Auto‑coding & Good Verbatims

Why this matters

When sites enter clear, clinical phrases (verbatims), your coding tool can auto‑match them to MedDRA terms with little or no manual work. That means:

• Fewer coding backlogs and fewer data queries back to sites.
• Cleaner analyses—because each record aligns to a single, well‑defined medical concept.
• Faster safety review—signals aren’t hidden behind vague phrases or raw numbers.

The opposite is also true: numbers alone (“180 mg/dL”), slang (“sugar high”), and stacked terms (“N/V/D x3 days”) confuse auto‑coding and end in manual coding or a query back to the site.

Core concepts & definitions

Plain‑language glossary (minimal abbreviations)

• Verbatim term — The exact text typed by the site to describe what happened (e.g., “Nausea”, “Injection site pain”). It is site data: the coder and data management (DM) never edit it; if it cannot be coded as written, DM sends the site a non‑leading query.

• Auto‑coding — The coding system matches the verbatim against dictionary terms and an approved synonym list (verbatims coded earlier) and proposes the code; a trained coder confirms the match and codes what did not match. The hit rate (first‑pass yield) is the share that auto‑codes: e.g., 52 of 80 is 65%.

• MedDRA — The medical dictionary commonly used for adverse events. It has multiple levels; two matter most here:

  • LLT (Lowest Level Term) — The dictionary entry coders usually select.
  • PT (Preferred Term) — The roll‑up term used for analysis and summaries.

• Clinical concept — A recognized medical idea such as a symptom (“Headache”), sign (“Fever”), diagnosis (“Pneumonia”), or event (“Fall”).

• Stacked concept — A string that contains more than one clinical concept (“Headache and fever”). These tend to fail auto‑coding; the site records each concept as a separate entry when the CRF expects one concept per line. A diagnosis reported with its own signs and symptoms is not stacked: the diagnosis alone is recorded.

• Ambiguous abbreviation — A short form that could mean different things (“MS” could be multiple sclerosis or mitral stenosis). Avoid these in verbatims.

Key idea: Verbatims should express one clear clinical concept. Numbers, time spans, and severity belong in dedicated fields—not inside the verbatim. This is guidance on how to write, not which term to pick: sites use plain medical words and need not know MedDRA.

The process, step‑by‑step

Goal: Build good phrasing into your training, CRF, and edit checks so sites succeed on the first try.

1. Write it into the CCG (CRF Completion Guidelines). Include a one‑page style guide with “Say this / Not this” examples and a short list of banned abbreviations (e.g., SOB, N/V/D, CP, HTN). In the verbatim field every term is written in full; abbreviations allowed elsewhere on the CRF are listed in the CCG with their meanings.

2. Teach the pattern during site training, before sites get EDC access.

   • Use symptom words when a diagnosis is not confirmed (“Shortness of breath” or “Dyspnoea” instead of “SOB”).
   • Use diagnoses when confirmed by an investigator (“Pneumonia”).
   • Use event terms for things that happened (“Fall”, “Infusion site extravasation”).
   • Keep one concept per record. If the patient had nausea and vomiting, create two verbatims if the CRF is structured that way.

3. Avoid common traps.

   • No raw numbers or units as the verbatim (“Glucose 180 mg/dL”).
   • No vague phrases (“Felt bad”).
   • No stacked concepts (“Headache/fever/chills”).
   • No local slang or unexpanded abbreviations.

4. Use on‑screen prompts in the EDC. Example help text to paste in the CRF field:

   Enter a single clinical term (e.g., “Nausea”, “Dyspnoea”, “Injection site pain”).
   Do not enter only numbers or units. Avoid abbreviations (e.g., SOB, N/V/D).
   If multiple symptoms occurred, record each as a separate entry.
   

5. Add simple edit checks.

   • Soft check (a warning the site can answer; it does not block saving): Flag entries that are numbers‑only or contain slashes suggestive of stacked concepts (e.g., “N/V/D”).
   • Info prompt: If the term includes only a lab value, prompt: “Please describe the finding in words (the diagnosis, or the test and the direction of change), not the value alone.”

6. Review early, then coach. In the first month of enrollment, sample 50–100 recent verbatims, measure the auto‑coding hit rate, and send a 1‑page “Top 5 Fixes” back to sites (with before/after examples).

What good looks like

• Clear, clinical terms: “Nausea”, “Dyspnoea”, “Hyperglycaemia”, “Injection site pain”, “Syncope”.
• No abbreviations: Write the full term (“Dyspnoea”, not “SOB”).
• One concept per verbatim: Split combined symptoms.
• Keep metadata out of the verbatim: Duration, severity, and units go to their own fields.
• Dictionary‑friendly spelling: Use standard medical spelling. MedDRA holds both British and American spellings as LLTs (“Diarrhoea”, “Diarrhea”), so either codes; a misspelling may not.

Worked examples

This section is self‑contained. It repeats minimal definitions so you don’t need to read elsewhere.

How matching works: Auto‑coding maps your text to a MedDRA LLT which rolls up to a PT used in analysis. The examples below show Bad → Better, with a brief Why.

1. Metabolic / Lab finding

• Bad: “sugar high 180”
• Better: “Hyperglycaemia” or “Blood glucose increased”
• Why: Numbers alone don’t express a clinical concept. “Hyperglycaemia” is a concept MedDRA recognizes. Diagnosis or test result (“Blood glucose increased”) is the investigator’s choice: they code to different MedDRA terms.

2. Respiratory symptom

• Bad: “SOB”
• Better: “Shortness of breath” or “Dyspnoea”
• Why: “SOB” is ambiguous and often fails auto‑coding. Either full term matches a MedDRA LLT directly; sites need not use the dictionary’s wording.

3. Non‑specific feeling

• Bad: “Felt bad”
• Better: “Fatigue” (if tiredness), or “Malaise” (if general discomfort), whichever the investigator documents
• Why: Use a clinical term that captures the symptom observed.

4. Stacked gastrointestinal symptoms

• Bad: “N/V/D x3 days”
• Better (split): “Nausea”; separate entry: “Vomiting”; separate entry: “Diarrhoea”
• Why: One concept per verbatim. Duration belongs in separate fields.

5. Injection site

• Bad: “Pain where shot was given”
• Better: “Injection site pain”
• Why: Dictionary‑friendly phrasing improves matching.

6. Event vs cause

• Bad: “Fall due to dizziness”
• Better: “Fall” (and, if captured separately) “Dizziness”
• Why: Don’t combine event and symptom. Each is a concept.

7. Device / procedure issue

• Bad: “IV blew”
• Better: “Infusion site extravasation”, if that is what the investigator documents
• Why: Use the recognized clinical term for the event.

8. Diagnosis only when confirmed

• Bad: “Flu?” or “possible flu”, when the investigator has not made the diagnosis
• Better: “Fever”; separate entry: “Cough”; separate entry: “Myalgia”
• Why: Do not coach to an unconfirmed diagnosis. Capture symptoms until a diagnosis is documented. Once the investigator diagnoses influenza, the site records that diagnosis in place of its symptoms; DM never asks for a diagnosis to be broken into symptoms.

9. Skin reaction

• Bad: “Rash on arm after drug”
• Better: “Rash on arm”; (causality is assessed elsewhere)
• Why: The verbatim is the clinical concept; relationship to study drug is assessed in other fields. Detail that changes the medical meaning stays (“Injection site pain”, not just “Pain”): accuracy comes before yield.

10. Vital signs only

• Bad: “BP 90/60”
• Better: “Hypotension” or “Blood pressure decreased”, as the investigator judges
• Why: Numbers alone are not a concept. The reading belongs on the vital signs form; it is an adverse event only if the investigator judges it clinically significant under the protocol.

Note: Exact PT names can differ by MedDRA version; the examples are phrased to be dictionary‑friendly across versions.

Pitfalls & misconceptions

• “Lab values alone are valid AE terms.” No. Record the clinical concept (e.g., “Blood potassium increased” or “Hyperkalaemia”). The value stays on the lab form; whether an abnormal result is an adverse event at all is the investigator’s call, not DM’s.
• “It’s okay to bundle symptoms to save time.” Bundling causes ambiguity and mis‑coding. Add separate entries if the CRF expects one concept per line.
• “We can guide sites to write the diagnosis we prefer.” Do not coach the medical conclusion. Use symptoms unless a diagnosis is documented by the investigator.
• “Severity, duration, and units belong in the verbatim.” Keep these in their designated fields; verbatim should be just the concept.
• “Abbreviations are fine if the site understands them.” Auto‑coding and downstream reviewers may not. Use full clinical terms. A coder may read through an unambiguous abbreviation; an ambiguous one becomes a query.

Practice in the wild (checklist)

• Add a one‑page style guide to the CCG with Say this / Not this examples.
• List banned abbreviations (e.g., SOB, N/V/D, CP, HTN) and give their acceptable full terms.
• Configure EDC prompts: on‑field help + soft checks for numbers‑only and slashes.
• Monitor auto‑coding hit rate in Month 1; share a “Top 5 Fixes” with sites.
• Sample early verbatims weekly for drift; retrain if needed.
• Query stacked entries when cleaning: the site splits them into separate entries. DM and coders never edit, split, or retype a verbatim.
• Document dictionary version (e.g., MedDRA 27.0) in the data management plan (DMP) and coding plan.

Key points to remember

• A verbatim should express one clear clinical concept (symptom, sign, diagnosis, or event) that maps cleanly to MedDRA.
• Use dictionary‑friendly, fully spelled‑out terms; avoid ambiguous abbreviations and local slang.
• Do not enter numbers‑only or raw vital/lab values. Capture the clinical concept instead (e.g., “Hypotension”, “Blood glucose increased”).
• Do not stack concepts. Record each concept as a separate entry when the CRF expects one concept per line.
• Keep duration, severity, units, and causality out of the verbatim; place them in their designated fields.
• Use symptoms until the investigator makes a diagnosis, then record the diagnosis in place of its symptoms; don’t coach the medical conclusion.
• Use standard medical spelling: British and American spellings both code; misspellings may not.
• Bake quality into operations: include a one‑page style guide and banned‑abbreviation list in the CCG, add on‑field prompts and soft checks in the EDC.
• Monitor early auto‑coding hit rate and send brief coaching with before/after examples to raise first‑pass yield.
• Strong, clinical verbatims improve first‑pass auto‑coding and let coders focus on true edge cases. DM and coders never edit a verbatim: unclear entries go back as queries.

Valid AE reported terms are clinical concepts that facilitate auto‑coding (e.g., “Nausea”, “Hyperglycaemia”); numeric values alone or vague phrases impede coding.