Complaint handling and adverse event reporting

Problem class

Traditional adverse event case processing can consume up to two-thirds of a pharmacovigilance budget. Some Marketing Authorization Holders process over 1 million ICSRs per year. Underreporting of adverse events has a median rate of 94% — meaning for every adverse event reported, roughly 15 go unreported.

FDA issued 105 warning letters in FY2024 — the highest in five years. QSR-related warning letters rose 58% from 2024 to 2025. Complaint handling (21 CFR 820.198) is among top enforcement priorities alongside CAPA and Design Controls.

Mechanism

Systematic capture, investigation, trending, and regulatory reporting of customer complaints and adverse events. Links complaint investigation to CAPA, risk management file updates, and recall decisions. The process runs from intake through assessment, investigation, MDR reportability evaluation, CAPA initiation (if systemic), resolution, and trend analysis.

Reporting systems. FDA MedWatch for voluntary reports; FAERS/AEMS for structured adverse event reports per ICH E2B; MAUDE database for medical device reports (publicly searchable since 1996 with 1,000+ device problem codes); eMDR for mandatory electronic submission. EU EUDAMED becomes fully mandatory from May 28, 2026, replacing national vigilance reporting. RASFF for EU food safety, CPSC for US consumer products. Global reporting timelines vary: Australia TGA requires 2-day reports for public health threats, 10-day for deaths/serious injuries. FDA mandates 5-workday reports for events requiring immediate corrective action.

AI transformation — NLP signal detection. A CNN + BERT model achieved 85% accuracy for serious adverse event detection, outperforming Logistic Regression (78%) and SVM (80%). MADEx (Medication and Adverse Drug Event Extraction) uses advanced NLP for extracting ADRs from electronic health records. vigiMatch, developed by the Uppsala Monitoring Centre (WHO), uses ML for duplicate report detection. The Netherlands Pharmacovigilance Centre Lareb uses NLP-driven prediction models for identifying serious adverse drug reactions. CIOMS Working Group XIV recommends treating PV AI agents "like medicinal products" with defined scope, capabilities, and limitations. Veeva launched AI Agents for Safety & Quality in April 2026 for narrative summary generation and Annual Product Quality Review drafting.

Required inputs

• Document Control (complaint intake procedures, MDR reportability evaluation forms)
• CAPA system (for systemic complaint patterns)
• Product Traceability (customer-lot linkage)

Produced outputs

• Complaint investigation records with MDR reportability decisions
• Mandatory regulatory reports (MDR, MedWatch, EUDAMED vigilance)
• CAPA triggers for systemic complaints
• Trending analysis for safety signal detection
• Risk management file updates
• Recall decision inputs

Industries where this is standard

• Medical devices (FDA 21 CFR 820.198 and 21 CFR Part 803 MDR — among top enforcement priorities)
• Pharmaceuticals (FAERS/AEMS per ICH E2B; EU EUDAMED fully mandatory May 28, 2026)
• Food safety (RASFF for EU, CPSC for consumer products)
• Automotive (IATF 16949 Clause 10.2.6)
• Any ISO 9001:2015-certified organization (Clause 9.1.2)

Counterexamples

• Not linking complaints to CAPA — systemic issues go unaddressed.
• Delayed adverse event reporting — regulatory risk with mandated 5-workday and 30-day deadlines.
• Not trending by failure mode — missing emerging safety signals.
• Complaint fatigue in high-volume environments where investigation quality degrades.
• Failure to evaluate MDR reportability — FDA requires documentation even when NO MDR is required.

Representative implementations

• CNN + BERT adverse event model — 85% accuracy for serious adverse event detection vs. Logistic Regression (78%) and SVM (80%).
• vigiMatch (Uppsala Monitoring Centre / WHO) — ML for duplicate adverse event report detection across global pharmacovigilance databases.
• Veeva AI Agents for Safety & Quality (April 2026) — narrative summary generation and Annual Product Quality Review drafting.
• Netherlands Pharmacovigilance Centre Lareb — NLP-driven prediction models for identifying serious adverse drug reactions.
• MADEx — advanced NLP for extracting ADRs from electronic health records.

Common tooling categories

Pharmacovigilance platforms (Veeva Safety, Oracle Argus, Aris-g), complaint management modules in QMS platforms, NLP signal detection tools, regulatory submission platforms (eMDR, EUDAMED), duplicate detection engines (ML-based).

Regulatory anchors

FDA 21 CFR 820.198 (complaint files), FDA 21 CFR Part 803 (Medical Device Reporting), ISO 13485:2016 Clauses 8.2.2–8.2.3, ISO 9001:2015 Clause 9.1.2, EU MDR Articles 87–92, IATF 16949 Clause 10.2.6.