Complete, Consistent, Enduring, Available

Part of Clinical Data Management and Integrity: Record Integrity and ALCOA+.

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

Why this matters

The four additional expectations are treated as an afterthought, which is a mistake, because they address failures the first five cannot detect.

The first five ask whether a record that exists is sound. These four ask whether the right records exist at all, whether they agree with each other, whether they will survive, and whether anyone can get to them. A record can be attributable, legible, contemporaneous, original and accurate, and the dataset containing it can still be unusable.

Complete

Everything is there, including the parts that are inconvenient.

Completeness means the record includes repeated measurements, failed attempts, the reading that was discarded, and the visit that did not happen. A dataset containing only successful assessments is not a clean dataset; it is a biased one.

The specific failure to watch for is the silent exclusion. A device that permits three attempts and transmits only the accepted one. A form with no way to record that an assessment was attempted and abandoned. A process where a sample that fails quality control is simply re-drawn with no record of the first.

None of these looks like data loss. Each produces a record of the successes with no trace of the rest, and the absence is invisible precisely because nothing marks it.

Consistent

The records agree with each other, and where they do not, the disagreement is itself recorded.

Inconsistency between two records of the same fact is information. A laboratory value differing between the vendor file and the clinical database means something happened, and whatever it was is worth knowing.

The failure is resolving inconsistency by making it disappear. Changing one value to match the other produces a consistent dataset that no longer records that the two ever disagreed, and destroys the evidence of whatever process caused it.

Consistency also means internal coherence: a visit date before enrollment, an end date before a start, a death date followed by a visit. Edit checks catch most of these, which is exactly what edit checks are for.

Enduring

The record survives for as long as it must, in a form that remains readable.

Retention periods are long, frequently far longer than the life of the software that created the record. A database in a proprietary format whose vendor no longer exists is a preservation problem that nobody notices until the record is needed.

Enduring also covers the mundane: paper that fades, thermal printouts that go blank within a few years, and the electronic record whose only copy is on a laptop.

Thermal paper is worth naming. It is still used by clinical devices, it produces a legible record on the day, and it can be blank within two years. A site that files the original thermal strip has complied with the originality expectation and will have nothing at all when an inspector asks.

Available

Someone who is entitled to see it can actually retrieve it, within a reasonable time.

A record that exists and cannot be produced is functionally absent during an inspection. An inspector asking for the audit trail of a specific field does not accept that it exists somewhere in an archive that takes three weeks to restore.

Availability also covers access control from the other direction: restricting access is right, and a restriction so tight that nobody currently employed can retrieve the record is a failure.

In practice

A data manager reviews a study's electronic outcome data, where participants complete a questionnaire on a handheld device.

The dataset has one response per participant per day, and it is clean and complete by every check.

She asks the vendor what happens when a participant opens the questionnaire and does not finish. The answer is that the partial response is discarded and the participant may start again.

Completeness fails. The dataset shows completed questionnaires and no trace of abandonment, and abandonment is meaningful: a participant too unwell to finish is a signal the dataset has quietly removed.

The fix is a configuration change and a variable, and it is possible now. In eight months, at lock, the analysis would have had a clean dataset with no way to know how often the instrument was attempted and not completed.

What this looks like when it goes wrong

A study retains its electronic data in the clinical database and its audit trail in the system's own archive. Both are backed up and both are retained for the required period.

Four years after the study closes, the database vendor is acquired and the platform is retired. Migration converts the current data cleanly into the new system.

The audit trails do not migrate. They are retained, as exported files, in the retired system's own format, which the new system cannot read and for which no reader remains licensed.

Every value is preserved. The record of how each value came to have its current content is not, which means the dataset can no longer show whether any value was ever changed. Endurance failed at a migration nobody thought of as a data integrity event.

Key takeaways