Samples / GLPI-103 / tmf/TMF-002_data_management_plan.md

Data Management Plan (DMP) — GLPI103-301

Field Value
Document ID TMF-DMP
Version 2.0 (full)
Study No. GLPI103-301
Standard ICH E6(R3); CDISC (CDASH/SDTM/ADaM); SCDM GCDMP
Confidentiality Confidential — TMF

Change History

Version Date Author Summary
1.0 2026-06-29 Data Management Initial DMP
2.0 2026-06-29 Data Management Full DMP — edit-check spec, data flow, coding, reconciliation, lock

1. Scope, Roles & Responsibilities

Governs data capture → cleaning → coding → standardization → lock for GLPI103-301.

Role Responsibility
Lead Data Manager DMP, edit checks, query oversight, lock
DB Programmer EDC build, edit-check programming, exports
Medical Coder MedDRA/WHO-DD coding
Biostatistician SAP alignment, derived data review
PV SAE reconciliation
QA Audit, lock approval

2. Systems

EDC (eCRF, CDASH-aligned), IWRS (randomization/IMP — TMF-RAND), central laboratory data feed (TMF-LAB), safety database. Access controlled, audit-trailed, 21 CFR Part 11 compliant.

3. Data Flow

Site eCRF entry → automated/manual edit checks → query issuance/resolution → external-data reconciliation (lab, IWRS, SAE) → medical coding → SDTM → ADaM → define.xml (DEF-301) → analysis (SAP-301). Soft/hard locks per visit cohort.

4. Data Validation Specification (Edit Checks)

Programmatic checks implemented by the Data Quality Engine (src/validation/dqe.py); outputs the DVS report and query log (outputs/validation/). Active rules:

Rule Description Domain Severity
EDC-001 HbA1c plausibility (4.0–15.0%) LB Query
EDC-002 FPG plausibility (2.0–25.0 mmol/L) LB Query
EDC-003 Weight plausibility (30–200 kg) VS Query
EDC-004 Blood pressure plausibility (SBP 70–250, DBP 40–140) VS Query
EDC-005 Visit-window deviation (>7 days from scheduled) Visit Query
EDC-006 Post-baseline eGFR < 45 mL/min/1.73m² LB Notable
EDC-007 ALT/AST > 3× ULN (ULN 40 U/L) LB Notable

Realized query volume (this dataset): EDC-005 visit-window 1,240; EDC-006 eGFR 27; EDC-007 hepatic 2; EDC-002 FPG 1 (outputs/validation/dvs_report.csv). "Notable" findings (EDC-006/007) trigger medical review and AESI follow-up.

5. Query Management

Queries are auto- or manually generated, routed to sites, tracked to closure with audit trail; aging/overdue queries are escalated (KRI in TMF-CMP). Target: critical queries resolved before each interim/lock.

6. Medical Coding

Adverse events → MedDRA (version fixed at study start; PT/SOC); concomitant medications → WHO-DD. Coding reviewed by medical monitor; auto-encoder + manual reconciliation; coding report approved before lock.

7. External Data Reconciliation

Central lab vs eCRF (key analytes), IWRS (treatment/strata) vs database, and SAE reconciliation between clinical and safety databases (PV) — all reconciled and documented before lock.

8. CDISC Standardization

Raw → SDTM (DM/SC/VS/LB/AE) → ADaM (ADSL/ADLB; ADaM-SPEC-301) with define.xml/SDRG/ADRG (DEF-301). Controlled terminology in English. Conformance via validate_cdisc.py.

9. Database Lock

Lock checklist: all data entered/cleaned; queries closed; coding approved; external data reconciled; SAE reconciliation complete; QC sign-off. Post-lock changes are controlled and documented. Unblinding follows lock per SAP-301.

10. Quality & Metrics

Double-programming/QC of derivations; data-quality metrics (entry timeliness, query rate/aging, error rate). The reproducible build (build.py) and gates (verify_pipeline, validate_cdisc, reconcile_csr) provide automated, repeatable QC evidence (QC-001).