What Lock Actually Commits You To

Part of Clinical Data Management and Integrity: Database Lock.

About 5 minutes. Free, in your language, no account needed.

Why this matters

Database lock is the moment the study's data stops being editable and becomes the thing the analysis runs on. In a blinded study it is followed by unblinding, and unblinding is irreversible in the way that matters: after it, nobody on the study can make a decision that is demonstrably independent of knowing the result.

Lock is therefore not an administrative milestone. It is a declaration that the data is as good as it is going to get, made at the last moment when that declaration can be made without anybody knowing the answer.

What lock is asserting

That the data is complete. Every expected record is present or its absence is documented and explained.

That known discrepancies are resolved or documented. Not that none exist, which is never true, but that every one has been examined and disposed of on a stated basis.

That external data has been reconciled. All transfers received, all reconciliations run, all differences addressed.

That coding is complete and reviewed.

That derivations produce what the analysis plan specifies, demonstrated rather than assumed.

That the decisions defining the analysis have been made while still blind: population membership, protocol deviations affecting analysis, and any exclusions.

That last one is the substantive point of the sequence. Deciding which participants belong in which analysis population is a judgment, and a judgment made after unblinding is not credible however honest it is, because it cannot be shown to be independent of the effect it has on the result.

The order matters

The sequence is deliberate and gets compressed under pressure, which is where it fails.

First, the data is cleaned and reconciled. Then the blind data review meeting takes place, at which the team examines the data without knowing allocation and makes the classification decisions: deviations, population membership, outliers, and any handling rules the plan left open.

Then those decisions are documented and approved. Then the database is locked. Then, and only then, the allocation is revealed.

Compressing this means making classification decisions after lock or, worse, after unblinding. A team that unblinds first and then decides which participants to exclude has made a decision that cannot be defended, whatever its merits.

What must be in place before lock

A checklist, documented and signed, with an owner per line. It typically covers query status by category, external data reconciliation, safety reconciliation, coding completion, derivation quality control, audit trail review, protocol deviation classification, population assignment, and the blind data review's documented outcome.

The value of the checklist is not the ticking. It is that each line has a name against it, so that "we thought that was done" cannot happen, which is the commonest lock failure.

In practice

A study approaches lock with the blind data review scheduled for the same week and the lock date fixed by an external commitment.

The data manager identifies a problem with the sequence rather than with any single item. The review is scheduled after the final data cleaning completes, which leaves two days to document and approve the classification decisions before lock.

Two days is not enough for the sponsor's own approval process, so either the decisions will be approved after lock or the lock will slip.

She raises it four weeks in advance, which is the entire contribution. Moving the review forward by ten days costs nothing at that distance and is impossible to arrange in the final week.

The alternative, which she has seen, is a lock that happens on time with the classification decisions approved retrospectively, and a set of approvals dated after the database was locked which nobody can explain at an inspection.

What this looks like when it goes wrong

A study locks on schedule. Three days later, during analysis, a programmer finds that eleven participants have a baseline value taken after first dose, because the visit was recorded as a screening visit but occurred on the dosing day.

The correct handling is a defined rule for which assessment is baseline in that circumstance. The analysis plan does not state one.

The team must now decide. They are unblinded, because unblinding followed lock by one day.

Whatever they decide, and they decide carefully and honestly, the decision was made by people who can see how each option moves the primary endpoint. There is no way to demonstrate otherwise, and the study report must disclose that the rule was established after unblinding.

The rule was a sentence. It could have been written at any point in the previous three years, and the blind data review is where it should have been settled.

Key takeaways