Building a Dictionary Strategy That Survives Audits

Version governance. Freeze dictionary versions per study and record effective-from dates (e.g., MedDRA 27.0; WHODrug Global B3/C3 Q2-2025). Document rationale for any mid-study upgrade (e.g., improved coverage for a new indication), impact assessment on key tables/figures/listings, and the re-coding plan. Maintain side-by-side outputs during transition and archive both versions and release notes in the Trial Master File (TMF).

Term selection hierarchy (MedDRA). Map verbatims to the most specific Lowest Level Term (LLT) that exactly reflects the clinical concept (including laterality, temporality, and anatomic details). The LLT rolls up to a Preferred Term (PT) for analysis. Ensure coders understand multi-axiality and Primary SOC logic so counts at SOC level reflect regulatory conventions.

Preferred and base names (WHO-DD). Normalize drugs to the correct Preferred Name or Preferred Base Name and ensure correct ATC level assignment. Capture route/form/strength when clinically relevant (e.g., oral vs IV corticosteroids). For combination products, confirm all ingredients are represented and that the ATC reflects the intended therapeutic class.

Synonym lists and stop-word policies. Maintain curated synonym lists (brand↔generic, frequent misspellings, country-specific trade names) and a stop-word list (e.g., “tablet,” “cream,” “mg,” “capsule”) to enhance auto-suggest without over-constraining free text. Version these lists, test their effect on auto-coding hit rates, and file them like any other configuration artifact.

Auto-coding thresholds and explainability. Define when the system can code automatically (exact/high-confidence matches) and when to route to manual review (low confidence, multi-match, ambiguous context). Auto-coding decisions must be deterministic and reproducible with logs showing the algorithm version, dictionary version, and chosen path.

Handling vague or non-current terms. Apply MedDRA guidance for non-current LLTs (map to current equivalents) and avoid vague concepts (e.g., “feeling unwell”) by querying for clarification when CtQ impact exists. For WHO-DD, query for generic ingredient when only a local brand is given and the mapping is unclear or the ATC class could change interpretability.

Specialized groupings. Pre-define Standardised MedDRA Queries (SMQs) and sponsor-defined baskets for medical review (e.g., hepatic events, hypersensitivity). Document which SMQ versions (narrow/broad) are used in the SAP and how sensitivity analyses will be conducted if dictionary versions change.

Intended-use validation. Treat coding platforms as computerized systems: retain requirements, risk assessment, test scripts/results, deviations, approvals, and user training evidence. Validate exports (coded value, code, and dictionary version) and audit-trail functionality (who coded/recoded, when, why) so you can reconstruct decisions during inspection at FDA/EMA/PMDA/TGA.

Day-to-Day Operations: From Verbatims to Coded Terms Without Breaking the Blind

Workflow and RACI. Publish who auto-codes, who manually codes, who performs medical coding QC, and who adjudicates disagreements. Segregate blinded and unblinded lanes; prohibit blinded coders from seeing kit maps or IMP brand names. Route IMP-related questions to unblinded pharmacy/IRT queues with access logs.

Authoring high-quality verbatims at the source. Train sites to provide clinically meaningful entries (diagnosis vs symptom, anatomical site, laterality, timing) and to avoid abbreviations unless standard. Provide job aids with examples (e.g., “Acute bacterial pneumonia” vs “Cough”). Better verbatims reduce queries and re-work.

Auto-coding and manual review. Use auto-coding for exact/safe matches. Manual coders confirm context and address edge cases (e.g., “MI” → “Myocardial infarction” vs “Mitral insufficiency” based on context). Require coders to document rationale for any non-obvious mapping and to request site clarification when ambiguity affects CtQs.

Quality control (QC) and agreement. Apply dual-coding or sampling (e.g., 100% QC for AEs/SAEs; 10–20% risk-based sample for concomitants), calculate inter-coder agreement (e.g., Cohen’s kappa), and review discrepancies in a weekly huddle. Track recurring disagreement patterns and update guidance/synonym lists to prevent re-occurrence.

Queries for clarification (minimum-necessary). When coders need more detail, send clear, non-leading questions: “Please confirm whether the event is a diagnosis (e.g., pneumonia) or a symptom (e.g., cough).” Avoid PHI; request certified copies only when essential, with redaction and provenance (system/report version, local time + UTC offset, user attribution).

Linkage to pharmacovigilance. Align AE coding with safety case processing so MedDRA versions match across the clinical database and the safety database. Reconcile AE/SAE records at a defined cadence; investigate discordant coding (e.g., different PTs selected for the same case). Ensure narrative redactions protect blinded roles and privacy.

WHO-DD specifics for concomitants. Normalize to preferred/generic names and check ATC class. For therapies with multiple indications (e.g., beta-blockers), capture indication in MedDRA if collected and ensure consistency with analysis conventions (e.g., grouping “prohibited/concomitant of special interest”). For combination products, verify that all ingredients map correctly and that dose/route are captured where clinically meaningful.

Procedures, devices, and product-quality issues. Use appropriate MedDRA concepts for procedures (“Procedure” SOC), device complications, and product-quality terms. Coordinate with the SAP if certain device-related PTs require bespoke analyses or listings.

Decentralized/Hybrid (DCT) realities. eCOA and telehealth may inject consumer drug brand names or colloquialisms. Enhance synonym libraries with local brands and common misspellings; add device-app version fields to support targeted clarification when language models on mobile keyboards introduce artifacts.

SDTM and downstream harmony. Ensure coded outputs map cleanly to SDTM (e.g., coded PT text in –DECOD, SOC in –SOC when populated, and dictionary/version metadata). Keep derivation specifications for groupings (SMQs, ATC levels) with traceability to the dictionary version used at analysis.

Making Coding Inspectable: Evidence, Metrics, and Pitfalls to Avoid

Documentation architecture. Keep a rapid-pull bundle in the TMF: coding SOPs and work instructions; version control records for MedDRA/WHO-DD; synonym/stop-word lists (with versions); auto-coding configuration; coder training/competency records; QC plans and results; discrepancy adjudication minutes; reconciliation logs with the safety database; certified sample exports; and audit-trail extracts showing who coded/recoded, when, and why. Include point-in-time configuration snapshots of dictionary versions at first patient in, each amendment, and database lock.

Program KPIs that prove control.

• Auto-coding hit rate (by domain) and false-positive rate confirmed by QC.
• Clarification query rate per 100 entries (target downward trend as verbatim quality improves).
• QC agreement (e.g., Cohen’s kappa ≥0.8 for AEs) and re-work rate after QC.
• Dictionary alignment—100% MedDRA version match between clinical and safety systems for reconciled cases.
• Turnaround time—median days from data entry to coded status for CtQ domains (AEs/SAEs, indications of interest).
• SMQ/ATC completeness—% of planned groupings generated without manual patches.
• Blinding/privacy hygiene—0 unmitigated blind leaks; same-day deactivation on role change; access logs reviewed.

Inspection-day playbook. Be ready to show: (1) the dictionary versions and effective dates; (2) how verbatims become coded terms (auto vs manual) with evidence; (3) how disagreements were resolved; (4) how coding aligns with safety and with SDTM/analysis; and (5) how blinding and privacy were protected. Demonstrate a re-run using a point-in-time configuration and produce the audit-trail and certified copies without vendor engineering.

Common pitfalls—and durable fixes.

• Dictionary/version drift → freeze versions; document rationales for upgrades; maintain side-by-side outputs during transition; QC a risk-based sample after re-coding.
• Ambiguous verbatims → train sites; provide examples and checklists; issue targeted clarification queries only when CtQ impact exists.
• Over-reliance on auto-coding → set conservative thresholds; audit low-confidence matches; measure false positives.
• Inconsistent MedDRA SOC counts → ensure Primary SOC logic is applied; educate analysts about multi-axiality; align SAP shells accordingly.
• Mis-mapped combinations in WHO-DD → verify all ingredients; confirm ATC class; lock route/form where it affects interpretation.
• Blind leaks via drug names → mask IMP names for blinded users; segregate unblinded workflows; log any key/kit-map access.
• Evidence gaps → export and file audit trails and configuration snapshots at key milestones; keep coder rationales with references.
• Privacy over-collection → apply minimum-necessary standard; redact PHI on certified copies; document cross-border mechanisms consistent with HIPAA/GDPR/UK-GDPR.

Quick-start checklist (study-ready coding framework).

• MedDRA and WHO-DD versions selected and frozen; effective dates recorded; release notes filed.
• Synonym and stop-word lists curated and versioned; auto-coding thresholds defined and tested.
• Blinded/unblinded coding lanes segregated; IMP masking rules in place; access logs reviewed.
• Coding SOPs, coder training, and QC plan (dual coding/sampling, adjudication) approved.
• Safety reconciliation cadence set; dictionary versions aligned between clinical and safety databases.
• SDTM mapping documented (coded text and code, dictionary/version metadata); SMQs/ATC groupings specified in SAP.
• Audit-trail exports and configuration snapshots rehearsed; certified samples filed in TMF.

Bottom line. Coding is not a clerical step; it is a scientific control that underpins safety review and efficacy interpretation. When dictionaries are governed, workflows are risk-proportionate, blinding and privacy are protected, and evidence is reproducible, your program will meet expectations across the FDA, EMA, PMDA, TGA, the ICH community, and the WHO.