- Published
- Updated
- Sources checked
Data cleaning within an EDC does not establish agreement with laboratory, safety or randomisation records. Reconciliation needs a defined population, comparable keys and an owner who can explain unmatched entries.
Define the comparison population
Create an interface register describing each transfer, subject identifiers, time zones, units and expected frequency. Separate a missing record from a legitimate timing difference. Agree which team may correct the originating system; avoid repairing an inconsistency only in a downstream spreadsheet.
Keep the mismatch trail
In safety reconciliation, compare event identifiers and agreed attributes using the study procedure. A mismatch log should show both values, extraction versions, decision and evidence of correction or accepted explanation. Keep clinical interpretation with appropriately qualified reviewers.
Reconciliation checks
- Check subject and visit key uniqueness before matching.
- Record expected files and investigate missing deliveries.
- Rerun affected comparisons after approved corrections.
Accept the explained population
Accept a reconciliation cycle when the compared population is accounted for and material discrepancies have a recorded disposition. Preserve extracts, matching rules and exception decisions so the comparison can be understood later without recreating private working notes. Specify whether each repeat delivery replaces the previous file or contains changed rows only; otherwise loading can create duplicate events or lose corrections.
Frequently asked questions
- Does reconciliation mean all systems use identical text?
- No. Compare agreed attributes while respecting the different purposes and terminology of each system.
- Can unmatched rows be dropped?
- Only through an authorised, justified disposition that preserves the reconciliation trail.