In this study, we applied pragmatic PBPK modeling and simulation to evaluate current DPF carbamazepine and VPA dosing recommendations in the treatment of epilepsy in children. By assessing predicted drug and primary metabolite concentrations against established therapeutic targets and toxicity thresholds, respectively, we confirmed that current Dutch national dosing recommendations, including step-up dosing, appear adequate for reaching appropriate levels for children of all ages. As our simulations show, due to the large interindividual variability in drug exposure between children, increasing the dose until each individual achieves a clinically desirable effect remains the preferred approach. Some fine tuning of the current dose recommendations can be advised, as discussed in more detail in the following.

First, we recommend initiating neonatal carbamazepine treatment with 10 mg/kg/day instead of 7 mg/kg/day. Other guidelines providing carbamazepine dosing recommendations for neonates are lacking, complicating the comparison of our suggested model-informed dose with other guidelines. In contrast, children aged 12–18 years may receive a higher starting dose (e.g., 400 mg/day instead of 200 mg/day). Our recommendation to increase initial carbamazepine doses in children aged 12–18 years compared with the current guidelines is logical from a pharmacokinetic viewpoint, aiming to achieve therapeutic concentrations earlier. Lexicomp suggests an initial carbamazepine dose of 200 mg twice daily in adolescents [35], which is in line with our recommendation to initiate treatment with 400 mg/day instead of 200 mg/day. Still, the standard of care is to initiate treatment with a low dose to reduce side effects. However, evidence in the literature regarding the side effects associated with higher starting doses remains poor and mostly anecdotal [36]. Drowsiness, ataxia, and vertigo are commonly observed upon therapy initiation or dose increase [37]. Yet, as the actual difference in incidences and administered dosages are not specified in the study of Pellock [37], it remains unclear whether a higher starting dose would lead to more side effects. As underlying data are ambiguous, we suggest that physicians should carefully weigh each patient’s need for rapid seizure control against the anticipated side effects and consequent compliance with therapy.

Furthermore, additional attention is warranted when higher doses are needed to achieve a beneficial effect, as the CBZE/carbamazepine ratio becomes less favorable. An inverse relationship between age and CBZE/carbamazepine ratio was observed in children aged 0–18 years, resulting in a higher CBZE/carbamazepine ratio than in adults [38, 39], which aligns with our findings. Additionally, with higher carbamazepine doses, auto-induction of CYP3A4 increases, leading to increased CBZE formation, whereas CBZE elimination remains constant, ultimately resulting in higher CBZE/carbamazepine ratios. Higher CBZE/carbamazepine ratios have been associated with a higher incidence of side effects [34, 40], but this could not be confirmed by other studies [41, 42]. CBZE undergoes epoxide hydroxylation to the inactive trans-10,11-dihydrodiol carbamazepine [43]. Little is known about the ontogeny of epoxide hydrolase. What is known is that its expression in the fetal liver is approximately 25–50% that of adult levels [44, 45], and hepatic enzymatic levels increase linearly with gestational age [46]. However, further information on its development in children is unknown, and it has been argued that the degradation of CBZE is age independent [47]. CBZE elimination was not mechanistically incorporated within the model as it is described as an oral clearance based on adult clinical data with a 30% coefficient of variance. Oral clearance follows allometric scaling (using an allometric exponent of 0.75 and a reference adult body weight of 70 kg), but whether this holds true for epoxide hydrolase activity is unknown. Despite this uncertainty, CBZE concentrations were predicted rather accurately with the pediatric model, exhibiting only a slight underprediction and a greater predicted variability than the observed data. Moreover, CBZE data were only available in children aged 4–17 years, and evaluated doses ranged from 6.75 to 10 mg/kg q12h. Therefore, a level of uncertainty in CBZE model predictions is present in children aged < 4 years and doses > 10 mg/kg q12h. Overall, we believe caution regarding elevated CBZE levels in younger children is warranted, especially following high carbamazepine doses.

For VPA, a dosing strategy of 20 or 30 mg/kg/day upon IR and ER administration seems to yield appropriate levels of total VPA, unbound VPA, and 4-ene-VPA across all simulated age groups. Interestingly, predicted VPA levels were slightly lower in the group aged 0–1 month than in the other pediatric age groups receiving the same bodyweight-based dose, indicating a higher predicted clearance when corrected for bodyweight. Clearance values observed in literature are highest within the group aged 1–2 years and decreased to adult levels at approximately 12 years of age [48, 49]. Comparing the predicted clearance across the pediatric age span with observed clearance values shows that the increase in clearance within the 1- to 2-year age span that was observed in one study (i.e., Hall et al. [50]) was not adequately captured by the model. This discrepancy was unnoticed during model verification as the included studies represented predominantly broad age ranges. One study used during the verification included children with a small age range (i.e., 0.5–1.5 years; n=7), and our predicted clearance values were slightly reduced compared with the observed clearance values, albeit still within an acceptable range (i.e., predicted/observed ratios were 0.76- to 0.86-fold) [51]. Additionally, predicted VPA concentrations accurately matched observed data. Although predictions were within acceptable ranges, the observed data indicated higher clearances or oral absorption in children aged 1–2 years, so these patients might require higher doses.

A total VPA concentration within the therapeutic window does not necessarily indicate that the unbound (reflecting active drug) concentration is within its therapeutic window. Several case reports and case series have indicated that hypoalbuminemia causes VPA toxicity despite a therapeutic total VPA level [19, 20]. Another study indicated that patients with elevated unbound VPA levels had significantly lower median serum albumin levels than those with unbound VPA levels within the therapeutic range [52]. Our simulations indicated that altering the albumin levels only changes total VPA levels and had no effect on unbound VPA levels due to compensatory alteration in clearances and fraction unbound. Yet, a VPA dose adjusted based on total VPA levels can indeed result in undesirable elevated unbound peak VPA levels, even within normal albumin ranges (i.e., −20%). Therefore, we recommend that unbound VPA concentrations are routinely monitored in patients with hypoalbuminemia and/or higher VPA doses (i.e., above 30 mg/kg/day) to monitor free VPA toxicity.

Several population pharmacokinetic models have previously been published for VPA. Three of these studies indicated that children aged 1–2 years require higher bodyweight-based VPA doses than do older children. According to Ding et al. [53], children aged 1 year should receive a VPA daily dose of approximately 20 mg/kg/day to reach a total VPA steady-state Ctrough of 50 μg/ml [53], whereas Gu et al. [54] recommended a 50 mg/kg/day dose to achieve therapeutic unbound VPA levels (almost 75% of the simulated population has a Ctrough > 4 μg/ml) [54]. Simulations from Tauzin et al. [55] suggested a dose of at least 40 mg/kg/day for patients who weigh ≤10 kg (equal to an age of < 1 year), yet half of the simulated patients remained underdosed. In contrast, our simulations suggest that most children aged 1 year reach therapeutic total and unbound VPA Ctrough levels with 30 mg/kg/day. A possible reason that we did not observe such an age-related effect with our PBPK model is that the fraction metabolized by an additional hepatic clearance input (non-mechanistic input value) is between 50% and 60% for children aged 0–5 years, potentially overshadowing the maturational effect of the included CYP and UGT ontogeny profiles in the model.

Our study has some limitations. First, the therapeutic windows used for both carbamazepine and VPA are based on clinical experience. Since the relationship between plasma level and efficacy is unpredictable for both drugs [29], no therapeutic drug monitoring is generally recommended [30, 56]. Therapeutic drug monitoring is used only in cases of unexplained side effects or lack of effect, adhering to the therapeutic windows presented here. This does not alter the fact that both lower and higher levels have been shown to be effective in practice. Nonetheless, a pharmacokinetic target is required to aim for during dose-finding simulations and, therefore, applied here as a directive.

Second, ASMs are often not administered as monotherapy and ASM drug–drug interactions are frequent. Additionally, ASMs typically have more DDIs than drugs of other therapeutic classes, posing a major challenge in treatment [57]. However, in this study, no DDI simulations were performed, and recommendations are made only for carbamazepine and VPA monotherapy. The effect of concomitant use of other ASMs can be explored with pragmatic PBPK modeling, as PBPK compound models are available for certain ASMs that include enzyme kinetics and inhibition/induction data, such as lamotrigine [58] and phenytoin (Simcyp software). A next step with PBPK modeling would be to integrate individual patient data (such as concomitant medication and disease state), allowing to obtain individual model results that can be applied in practice. However, thorough monitoring is advised to ensure the doses are safe and effective.

Third, the model was verified under both single-dose and multiple-dose (i.e., steady-state) conditions. As indicated in the Results section, the multi-dose simulations for carbamazepine performed much better than the single-dose simulations. Nevertheless, dose simulations following the current DPF recommendations were conducted to reflect the clinical setting (i.e., including both starting and maintenance doses). The overall underprediction of clearance observed in the single-dose simulations should be taken into account when interpreting the simulation results. This also resulted in our recommendation that neonates should receive a starting dose of 10 mg/kg/day (high end of current dosing range).

Fourth, many papers have been published regarding carbamazepine and VPA model verification in children (i.e., 10 and 15 studies, respectively), yet certain age ranges are still lacking. We lack adequate pharmacokinetic data in children for single-dose carbamazepine between the ages of 1 and 5 years, pharmacokinetic data for multi-dose carbamazepine for ages < 4 years, and pharmacokinetic data for VPA ER for ages < 4 years. Conducting clinical studies to fill these information gaps, if feasible, may take years to conduct. Since we rely on well-validated PBPK models based on virtual physiology that is constantly refined over time and by delineating relevant pharmacokinetic mechanisms, we provide pragmatic solutions by extrapolating modeling predictions for the age groups without observed data. Yet, in terms of model performance, despite the extrapolation, the reliability in model performance was high as our predictions are in line with current clinical practice.