Pharmacology Journal

Carbamazepine (CBZ) is considered as the primary drug for the treatment of partial and tonic-clonic seizures. It is also the drug of choice for trigeminal neuralgia and frequently used for treating bipolar depression. The originator patent expired long ago (in 1986) and there are now several generic alternatives. However, after the introduction of the generic competitors concerns have arisen about the safety and therapeutic equivalence of generic CBZ tablets.

Several cases have been published about the occurrence of adverse effects after switching from one CBZ product to another. Crawford et al. have reported that approximately 10% of patients complained especially about central nervous system (CNS)-related adverse effects when they were switched to one of the generic products.

Neuvonen has demonstrated that the occurrence of CNS-related adverse effects is independent of the extent of absorption, but apparently related to the rate of absorption. There were patients who tolerated high concentrations when the absorption rate was low but experienced adverse effects at low but rapidly rising concentrations. In steady-state observational studies, the fluctuation of the CBZ concentration appeared to correlate with CNS-related adverse effects. To account for these observations, Riva has proposed a threshold toxicity model for characterizing the dependence of adverse effects on CBZ concentrations. Other clinical studies have also consistently found that adverse CNS effects depend on absorption rates, but have provided no explanation.

The reports of clinical problems with approved generic products also raised concerns of the drug regulatory agencies. In a Food and Drug Administration (FDA)-sponsored study, Meyer et al. explored the relationship between the dissolution rate and concentration profiles of CBZ tablets in subjects with known clinical failures. The relationship between CBZ concentration profiles and clinical effects has been investigated in healthy volunteers and patients. The study of Olling et al. was a four-arm bioequivalence (BE) trial. They compared three generic formulations with the brand-name product. Their goal was to determine which absorption rate-dependent BE metric had the best discriminatory power, including the prediction of adverse effects. Four BE metrics were compared: maximum concentration (C_max), time of the peak concentration (T_max), C_max/AUC (where AUC is the area under the concentration–time curve), and the area until T_max [partial AUC (AUCP)]. The study was inconclusive. Although there was a tendency to have more adverse effects with the faster absorbing preparations, and the four formulations were clearly bioinequivalent, statistically significant differences of adverse effects among formulations could not be demonstrated. No statistically significant, clear relationship was found between BE metrics characterizing absorption rates and the frequency of the adverse effects.

In the present study a pharmacokinetic–pharmacodynamic (PK–PD) model is proposed to characterize acute CBZ toxicity based on the data of Olling et al.. By building a PK–PD model, an attempt is made to answer the original question posed by Olling et al. (‘which metric is the most sensitive indicator of toxicity?’) and also to gain insight into the occurrence and mechanism of the absorption rate-dependent pharmacological response. In addition, the relationships between the PK metrics for BE and the PD metrics of toxicity will be explored, together with their clinical consequences.

Issues in statistical methodology

Olling et al. tried to find possible correlations between the occurrence of adverse effects and PK characteristics, especially those reflecting the rate of absorption (C_max, AUCP). However, although adverse effects were pooled together, Olling et al. could not achieve the goal stated above, and were unable demonstrate significant relationships between adverse effects and PK metrics. We have also failed, using conventional statistical techniques such as logistic regression. However, a combination of modern exploratory data analysis such as GAM with mixed-effect PK–PD modelling has shown a highly significant relationship using the results of a relatively small, conventional size BE trial. The advantages of model-based drug development are well recognized by the industry and model-based assessment of new drug applications is now an integral part of regulatory work. Even though rigorous comparison of model-based and classical statistical analysis is lacking, our results show that a model-dependent approach may be more powerful than model-independent statistical analysis.

Issues in clinical pharmacology

Dodrill carefully reviewed in 1991 the behavioural effects of antiepileptic drugs and concluded that ‘the area of behavioural effects of antiepileptic drugs is poorly defined, lacks recognized and validated methods of assessment, and has suffered from a number of methodological limitations, especially including the use of experimental designs which have led to the contamination of drug effects and subject effects’. Although there have been significant improvements since 1991 in the methodology of clinical trials, and our understanding of the molecular mode of action of CBZ has significantly improved, fundamental PK and PD data of such a widely used and well-known drug as CBZ are still lacking. For example, Benet could find no data concerning the within-subject variation of CBZ or digoxin.

Our literature search also yielded very little quantitative information about the PK–PD properties of CBZ. It is well known that slow-release CBZ formulations have fewer neurological side-effects, but this has been attributed to such a vaguely defined pharmacological term as ‘increased fluctuation’. The demonstration of the development of fast tolerance has been unsuccessful. We have succeeded in showing the relationship between release rates and the occurrence of adverse effects by re-evaluating the data of a carefully planned BE trial with contemporary, model-driven statistical analysis. The strong, direct dependence of the adverse effects on the absorption rate can be explained by an examination. Fast absorption is unfavourable not only because it increases the concentration in the absorption phase, but also because these elevated concentrations occur earlier when the sensitivity is higher. These two factors multiply the effects of each other, and the net outcome explains the results.

The established tolerance model has direct clinical relevance. CBZ is one of the few drugs where therapeutic monitoring is strongly advised. However, its therapeutic range is not clearly defined, although the tolerance model predicts that, at least in regard to acute neurological adverse effects, such a range does not exist. The concentration data could be interpreted only by noting that subjects may complain about toxic effects at lower concentrations during the absorption phase, when PD sensitivity to the adverse effects is high, in comparison with the elimination phase, when sensitivity is low.

Drug regulatory issues

There is ongoing discussion of whether different BE regulations are needed for narrow therapeutic index drugs and, if so, which drugs belong to this category. One of the questions is whether CBZ should be considered as a narrow therapeutic index drug. Health Canada, in its recently issued guideline, has set more stringent BE requirements for narrow therapeutic index drugs. For example, the 90% CI for the AUC ratio in such cases should be within the 0.90–1.12 interval. However, CBZ is not listed among the narrow therapeutic index drugs, contrary to an earlier guideline. The European Guideline on BE also recommends tightening of the regulatory limits in the case of certain narrow therapeutic index drugs, but specific examples are not given. The question is decided by the competent national authorities. For example, the Danish Medicines Agency asserts that if the usual acceptance limits (0.80–1.25) are applied to CBZ, then therapeutic problems could occur with substitutions of the preparations.Therefore, a generic CBZ tablet should not be substituted with the original unless the relevant confidence limits for AUC and C_max are between the limits of 0.90 and 1.10.

casts doubt on whether these approaches are scientifically sound and viable, showing there could be a difference in relative risk of 50% between generic products even though their C_max ratio is still very much between the regulatory limits. A similar conclusion could be reached after considering the CIs for the metrics. Based, it is suggested that, at least in the case of CBZ, instead of tightening the regulatory criterion, a more reasonable approach would be to use a PK metric which is more sensitive to possible adverse effects.

The recently introduced concept of early exposure and its suggested measure (AUCP) seem to fit this purpose. AUCP is a more sensitive indicator of rate-dependent toxicity, particularly in the case of CBZ when development of the observed pharmacological response is much faster than the absorption process itself. The FDA observes in its current regulatory Guidance that, for some drugs, comparisons of early exposure could be important, and recommends that the ratio of partial AUCs is a relevant metric in these cases. The present study of the PK–PD properties of CBZ supports the regulatory statements of the FDA.

The suggestion that partial AUC could serve as an index of early exposure in BE studies would be useful only if the corresponding regulatory limits were established. The present study intends to encourage the development of a relevant CI for the ratio of partial AUCs.

demonstrates that the therapeutic risk of switching (the Rmax ratio) is strongly underestimated by ratios of C_max and, to a lesser extent, of partial AUC. Therefore the use of AUCP, while not ideal, is much preferable to indicate rate-related toxicity. It is acceptable to the FDA even if it has not been widely applied. Thus, our results suggest that more consideration should be give to using AUCP for the evaluation of BE of narrow therapeutic index drugs.

Generally, it is suggested that more consideration should be given to using AUCP for evaluating the bioequivalence of narrow therapeutic index drugs. For example, experience in renal transplantation has shown that ciclosporin exposure during the absorption phase (AUC_0−4) is the critical factor for optimizing immunosuppression. Thus, a partial AUC was indeed found to be a clinically relevant and useful PK metric.

This study has shown quantitatively the dependence of neurological adverse effects on the absorption rate of CBZ in demonstrations of the development of tolerance, and has re-evaluated the BE study of Olling et al. by the exploratory data analytical technique of generalized additive modelling (GAM) followed by mixed-effect population PD–PK analysis. The development of tolerance was fast and had a characteristic half-life of 2.33 h, with an initial EC50 of 2.29 mg l^−1.

The study has also explored the concepts of PD and PD–PK sensitivities of metrics using an actual clinical dataset and has demonstrated that C_max has both poor PK and PD–PK sensitivities, whereas partial AUC exhibits better properties in this regard. Therefore, partial AUC should be applied in regulatory decisions concerning the BE of CBZ formulations.

The converse of the above conclusions is that two CBZ products can yield clinically important differences of acute toxicity even if they are declared to be bioequivalent on the basis of sufficiently similar PK metrics.

Filed under: Drug Safety