cardiac surgery there is a paucity of data on PHB PK in this population. Acute kidney injury frequently occurs (19) and may alter PHB excretion. Marginal cardiac output may affect liver perfusion, thereby limiting drug metabolism. The presence of a cardiac shunt may create an imbalance between the systemic and pulmonary circulations, which may impact PK parameters. The presence of extracorporeal devices, such as extracorporeal membrane oxygenation (ECMO) and renal replacement therapies (RRTs), may further alter PHB disposition. Finally, concomitant administration of additional neuroactive medications commonly used in this population, including (fos)phenytoin and midazolam, can alter PHB CL, either through inhibition or induction of its metabolism (16).

Rapid and efficient treatment of seizures may improve neurodevelopmental outcome in neonates following cardiac surgery (2, 20), but the optimal PHB dosing regimens remain unknown. We aimed to characterize PHB PK and to provide dosing recommendations in this population.

Methods

This was a single-center retrospective study of neonates (≤ 30 days of age, corrected gestational age ≤ 44 weeks) who were treated with intravenous (IV) PHB following cardiac surgery at Children’s Hospital of Philadelphia. The Institutional Review Board approved the study. We included subjects if they received ≥1 dose of IV PHB and had ≥1 PHB plasma concentration drawn for therapeutic drug monitoring (TDM) per standard of care from June 15, 2012 to October 15, 2018. Data collected included patients demographics (age, weight, height, birth weight, gestational age), surgery related data (primary cardiac anomaly, procedure performed, CPB times), laboratory values (blood urea nitrogen [BUN], serum creatinine [SCR], albumin, and alanine aminotransferase [ALT]), PHB plasma concentrations, and concomitant medications known to alter PHB metabolism (pantoprazole, midazolam, [fos]phenytoin)(16). PHB dosing data were obtained from the medical record and included dose amounts, route of administration, and date/time. When applicable, the details of ECMO and RRT were also collected. Data were collected through the duration of IV PHB therapy during neonatal hospitalization.

Drug Dosing and PK Samples

The current recommended initial IV PHB dosing consists of a loading dose of 15–20 mg/kg administered at a maximum rate of 1 mg/kg/minute, with additional boluses of 5–10 mg/kg every 15–20 minutes as needed to terminate seizures or reach therapeutic concentrations up to a maximum total dose of 40 mg/kg (21). Loading doses are typically followed by maintenance doses of 3–6 mg/kg/day (21). Maintenance doses were defined as any scheduled dose given with the intent of maintaining a desired concentration. Initial loading doses were identified as single doses administered before initiation of scheduled maintenance dosing. Hypotension is a known dose-dependent adverse effect of PHB (21) so the initial loading bolus in children with hemodynamic instability is often divided into smaller doses of 2.5–10 mg/kg. Additional boluses were defined as single doses administered after initiation of scheduled maintenance dosing. Sampling was performed per standard of care by the clinical team for TDM. Comparative statistics between administered doses in neonates with or without ECMO were performed using Wilcoxon’s rank-sum test.

PK Analysis

PHB disposition was estimated using a population PK analysis (NONMEM software, version 7.3.4; ICON Clinical Research LLC). All models were run with the First-Order Conditional Estimation with Interaction (FOCEI) method. Goodness-of-fit diagnostic plots were generated in R. Output was summarized using R through the RStudio platform (version 1.1.456) and STATA (version 14.2). The goodness-of-fit for each run was assessed by examining the following criteria: visual inspection of diagnostic scatter plots, the precision of the parameter estimates, successful minimization, relative changes in Akaike Information Criteria (AIC) and Objective Function Value (OFV), and differences in interindividual and residual variabilities.

One- and two-compartment models with linear disposition were evaluated. For a one-compartment model, the estimated parameters included CL and V. For a two-compartment model, the estimated parameters included CL, intercompartmental clearance (Q), central (Vc) and peripheral volumes (Vp) of distribution. Interindividual variability was assessed using an exponential model. Residual variability was evaluated using an additive, a proportional, and a combined (additive and proportional) model. The potential effects of covariates were evaluated if a relationship was suspected based on physiological plausibility. The relationship between the initial PK parameter estimates and their random effect with covariates was evaluated using visual inspection of scatter and box plots (continuous and categorical, respectively). Body weight was included as a covariate for clearances and volumes of distribution and was included in the base model. The relationship between weight and PK parameter was characterized using a fixed allometric relationship normalized on our population median weight, where the allometric exponent was fixed at 0.75 for clearances and 1 for volumes. Covariates were assessed for collinearity using scatter plots, coefficients of correlation (>0.5), and relative changes in final estimates, AIC and OFV during model building (22). Covariates that were evaluated included postnatal age, postmenstrual age (defined as postnatal age + gestational age), postoperative day, the presence of ECMO and RRT (as dichotomous variables), single ventricle physiology, CPB time, concomitant medication of interest (midazolam, [fos]phenytoin, and pantoprazole), SCR, and albumin. Postmenstrual age was evaluated both as a continuous variable and as part of a sigmoid maturation model (Hill equation)(23). The Hill equation was tested both using estimated and fixed parameters (Hill coefficient and maturation half-life [TM50]) based on literature values (23). Additionally, as PHB induces hepatic microsomes when administered for approximately one week, the number of days receiving PHB was evaluated on CL if PHB was administered for ≥7 days (24). The relationship between PK parameters and continuous covariates were explored using power models normalized on the covariate median value. Addition of covariates to the model was made using a stepwise forward additive and a backward elimination approach, using a p-value of 0.05 (ΔOFV=3.84) for inclusion and 0.01 (ΔOFV=6.63) for removal. Collinear covariates were not included together in the multivariate step of the model to limit competitive bias (22). The model fit was evaluated using successful minimization, goodness-of-fit plots, plausibility and precision of parameter estimates, bootstrap procedures (1,000 replicates) and standardized visual predictive check.

Simulations

The final PHB PK model was used to assess dose-concentration relationship. The effect of significant covariates was illustrated through simulations where different loading and maintenance doses were administered to virtual neonates with similar baseline characteristics than our population. Simulations included 1,000 replicates and were summarized by median value with 95% confidence interval. Simulated loading doses ranged from 10–80 mg/kg/dose given at a rate of 1 mg/kg/min and simulated maintenance doses ranged from 3 to 6 mg/kg given every 12h. Goal concentration was defined as 20–40 mg/L for both initial (peak concentration two hours following a loading dose) and sustained efficacy at estimated steady state (trough concentrations after 20 days of therapy) (7, 16, 25). The percent of peak concentrations within the goal of 20–40 mg/L after the initial loading dose was determined. Optimal dosing regimens were selected based on their ability to reach goal concentration. Finally, dosing guidance based on the simulations were compared to actual doses administered to assess whether dosing would change based on model specific guidance.

Results

Study Population

We included 38 neonates. One neonate was found be an extreme outlier based on the PK profile with observed concentrations lower than expected. None of the tested models could accurately characterize PHB disposition for that patient (overprediction of observed PHB concentrations). Table 1 provides a comparison of baseline characteristics between this neonate and the remainder of our cohort. This neonate was hemodynamically unstable with ongoing bleeding at the time of PHB administration. Over the day preceding PHB first dose administration, he received >2 liters in blood products, representing 7 to 8 times his total circulating blood volume (26). This neonate was considered an outlier and was excluded from further analysis and model development, yielding 37 neonates for analysis.

Table 1.

Baseline Characteristics and PK Parameters

Baseline Characteristics	Neonates Included in Model Development1 (N=37)	Neonate Excluded from Model Development (N=1)	Entire Cohort1 (N=38)
Postnatal age (days)	5 (0, 26)	23	5.5 (0, 26)
Postmenstrual age (weeks)	39 2/7 (28 2/7, 42 6/7)	41 4/7	39 2/7 (28 2/7, 42 6/7)
Height (cm)	49 (33, 56)	52	49.2 (35, 56)
Weight (kg)	3.2 (1.3, 3.8)	3.5	3.2 (1.3, 3.8)
Cardiopulmonary bypass time (min)	48 (34, 179)	255	49.5 (34, 255)
DHCA, n (%)	28 (75.7%)	yes	29 (76.3%)
DHCA time (min)	40 (18, 104)	34	40 (18, 104)
Single ventricle, n (%)	20 (54.05%)	no	20 (52.6%)
ECMO, n (%)	12 (32.4%)	yes	13 (34.2%)
RRT, n (%)	2 (5.4%)	yes	3 (7.9%)
Postoperative day during PK sampling	3.4 (−13.8, 52.7)	2 (1.5–3.5)	3.1 (−13.8, 52.7)
Albumin value (mg/dL)	2.5 (1.6, 3.7)	2.3	2.5 (1.6, 3.7)
SCR value (mg/dL)	0.6 (0.2, 2.1)	0.3	0.6 (0.2, 2.1)
ALT value (U/L)	32 (14, 404)	54	33 (14, 404)
Co-medication
Midazolam, n (%)	8 (21.6%)	yes	9 (23.6%)
Pantoprazole, n (%)	15 (40.5%)	yes	16 (42.1%)

¹Median (range) unless otherwise specified

ALT: alanine aminotransferase, DHCA: deep hypothermic circulatory arrest, ECMO: extracorporeal membrane oxygenation, RRT: renal replacement therapy, SCR: serum creatinine

The median (range) weight, postmenstrual age, and postnatal age were 3.2 kg (1.3–3.8), 39 2/7 weeks (28 2/7–42 6/7), and 5 days (0–26), respectively (Table 1). Throughout the course of PK sampling, 12 neonates (32%) received ECMO support and 2 neonates (5%) received RRT (continuous veno-venous hemodiafiltration in one neonate [3%], and peritoneal dialysis in one neonate [3%]; Table 2). During PK sampling, 15 neonates (41%) received concomitant pantoprazole therapy, 8 (22%) received concomitant midazolam therapy, 1 (3%) received concomitant phenytoin therapy, and 3 (8%) received concomitant fosphenytoin therapy.

Table 2.

PK Sampling during Extracorporeal Support

Sampling Pattern on Extracorporeal support	Number of Samples, n (%)
ECMO
Subjects who underwent PK sampling only while on ECMO (n=8)	33 (21%)
Subjects who underwent PK sampling during and after ECMO (n=4)
On ECMO	13 (8%)
Post-ECMO	13 (8%)
Total number of samples on ECMO	46 (29%)
RRT
Subjects who underwent PK sampling before and during CVVDDF (n=1)
Pre-CVVHDF	3 (2%)
On CVVHDF	1 (1%)
Subjects who underwent PK sampling before and during PD (n=1)
Pre-PD	11 (7%)
On PD	11 (7%)
Total number of samples on RRT	12 (8%)

CVVHDF: continuous veno-venous hemodiafiltration, ECMO: extracorporeal membrane oxygenation, PD: peritoneal dialysis, PK: pharmacokinetics, RRT: renal replacement therapy

PHB Dosing and PK Samples

Neonates received a median (range) of 3 (1, 11) initial loading doses of 10 mg/kg (2.4, 28.9) prior to maintenance doses, resulting in a total initial loading amount of 30.6 mg/kg (4.8, 70). Although not statistically significant, there was a trend toward neonates on ECMO receiving a higher total initial loading amount (39 mg/kg [10.1, 60.1] vs. 25.1 mg/kg [4.8, 70], p=0.07). Fourteen neonates (38%) continued to have seizures despite reaching PHB concentrations >20 mg/L and received additional initial loading doses after the achievement of goal concentration. Initial loading doses were followed by median maintenance doses of 2.44 mg/kg (1, 10.2) every 12–24h. Three neonates received a median of 2 (1, 3) additional boluses of 9.95 mg/kg (3.1, 10) following initiation of maintenance doses. In one neonate, seizures recurred shortly after initiation of maintenance therapy despite reaching goal concentration, and an additional bolus was given less than 24 hours following initial loading. In the remaining two neonates, additional boluses were given following recurrence of seizures more than two weeks following initial loading doses. In one of those two neonates, PHB concentrations had dropped below 20 mg/L, whereas they remained within target for the remaining neonate. The median duration of IV PHB treatment was 8.2 days (0.2, 51.5), and plasma concentrations were obtained after a median of 1.7 days of treatment (0.03, 51.4). Including 2 neonates (6%) who had seizures and were treated with PHB prior to their cardiac intervention, PK samples were obtained at a median of 3.4 days (−13.8, 52.7) following surgery. A total of 159 PK samples were collected with a median of 5 (1, 22) samples per patient. None of the PK samples were below quantification limit.

Population PK Model

A one-compartment model with first-order elimination best described the data. Progression of model development is shown in Supplemental Digital Content, Table S1. As very few neonates (<10%) received concomitant treatment with either phenytoin or fosphenytoin, the effect of those concomitant medications was not tested as covariates in the model. Similarly, the impact of RRT on PK parameters was not tested (<10% of our population on RRT). After accounting for weight, postnatal age as a covariate on CL, and ECMO and albumin as covariates on V resulted in a significant reduction in the OFV/AIC and improvement in the model fit. No other covariates reached statistical significance. Interindividual variability for CL was estimated at 13.6%, although it was estimated with poor precision (residual standard error [RSE]=92% and confidence interval crossing null value). Interindividual variability for V was estimated at 15.8% (RSE=42%). A proportional error model was used to describe the random residual variability and was estimated at 15.9%. A bootstrap analysis confirmed the stability of our model, as shown in Table 3. The precision of our covariate effects was also validated by the bootstrap analysis, with the exception of albumin on V (confidence interval crossing null value, Table 3). Goodness-of-fit plots indicated a good fit for the final model, with adequate prediction of individual and population concentrations, as presented in Figure 1. The standardized visual predictive check indicated good prediction of our final model, with 16 observed values (10%) outside of the 95^th confidence interval (Supplemental Digital Content, Figure S2).

Table 3.

Final Model and Bootstrap Analysis¹

	Final Model				Bootstrap Analysis (N=998)
	Point Estimate	RSE (%)	95% CI	CV (%)	2.5th percentile	Median	97.5th percentile
CL (mL/h) for a 3.2kg neonate	14.4	6.22	12.6, 16.2		12.2	14.6	18.1
V (mL/3.2kg)	3050	4.10	2810, 3300		2810	3049	3300
PNA on CL	0.25	19.2	0.15, 0.34		0.006	0.24	0.4
ECMO on V	1.21	8.15	1.02, 1.40		1.05	1.22	1.45
Albumin on V	−0.43	48.2	−0.84, −0.02		−0.88	−0.44	0.002
Interindividual variability on CL	0.02	91.9	−0.01, 0.05	13.6	0.000002	0.02	0.06
Interindividual variability on V	0.03	42.4	0.004, 0.05	15.8	0.006	0.02	0.05
Residual variability2	0.03	24.6	0.01, 0.04	15.9	0.01	0.02	0.04

¹Parameter estimates are for a 3.2kg neonate of 5 days old, with no extracorporeal membrane oxygenation (ECMO) and an albumin value of 2.5 mg/dL

²Proportional error

CL (mL/h)=14.4 x (WT/3.2)^0.75x (PNA/6)^0.25

V (mL)=3050 × 1.21 (if on ECMO) x (ALB/2.5)^−0.43

BSV: between-subject variability, CL: clearance, CV: coefficient of variation, ECMO: extracorporeal membrane oxygenation, PNA: postnatal age, RSE: relative standard error, V: volume of distribution

Figure 1. Diagnostic Plots for the Final Model.

Final phenobarbital diagnostic plots: observed versus individual predicted concentrations (A), or population predicted concentrations (B), and weighted residuals (C) or conditional weighted residuals (D) versus population predicted concentrations.

Our population estimate for CL was estimated at 14.4 mL/h for a typical 3.2kg neonate of 5 days old. The model predicted an increase in CL with postnatal age. For example, at 0, 10, and 20 days old, the estimated CL for a 3.2kg neonate were 9.2, 16.8, and 19.7 mL/h, representing a 114% increase in CL over the first 20 days of life. Our population estimate for V was 3050 mL for a typical 3.2kg neonate without ECMO and with an albumin value of 2.5 mg/dL. The presence of ECMO increased the estimated V by 22%, and normalization of albumin values from 2.5 mg/dL to 3.5 mg/dL decreased the estimated V by 13%.

Simulations

Using the final PHB PK model derived in the current study, different loading doses were simulated in a 3kg neonate with and without ECMO, using albumin values of 2.5, 3, and 3.5 mg/dL (Figure 2). Because inter-individual variability on CL could not be characterized with precision (RSE>90% and confidence interval crossing null value), the effect of covariates on maintenance doses was not evaluated, and simulations were conducted using a 3kg neonate without ECMO and with an albumin of 2.5 mg/dL (our population median value).

Figure 2. Simulated Peak Phenobarbital Concentrations.

Simulated peak phenobarbital concentrations 2 hours post-dose with and without ECMO using albumin values of A) 2.5 mg/dL, B) 3 mg/dL, and C) 3.5 mg/dL. The gray box represents peak goal concentrations between 20–40 mg/dL.

In a neonate with or without ECMO, loading doses of 30 mg/kg showed the highest proportions of simulated concentrations within the goal range with albumin values of 2.5 and 3 mg/dL (with ECMO: 83.1% [albumin=2.5 mg/dL], and 86.4% [albumin=3 mg/dL]; without ECMO: 85.7% [albumin=2.5 mg/dL], and 78.1% [albumin=3 mg/dL]. With an albumin value of 3.5 mg/dL, following loading doses of 20 mg/kg in a neonate without ECMO and 30 mg/kg in a neonate with ECMO, 77.3% and 88.1% of the simulated concentrations were within the goal concentration range, respectively. In neonates without ECMO and with an albumin value of 3.5 mg/dL, loading doses of 20 and 30 mg/L had similar attainment rate, reaching our goal concentrations in 77.3% and 70.8% of the simulated concentrations, respectively. Similarly, in neonates with ECMO and with an albumin value of 2.5 mg/dL, loading doses of 30 and 40 mg/kg reached similar concentrations, with goal concentration attainment rate across repetitions of 83.1% and 76.6%, respectively. Maintenance doses of 4 to 5 mg/kg/day administered every 12h reached our goal trough concentrations (Figure 3).

Figure 3. Simulated Phenobarbital Concentrations in a Typical Neonate.

Simulated phenobarbital concentrations over 20 days for a 3kg neonate without ECMO and with an albumin value of 2.5 mg/dL. The gray box represents goal concentrations between 20–40 mg/dL.

Discussion

To our knowledge, this is the first PHB population PK study in neonates following cardiac surgery and the first to include neonates with and without ECMO. A one-compartment model best described our data as previously reported in neonates, infants, and children (8–10, 13, 15, 16, 18). Our population estimates for CL and V were in the lower and upper range of previously reported values in neonates, respectively (8, 10, 13, 16, 18). Clinically, this resulted in neonates with hypoalbuminemia and/or ECMO exposure displaying higher V and thus requiring higher PHB loading doses to attain goal concentrations.

Few data describe PHB PK in patients undergoing ECMO support (27). Our final model included 46 PK samples from 12 neonates undergoing ECMO. Pokorna et al. previously described PHB PK in 7 neonates on ECMO (14), and the study reported a higher CL (8 mL/h/kg vs. 4.5 mL/h/kg) and a lower V (460 mL/kg vs. 1153 mL/kg) when expressing the data per kilogram of body weight for comparison. This may represent the expected variability in PK parameters in neonates, as all values are within the range of previously published values in neonates not supported by ECMO (8–13, 15, 17, 18). However, another potential explanation for the difference in CL/kg involves the presence of organ dysfunctions in our population. PHB is metabolized by the liver via cytochrome P450 pathways with about a quarter of the dose excreted unchanged by the kidneys (7). In our study, neonates on ECMO had severe cardiac dysfunction which potentially led to hypoperfusion-related insult to the liver and the kidneys prior to cannulation. The low albumin and high SCR values in our population support the presence of organ dysfunction, although those are unreliable markers of liver and kidney functions in children (28, 29). Similarly, an important inflammatory state in our cohort may explain our observed higher V/kg. Critical illness promotes an inflammatory reaction which increases V notably via systemic capillary leak syndrome and fluid resuscitation (30).

Postnatal age was a significant covariate on CL in our model, which is consistent with previously published literature (8, 10, 13, 15). In addition to the expected enzyme maturation with time, the increase in CL with age in our population can be explained by the resolution of organ dysfunctions occurring during the perioperative period. Postmenstrual age has been associated with CL (16). The addition of postnatal age to our model resulted in a better fit than the addition of postmenstrual age. Our relatively advanced median postmenstrual age (39 2/7 weeks) may have prevented us from characterizing the impact of prematurity on CL. The rapid maturation of metabolic pathways following birth, independently of gestational age, may also explain the greater impact of postnatal age on CL when compared to postmenstrual age (31). PHB has been shown to induce its own clearance over time through induction of hepatic enzymes (24). Our short duration of treatment (median of 8.2 days) may have precluded characterization of this effect.

V was increased in neonates undergoing ECMO. This finding is supported by the fact that higher doses are needed to reach goal concentrations while on ECMO (32). Many factors may contribute to the increased V while on ECMO. Neonates typically require a large volume of blood products to prime the circuit and maintain hemostasis throughout the ECMO course. In addition to blood products, fluid boluses may be needed to maintain hemodynamic stability and ECMO flow, thereby contributing to hemodilution. The effect of hemodilution was well demonstrated in the current study: our outlier neonate, after receiving multiple transfusions, had a higher V compared to our cohort’s value. Direct drug adsorption by the ECMO circuit could also contribute to the increased V. However, the latest published ex vivo study demonstrated stable PHB concentrations in the ECMO circuit (33).

Our results suggest that V increases as albumin decreases. Although the clinical significance of this finding remains unclear, as shown by the high residual standard error of our point estimate and the bootstrap analysis, we believe this finding is of interest. It is possible that hypoalbuminemia may be a surrogate for fluid status. Fluid shifts caused by significant fluid losses, subsequent fluid resuscitation, and capillary leak are frequent in the early postoperative period following neonatal cardiac surgery and may result in hypoalbuminemia. As a positive fluid balance is associated with an increased V, the association between hypoalbuminemia and increased V may, in fact, represent an association between positive fluid balance and increased V. Significant fluid shifts occur more frequently in the early postoperative period, but postoperative day was not found to be a significant covariate on V. We hypothesize that differences in postoperative fluid status and fluid management based on (1) the clinical stability and cardiac physiology of individual neonates, and (2) different fluid management strategies depending on the preferences of the clinical team may have prevented us to characterize an association between postoperative day and V. Alternatively, the increased V associated with low albumin may be explained by extremely low protein binding. Hypoalbuminemia, by decreasing the bound fraction, can result in lower total concentrations, higher free concentrations, and an apparent increase in V (34), but this is typically observed with highly protein-bound molecules. PHB exhibits low to intermediate protein-binding in neonates with 28–36% of the total concentrations bound to proteins (35). Therefore, changes in albumin concentrations would not be expected to result in significant changes in V. However, important changes in protein binding were previously associated with a clinically significant increase in free PHB concentrations in an adult with combined hypoalbuminemia and hyperbilirubinemia (36). Similarly, it is possible that large changes in protein binding impact PHB disposition in critically ill neonates.

Our PHB dose-concentration simulations indicated that loading doses of 20–30 mg/kg and maintenance doses of 4–5 mg/kg/day achieved our goal concentration of 20–40 mg/dL. This is consistent with published dosing recommendations (21) and previous studies (9, 16, 18). This is also consistent with our results, as most neonates received repetitive loading doses to reach a median total loading amount of 30.6 mg/kg. Considering that neonates received median loading doses of 10 mg/kg (2.4, 28.9) in the current study and that timely and effective treatment is associated with better outcomes (37), PHB dosing in our population may have been optimized with faster administration of larger doses, assuming hemodynamic and respiratory stability. Alternate antiseizure medications that allow for more rapid administration of higher doses without concern for hemodynamic side effects may be considered in clinically unstable patients.

PHB PD requires better characterization. PHB concentrations required for seizure cessation are not well determined and may vary according to individual characteristics, with some patients responding only to very high concentrations (38, 39). Such variability was observed in our study, with 38% of neonates requiring additional loading doses despite reaching goal concentration. Moreover, as mentioned earlier, significant alterations in protein binding may lead to changes in free and total concentrations. Monitoring free concentrations, which correlate better with PHB efficacy (40), could therefore be valuable under certain circumstances.

Our study has limitations. First, we could not characterize interindividual variability on CL with precision. This is probably due to the retrospective nature of our study, with very few PK samples drawn later in the dosing interval. Second, our results are only applicable to neonates with similar baseline characteristics as our cohort. Indeed, we had to exclude one neonate with massive bleeding from our PK analysis as our model could not fit the data appropriately. This is most likely because (1) the rest of our cohort was homogenous and displayed similar characteristics, and (2) the impact of one outlier neonate was exacerbated by the relatively small number of patients. Importantly, this outlier patient illustrates the impact of massive transfusions and hemodilution on PK and therapeutic drug monitoring should be considered to optimize drug exposure in those circumstances. Third, free concentrations were not available, and correlation between free concentrations and efficacy was not possible. Fourth, this study lacks safety data, which is a major concern when administrating PHB to critically ill neonates.

This study also has considerable strengths. To our knowledge, it is the first study on PHB PK in neonates following cardiac surgery, and it includes neonates on ECMO. It also is the first PK study to describe an association between albumin and PHB V. The significance of this finding is unclear and may represent the impact of altered protein binding or positive fluid balance and warrants further studies. Finally, this study provides useful PHB dosing guidance.

Conclusion

In neonates following cardiac surgery, PHB CL and V were in the lower and upper range of previously reported values in neonates, respectively. In neonates without ECMO, loading doses of 30 mg/kg and 20 mg/kg reached goal peak concentration with albumin values ≤ 3 mg/dL and 3.5 mg/dL, respectively. In neonates on ECMO, loading doses of 30 mg/kg achieved goal peak concentration. Maintenance doses of 4 to 5 mg/kg/day sustained goal trough concentration. Postnatal age, ECMO, and albumin values should be considered when dosing PHB in critically ill neonates following cardiac surgery. A better understanding of PHB PD and further PK data focusing on the impact of altered protein binding, fluid status and fluid management strategies in neonates are needed.