endpoints, populations, and methods

Clinical Study Report (CSR): Must include detailed descriptions of all analyses and their results

ADaM Datasets: Analysis-ready datasets following CDISC standards

SDTM Datasets: Raw data organized into domains (e.g., demographics, labs, AEs)

Listing Tables and Figures (TLFs): Detailed summaries and displays of key outcomes

Metadata and define.xml files: Used to interpret datasets during electronic submissions

Reviewers often use internal software to reproduce key analyses independently.

Regulatory Guidelines Governing Statistical Review

• FDA: Follows ICH E9 (Statistical Principles), 21 CFR Part 314, and publishes detailed review memos in Drugs@FDA
• EMA: Uses CHMP guidelines for biostatistics and requires Module 2.7 summaries in the Common Technical Document (CTD)
• PMDA: Conducts in-depth statistical evaluation in parallel with clinical review and often requests mock-up tables during pre-NDA meetings
• CDSCO: Expects alignment with GCP and NDCTR 2019 and may ask for on-site validation of statistical code

All agencies emphasize traceability, transparency, and pre-specification in statistical methodology.

Common Focus Areas During Phase 3 Statistical Review

1. Population Definitions

Reviewers scrutinize how populations were defined:

• Intent-to-Treat (ITT): All randomized subjects
• Per Protocol (PP): Subjects who completed treatment as planned
• Safety Population: All subjects who received at least one dose

Inconsistencies between protocol, SAP, and CSR may trigger major queries.

2. Multiplicity Control

Multiple endpoints or comparisons can inflate the Type I error rate. Agencies require clear hierarchical testing strategies, gatekeeping procedures, or adjustments (e.g., Bonferroni, Hochberg) to control false positives.

3. Handling of Missing Data

Missing outcomes can bias results. Reviewers assess:

• Imputation methods: LOCF, MMRM, or multiple imputation
• Sensitivity analyses: To test robustness of assumptions
• Extent and reason for missingness

4. Protocol Deviations and Interim Analyses

Any deviation from the statistical plan, unblinding, or interim decision must be justified and documented. Adaptive designs are especially scrutinized for alpha control and independence of decision-making.

5. Subgroup Analyses

Subgroup effects (e.g., gender, race, age) must be interpreted with caution. Agencies want:

• Interaction tests to confirm differential treatment effects
• Pre-specified subgroup definitions
• Control for multiplicity if subgroups drive regulatory decisions

Statistical Queries and Information Requests

After initial review, agencies often send Information Requests (IRs) or Clarification Questions. These may include:

• Request to re-analyze data using different population or method
• Clarification of discrepancies between datasets and CSR
• Request for additional sensitivity or supportive analyses
• Submission of raw analysis codes for validation

Timely and well-documented responses from the sponsor’s biostatistics team are critical to avoid review delays.

Tools and Software Used by Reviewers

• JMP and SAS: For analysis reproduction and visualizations
• R and Python: Increasingly accepted for statistical validation
• Integrated Review Systems: Used by FDA to compile and cross-check submission modules
• OpenCDISC Validator: Checks CDISC compliance of SDTM and ADaM datasets

Sponsors should conduct internal pre-submission data checks using the same tools to minimize errors.

Best Practices to Prepare for Statistical Review

• Ensure consistency: Between SAP, CSR, TLFs, and datasets
• Avoid post-hoc changes: All modifications must be clearly explained
• Document assumptions: In analysis models, transformations, and imputation
• Prepare a statistical reviewer’s guide: To accompany the data package and explain derivations
• Conduct mock audits: To simulate potential reviewer questions and validate responses

Case Study: Statistical Review in a Diabetes Phase 3 Program

In a recent Type 2 Diabetes Phase 3 submission, FDA reviewers identified an issue where the SAP had not pre-specified the method of handling missing HbA1c values. Although the sponsor had used a valid MMRM approach, the lack of documentation led to a major query and delayed approval.

The sponsor subsequently submitted revised SAP documentation and sensitivity analyses, which resolved the concern. This case underscores the importance of statistical transparency and protocol adherence.

Final Thoughts

Statistical review is one of the most detailed and rigorous components of Phase 3 regulatory submissions. Agencies must be confident that your conclusions are supported by valid, reproducible, and appropriately analyzed data. By understanding reviewer expectations and adopting best practices, sponsors can ensure smoother reviews and build trust in their development program.

At ClinicalStudies.in, mastering the statistical review process prepares you for impactful roles in biostatistics, clinical data science, regulatory writing, and submission management.