Clinical data cleaning is the process of finding and resolving errors, omissions and inconsistencies in trial data so that the final dataset is accurate and complete enough to analyse. It runs throughout the trial, not only at the end.
Free sandbox · No credit card · 21 CFR Part 11 aligned
Clinical data cleaning
noun · clinical research
Identifying, querying and resolving discrepancies in clinical trial data until the dataset is ready to be locked and analysed.
Related terms
In short
Explained
Data cleaning is a core part of clinical data management. It aims to make sure that what is in the database matches what happened to the participant. The rules for what to check are normally written down in advance, in a data management plan and a data validation plan; see the data management plan template and the data validation plan template. Our clinical data cleaning guide covers the working routine.
A discrepancy is any data point that fails a check or looks implausible. The data manager or monitor raises a query to the site, the site checks the source and either corrects the value or confirms it, and the query is closed. Everything is recorded so that an inspector can reconstruct the history.
Range checks flag values outside a plausible or protocol-allowed range, such as a systolic blood pressure of 1200. Consistency checks compare fields on the same or different forms, for example an adverse event start date before consent, or a visit date before the previous visit. Missing-data checks catch required fields left blank. Cross-form and cross-system checks reconcile data held elsewhere, such as safety cases or central lab results.
Cleaning looks at the data. Monitoring looks at conduct and compliance, including source data verification. Medical coding assigns standard dictionary terms to events and medicines. They overlap in practice, and the plan should say who owns which step.
Cleaning is not data manipulation. Changes are made to correct documented errors, with a reason, and not to improve results. Blinded teams should not use unblinded information while cleaning. Missing values are queried, not guessed.
Examples
Illustrative checks of the kind a data manager might write. Values are made up for demonstration.
| Check | Example finding | Typical action |
|---|---|---|
| Range | Heart rate recorded as 720 bpm at Visit 3 | Auto-query to site; likely extra digit |
| Date logic | AE onset date earlier than consent date | Query: confirm date or whether it is medical history |
| Missing value | Weight blank at baseline, BMI cannot be calculated | Query for value or "not done" reason |
| Cross-form | Medication recorded for an AE, but no matching con-med row | Query to complete concomitant medications |
| Duplicate | Two enrolment records with the same subject ID | Check site log and merge or correct |
Workflow
List edit checks and manual review rules before first participant in, in line with the protocol.
Automated checks fire when the value is typed, while the site still has the participant and source to hand.
Data management reviews outliers, trends and patterns that rules did not catch.
Raise queries, track age and status, and resolve with a documented answer.
Reconcile with safety data and external data, close queries and move toward lock.
Create a form, set a range rule and enter a bad value in the free sandbox to watch the query appear. No credit card.
In Capture
Capture edit checks (range high, range low and custom value rules, with configurable priority) raise an auto-query when a site enters a value that breaks a rule. Skip logic keeps irrelevant fields out of the way, calculated fields such as BMI and eGFR are shown read-only, and the audit trail records timestamp, user, old value, new value and reason for every change. Source data verification can be set per field.
Edit checks / auto-queries
2Type
Range High
Operator
Greater than
Value
180
Priority: High
Type
Range Low
Operator
Less than
Value
80
Priority: Normal
Query raised automatically
Value 192 violates limit (180). Please verify.
Finding and resolving errors, gaps and inconsistencies in clinical trial data so the dataset is accurate and complete enough for analysis and database lock.
Validation is the set of rules and checks that detect problems. Cleaning is the wider process of acting on what those checks find, including queries and resolution.
A question raised about a specific data point, sent to the site for confirmation or correction. Queries are tracked until closed.
Before the first participant is entered, when the checks are designed, and it continues throughout the study until database lock.
Yes, if it corrects a documented error and the change is traceable with who, when, old value, new value and reason. Silent overwriting is not acceptable.
Data managers write checks and raise queries, monitors verify against source, and site staff answer queries. Smaller academic trials often combine these roles.
Keep exploring
Clinical data cleaning guide
The working routine.
Edit checks software
Catch errors at entry.
Query management
Track queries to closure.
Database lock
The end of cleaning.
Source data verification
Checking against source.
Data validation plan template
Write the rules down.
Clinical data management software
The tools involved.
Free sandbox with edit checks, queries and audit trail. No credit card.