Discussion

This study examined the relationship between concurrent ASM and the seizure reduction efficacy of the KD in children with drug-resistant epilepsy. Five anti-seizure medications, vigabatrin, lacosamide, lamotrigine, topiramate, and levetiracetam, had a positive association with the seizure reduction efficacy of the diet. These ASM demonstrated statistically significant odds ratios in logistic regression models, both with and without auxiliary clinical parameters such as age, etiology, seizure type, and diet type. We did not find a negative association of efficacy with any ASM used. The only non-ASM predictor with significant odds ratio was the modified Atkins diet which was associated with worse seizure reduction efficacy of KD compared to the Classic diet.

Although several ASMs appeared positively associated with KD response, the heterogeneity of their mechanisms of action argues against a biologically coherent pharmacodynamic synergy with the KD. These agents act through distinct pathways, including enhancement of γ-aminobutyric acid (GABA) signaling (vigabatrin, topiramate), modulation of voltage-gated sodium channels (lacosamide, lamotrigine), and regulation of synaptic vesicle neurotransmitter release (levetiracetam). ^{Reference Rogawski, Löscher and Rho5} Given this mechanistic diversity, the observed associations are more plausibly explained by confounding factors – such as epilepsy syndrome, age at treatment initiation, and treatment selection bias – rather than true synergistic effects. However, the ketogenic diet may act synergistically with anti-seizure medications by converging on shared molecular pathways that regulate neuronal excitability. For example, both vigabatrin and the ketogenic diet have been shown to influence the mTOR signaling pathway, potentially amplifying effects on epileptogenesis and seizure control. ^{Reference Zhang, McDaniel, Rensing and Wong6,Reference McDaniel, Rensing, Thio, Yamada and Wong7}

The anti-convulsant mechanism of the KD is not well understood. The KD causes a shift in the energy dynamics of the body towards the utilization of ketone bodies as the main source of fuel. The ketone bodies acetone, aceto-acetate, and beta-hydroxy butyrate have anti-convulsant properties in vitro. ^{Reference Simeone, Simeone, Stafstrom and Rho8} KD has been shown to modulate mitochondrial metabolism, reactive oxygen species production, enhanced synthesis and release of the inhibitory modulator adenosine, opening of K(ATP) channels, and promotion of GABAergic inhibition through GABAβ receptors. ^{Reference Murakami and Tognini9} The ketone bodies β-OHB and acetoacetate act as signaling molecules and epigenetic modifiers with action on histone acetylation, hence promoting transcriptional activation. ^{Reference Shimazu, Hirschey and Newman10} The KD has been associated with increased polyunsaturated fatty acid levels, which have been shown to cause activation of peroxisome proliferator-activated receptors (PPARs), and hyperpolarization of neurons, leading to anti-convulsant effect. ^{Reference Abdallah11} Effect of the KD on the gut microbiome has also been postulated as a potential anticonvulsant mechanisms of the KD. ^{Reference Olson, Vuong, Yano, Liang, Nusbaum and Hsiao12}

Pharmacokinetic interactions between KD and ASM may include the effect of KD on the absorption, distribution, metabolism, and excretion of ASM, affecting serum ASM concentrations. ^{Reference Tayutivutikul, Wanleenuwat, Panapongvasin, Klajing and Iwanowski13,Reference Pedersen, Kverneland and Rudi14} Decrease in serum levels of ASM has been noted in patients on ketogenic diet therapies ^{Reference Pedersen, Kverneland and Rudi14–Reference Kverneland, Taubøll, Selmer, Iversen and Nakken16} In addition, cytochrome p450 enzymes may be upregulated as a consequence of high fat intake, leading to reduced serum concentrations of ASMs metabolized by these enzymes. ^{Reference Tayutivutikul, Wanleenuwat, Panapongvasin, Klajing and Iwanowski13}

In a previous retrospective study, children receiving phenobarbital in combination with the KD were significantly less likely to have a >50% seizure reduction. ^{Reference Morrison, Pyzik, Hamdy, Hartman and Kossoff2} Conversely, those receiving zonisamide in combination with the KD at onset were more likely to have a >50% reduction. Since this study, physicians are potentially cautious about prescribing phenobarbital with the KD. We did not find a negative association of concurrent phenobarbital with KD. The rationale for reduced KD efficacy with phenobarbital is not clear. A possible mechanism is that phenobarbital is a potent inducer of hepatic cytochrome P450 enzymes. This enzyme induction may alter systemic energy metabolism by enhancing gluconeogenesis and carbohydrate utilization, thereby making it more difficult to achieve or maintain a state of therapeutic ketosis. We could not assess the effect of concurrent zonisamide, as zonisamide is not easily available in Canada. Concomitant lamotrigine was found to reduce the anti-seizure efficacy of the diet in a retrospective study of 71 children with refractory epilepsy started on KD. ^{Reference van der Louw, Desadien, Vehmeijer, van der Sijs, Catsman-Berrevoets and Neuteboom17} The likely mechanisms postulated were reduced level of ketosis when on lamotrigine, or sub-therapeutic lamotrigine levels as a result of KD-induced UGT1A6 enzymatic activity, which increases the glucuronidation of lamotrigine to inactive metabolites. ^{Reference Armeno and Kossoff18}

There are many clinical confounders, both known and unknown, which may influence seizure control on KD. Details of etiology and electroclinical syndrome characterization were not consistently available in our cohort and thus not a variable of analysis, but may contribute to outcomes on KD. For example, vigabatrin appears to have a positive effect on KD efficacy. However, vigabatrin is mostly started in children with infantile spasms, a condition which is well known to respond to the KD. ^{Reference Prezioso, Carlone, Zaccara and Verrotti19} A similar positive association may also be observed in other developmental and epileptic encephalopathies, such as Dravet syndrome and Lennox–Gastaut syndrome, in which both selected anti-seizure medications (e.g., topiramate, levetiracetam) and the ketogenic diet are known to reduce seizures. This raises the issue of targeted, non-random ASM selection in children with developmental and epileptic encephalopathies – a major cause of pediatric drug-resistant epilepsy – which may confound observed associations. ^{Reference Samanta, Bhalla and Bhatia20,Reference Xu, Lin, Perry and Nascimento21} Furthermore, regression analyses may not fully account for heterogeneity in epilepsy syndromes and underlying disease severity.

We could not study the effect of several novel ASMs such as cenobamate, cannabidiol, and fenfluaramine, as these have only recently become available for clinical use in Canada. However, our study does expand the total number of ASM that have been investigated on this topic from six to ten. Additionally, our data may include outcome measurement bias. Seizure outcomes were caregiver-reported, dietary adherence and degree of ketosis were not consistently quantified, and clinical decision-making influencing ASM selection could not be fully accounted for. Other limitations include lack of data on ketone levels, diet adherence, and ASM dosage and levels; important determinants of diet and ASM efficacy. A possible selection bias also exist as we only selected children with no ASM dose change for at least 3 months, likely selecting more stable patients, and possibly better responders.

Lastly, while regression analyses are useful in medical research, a more appropriate analysis to determine the impact of individual ASM on KD efficacy would be bivariate analysis (e.g., Fisher’s exact test) of monotherapy. A previous retrospective study on the effect of concurrent ASM on the efficacy of KD in 115 children attempted such an analysis. ^{Reference Morrison, Pyzik, Hamdy, Hartman and Kossoff2} Unfortunately, in both our study and the previous study, relatively few drug-resistant epilepsy patients were on monotherapy; performing such a statistical test on individuals using polytherapy violates the independence and mutual exclusivity assumptions of bivariate analyses, thus, we believe a regression model was the most appropriate option available. The modest sample size and number of treatment responders raise the possibility of model overfitting, particularly in the full regression model and for ASM that were near our sampling size minimum of 10 patients. Model estimates are imprecise and may reflect model instability. Although regularization techniques were applied, external validation was not possible. With this in mind, we reported the full models, which often have better reproducibility. ^{Reference Steyerberg22}