Published on 16/11/2025
EDC–PV Safety Reconciliation: Designing Reliable Processes That Regulators Recognize
Why EDC↔PV Reconciliation Exists—and What Regulators Expect to See
Safety data reconciliation is the systematic comparison of safety facts recorded in the clinical database (EDC) against the pharmacovigilance (PV) safety system (e.g., Argus/ARISg). Its purpose is to ensure that every reportable safety event is captured, assessed, and—where required—submitted correctly and on time. Regulators including the U.S. FDA, the EMA, Japan’s PMDA, Australia’s TGA, and the harmonized expectations under the Why the two systems diverge. EDC is optimized for protocol conduct and statistical analysis; PV is optimized for case processing and expedited reporting. Differences arise because EDC captures study-centric information (e.g., visit windows, endpoints) while PV captures regulatory-centric information (e.g., seriousness criteria, expectedness vs RSI, distribution history). Data enter along different paths and clocks: “day 0” (PV clock start) is the date the sponsor or its agent becomes aware of a valid case; EDC entries may lag or precede that awareness. Without disciplined reconciliation, mismatches in onset dates, seriousness, causality, or outcomes can cause late SUSARs, incorrect denominators, and inspection findings. What authorities expect. Auditors want to see: (1) a written procedure defining scope, frequency, roles, and metrics; (2) a controlled crosswalk of fields compared; (3) reconciliation logs with timestamps (recorded as local time + UTC offset) and aging; (4) queries to sites/partners with documented resolution; (5) traceability between EDC SAE forms and PV case numbers; and (6) CAPA for recurring discrepancies. They also look for dictionary discipline (MedDRA/WHO-DD versions) and expectedness anchored to the correct RSI/label at the time of onset, consistent with ICH E2A/E2D. Scope beyond SAEs. While serious adverse events (SAEs) are core, mature programs reconcile a broader set: Important Medical Events (IMEs), adverse events leading to discontinuation, death records, pregnancies, medication errors, lack of efficacy in life-threatening disease, special interest AESIs, and device malfunctions in combination-product studies. The reconciliation plan states which event classes are in scope and the data elements to be matched for each class. Risk-based proportionality. The cadence and depth should reflect study risk: early-phase single-site trials may reconcile monthly; pivotal, multi-region trials with expedited risk (oncology, vaccines, cell/gene therapies) often reconcile weekly or even rolling. Frequency should align with expedited-reporting clocks under the FDA (e.g., 21 CFR 312.32), the EU Clinical Trials Regulation/EudraVigilance processes under the EMA, and national expectations at PMDA and TGA. Blinding and independence. For blinded studies, reconciliation should be arm-agnostic. Only independent, unblinded personnel (per DSMB/IDMC charter) may view arm-level safety if required. Operational teams see subject-level facts stripped of treatment identity; all access is logged. Define the field crosswalk. Start with a reconciliation specification listing the exact variables to compare, their owners, and matching rules. Typical core elements include: Matching logic that survives reality. One SAE can generate multiple PV updates (follow-ups), and a single clinical episode can be represented in EDC as several records (e.g., “pneumonia,” “hospitalization,” “fever”). Define linkage rules: use a composite key (subject + site + onset ± 1–2 days + primary PT family) and allow a configurable window for onset differences (time-zone entry, clock skew). Keep a case number crosswalk table to persist matches over time. Frequency and timing. Establish a steady cadence (weekly/biweekly/monthly) and event-driven checks (e.g., 48 h after any SAE entry, pre-data lock, and pre-IB/RSI update). Hard-stop reconciliations are required at Data Lock Points (DLPs) for DSUR/PBRER cycles to ensure counts and narratives match aggregate reports. Tools and automation. Build or buy a reconciliation utility that extracts deltas from EDC and PV, normalizes formats (dates, time zones, country codes), and compares fields with rule-based tolerances. The tool should generate exceptions with reason codes (missing in PV; missing in EDC; mismatch onset; mismatch seriousness; mismatch expectedness; duplicate risk) and route them to owners with due dates. Integrate dashboards to show open exceptions by site, age, and severity; export a rapid-pull log for inspectors. Governance and RACI. Clarify who is Responsible (EDC data manager for clinical fields; PV operations for case facts; medical monitor/safety physician for adjudications), Accountable (study safety lead), Consulted (site, CRO), and Informed (QA). Define escalation rules (e.g., discrepancies > 7 days old → safety lead; > 14 days → governance committee) and document decision rights for changes to case classification (seriousness, expectedness, causality). Dictionary discipline. Pin MedDRA versions in both systems; time-box upgrades and run impact analyses to quantify remaps and AESI retrieval changes. Ensure WHO-DD versions align for concomitant medications. Display versions and effective dates in reconciliation reports to prevent “silent drift.” Privacy and traceability. Reconciliation outputs should exclude direct identifiers; keep linkable keys under access control and lawful bases (GDPR/UK-GDPR, HIPAA where applicable). Time-stamp every extraction and decision with local time + UTC offset to simplify multi-region audits by the FDA/EMA/PMDA/TGA and meet ICH documentation expectations. Step-by-step execution. Classic discrepancy patterns—and what to do. Query management that respects timelines. Use templated, topic-specific queries (e.g., DILI/Hy’s Law, anaphylaxis, pregnancy). Set due dates that preserve expedited clocks; escalate non-responses; document all outreach attempts. For global trials, provide translated query text and maintain culturally clear instructions for sites. Root-cause analysis (RCA) and CAPA. Trend discrepancies by site, CRO, and category. For systemic issues (e.g., onset date errors at multiple sites), conduct RCA: training gaps, confusing form design, or dictionary drift? Implement targeted training, form tweaks (e.g., tooltips explaining “first symptom vs diagnosis”), or system controls (mandatory fields, date validations). Track CAPA effectiveness (post-implementation error rate). Synchronizing with aggregate reporting. Before DSUR/PBRER DLPs, perform hard-stop reconciliation; lock counts and narratives; record extraction timestamps (local time + UTC offset). Ensure MedDRA versions in tabulations match PV. Any post-DLP case updates should be described in the “events after DLP” section without retro-changing counts. Blinded safety and arm-agnostic dashboards. Reconciliation reports used by blinded teams should omit treatment labels. Comparative or arm-level checks (e.g., EAIR differences) are run by the independent unblinded statistician/physician per charter; only decisions (continue/stop/enrich) flow back. Inspection-ready evidence bundle. Keep a “rapid-pull” index that surfaces within minutes: Program-level KPIs that matter. Common pitfalls—and durable fixes. One-page checklist (study-ready reconciliation). Bottom line. Reconciliation is not an administrative chore; it is a regulatory control that protects participants, ensures accurate expedited reporting, and underpins credible aggregate submissions. With clear specifications, disciplined cadence, robust tools, and documented adjudication, sponsors can demonstrate sustained control to authorities across the FDA, EMA, PMDA, and TGA, consistent with ICH principles and the WHO’s public-health mission.Designing a Fit-for-Purpose Reconciliation Framework
Running the Engine: Matching, Queries, and Resolving the Hard Problems
Proof of Control: Evidence, KPIs, Pitfalls, and a One-Page Checklist