Coding Workflows, QC & Change Control

Why this matters

Clinical terms typed by sites (“verbatims”) must be translated into standard codes so regulators, statisticians, and safety reviewers can analyze them consistently. If different trials use different dictionary versions or inconsistent conventions, pooled analyses break: categories shift, counts change, and you may need to recode large datasets under time pressure. A planned workflow with clear conventions, QC (independent review), and formal change control prevents rework and protects submission timelines.

Core concepts & definitions

Verbatim term The exact text entered by the site (e.g., “heart attack”, “ASA”, “Tylenol PM”). Good quality verbatims reduce ambiguity.

Dictionary A standardized vocabulary used to map verbatims to codes:

• MedDRA (Medical Dictionary for Regulatory Activities): used for clinical events (adverse events, medical history, indications).
• WHODrug Global (often written “WHO Drug”): used for medications (concomitant and prior medications), including product, ingredient, and Anatomical Therapeutic Chemical (ATC) classification.

Version Dictionaries are updated regularly: MedDRA and WHODrug Global each release twice a year (March and September). The version (e.g., MedDRA 27.0) must be recorded and controlled, especially when pooling across studies.

Coding Assigning a dictionary code to a verbatim. This may involve auto‑coding (the system matches the verbatim to a dictionary term or approved synonym and proposes the code) and manual coding (a trained coder confirms the match or codes what did not match). A medical reviewer advises when the coder cannot settle the term; an unclear verbatim goes back to the site as a query.

Conventions Study/company rules that resolve ambiguities and drive consistency. Examples:

• How to handle abbreviations (read “MI” as “myocardial infarction” only when the entry leaves no doubt; otherwise query the site. The verbatim itself is never rewritten).
• Combined concepts: use one MedDRA combination term if it exists; otherwise each separate concept gets its own code (“headache and nausea” → two coded terms). No reported concept is dropped.
• Non‑clinical entries (“study visit” as reason for hospitalization → do not code as AE).

Conventions are documented, versioned, and covered in coder training.

QC sampling (Quality Control) Independent review of a subset of coded terms. Risk‑based rules (e.g., higher sampling for new coders or high‑impact categories) help balance quality and speed. The reviewer is never the person who coded the term. The plan sets an acceptance threshold: e.g., 6 discrepancies in a 200‑term sample is a 3% error rate; above a 2% threshold, that triggers retraining and a larger sample.

Version alignment Using the same dictionary version for all studies intended to be pooled. If versions differ, plan recoding to a common target and maintain an audit trail.

Recoding Re‑applying codes after a dictionary upgrade or convention change. Requires planning, delta listings (what changed), and archiving before/after states.

Audit trail A traceable record of who coded what, when, with which version and rationale—supporting compliance and inspections. The system creates it automatically with a time stamp, and a changed code never overwrites the earlier one.

The process, step‑by‑step

1. Select dictionaries & versions

   • Choose MedDRA for events and WHODrug for medications.
   • Record the exact version and rationale in the data management plan (DMP) and coding plan.
   • Agree the same with every vendor that codes or sends coded data (e.g., a contract research organization (CRO) or central lab) before the first transfer: dictionary and version, the codelists (allowed values for coded fields such as units), who codes, and the upgrade schedule. Record it in the DMP/coding plan and the data transfer specification; the sponsor still oversees delegated coding.

2. Write & publish conventions

   • Include examples of good vs. poor verbatims.
   • Cover abbreviations, multi‑concept phrases, misspellings, and non‑events.
   • Train coders on the conventions before they code, and sites on clear verbatim entry (one event per line, no abbreviations), without coaching them toward dictionary terms. Store conventions under version control.

3. Prepare the data for coding

   • Ensure context fields are captured to improve accuracy:

     • For events (MedDRA): onset date, seriousness, indication (if relevant).
     • For medications (WHODrug): indication, route, dose, unit, frequency (these qualifiers help the coder choose the right product and ATC class).

4. Run the coding workflow

   • Intake verbatims → auto‑code attempt (system suggestions) → manual review/confirmation by a trained coder.
   • Escalate unclear terms: data management (DM) queries the site. Only the site changes its verbatim; the coder never edits it.
   • QC sample per plan (e.g., 10% for new coders until stable, then 5%; 100% for pre‑defined high‑risk categories).

5. Plan for pooled analyses early

   • For programs with multiple trials, define a target dictionary version (e.g., “All cardiovascular (CV) studies will use MedDRA 27.0 by Q3”).
   • If a feeder study is on an older version, plan recoding and testing ahead of the pool.

6. Change control & upgrades

   • For any dictionary upgrade or convention change, perform an impact assessment (what shifts, which outputs are affected).
   • Decide recode now vs. lock to current; document rationale.
   • If recoding: work under an approved change request, run delta listings, re‑run QC sampling on impacted terms, and archive both states. Recoding changes only the codes, never the site’s verbatim; one analysis uses a single dictionary version.

7. Finalize & archive

   • Confirm that every term is coded and coding queries are closed, then freeze coding at database lock with the version recorded.
   • Archive: conventions (by version), coding decisions, QC results, change requests, and before/after code versions.

What good looks like

• A Coding Convention Manual with examples, “do/don’t” guidance, and a change log; coders trained on it before they code.
• A risk‑based QC plan (clear sampling rules, acceptance thresholds, and re‑training triggers).
• Program‑level version strategy (e.g., “All pooled CV studies normalized to MedDRA 27.0”).
• Automated delta listings available for any recode event, reviewed and signed off.
• Audit‑ready records: who coded, who QC’d, when, with which version, and why.

Worked examples

Example 1 — Aligning event coding for a pooled cardiovascular (CV) meta‑analysis

Context: Two CV outcome trials will be pooled. Trial A used MedDRA 26.1; Trial B used MedDRA 27.0.

Problem: Suppose the impact assessment shows that some ischemic‑event terms used in Trial A are grouped differently in 27.0. Mixing versions could change totals in pooled tables.

Action:

1. Set MedDRA 27.0 as the program target.
2. Recode Trial A from 26.1 → 27.0 in a sandbox copy.
3. Generate delta listings (verbatim, old code, new code, hierarchy changes).
4. Apply focused QC (100%) on event classes known to shift (e.g., myocardial infarction, stroke).
5. Obtain clinical and biostat sign‑off; archive both versions with rationale.

Outcome: The pooled datasets now align at 27.0, enabling clean pooling with transparent change documentation. If Trial A is already locked and reported, its own database and study report stay at 26.1; every output states its version.

Example 2 — Medication coding with WHODrug using qualifiers

Context: A site enters “ASA 81” for a concomitant medication.

Risk: “ASA” could mean acetylsalicylic acid (aspirin) but could also be an abbreviation for something else in certain contexts.

Action:

1. Check qualifiers: indication (e.g., “cardiovascular prevention”), dose (81), unit (mg), route (oral), frequency (once daily).
2. Use qualifiers to code to the correct WHODrug record (ingredient level if product unclear). If they do not show which medication is meant, DM queries the site.
3. Apply convention: if product brand is unknown, code to ingredient; avoid guessing brands.

Outcome: Accurate, reproducible coding and clear traceability (verbatim + qualifiers → code). The indication also sets the ATC class: aspirin for cardiovascular prevention is an antithrombotic; for pain, an analgesic.

Example 3 — Convention change mid‑study (and how to control it)

Context: Mid‑study, the team clarifies a convention: an entry reporting two separate concepts (e.g., “headache and nausea”) will no longer be coded to the first term only. Each concept gets its own code, so DM queries the site to report the two events on separate lines.

Risk: Early records were coded inconsistently.

Action:

1. Raise a change request describing the convention update and the reason.
2. Assess impact: how many terms are affected? Which tables/listings change?
3. Once the sites have answered the queries, recode impacted records and run delta listings.
4. Perform targeted QC (e.g., 100% of changed terms).
5. Retrain sites and coders; update the convention manual with a new version.

Outcome: Uniform application of the updated rule with documented traceability.

Pitfalls & misconceptions

• Mixing dictionary versions inside the same pooled output Why it hurts: totals shift unpredictably. Fix: pick a target version; recode before pooling.

• Relying solely on auto‑coding Why it hurts: edge cases, abbreviations, and multi‑concept phrases are misclassified. Fix: manual review + risk‑based QC.

• Silent convention changes Why it hurts: inconsistent coding over time. Fix: use change control, retrain, and recode (“back‑code”) impacted data.

• Poor verbatim capture Why it hurts: ambiguity and rework. Fix: site training and form guidance (sites write terms in full: no “?”, no “rule out”, no abbreviations).

• No qualifiers for medications Why it hurts: wrong product/ingredient mapping. Fix: collect indication, route, dose, unit, frequency.

• Skipping delta listings after recoding Why it hurts: reviewers can’t see what changed. Fix: always produce and archive deltas.

Practice in the wild (checklist)

• Publish conventions with version control; train coders on them and sites on clear verbatim entry.
• Capture qualifiers (especially for medications): indication, route, dose, unit, frequency.
• Define QC sampling rules; track discrepancies and retrain if thresholds are missed.
• Set a program target version for pooling; plan recoding windows.
• Run delta listings for any dictionary upgrade or convention change.
• Archive before/after coding states with who/when/why (audit trail).
• Coordinate with vendors early on dictionaries, versions, and codelists to avoid rework.

Key points to remember

• Use the right dictionaries & record versions: MedDRA for clinical events; WHODrug for medications. Capture the exact version in the DMP/coding plan and audit trail.
• Align versions for pooling: Set a program target version early and recode to it before pooling to prevent shifting counts.
• Conventions reduce ambiguity: Document, version‑control, and train on rules for abbreviations, multi‑concept phrases, and non‑events so coding is consistent.
• Qualifiers make medication coding precise: Collect indication, route, dose, unit, frequency; code to ingredient when brand is unknown—avoid guessing.
• Auto‑coding accelerates; QC assures: Keep humans in the loop. Use manual confirmation and risk‑based QC sampling with clear thresholds; escalate unclear verbatims.
• Change control with evidence: For dictionary upgrades or convention updates, run an impact assessment, produce delta listings, apply targeted re‑QC, retrain, and archive before/after states.
• Plan early and coordinate: Align with vendors on dictionaries, versions, codelists, file‑exchange cadence, and audit‑trail expectations; capture agreements in the DMP/coding plan.
• Be audit‑ready: Maintain traceability for who coded, who QC’d, when, with which version and rationale; freeze coding at lock and archive conventions, decisions, QC results, change requests, and code versions. The coder and DM never edit the site’s verbatim; unclear terms go back as queries.