F. Mercier, L. Claret, R. Bruno
Pharsight Consulting Services, a Certara™ Company, St. Louis, MO, USA
Objectives: In recent years, the adjusted indirect comparisons (ITC) method has been widely used to compare competing treatments in the absence of direct evidence about their relative performance. For instance, if two treatments A and B are compared against a common comparator O via two distinct sets of randomized trials, ITC can be used to derive an indirect estimate of the relative effect of A versus B. Borrowing data from literature, we discuss two ITC methods to estimate the odds-ratio (OR) of side-effects in patients treated for pain.
Methods:Taking the example of a binary response variable, two approaches can be considered to obtain indirect estimates: (1) model in the response ‘domain’ (e.g. proportion of events) in each treatment group and derive the effect size (OR) between A and B; (2) derive the effect size (A vs. O and B vs. O) from arm-level data and model it to extract the effect size of interest (A vs. B). Pharmacometricians are referring to the first approach as ‘model-based meta-analysis’ while statisticians and/or epidemiologists use the second simply called ‘meta-analysis’. We apply and compare these two approaches in the analysis of AEs frequency (constipation, nausea, dizziness, vomiting, somnolence), as well as frequency of drop-out for AE in 44 randomized controlled trials (representing approximately 14000 patients) evaluating various treatments for pain management.
Results: The two approaches give consistent results. Taking constipation as an example, the comparison of A vs. B gives an OR (95%CI) of 1.53 [1.03, 2.23] using the response ‘domain’ model and 1.54 [1.04, 2.27] using the effect size ‘domain’ model. Frequencies of constipation and vomiting frequency appear to be higher with A compared to B. The meta-analysis does not reveal any other significant difference in AE or drop-out for AE frequencies between A and B.
Conclusions: Response or effect-size model-based indirect treatment comparisons lead to similar results. While the response-based models offer more flexibility, effect-size based model are easier to implement and to interpret. Both approaches can be considered as complementary.
Reference: PAGE 21 () Abstr 2493 [www.page-meeting.org/?abstract=2493]
Poster: Estimation methods