Drug Interactions

From: Janet.Wade@mpa.se Subject: Drug Interactions Date: 17 Dec 1997 07:02:31 -0500 Hello there! Whilst this may not directly deal with all the points Gary Maier raised in his e-mail with respect to drug interactions and the population approach, maybe the following short summary of what has happened in the last two years within the European regulatory environment will be useful. The inclusion of population analyses in the documentation submitted is increasing, and almost doubled in 1997 versus 1996. By far, the most frequent use made of the results of the population analyses submitted is in the area of drug interactions. The population analysis results have been used both to support drug label claims of the lack of interactions, and to quantify interactions (with dosage adjustments if needed). Indeed, the great use to which the population approach has been put to in this respect prompted the inclusion of a section addressing the use of the approach in detecting drug interactions in the new CPMP guideline for drug interactions (and, I think, which should be adopted at the CPMP meeting that is ongoing this week). I have enclosed a copy of the latest version that I have of that text, at the end of this letter. With regard to false positives, we have been applying a somewhat arbitrary rule of thumb that if an interaction is detected (on clearance for example) but its magnitude is less that the estimated intersubject variability in clearance, then that is not included in the drug label. Obviously this depends on where the interaction is on the concentration response curve, but we are often in ignorance of that so we do the best we can. This arbitrary rule of thumb has also not been universally applied, sometimes interactions that are of a smaller magnitude have been included in the drug label in the interests of maintaining European harmony. With regard to false negatives, we always like/want to see the confidence interval associated with the lack of effect, so as to have some idea of the reliability of the results. I mentioned this fact at a meeting I attended earlier this month and was told by Leon Aarons that we should also be interested in the confidence interval of the confidence interval, maybe someone other than me can address this point! Going back to false negatives, we also look at how many subjects are taking the potentially interacting drug, although what is deemed a sufficient number is currently judged on a case by case basis (sorry for that piece of regulatory rhetoric!). What follows next is the section dealing with the population approach in the new CPMP guideline on drug interactions. Janet R. Wade Swedish Medical Products Agency Population Studies It is often valuable to include a population approach in Phase II/III clinical trials to screen for pharmacokinetic drug interactions. Valuable additional information is then obtained from studies that are performed for other reasons. It is, however, important to remember that in the context of the population approach, these studies may not randomised and can therefore be subject to the usual bias of observational studies. The outcome should mainly be used as hypothesis-generating and the best use of this approach can probably be made to highlight unsuspected interactions and possibly to confirm absence of suspected interactions. The successful use of such an approach is highly dependent upon the protocol inclusion criteria so that: 1. Sufficient numbers of patients taking the potentially interacting drug exist (to avoid falsely not finding an interaction). 2. Information should be available as to when the interacting drug was taken, and should be within a reasonable time frame with respect to when the test drug was administered. In order to avoid drawing incorrect conclusions, in particular false negative conclusions, certain aspects of analysis need special attention, because the statistical models and computational procedures used to analyse population pharmacokinetic studies can be particularly complex. It is important to ensure that the particular models chosen and procedures used are reliable, and are appropriate for the statistical distribution of the data. In addition, the influence of potential confounding factors, such as age or other demographic or pathophysiological characteristics, on the results and conclusions should be checked, bearing in mind the potential lack of randomisation and the possibility for bias. A confidence interval associated with the estimate of the interaction should be presented. This is particularly important if no significant interaction is detected in order to permit an assessment of the degree of interaction potentially excluded.