ICH E17 Guideline attention to quality during trial planning, investigator training and trial monitoring will help achieve the consistently high trial quality required for a successful MRCT. 2.2.7 Statistical Analysis Planning ICH E9 provides general statistical principles for planning and conducting statistical anal- yses of randomised clinical trials. Aspects of analysis planning that are particularly im- portant for MRCTs are described below. The analysis strategy should be planned to en- able the qualitative and/or quantitative evaluation of benefit/risk (ICH M4E) across re- gions or important subpopulations represented in the MRCT.
Obtaining Regulatory Input on Analysis Strategy It is recommended to have early discussions with the regulatory authorities involved in the MRCT, and to obtain their agreement on the proposed analysis strategy. The standard is to specify a single primary analysis approach in the statistical section of the study pro- tocol to be agreed upon with the authorities in advance of initiating the trial. If in an MRCT, regulatory requirements for the primary analysis strategy differ due to well-jus- tified scientific or regulatory reasons, the analysis strategies planned to satisfy the differ- ent requirements should be described in the study protocol.
If, in addition, a statistical analysis plan is required as a separate document by more than one regulatory authority, a single analysis plan integrating the different regulatory requirements should be devel- oped. The analysis strategy for an MRCT should be finalized prior to unblinding of treat- ment assignments, if applicable, to ensure trial integrity.
Primary Analysis In planning an MRCT, the primary analysis strategy should carefully consider (1) the target population, (2) the endpoints/variables of primary interest, (3) the relevant intrinsic and extrinsic factors in the multi-regional, multi-subpopulation context and (4) the popu- lation-level summary of data required to describe the treatment effect. For most MRCTs, the primary analysis will correspond to a test of the hypothesis about the treatment effect and the estimation of that effect, considering data from all regions and subpopulations included in the trial. If randomisation is stratified by region, the primary analysis should adjust for regions 20