This is a ready-to-use checklist to protect prespecification, the firewall that stops data-driven analysis choices from inflating false-positive rates. It confirms the statistical analysis plan (SAP) was finalized and version-controlled before unblinding, that the estimand and analysis rules are unambiguous, and that database lock and unblinding are controlled, evidenced events. Replace every <<FILL: ...>> placeholder, mark each item Pass / Fail / NA, and sign it. A filled specimen follows. This is educational reference content, not regulatory advice.
Why this checklist earns its place
Prespecification is an inspectable fact: the SAP version history and its timing relative to unblinding are exactly what an inspector examines. A SAP finalized after database lock or after unblinding, even if benign, removes the firewall and invites the question of whether choices were data-driven. See ICH E8(R1) and E9 trial design and statistics.
Header
| Field | Entry |
|---|---|
| Checklist number | <<FILL: CL-ID>> |
| Protocol / SAP reference | <<FILL>> |
| Milestone | Database lock readiness / Unblinding readiness (<<FILL>>) |
| Date | <<FILL>> |
Section A: SAP prespecification
| # | Item | Pass / Fail / NA | Evidence |
|---|---|---|---|
| A1 | SAP finalized and approved before unblinding | <<FILL: SAP version, approval date>> | |
| A2 | SAP under formal version control; version history retained | ||
| A3 | Primary estimand fully specified (five attributes) and in the SAP | <<FILL: link to estimand spec>> | |
| A4 | ICE handling strategy defined per anticipated event | ||
| A5 | Analysis populations (FAS, PPS, Safety) defined with unambiguous classification rules | ||
| A6 | Primary estimator and missing-data method aligned to the ICE strategy | ||
| A7 | Multiplicity strategy prespecified (hierarchy / gatekeeping / alpha allocation) | ||
| A8 | Sensitivity analyses for the primary estimand prespecified |
Section B: Unblinding and lock controls
| # | Item | Pass / Fail / NA | Evidence |
|---|---|---|---|
| B1 | Unblinding timestamp is a controlled, evidenced event | ||
| B2 | No SAP change after unblinding without a flagged, justified, documented exception | ||
| B3 | Patient classification into analysis populations finalized before unblinding, per prespecified rule (no post-unblinding reclassification) | ||
| B4 | DMC access to unblinded data governed by a charter that does not compromise the prespecified analysis | ||
| B5 | Database lock criteria met (queries resolved, reconciliations complete, CtQ data complete) | ||
| B6 | Audit trail on SAP and EDC supports the prespecification timeline |
Section C: Data readiness
| # | Item | Pass / Fail / NA | Evidence |
|---|---|---|---|
| C1 | Estimand-critical (CtQ) data complete to the prespecified tolerance | <<FILL: missing rate vs limit>> | |
| C2 | Post-ICE data collected where the estimand requires it (casebook did not close on discontinuation) | ||
| C3 | Key reconciliations complete (SAE, external lab, IRT) |
Sign-off
| Role | Name | Signature | Date |
|---|---|---|---|
| Biostatistician | <<FILL>> | ||
| Data management | <<FILL>> | ||
| Clinical QA | <<FILL>> |
Filled specimen
Example at database-lock readiness for a phase 3 trial.
| # | Item | Result | Evidence |
|---|---|---|---|
| A1 | SAP finalized before unblinding | Pass | SAP v3.0 approved 12-May; unblinding scheduled post-lock |
| A3 | Estimand fully specified | Pass | Primary estimand, five attributes, SAP section 4 |
| A6 | Estimator aligned to ICE strategy | Pass | MMRM, MAR consistent with treatment policy |
| A7 | Multiplicity prespecified | Pass | Fixed hierarchy, SAP section 7 |
| B3 | No post-unblinding reclassification | Pass | Population classification locked in a dated pre-unblinding memo |
| C1 | CtQ data complete to tolerance | Pass | Week-24 primary endpoint missing rate 5.1%, below the 8% limit |
| C2 | Post-ICE data collected | Pass | 96% of discontinued patients have a week-24 measurement |
The two rows that carry the most weight in an inspection are A1 and C2: the SAP was locked before unblinding (the firewall held), and the post-discontinuation data the treatment-policy estimand demands were actually collected (the estimand was not quietly broken). Both are objective, dated facts.
Common inspection findings this checklist prevents
- SAP finalized after database lock or after unblinding, removing the prespecification firewall.
- Patients reclassified between analysis populations after unblinding, with no prespecified rule.
- A missing-data method chosen independently of the ICE strategy.
- Post-ICE data not collected, so the treatment-policy estimand cannot be estimated as defined.
- No evidence trail linking the SAP version to the unblinding timestamp.
How to adapt this checklist
- Run Section A before you approach unblinding; A1 and A2 are the firewall.
- Treat the unblinding timestamp as a controlled event and evidence it (B1).
- Check estimand-critical data completeness against the numeric tolerance from your CtQ register (C1).
- Keep the signed checklist with the SAP version history and the database-lock package.