Combining studies can increase the available evidence, but it also combines differences in populations, treatment exposure, collection methods, endpoint definitions, and follow-up. A larger dataset does not automatically represent one coherent analysis question.
The FDA’s integrated summary of effectiveness guidance provides submission context. The approach below is a practical recommendation for assessing compatibility before constructing an integrated analysis package.
1. Build a study compatibility map
Compare the concepts that determine interpretation: eligibility, treatment conditions, endpoints, assessment schedules, populations, coding conventions, and data maturity. Record where definitions are aligned, reconcilable, or materially different. An identical variable name is weak evidence of equivalence.
For a hypothetical safety pool, two studies may use different treatment-emergent windows. Decide whether a harmonized rule is scientifically and operationally justified and whether the necessary source information exists. Preserve the original study-specific definitions alongside the integrated decision.
2. Define the pooled question
Specify which studies and participants contribute to each analysis and why. Consider differences in design and comparators before presenting a pooled treatment contrast. Descriptive aggregation and inferential pooling are different activities and should be labeled accordingly.
ICH E9(R1) is a useful reference for the treatment-effect question. Our recommendation is to write an integrated analysis intent record before programming, including limitations introduced by study heterogeneity.
3. Preserve study provenance
Retain study identifiers and the original analysis context. A reviewer should be able to inspect both the integrated value and its study-specific source or derivation. Harmonization should not make discrepancies disappear without explanation.
CDISC ADaM informs the analysis-data layer. Use a controlled harmonization specification that identifies the source concept, integrated representation, transformation, and known limitation. Test one participant-level path from each contributing study.
4. Reconcile the delivered results
Check study-specific and integrated denominators, exposure totals, categories, and exclusions. Differences may be justified, but they should be explained. Examine whether coding updates or data refreshes affect some studies differently from others.
Keep a release manifest identifying the contributing study versions and integrated programs. A late update to one study should trigger a targeted impact assessment rather than an undocumented refresh of the entire pool.
The strongest integrated package makes compatibility and limitations visible. It allows reviewers to understand what was combined, what was transformed, and which conclusions depend on assumptions about comparability. Integration should increase access to evidence without obscuring the differences that shape its meaning.
