Chapter 10 Pre-Database Lock Activities
The Statistician’s Final Accountability Before Results Are Frozen
The period immediately preceding database lock represents the final opportunity for statistical judgment to materially influence the integrity, interpretability, and regulatory defensibility of trial results.
At this stage, no new analyses are being invented.
Instead, the statistician’s responsibility is to confirm alignment—between plans and reality, definitions and data, and analytical intent and execution.
Once the database is locked, unresolved issues become explanation problems rather than fixable problems.
10.1 Why the Pre-Lock Phase Is Statistically Critical
The pre-lock phase is often misinterpreted as an administrative checkpoint.
From a statistical perspective, it is a decision boundary.
Before database lock: - SAP assumptions are confronted with real data - Analysis datasets transition from conceptual definitions to fixed objects - TFL shells become binding deliverables
After database lock: - Analysis definitions are frozen - Population flags are no longer negotiable - Deviations must be justified rather than corrected
For the biostatistician, this is the last point at which technical authority can prevent downstream analytical risk.
10.2 Final Review of TFL Shells
10.2.1 Why TFL Shells Must Be Finalized Before Lock
TFL shells are not formatting artifacts.
They formally define:
- What endpoints will be summarized
- Which populations will be analyzed
- How data will be stratified or grouped
- What comparisons will be reported
If a shell cannot be populated cleanly using existing analysis datasets, the issue lies upstream—in definitions, not presentation.
10.2.2 Statistical Verification of TFL Shells
During final TFL shell review, statisticians must confirm that: - Titles, footnotes, and population labels align with the SAP - Endpoint definitions match available ADaM variables - Planned stratifications and subgroup analyses are derivable - No post-hoc endpoints or populations are implicitly introduced
A practical rule: > If a shell requires interpretive explanation to understand how it will be filled, it is not ready for database lock.
10.3 Confirming Consistency Between SAP and Actual Data
10.3.1 The Required Reality Check
The SAP defines an idealized analytical framework.
The collected data reflect operational reality.
Before database lock, statisticians must explicitly assess: - Existence of all SAP-defined variables - Adequacy of data granularity for planned analyses - Continued defensibility of SAP assumptions
This assessment focuses on material consistency, not perfection.
10.3.2 Common Sources of Misalignment
Frequent discrepancies include: - Planned visits that rarely occurred - Endpoints with higher-than-anticipated missingness - Covariates inconsistently collected across subjects - Time windows misaligned with actual visit timing
When misalignment is identified, statisticians must determine whether: - The SAP can be applied as written - A justified SAP amendment is required before lock - The issue can be addressed through pre-specified sensitivity analyses
Silence at this stage constitutes implicit endorsement.
10.4 Confirmation of Analysis Dataset Definitions
10.4.1 Transition to Lock-Ready Analysis Datasets
By the pre-lock phase, analysis datasets (typically ADaM) must be: - Fully specified - Deterministically derivable - Consistent with the SAP
This includes finalization of: - Dataset structures - Variable derivations - Population flags - Parameter-level metadata
10.4.2 Statistical Confirmation Responsibilities
Statisticians should explicitly verify that: - Each analysis dataset aligns with SAP specifications - Key variables are uniquely and reproducibly derived - Population flags are deterministic and traceable to source data - No analysis-critical derivation depends on post-lock decisions
A guiding principle: > If a derivation still depends on judgment calls, it is not ready for database lock.
10.5 Statistical Sign-Off Prior to Database Lock
10.5.1 Meaning of Statistical Sign-Off
Statistical sign-off is not a procedural formality.
It represents a professional assertion:
Based on statistical review, the data and definitions are sufficient to support the planned analyses as specified.
It does not imply: - Data perfection - Guaranteed favorable results - Absence of sensitivity analyses
It confirms that analyses will be: - Reproducible - Interpretable - Defensible to regulators and auditors
10.5.2 Scope of Statistical Accountability
By providing sign-off, the statistician confirms that: - TFL shells are executable and SAP-consistent - SAP assumptions remain materially valid - Analysis dataset definitions are finalized - Outstanding issues are documented and accepted
Concerns not documented before lock will be assumed not to exist.
10.6 Common Pitfalls in the Pre-Lock Phase
Experienced statisticians remain alert to: - Deferring issues with “we will explain it in the CSR” - Last-minute shell changes without SAP updates - Ambiguous population definitions left unresolved - Pressure to sign off without sufficient review time
A critical lesson: > Schedule pressure does not diminish statistical responsibility.
10.7 Chapter Summary: Database Lock as a Line of No Return
Database lock marks a definitive boundary.
Before it, statisticians can prevent analytical failure.
After it, they can only manage interpretive risk.
The statistician’s value at this stage lies in: - Enforcing clarity before commitment - Challenging misalignment before irreversibility - Protecting analytical credibility under time pressure
Key takeaway:
Database lock freezes not only the data, but every statistical decision left unchallenged.