Even with careful design and appropriate execution, unexpected regional differences may still be observed. If potential clinically relevant differences in treatment effects across regions are observed, a structured and holistic exploration should be utilized to investigate sources of differences. It is also important that consistency evaluation should be conducted using a descriptive framework, rather than a hypothesis testing framework.

Clinical Relevance

Under the MRCT context, the difference across regions or subgroups may be quantitively large but not clinically relevant considering limited sample size and random highs/lows. Points of consideration include target population, burden of disease, study endpoints, medical practice, sample size, and metrics of treatment effect. If any difference between regions is observed, the first step should be to evaluate whether the regional differences are clinically relevant in terms of size and clinically meaningfulness.

Disease and Treatment

The summary of epidemics, diagnostics, and treatments in the design and planning stage should be re-examined using the most recent data at the interpretation stage. In some cases, important changes such as trial results update, drug approval, or guideline updates may occur in some regions or countries during the trial. These changes can impact trial execution (withdrawal from trial/treatment, subsequent treatment usage), and subsequently confound the treatment effect.

Clinical Pharmacology

If clinically significant regional differences in drug exposure, PD and/or dose-exposure-response (PD, safety, efficacy) are identified in clinical studies, in vitro and/or in vivo mechanistic studies should be conducted to explore the impact of PK and PD-related factors, as exemplified in Sect. 2.2. Population PK and PD, M&S and other methods may also be used to analyze the possible impact of each covariate.

Biological Plausibility

Biological plausibility denotes the degree of causality and relevance of a particular effect on the treatment effect across regions or subgroups that can be predicted or expected. Such anticipation is based on clinical, pharmacological, and mechanistic considerations associated with intrinsic and extrinsic factors. They are also not directly quantifiable or measurable unless they have been accounted at the planning stage. In general, large regional differences are not expected for agents with local effects and targeted therapy for certain genetic mutations. Treatment effects of some drugs may be associated with baseline weight or BMI, baseline risk, histology, and biomarker expression levels, rather than directly with ethnicity or region. This biological plausibility can be used to investigate efficacy and safety, as well as differences between groups, or treatment and control groups.

Enrollment and Sample Size

In the consistency evaluation exploration, detailed enrollment should be investigated, including overall and regional enrollment start time, enrollment status and end time, regional sample size allocation, proportion of regions/countries of interest and sample sizes of each treatment group. Enrollment affects both exposure time and follow-up time, and the proportion and balance of sample size between groups also affect statistical uncertainty.

Baseline

Consistency of the baseline of regional population with the overall population between the treatment and the control group should always be evaluated. In addition to pre-defined predictive and prognostic factors, special attentions should be paid to variables where baseline differences and treatment differences were observed. Multivariate models may be considered to adjust for important variables to understand whether differences are caused by baseline imbalances [9]. When the regional sample size is small with many variables, multivariate models may not be applicable. In that case, population resampling method could be considered to assess whether the treatment effect with similar baseline characteristics is consistent between the region and the overall population.

Exposure, Follow-up, and Disposition

Exposure, follow-up, and disposition including reasons of trial discontinuation and treatment discontinuation should be summarized in detail. For the exposure of combined therapies, it may be considered to include the exposure of each component in the combination therapy. It is also recommended to evaluate the similarity of exposure and study follow-up in the regional and overall population. Differences need to be investigated further about whether there are impacts on the treatment effect and whether exposure or follow-up adjusted analyses should be considered for evaluation.

Internal Consistency

The strength to support internal consistency is reinforced if internal consistency is demonstrated across with different analytical methods, shown in the different study endpoints (e.g., primary, secondary, and other supportive endpoints); evidenced by generally consistent and stable subgroup results; observed with consistent trends over time. In reality, some observations may appear to indicate potential internal inconsistency but become consistent after further investigation. Details of some examples could be found in appendix 1.2.

External Consistency

External consistency can be assessed by examining consistency between similar studies, consistency with historical/external data, comprehensive analysis of efficacy, or meta-analysis, as appropriate. Before examining external consistency, it is important to examine whether the trial conditions are similar in terms of treatment regimens, study population, endpoint measures and their summary metrics, as well as intercurrent events and multi-regional context. It is important to identify the potential sources of observed external inconsistency (endpoints, subgroups, analysis methods, treatment differences or certain treatment groups) and adjust properly. Details of some examples could be found in appendix 1.3.

Statistical Uncertainty

Statistical uncertainty arises from the play of chance. When multiple regions, countries, or subgroups are included in an MRCT, the play of chance can result in observation of seemly inconsistency of treatment estimates (some random highs and lows), particularly when regional sample sizes are small or with too many regions. In such cases, pooling strategies or data-borrowing methods such as Bayesian methods [10] could be considered to reduce variability and increase estimation robustness. The treatment-by-region interaction test could be used to help assessment. In addition, the use of graphics, such as funnel plots, can also be used to display the expected estimate variations (e.g.: 95% CI) under different sample sizes [11]. This helps to better understand regional variability and facilitates further investigation.

Safety Analysis

When notable differences in safety measures, such as incidence, severity, or category of AEs, are observed between the regional population and the overall population, these differences need to be analyzed further to determine the potential cause. Potential underlying reasons could be caused by PK exposure, intrinsic and extrinsic factors (such as patient weight, baseline, regional medical practice and AE management, concomitant medication like the use of traditional Chinese medicine). If necessary, these analyses could also be combined with previous historical data in regional population. If there are differences in exposure time between regional population and the overall population, exposure-adjusted analyses can be used for safety consistency evaluation, particularly for certain events associated with exposure time.