What reconciliation means in clinical data management
Modern trials collect data in many systems: the EDC for case report forms, a separate safety database for serious adverse events, central laboratories, ECG and imaging vendors, ePRO devices, and randomization systems. Reconciliation is the repeated comparison of overlapping data across those systems to make sure they describe the same subjects, visits and events in the same way.
It is not a one-time cleanup. Reconciliation runs throughout the study and must be completed before database lock, because the analysis, the clinical study report and regulatory safety reporting all depend on the sources agreeing.
SAE reconciliation
Why it is needed
Serious adverse events are usually reported twice: on the CRF in the clinical database, and through expedited reporting into a separate safety database. The SCDM Good Clinical Data Management Practices chapter on SAE data reconciliation (republished in the Journal of the Society for Clinical Data Management, December 2023) defines it as comparing key safety variables between the two databases to confirm the events match, repeated during the study.
What to compare
The GCDMP lists, among others:
| Category | Fields |
|---|---|
| Identification | Protocol, investigator, subject and randomization number |
| Demographics | Initials, date of birth, sex, race |
| Event | Event number, diagnosis, verbatim term, coded/preferred term |
| Dates | Onset, resolution, date of death |
| Assessment | Outcome, severity, causality, action taken with study drug |
How to run it
- Reconcile several times during the study, aligned with data receipt, safety updates and interim or final reports.
- Set a cutoff after which no new SAEs are added to the clinical database.
- Confirm data entry, validation and dictionary coding are complete before each cycle.
- Program the comparison where feasible, validate it once and rerun it.
- Agree turnaround times among data management, drug safety and clinical operations.
- Before database lock, return and integrate all queries, finish reconciliation for every event, and issue written notice of completion.
The GCDMP accepts some differences, such as minor terminology variations or start dates that legitimately shift when an event becomes serious, as long as they are understood and documented. Unresolved inconsistencies go into a data handling report.
Lab reconciliation
When samples go to a central laboratory, results arrive as an external data transfer rather than through the CRF. Lab reconciliation checks that every sample the site says it collected matches a result from the lab, and vice versa. Typical checks include subject and visit identifiers, sample or accession numbers, collection dates and times, and unscheduled or repeat samples. Missing results, orphan results (no matching visit) and date mismatches are the usual findings.
The SCDM External Data Transfers chapter treats laboratory and PK/PD data as core external data types and recommends primary keys for sponsor, study, site, subject, visit and sample, with secondary keys such as sex, date of birth, visit date and sequence number.
External and vendor data reconciliation
The same principles apply to ECG, imaging, vital-sign devices, electronic diaries and other vendor data. The GCDMP chapter’s minimum standards include:
- Written transfer specifications agreed with each vendor: file formats, field types, date formats, handling of ”<” and ”>” values, missing data and critical variables.
- Test transfers during database setup to validate generation, transmission, loading and screening.
- Key-based matching on subject, visit and sample identifiers, with subject and treatment-code ranges provided in advance.
- Duplicate checks as data arrive.
- Version-controlled files carrying origin, creation date, send date and record count, with an audit trail and no deletion of entered data.
- Blinding safeguards for primary efficacy data.
- Regular vendor audits and a formal data clarification process.
Reconciliation “during the study conduct or before database lock” is expected, and sponsors can share demographic or randomization listings with vendors for independent reconciliation.
Software and AI in reconciliation
New 2026 searches ask for the “best software for clinical data reconciliation” and AI approaches. Platforms increasingly aggregate data from multiple systems. For example, Medidata launched Clinical Data Studio in June 2024 with AI-assisted reconciliation and anomaly detection, claiming review and reconciliation “up to 80 percent faster” (a vendor claim). When evaluating any tool, check that matching logic is specified and validated, that every discrepancy has a traceable resolution, and that outputs can be produced as inspection-ready evidence.
Common weaknesses
- Reconciliation done only once, just before database lock, leaving no time to query sites.
- Coding not finished, so event terms cannot be compared.
- No written transfer specification, so vendors change formats mid-study.
- Orphan lab results ignored because they do not map to an EDC visit.
- Accepted discrepancies not documented with a rationale.
- Reconciliation programs never validated, or changed without version control.
- No signed completion notice before lock.
Checklist: reconciliation readiness for database lock
- Reconciliation plan defines sources, fields, frequency and owners
- Data transfer specifications signed for every external vendor
- Test transfers completed and documented
- SAE cutoff date agreed and communicated
- All SAEs reconciled between clinical and safety databases
- Lab and vendor data matched on subject, visit and sample keys
- Orphan and missing records resolved or explained
- Accepted discrepancies documented in the data handling report
- Written notice of reconciliation completion issued before lock
How AskDona helps
Reconciliation leaves a trail of plans, transfer specifications, discrepancy listings, queries and sign-offs, and inspectors ask to see it. AskDona reads that evidence against your data management plan and the GCDMP practices you follow, checks that each discrepancy has a documented resolution and that completion was signed off before lock, and cites the exact file and row. Run a sample reconciliation evidence review before database lock on a current study. AskDona shows the evidence; your data management and safety teams decide.
Editor’s note: Practices come from SCDM GCDMP chapters republished in 2023; the underlying chapters were last substantively revised between 2008 and 2009, so newer practices such as risk-based data review are not fully covered there. The lab reconciliation checks reflect common practice built on the GCDMP external-data chapter. The Medidata figure is a vendor claim and is cited only as an example of market direction.
FAQ
What is SAE reconciliation in clinical data management? Comparing serious adverse event data between the clinical database and the safety database, repeatedly during the study and completely before database lock, so both describe the same events.
What is lab reconciliation? Matching samples recorded at sites with results received from the laboratory, using subject, visit, sample and date keys, and resolving missing or orphan records.
What is external data reconciliation? The same matching process for any data from outside vendors, such as ECG, imaging or ePRO, governed by written transfer specifications.
When should reconciliation be done? Throughout the study at planned intervals, and completed with written confirmation before database lock.
Can reconciliation be automated? Comparison programs and AI tools can find mismatches quickly, but programs should be validated and resolutions documented and reviewed by people.
References
All pages opened and checked by October 19, 2026.
| Source | Type | Used for |
|---|---|---|
| SCDM GCDMP: Serious Adverse Event Data Reconciliation | Professional society practice | SAE fields, timing, roles, standards |
| SCDM GCDMP: External Data Transfers | Professional society practice | Lab and vendor data keys, specifications, standards |
| Medidata Clinical Data Studio launch | Vendor press release | AI reconciliation trend |