Data Cleaning and Quality Control
Data cleaning is the systematic review that makes data consistent, complete, and analyzable. It combines automated edit checks, manual review listings, medical review, and cross-source reconciliation. Candidates must know the cleaning categories: completeness (missing visits/forms), consistency (cross-field logic), plausibility (clinically improbable values), and conformity (coding/format standards).
Quality control includes both QC (checking the work) and QA (the system that ensures quality). Key tools: data review listings, SAS/programmatic checks, medical monitoring review, and the data review meeting before lock. Understand outliers - statistically unusual values are not automatically errors; they must be investigated and, if confirmed correct, retained. Understand protocol deviations versus data errors: a deviation is a departure from the protocol (e.g., out-of-window visit), which may be valid data but flagged.
Traps: deleting or "fixing" an outlier without site confirmation; treating a protocol deviation as a data entry error; assuming cleaning ends at the last query (medical review and listings continue until lock); and confusing intent-to-treat population logic with cleaning logic. Another trap is believing 100% source data verification is required - risk-based approaches allow targeted SDV.
Rule of thumb: investigate, don't erase. Any value change needs a documented reason and a query trail. Worked example: a subject's systolic BP is 260 mmHg. The system flags it; the data manager queries the site; the site confirms a transcription error (should be 160). The correction is made at the site, the query is closed, and the audit trail preserves both values.