Real‐world interpatient variability in the pharmacokinetics of levetiracetam

Abstract

Background

Levetiracetam (LEV) is an antiepileptic drug (AED) used to treat a variety of seizures in adult and pediatric populations. It is an ideal AED due to its favorable pharmacokinetic (PK) and pharmacodynamic profile and lack of interactions with other AEDs.

Methods

This retrospective cohort study was designed to identify covariates that affect LEV clearance and volume of distribution and to generate a population PK model. Adults with a seizure history receiving LEV during hospital admission with a minimum of one serum LEV concentration available were included in the study. Population PK modeling and covariate testing was performed with MONOLIX Suite 2020R1 (Lixoft, France).

Results

A total of 162 serum concentrations were collected from 143 patients. Age, sex, body weight descriptors, serum creatinine, creatinine clearance (CrCL), serum albumin, liver enzymes, and total bilirubin were evaluated. Body surface area (BSA) was a significant covariate for the apparent volume of distribution (V/F). The exclusion of BSA as a covariate of V/F increased the objective function value (OFV) 5.6. CrCL was a significant covariate of apparent plasma clearance (CL/F). The exclusion of CrCL increased the OFV by 18.16 and significantly increased the root square error (RSE) % of the between‐subject variabilities of the parameters.

Conclusion

LEV clearance is an effective predictor of serum concentration. CrCL was a significant covariate influencing LEV clearance, and BSA was found to influence the volume of distribution. Further studies are needed to determine the effect of body weight descriptors on LEV clearance and, ultimately, outcomes.

Abbreviations

AdjBW

adjusted body weight

AED

antiepileptic drug

ALT

alanine aminotransferase

AST

aspartate aminotransferase

AUC

area under the curve

BIC

Bayesian information criteria

BMI

body mass index

BSA

body surface area

CL

clearance

CL/F

apparent plasma clearance

C_max

maximum plasma concentration

CrCL

creatinine clearance

ESRD

end‐stage renal disease

IBW

ideal body weight

ICU

intensive care unit

ILAE

International League Against Epilepsy

LBW

lean body weight

LEV

levetiracetam

OFV

objective function value

PK

pharmacokinetic

RSE

root square error

TDM

therapeutic drug monitoring

V/F

apparent volume of distribution

VPC

visual predictive checks

1. INTRODUCTION

Levetiracetam (LEV) is a second‐generation antiepileptic drug (AED) widely prescribed to treat epileptic seizures [1, 2, 3]. LEV is an ideal AED due to its favorable pharmacokinetic (PK) and pharmacodynamic profile and lack of interactions with other AEDs since it is not metabolized via CYP450 enzymes [1, 4]. Additionally, LEV demonstrates fewer adverse effects, most commonly reported being somnolence, asthenia, coordination difficulties, and dizziness [1, 4, 5]. One unique although less common side effect noticed with LEV include behavioral changes ranging from agitation to suicidality [1, 6]. Creatinine clearance (CrCl) has a significant influence on LEV PK; however, other covariates such as age, weight, gender, race, and concomitant medications may also have an effect [4, 7, 8].

Therapeutic drug monitoring (TDM) of AEDs is a beneficial tool in clinical practice used to prevent the occurrence of seizures, minimize adverse effects, and optimize therapy outcomes in individual patients [9, 10, 11]. According to current guidelines, routine TDM for LEV is not required since LEV exhibits linear PK, predictable changes in serum concentrations following dose change, and low rates of side effects [11, 12, 13]. Nevertheless, some clinicians collect serum concentrations since TDM can be beneficial in patients where the PK may be altered or noncompliance is suspected [11]. Special patient populations, including older adults, pediatric patients, renally impaired, and critically ill patients in the intensive care unit (ICU), may also benefit from TDM of LEV due to their altered PK profiles [9, 11, 13]. The reference range of 12–46 mcg/dL was selected based on the recommendations of the International League Against Epilepsy (ILAE) [12].

LEV is rapidly absorbed after oral administration with 100% bioavailability and reaches steady‐state concentrations within 2 days [1]. LEV is not bound to plasma protein and is mainly eliminated via kidneys (66%), and clearance is reduced in patients with severe renal disease (CrCL < 30 mL/min) [1]. Therefore, dose adjustments may be necessary in patients with renal disease. The drug is also metabolized in plasma through the hydrolysis via type B esterases to inactive metabolites [1, 14].

Although LEV is considered a weight‐neutral agent, there have been controversial findings of the effect of body weight on LEV PK profile [4]. The most common covariate affecting the volume of distribution reported in the literature was total body weight [7, 8, 15, 16, 17, 18, 19]. Hernandez‐Mitre et al. reported a relationship between BSAand volume of distribution [20]. Pigeolet et al., Chhun et al., Ito et al., and Rhee et al. identified a direct relationship of weight on LEV oral clearance [15, 16, 17, 18]. Rhee et al. have suggested that doses of LEV should be adjusted according to the patient's body weight, especially in patients with significantly high or low body weight [16]. Currently, LEV is weight‐dose normalized only in the pediatric population, but certain adult populations (obese or underweight) could potentially benefit from weight‐based dose adjustment [1]. The results of Pigeolet et al. are consistent with those of Rhee et al., showing that body weight is a significant covariate affecting CL/F of LEV [15, 16]. Specifically, their findings showed a 35% increase in the maximum plasma concentration (C_max) and a 16% increase in area under the curve (AUC) both of which were significant when the body weight decreased from 70 to 40 kg [15, 16]. In contrast, more recent studies by Karatza et al. and Hernandez‐Mitre et al. did not report any significant correlation between weight and clearance (CL) of LEV [20, 21]. As mentioned by Hernandez‐Mitre et al. and Alzuela et al., LEV PK needs to be clarified, especially the effect of weight on CL to determine whether higher doses of LEV are needed in obese patients [20, 22].

This study aims to develop a population PK model through the application of non‐linear mixed‐effects modeling with real‐world clinical data to determine the effect of covariates on LEV CL and volume of distribution.

2. METHODS

2.1. Data source

Subjects for this study were identified using electronic health records from a 365‐bed academic community medical center located in New Jersey, USA, between January 1, 2017, and March 3, 2021. LEV serum concentrations were extracted from the medical records. Dosing regimens ranged from 250 mg to 2000 mg twice a day and were administered intravascularly or extravascularly (oral, gastric tube, nasogastric tube). The protocol was granted expedited approval by the Rutgers Biomedical and Health Sciences (PRO2021000173) and Robert Wood Johnson University Hospital Somerset (IRB21‐10) Institutional Review Boards.

2.2. Inclusion and exclusion criteria

Adults (≥ 18 years of age) with seizures who were managed with and have received LEV during hospital admission and with at least one serum LEV concentration post‐admission available were included in the study. Patients were excluded from the study if they were pregnant, had alanine aminotransferase (ALT) and aspartate aminotransferase (AST) levels greater than three times the normal upper limit (120 and 168 U/L respectively), or serum creatinine greater than 2.5 mg/dL which usually indicated end stage renal disease (ESRD) [23]. These patients were excluded to limit the potential for confounding due to altered PK profiles.

2.3. Pharmacokinetic modeling

Available data were analyzed using a population modeling approach in MONOLIX Suite 2020R1 (Lixoft, France) to capture the mean population parameters and interpatient variability. Several models with one or two distribution compartments, first‐ or zero‐order absorption with and without a lag‐time, and linear elimination were tested. Models were compared based on model fit, goodness‐of‐fit plots, precision of parameter estimates, and the value of the objective function.

Following establishing the final base model the effect of population covariates on model parameters was evaluated. Available covariates included age, obesity status (body mass index [BMI] ≥ 30), sex, total body weight, height, BMI, BSA, adjusted body weight (AdjBW), ideal body weight (IBW), lean body weight (LBW) calculated using Boer equation, free fat mass calculated by subtracting lean body weight from total body weight, serum creatinine, CrCL, concomitant enzyme‐inducing AEDs (phenytoin, carbamazepine, oxcarbazepine, and phenobarbital), concomitant enzyme‐inhibiting AEDs (valproic acid), serum albumin, AST, ALT, and total bilirubin. CrCL was calculated using the Cockcroft–Gault equation with actual serum creatinine and actual body weight. Continuous covariates were centered using corresponding median values. Covariates were selected via forward inclusion and then backward elimination method with the required p value of <0.05 and <0.01, respectively, to remain in the model. Model selection was based on reduction of the objective function value (OFV, defined as −2 times log of the likelihood), Bayesian information criteria (BIC), inter‐subject variability (omega), goodness‐of‐fit graphs, and the residual standard error of the parameter estimates. Furthermore, in consideration of the allometric rule delineating the dependence of CL and V on BW, the model incorporating this factor was compared to the final model.

2.4. Dosing simulation

The simulation was performed using the Simulix function in MONOLIX and the estimated population parameters of the final model obtained in the pharmacokinetic modeling portion (Table 2). One thousand five hundred of identical virtual patients were assigned to receive one of the LEV regimens, and simulation was done via resampling model. The results were stratified based on the CrCL ranges for presentation purposes. CrCL ranges selected for the simulation were <30 mL/min, 30–45 mL/min, 45–60 mL/min, 60–90 mL/min, 90–120 mL/min, and >120 mL/min. These ranges were selected based on cut off numbers that are commonly used to describe different stages of renal function [24]. While a CrCL exceeding 120 falls outside the physiologically normal range, computations employing the Cockcroft‐Gault equation may yield values exceeding 120 in individuals with obesity.

TABLE 2.

Final pharmacokinetic parameters of levetiracetam.

Fixed effect parameters	Final model
	Value (% RSE)
Ka (h−1)	2.31 (51.2)
V/F (L)	29.4 (8.67)
β_ V_BSA	0.46 (65.3)
CL/F (mL/min)	2.04 (31.6)
β_CL_CRCL	0.42 (21.4)
Inter‐individual variability
omega_Ka	2.11 (27.1)
omega_V	0.47 (26.9)
omega_CL	0.38 (12.7)
Residual error
Cc	0.34 (10.4)

3. RESULTS

3.1. Demographic data

A total of 162 serum concentrations were available from 143 patients (77 male). Patients ranged from 19 to 95 years of age and the median weight was 73 kg (range, 34.4–151.3). Table 1 summarizes the demographic characteristics of patients included in this study. Seventy‐eight patients received LEV extravascularly and dosing regimens including 250 mg (n = 8), 300 mg (n = 1), 500 mg (n = 45), 750 mg (n = 26), 1000 mg (n = 51), 1250 mg (n = 1), 1500 mg (n = 10), and 2000 mg (n = 1) were administered twice daily.

TABLE 1.

Baseline characteristics of the population included in the study.

Characteristics	Median (min–max)
Number of patients, N	143
Number of serum samples, N	162
Number of samples per patient	1.1 (1–5)
Sex (male/female)	77/66
Obesity a
Yes	36
No	107
Age (years)	63 (19–95)
Weight (kg)	73 (34.4–151.3)
Body mass index (kg/m2)	26 (12.6–52.2)
Body surface area (m2)	1.81 (1.31–2.51)
Adjusted body weight (kg)	64.9 (34.4–100.2)
Ideal body weight (kg)	61.5 (42.7–89.1)
Lean body weight (kg) b	53.4 (33.4–87.8)
Extra fat mass (kg)	18.9 (0.2–63.5)
Serum creatinine (mg/dL)	0.71 (0.18–2.22)
Creatinine clearance (mL/min) c	96 (18–377)
Liver function tests
ALT (U/L)	18 (6–133)
AST (U/L)	21 (7–97)
Albumin (g/dL)	3.8 (0.69–4.2)
Total bilirubin (mg/dL)	0.42 (0.16–3.8)
LEV administration
Intravascular	64
Extravascular	79

^a Defined as BMI ≥30.

^b Calculated using Boer equation.

^c Calculated using Cockcroft–Gault equation.

3.2. Population pharmacokinetic modeling

The observed serum concentrations of LEV are displayed in Figure 1. A one‐compartment model with a first‐order absorption, no absorption lag time, and linear elimination was selected as the base model. Models with two compartments and zero‐order absorption were also tested; however, they did not improve the fit. A proportional error model was selected for describing the residual variability. No correlations between random effects were observed. The model provided a good description of the data (Figure 2), and all parameters were estimated with sufficient precision (Table 2). The bioavailability parameter, F, estimated at 0.94 in this model, was fixed to 1 given their proximity.

FIGURE 1.

Observed individual levetiracetam concentrations (mcg/mL).

FIGURE 2.

Population (A) and individual (B) predictions versus observations goodness of fit plot and visual predictive check (VPC) plot (C).

Covariates were initially analyzed using Pearson's correlation and the Wald test for significance. Covariates were included in the model if the p value was <0.05, and the model showed an improved fit compared to the best model without a given covariate. This procedure was repeated until no more covariates could be added. Then, the potential covariates were tested using backward elimination. For a covariate to be considered significant and remain in the final model, a p value of <0.01 was required. BSA was found to be a significant covariate for V/F. The exclusion of BSA as a covariate of apparent volume of distribution (V/F) increased OFV by 5.6. Modeling runs were conducted incorporating both BSA and BW as covariates to evaluate their impact on the variability of Vd (omega V). Including BSA as a covariate resulted in omega_V reduction from 59 to 47%, whereas including BW reduced it to 55%. Consequently, BSA was determined to serve as a superior predictor of Vd within our model. CrCL was a significant covariate of CL/F. The exclusion of CrCL increased the OFV by 18.16 and significantly increased the RSE % of fixed parameters.

The final model parameters were described by the equations below:

$${\mathrm{k}}_{\mathrm{a}}\left({\mathrm{h}}^{-1}\right)=2.31$$

$$\mathrm{V}/\mathrm{F}\left(\mathrm{L}\right)=29.4{\left(\frac{\mathit{BSA}}{1.8}\right)}^{0.46}$$

$$\mathrm{CL}/\mathrm{F}\left(\mathrm{L}/\mathrm{hr}\right)=2.04{\left(\frac{\mathit{\text{CrCL}}}{95.5}\right)}^{0.42}$$

Additionally, the final model, incorporating BSA and CrCL as covariates for V and CL, respectively, was compared to a model considering the allometric effect of body weight (BW) on V and CL. Scaling parameters using the allometric rule resulted in a 6% reduction in omega_V and a 2% reduction in omega_CL, compared to 12% and 4% reductions respectively in the final model.

3.3. Simulation

Simulations were performed using estimated population parameters derived from the final model. The simulation involved 1500 subjects, employing resampling technique from the population, with subsequent stratification of results based on CrCL bands. All parameters were kept constant, except for the CrCL. CrCL was selected as a covariate due to its observed effect on CL/F. The selected dosing regimens for this simulation portion were 250 mg, 500 mg, 750 mg, 1000 mg, 1500 mg, 2000 mg, and 3000 mg (Figure 3).

FIGURE 3.

Predicted levetiracetam (LEV) concentrations (mcg/mL) versus time (h) graphs of the simulated oral LEV twice daily dosing regimens of 250 mg, 500 mg, 750 mg, 1000 mg, 1500 mg, 2000 mg, and 3000 mg stratified based on creatinine clearance (CrCL). The red lines indicate therapeutic range of LEV (12–46 mcg/mL). Black line represents the median concentrations. Blue areas represent the 95% interval.

Based on the simulation results, it was observed that patients with higher CrCL necessitated higher doses of LEV to sustain concentrations within a specified range. Doses ranging from 1500 to 2000 mg twice daily allowed to maintain LEV concentrations within the therapeutic range (12–46 mcg/dL) in patients with CrCL greater than 120 mL/min. In contrast, patients with CrCL less or equal to 30 mL/min, indicating renal failure, were within the therapeutic range while on LEV 750–1000 mg twice daily. Higher doses (more than 1500 mg twice daily) placed this population into a supratherapeutic window. Doses as high as 3000 mg twice daily were inappropriate for all subgroups since estimated levels exceeded therapeutic threshold of 46 mcg/dL.

4. DISCUSSION

The model in this study was developed from the data of 143 hospitalized patients who received LEV and had at least one LEV serum concentration measured with a known time of the last administered dose. A one‐compartment model with first‐order absorption and linear elimination best described the LEV pharmacokinetic profile. A proportional error model was selected to describe the residual variability. The absorption rate constant (2.31 h⁻¹), apparent volume of distribution (29.4 L), and apparent clearance (2.04 L/h) were comparable to those reported in the literature.

In the final model, CrCL was a significant covariate of CL/F of LEV, consistent with previous studies [7, 16, 17, 20]. LEV is mainly eliminated through the kidneys with about 66% of the administered dose eliminated unchanged in the urine [1]. Therefore, the effect of CrCL on LEV CL/F was expected. Our study population included patients with both reduced and normal renal function as described by CrCL <30 and >90 mL/min, respectively [24]. CrCL was shown to be a significant covariate and the inclusion of CrCL in the final model reduced the OFV by 18.16 (p < 0.01). The exponent on centered CrCL was estimated as 0.42.

Out of all bodyweight descriptors tested (actual body weight, adjusted body weight, BMI, BSA, ideal body weight, lean body weight, and extra fat mass), only BSA showed a significant effect on V/F. Weight was not found to be a significant covariate of LEV CL/F. There are conflicting data on the effect of weight on LEV CL/F in the literature. Some previous studies reported a correlation between weight and LEV CL/F since CrCL increases with increased weight [25, 26]. Meanwhile, studies by Lima‐Rogel et al., Karatza et al., and Hernandez‐Mitre et al. did not find any correlation between weight and CL/F [19, 20, 21]. A study by Hernandez‐Mitre et al. also reported the effect of BSA on V/F [20]. This correlation can be explained by the fact that LEV does not distribute in adipose tissue [20]. Although the inclusion of BSA as a covariate for V/F in this study showed significant effects, this finding is unusual and warrants further investigation. While BSA incorporates body weight into its calculation and may better account for variability in total volume and adiposity compared to other weight descriptors, its utility as a covariate for V/F in LEV PK remains uncertain. The observed effect of BSA in this study contrasts with prior research, and further studies are needed to determine whether this represents a true correlation. Previous studies have reported conflicting results regarding the effect of body weight on V/F, with some identifying actual body weight as a significant factor, while others have not consistently demonstrated this relationship [7, 8, 15, 16, 17, 18, 19]. Although BSA may offer a better approach than actual body weight or BMI for determining the appropriate dose, given its ability to account for variability in volume and adiposity, its role as a covariate for V/F in LEV PK requires further investigation [27]. Additionally, although the RSE for BSA was relatively high, the model still provided a better fit compared to other covariates, which justifies its retention in the final model.

Other significant covariates reported in the past studies include sex, the daily dose of LEV, and concomitant use of enzyme‐inducing AEDs [7, 8, 15, 17]. In this study, the concomitant use of enzyme‐inducing (phenytoin, carbamazepine, oxcarbazepine, and phenobarbital) or enzyme‐inhibiting (valproic acid) AEDs did not affect CL/F or LEV. The mechanism by which CYP450 inducing agents influence the CL/F remains unknown since LEV metabolism is independent of the liver [1]. Furthermore, we did not find serum albumin, AST, ALT, and total bilirubin to be significant covariates on any of the model parameters.

In this study, there was also a negative correlation between age and CL/F which is reasonable since CrCL declines with age [28]. This correlation, however, was not significant, and age was not included in the final model. The effect of age may have been sufficiently represented by the inclusion of CrCL since age is a variable included in the Cockcroft–Gault equation.

There were no covariates that had an effect over k _a . The reported k _a values in the literature range from 0.616 to 4.80 h⁻¹ [7, 8, 15, 16, 17, 18, 20]. In previous studies, there were differences reported in k_a based on whether LEV was taken on an empty stomach or not, k _a  = 4.80 h⁻¹ and k _a  = 2.44 h⁻¹ respectively [15]. This finding implies that food slows down the rate of absorption; however, it does not affect the extent of absorption [1]. In this study, we were not able to confirm the feeding state of patients; therefore, we were not able to confirm the effect of food on absorption.

Based on the simulation results, LEV dose should be adjusted based on CrCL. Patients with decreased CrCL may require lower doses of LEV to maintain LEV concentrations within the therapeutic range of 12–46 mcg/dL. High doses of LEV (2000–3000 mg twice daily) placed patients into supratherapeutic range and although LEV overdose is mainly asymptomatic, changes in mental status have been documented [29, 30]. As mentioned by Karatza et al., if the higher doses of LEV are unable to provide sufficient seizure control, a different AED should be added rather than continuing to escalate the dose [21]. Moreover, a paradoxical effect of increased seizure activity has been seen with high doses of LEV in pediatric population; therefore, high LEV doses should be used with caution [31].

One of the major limitations of this study was the small number of LEV serum concentrations per patient. However, since the data were collected from medical health records and LEV serum concentrations are not routinely quantitated due to predictable PK profile, this study represents the application of PK modeling in a real‐world clinical environment [11]. PK modeling is beneficial for determining appropriate dosing based on patient characteristics and to achieve more accurate therapeutic concentrations. Similarly, data for time to seizure was unavailable since some of these patients were not monitored at the site. Time to seizure would have been an important piece of information to determine the effectiveness of LEV beyond serum concentration as breakthrough seizures may occur despite levels being within the therapeutic range.

5. CONCLUSIONS

This study demonstrates that therapeutic drug monitoring concentrations from real‐world clinical data can be leveraged to build a population pharmacokinetic model. A one‐compartment model with first‐order absorption, linear elimination, no lag time, and a proportional error model best described the LEV PK data. Further studies are needed to determine the effect of body weight descriptors on LEV clearance and ultimately outcomes.

AUTHOR CONTRIBUTIONS

Luigi Brunetti, Leonid Kagan, and Alina Chykharivska conceptualized and designed the study and contributed to data interpretation. Alina Chykharivska conducted the analyses. Luigi Brunetti and Alina Chykharivska drafted the initial manuscript. All authors critically reviewed the manuscript for intellectual content and made revisions as needed. They all approved the final version of the manuscript and agree to be accountable for all aspects of the work, ensuring that any questions related to the accuracy or integrity of any part of the study are appropriately investigated and addressed.

CONFLICT OF INTEREST STATEMENT

Luigi Brunetti has received consulting fees for unrelated work from Horizon Blue Cross Blue Shield of New Jersey and research grants from CSL Behring and Merck and Company.

ETHICS STATEMENT

All trial aspects were conducted in accordance with the ethical principles of the Good Clinical Practice Guidelines and the declaration of Helsinki. This study was granted expedited approval. The protocol was granted expedited approval by the Rutgers Biomedical and Health Sciences (PRO2021000173) and Robert Wood Johnson University Hospital Somerset (IRB21‐10) Institutional Review Boards.

Chykharivska A, Kagan L, Wagner M, Brunetti L. Real‐world interpatient variability in the pharmacokinetics of levetiracetam. Fundam Clin Pharmacol. 2025;39(2):e13059. doi: 10.1111/fcp.13059

DATA AVAILABILITY STATEMENT

The data are held securely by the authors and reasonable requests for access will be considered.