However well a study is designed, missing values, inconsistent dates, wrong units, duplicates and protocol deviations are inevitable. Data cleaning is the systematic process of finding, reviewing and resolving them before analysis, so the final database reflects what actually happened during the investigation.
Chief Data Officer
Biometrics & Data Systems
in clinical data systems, biometrics and statistical reporting
Head of Quality & Compliance
ISO 14155 & data integrity
in quality systems and inspection readiness for device studies
Chief Operating Officer · former TÜV SÜD Senior Reviewer
in device evaluation and Notified Body review
The goal is not to correct values into something tidier. It is to ensure the recorded information is accurate, complete, consistent, logical, traceable and ready for analysis, with the original entry preserved and every correction documented.
The strategy for all of this is set in the data management plan, as part of the wider clinical data management scope.
Data managers see the same categories across investigations. Knowing them in advance is what allows the edit checks to be written before the data arrives.
A visit was incomplete, a form was overlooked, laboratory results were delayed, or entry was simply missed. Every gap is evaluated to decide whether clarification is needed.
Informed consent recorded after enrollment. Follow-up before baseline. Implantation before surgery. Adverse events before treatment. Most of these are catchable automatically.
Pregnancy recorded for a male participant. An explantation with no prior implantation. A participant marked withdrawn who keeps attending scheduled visits.
Negative age, implausible blood pressure, heart rate outside physiological limits, device measurements in the wrong unit. Not every unusual value is wrong, but each one is verified.
Duplicate adverse events, repeated laboratory results, several device identifiers for the same implant. Usually the result of entry from more than one system.
Serial and lot numbers, software versions, malfunctions, corrective actions, explantation details. Structured capture in the eCRF design is what prevents most of these.
Query response times, vendor delivery calendars and national data rules differ across a multinational investigation, and they shape how the cleaning effort is scheduled.
Data protection rules, hosting requirements and registry access vary by country and are settled before the database is built, not after the first patient is enrolled.
Technology finds what is checkable. People find what is odd. Neither alone produces a clean database.
Automation gives sites immediate feedback and takes the routine load off the team. Human review remains essential for anything requiring clinical interpretation, and it works best alongside clinical monitoring, which sees the same signals from the site side.
Risk-based cleaning concentrates review where an error would actually change something, and handles the rest through automation and targeted checks.
The full reasoning, including how the risk assessment is documented and revised, is on the risk-based data management page.
Quality is not a feeling about the database. These indicators are reviewed on a regular cycle and they give early warning long before database lock is at risk.
The count, and more importantly the age distribution. Old open queries are the ones that will hold up the lock.
Average turnaround by site. A site that takes six weeks to answer needs a conversation, not more queries.
Overall and for critical variables specifically. The second number is the one that matters.
The gap between visit and entry. A widening gap is usually a site resourcing problem showing up early.
A high rate can mean poor data, or it can mean the edit checks are too aggressive. Both are worth knowing.
Which external datasets are current, and which are waiting. This is a frequent cause of a delayed lock.
Device investigations receive data from central laboratories, imaging vendors, wearables, ePRO platforms, device telemetry and randomization systems. Each dataset is reconciled against the clinical database on a defined cycle.
The mistakes that cost most. Waiting until the end of the study to review data, generating excessive queries, ignoring recurring site errors, skipping external reconciliation, reviewing every variable with the same scrutiny, documenting review activities poorly, and letting data management and clinical operations work from different pictures.
Cleaning ends at clinical database lock for a medical device study, which is where the data management team takes over formally.
Query history, data review logs, coding decisions, reconciliation reports, database review records and quality control activities. Complete documentation is what demonstrates the data was managed according to predefined procedures, and it is the part an inspector will ask to see years after the study closed.
Published work and client evidence, not claims.
Where continuous cleaning sits in the wider data lifecycle.
Validation refers to the automated or predefined checks that flag potential errors at entry, such as missing values or out-of-range measurements. Cleaning is the broader process of reviewing, investigating and resolving discrepancies throughout the study, including query management, coding, external reconciliation and manual clinical review.
As soon as the first participant is enrolled. Continuous review distributes the workload and means issues are found while they can still be corrected, which is what shortens database lock.
The clinical data manager coordinates it, working with investigators, clinical research associates, medical monitors, statisticians and the sponsor.
No. Corrections are made through documented queries and recorded in the audit trail. The original entries are preserved, which is what Good Clinical Practice and the ALCOA+ principles require.
When sites start answering without reading, there are too many. Query volume is not a quality metric, and a high rate often points at the edit checks rather than the data.
Yes. The assessment looks at open query age, missing critical data, external reconciliation status, coding progress and validation documentation, and produces a remediation plan with a realistic lock date.
Send us the study details and access to the current metrics. We come back with an assessment of open query age, missing critical data, reconciliation status and coding progress, and a realistic view of what stands between you and a lock date.
Your documents are reviewed confidentially. An NDA can be put in place before we receive any technical or clinical information. You can also reach the team through the contact page.
Whitepaper, BSI and Eclevar
on the EU MDR
Client testimonial,
RegenLab