ICH E17 Guideline using appropriate statistical methods. If some regions were pooled based on intrinsic and/or extrinsic factors, or if pooled subpopulations were defined for stratification pur- poses during trial planning, then this pooling should be reflected in the analysis. Examination of Consistency across Regions and Subpopulations The statistical analysis strategy should include the evaluation of the consistency of treat- ment effects across regions and subpopulations. For this purpose, consistency in treat- ment effects is defined as a lack of clinically relevant differences between treatment ef- fects in different regions or subpopulations.
Various analytical approaches to this evalu- ation, possibly used in combination, include but are not limited to (1) descriptive sum- maries, (2) graphical displays (e.g., forest plots), (3) model-based estimation including covariate-adjusted analysis and (4) test of treatment-by-region interaction. There are strengths and limitations to any method (e.g., interaction tests often have very low power), and these should be carefully considered during analysis planning. If clinically relevant differences in treatment effects among regions are observed, a struc- tured exploration of these differences should be planned. The exploration could proceed in the following steps: 1.
Factors known a priori to vary among regions and hypothesized to be prognostic or predictive should be planned for and evaluated in the analysis model. Examples of intrinsic and extrinsic factors likely to be prognostic include disease severity, race, other subject characteristics (e.g., smoking status, body mass index), medical practice/therapeutic approach (e.g., different doses of concomitant medications used in clinical practice) or genetic factors (e.g., polymorphisms in drug metabo- lising enzymes), that are well-established for the disease or therapy and suggested from early stages of investigation. 2.
Even with careful planning, unexpected regional differences may be observed, and post-hoc analyses should be used for further investigation. Factors known to be prognostic for the disease would be examined first, because they are often found to be predictive of differential treatment effects as well. If the distribution of a prognostic factor is found to differ between regions, then apparent regional differences in treatment effects may be explained by differences in the prognostic factors. 3. Regional differences not explained by examination of known factors may require 21