CPCD cheat sheet

21 CFR Part 11 governs electronic records and signatures: audit trails, access controls, and validated systems are required for eCRF data to be considered reliable.

ICH E6 Good Clinical Practice defines the sponsor's responsibility for data quality and integrity; the data manager operationalizes these obligations through documented processes.

ALCOA+ principles: Attributable, Legible, Contemporaneous, Original, Accurate, plus Complete, Consistent, Enduring, and Available.

ICH E2A defines serious adverse event criteria: death, life-threatening, hospitalization or prolongation, persistent disability, congenital anomaly, or important medical event.

Source documents are the original records (hospital charts, lab printouts); the eCRF is a derived record and must be reconcilable to source.

The data management plan is a living document; version control and documented approval are required before it governs study conduct.

Inspection readiness requires retrievable audit trails, documented training records, and traceability from protocol to database to analysis.

Data privacy laws (GDPR, HIPAA) constrain transfer of identifiable subject data; de-identification and coding of subjects are standard controls.

Start-up sequence: protocol review, CRF/eCRF design, database build, UAT, DMP finalization, site training, first-patient-in readiness.

The DMP specifies data flow, source-to-CRF mapping, edit check strategy, SAE reconciliation plan, coding dictionaries, and database lock procedures.

A data transfer specification defines file format, structure, frequency, encryption, and reconciliation rules for external data (labs, ECG, PK).

User acceptance testing must be documented with test scripts, expected results, actual results, and sign-off before production release.

Site training on the eCRF should be documented per user; training records support inspection readiness and system access justification.

CRF completion guidelines reduce query volume by defining expected formats, units, and handling of missing or not-done data.

Protocol review identifies endpoints, visit schedule, and derived variables so the database structure supports the statistical analysis plan.

Risk-based monitoring and data review plans allocate source data review and SDV according to criticality of variables.

Collect only data required by the protocol and analysis plan; each field must map to an endpoint, safety assessment, or operational need.

Design fields to match source document structure to reduce transcription error and query generation.

Use controlled terminology (codelists) for categorical fields; avoid free text where standardized options exist.

Edit checks are built at field, form, and cross-form level; range checks, format checks, and consistency checks are core types.

Derived fields should be computed by the system rather than entered manually to avoid entry error and inconsistency.

Units must be specified on the form and stored separately from values to allow unit conversion and cross-site comparability.

Partial dates require defined handling rules (e.g., imputation strategy) documented in the DMP before database build.

Database build includes audit trail configuration, user roles and permissions, and query workflow definition.

Data entry modes: single entry, double entry with comparison, and direct eCRF entry; double entry is used where error risk is high.

A discrepancy (query) is generated when data fails an edit check or a manual review identifies inconsistency; each query requires a documented response and resolution.

Query lifecycle: open, sent to site, responded, reviewed, closed; aging metrics track query resolution timeliness.

Automatic edit checks fire at data entry; manual review checks are performed by data managers during ongoing review.

Silent or self-evident corrections must follow a predefined, documented convention and be traceable in the audit trail.

Discrepant values should never be overwritten without retaining the original; corrections require reason and user attribution.

Query text must be specific and actionable; vague queries increase resolution time and repeat queries.

System validation includes edit check testing in UAT; failed checks are logged, corrected, and retested before go-live.

Data cleaning includes range checks, consistency checks, cross-form checks, and review of free-text and comment fields.

Medical review of data (labs, AEs, concomitant medications) complements statistical/data-management cleaning.

Outlier review distinguishes data entry error from true biological variation; source verification resolves the distinction.

Key performance indicators for cleaning include query rate per CRF page, query resolution time, and SDV completion percentage.

Data review meetings document decisions on protocol deviations, eligibility, and data handling; minutes support audit trail of judgment.

Listings review for missing visits, missing forms, and out-of-window visits is a standard cleaning activity.

Duplicate subject or duplicate AE/medication records are identified through sorting and deduplication checks.

Final cleaning precedes database lock; unresolved issues are documented in a lock checklist with justification.

MedDRA is used for coding adverse events and medical history; it has a five-level hierarchy: LLT, PT, HLT, HLGT, SOC.

WHO Drug Dictionary (WHODrug) is used for coding concomitant medications and prior medications, with ATC classification.

Coding should be performed on verbatim terms; the verbatim term is retained and the coded term is added, not substituted.

MedDRA version must be consistent within a study; version changes require documented re-coding strategy and impact assessment.

Auto-coding tools propose matches; manual review and approval by qualified personnel are required for accuracy.

Standardized units (e.g., SI units) and standardized date formats support pooling and analysis across sites.

Coding consistency checks identify the same verbatim term coded to different concepts across subjects.

The coding plan or DMP specifies dictionary versions, coding conventions, and who performs and reviews coding.

SAE reconciliation compares the safety database against the clinical database for AE terms, dates, severity, and causality.

Reconciliation frequency should be defined (e.g., monthly and pre-lock) with documented discrepancy resolution.

Seriousness criteria per ICH E2A: death, life-threatening, inpatient hospitalization or prolongation, disability, congenital anomaly, medically important event.

Expedited reporting timelines: 7 calendar days for fatal or life-threatening unexpected suspected adverse reactions, 15 days for other serious unexpected suspected reactions.

Pregnancy and overdose cases are typically handled as special situations requiring safety review even without an AE.

Blinded studies require unblinding procedures and controlled access to treatment assignment data for safety reporting.

The DSMB/DMC receives safety data independently; data management supports the data cuts and unblinding logistics.

Safety data flow must preserve the audit trail from source to safety database to regulatory submission.

Database lock criteria: all queries closed or justified, SAE reconciliation complete, coding complete, data review sign-off, and documented approvals.

Soft lock (data freeze) allows limited changes with documented rationale; hard lock prohibits further data entry.

Post-lock changes require a formal unlock procedure with documented reason, approval, and audit trail.

Archival must preserve data, metadata, audit trails, and system documentation for the required retention period (often 2 years after last marketing approval or longer per local law).

Archival formats must remain readable; migration plans are needed when systems are decommissioned.

Close-out includes return or destruction of data, final reconciliation, and transfer of the database to the sponsor or archive.

The clinical study report depends on the locked database; any post-lock change must be reflected in the analysis and documented.

Lock checklists and sign-off records are inspection-critical evidence that lock was controlled and justified.

A validated computerized system per 21 CFR Part 11 and GAMP 5 has documented requirements, testing, and change control.

Audit trails must capture who changed what, when, and why; audit trails cannot be disabled by users and must be reviewable.

Role-based access control enforces least privilege; access is granted per role and revoked at study close or staff departure.

Electronic signatures must be attributable, linked to the record, and include the signer's identity and date/time.

Data integrity controls include backup, disaster recovery, and periodic restore testing.

System change control requires impact assessment, testing, and approval before release to production.

eSource and direct data capture shift source documentation into the system; validation and audit trail requirements still apply.

Data migration between systems requires verification that values, relationships, and audit trails are preserved.

Risk-based quality management (RBQM) focuses monitoring and cleaning effort on critical data and processes.

Data quality dimensions: completeness, accuracy, consistency, timeliness, and validity.

Central statistical monitoring uses statistical tests to detect anomalous site or subject patterns (e.g., digit preference, duplicate values).

Data integrity issues include fabrication, falsification, and unauthorized alteration; detection relies on source verification and pattern review.

System access reviews should be periodic and documented; dormant accounts are a common inspection finding.

Backup and recovery testing frequency and retention should be defined in the DMP and IT procedures.

Data flow diagrams in the DMP map each data source to the database and to the analysis dataset.

Interoperability standards (CDISC SDTM/ADaM) define how collected data are structured for submission.

CDISC SDTM structures submission data into domains (DM, AE, CM, LB); ADaM structures analysis datasets.

Data validation for submission includes checking against SDTMIG rules and controlled terminology versions.

Electronic data transfer between sponsors and CROs must be documented with file specifications and reconciliation logs.

System decommissioning requires data extraction, verification, and archival in a readable format with documented migration.

ePRO and wearable device data require specification of timestamps, time zones, and handling of incomplete transmissions.

Data lakes and analytics environments must preserve the controlled, validated source of record for regulatory data.

Vendor oversight includes qualification, audit, and documented service-level agreements for data systems.

Data integrity findings often stem from shared logins, disabled audit trails, or undocumented corrections; controls must address each.

The data manager's role in inspection includes producing audit trails, query logs, DMP versions, and training records on request.

Findings frequently relate to inadequate oversight of vendors, incomplete audit trails, and undocumented data corrections.

CAPA (corrective and preventive action) plans document root cause, corrective action, and effectiveness verification.

Standard operating procedures define required steps for data handling; deviations require documented justification.

Training records must demonstrate that personnel performing data tasks are qualified for those tasks.

Metrics and trend reports (query rates, lock timelines) support continuous improvement and inspection defense.

Data governance assigns accountability for data quality across functions (clinical, data management, biostatistics, safety).

Documented risk assessments justify the intensity of monitoring, SDV, and cleaning applied to each data element.