Coding clinical terms with MedDRA

Clinical coding converts free-text descriptions into standard terms that can be retrieved, grouped and analysed. MedDRA supplies a medical terminology and hierarchy for coding clinical data across the regulatory lifecycle.

From verbatim descriptions to coded data

MedDRA stands for Medical Dictionary for Regulatory Activities and is an international medical terminology used by regulatory authorities and the regulated biopharmaceutical industry from pre-marketing development to post-marketing surveillance. ¹ Sponsors or authorised companies code medical history, diseases and adverse events after data collection to support storage, retrieval, presentation and analysis. ²

An investigator's free-text description is a verbatim or investigator-reported term; descriptions of the same event can vary between investigators, sites, languages and spelling conventions. ³ Trained coding staff align each verbatim term with a MedDRA lowest level term (LLT), allowing several descriptions of one medical concept to be analysed together. ⁴ Statistics teams generate tables and listings from coded data, and medical writers present the results in clinical study reports and posters. ⁵

For example, “rash left arm”, “small rash on left lower arm” and “rash-like skin change lower left arm” can all map to the LLT “rash over arms”, which maps to the preferred term “rash”. ⁴ Aggregation avoids splitting the same medical phenomenon among categories based on its wording and supports comparison across regions. ⁴ Three events in an illustrative trial case were coded as follows: ⁶

Verbatim term	Lowest level term	Preferred term	System organ class
Rash on left lower arm	Rash over arms	Rash	Skin and subcutaneous tissue disorders
Fainted due to hypoglycaemia and standing up quickly	Hypoglycaemia	Hypoglycaemia	Metabolism and nutrition disorders
Sprained ankle	Ligament sprain	Joint injury	Injury, poisoning and procedural complications

⁶

Scope and hierarchy

MedDRA includes diseases, diagnoses, signs, symptoms, therapeutic indications, investigation names and qualitative results, medical and surgical procedures, medical, social and family history, medication errors, product quality issues, device issues and terms from other terminologies. ^{7, 8} It is not a drug dictionary and excludes patient demographic terms, frequency qualifiers, numerical results and severity descriptors. ^{8, 9} Thus “penicillin 500 mg orally three times daily” and “a 57-year-old male” fall outside its scope, whereas rheumatoid arthritis, joint pain, penicillin allergy, device-related infection and impaired driving ability fall within it. ¹⁰

The hierarchy has five levels, from specific observations to broad groupings: ⁹

1. Lowest level term (LLT): the coding level, reflecting recorded observations, synonyms, lexical variants, alternative word orders and abbreviations; each LLT links to one preferred term.
2. Preferred term (PT): a distinct medical concept, such as a symptom, sign, disease, diagnosis or therapeutic indication; every PT has at least one LLT, including itself.
3. High level term (HLT): a grouping of PTs related by anatomy, pathology, physiology, aetiology or function.
4. High level group term (HLGT): a grouping of HLTs within a system organ class.
5. System organ class (SOC): the broadest level, organised by anatomical or physiological system, aetiology, manifestation site or purpose, with a class for social circumstances.

^{11, 12}

Product issues became the 27th SOC in March 2016. ^{9, 12} Coding uses LLTs, while safety analyses commonly use PTs grouped by SOC. ^{4, 13} MedDRA terminology and the National Cancer Institute Common Terminology Criteria for Adverse Events serve different purposes: the latter grades severity, and its terms can be mapped to MedDRA without covering every MedDRA term. ^{14, 15, 16}

Primary and secondary classes

MedDRA is multiaxial: one concept can be represented in several SOCs, permitting retrieval and presentation through different classifications. ¹⁷ The links are predefined and should not be changed locally; a primary-SOC change requires an MSSO change request. ^{17, 18} Each PT has one primary SOC and may have secondary SOCs; primary allocation prevents double counting across SOCs in an analysis. ¹⁹ For example, influenza has infections and infestations as its primary SOC and respiratory, thoracic and mediastinal disorders as a secondary SOC. ¹⁹

Primary allocation normally follows the manifestation site for diseases, signs and symptoms, with exceptions for congenital or hereditary anomalies, infections and neoplasms. ²⁰ Congenital and hereditary anomalies belong primarily to congenital, familial and genetic disorders, infections to infections and infestations, and neoplasms to neoplasms benign, malignant and unspecified, including cysts and polyps. ²⁰ A cyst or polyp with a specific site instead has the manifestation-site SOC as primary and the neoplasm SOC as secondary. ^{20, 21} Where several exceptions apply, priority is congenital, familial and genetic disorders first, neoplasms second, and infections and infestations third; secondary SOC allocation has no priority order. ²¹

PTs in investigations, surgical and medical procedures, and social circumstances occur in only one SOC and are not multiaxial. ²² Secondary-class analyses can provide a fuller view of events dispersed among primary classes. ¹⁷ They cannot resolve every separation: “platelet count decreased” belongs to investigations, while “thrombocytopenia” belongs to blood and lymphatic system disorders. ²³

Term selection

Coding quality depends on source-data quality, consistent procedures, quality assurance and review by qualified people. ²⁴ MedDRA has a predefined hierarchy; a concept not adequately represented requires an MSSO change request rather than a local alteration. ²⁵

The selected term should be the current LLT that most closely and accurately reflects the verbatim description, using medical judgement when no exact match exists and checking the hierarchy above it. ^{25, 26} Ambiguous wording should first be clarified with the reporter: a report of “cold” should not be assigned to a disease or a sensation without clarification. ^{27, 18}

All reported information should be coded without adding an unreported diagnosis. ²⁶ For abdominal pain, increased serum amylase and increased serum lipase, the appropriate terms describe those three observations; coding pancreatitis would add a diagnosis not reported. ²⁶ Separate terms can preserve information in combined descriptions: diarrhoea and vomiting require terms for both, and a wrist fracture due to a fall requires terms for the fracture and the fall. ²⁸ Medical judgement is needed when splitting a description so that information is not lost. ^{28, 29}

When no term combines an event and its body site, the event generally takes priority: skin rash on the chest can be coded as skin rash. ²⁹ The body site can take priority when it changes the meaning; for cyanosis at an injection site, “injection site reaction” preserves the local nature of the event better than the general term cyanosis. ³⁰ If an event occurs at several sites and their LLTs share one PT, one LLT should capture the event as accurately as possible: rash on the face and neck can use skin rash, and oedema of the hands and feet can use oedema of extremities. ^{30, 31}

Death, disability and hospitalisation are patient outcomes and are not usually adverse reactions or events themselves; an outcome term is appropriate when it is the only reported information or adds significant clinical information. ^{32, 28} Neoplasm coding distinguishes cancer and carcinoma as synonyms, tumour as neoplasia, and lump or mass as not necessarily neoplasia. ³³ Unclear neoplasm type should be clarified with the reporter, and difficult or unusual neoplasms warrant medical expertise and the MedDRA Introductory Guide. ³³

Investigation results should be coded individually as reported rather than combined into an unreported inclusive term. ^{33, 34} “Abnormal liver function test” supports that LLT, whereas separately reported increases in alkaline phosphatase, SGPT, SGOT and LDH require four terms. ³⁴

Coding specifications and versions

Automatic coding still requires manual coding and review, because coders differ in medical knowledge and MedDRA offers many term choices. ³⁵ Abbreviations can defeat automatic matching: “allergic to CATS” may be matched to cats when CAT refers to a computerised tomogram. ³⁶ Companies should base their coding specification on the MedDRA Points to Consider documents and define the following: ^{36, 37}

• Medical terms to be coded, such as adverse reactions, adverse events and medical history.
• Coding dictionaries.
• Coding process.
• Coding review and approval.
• Handling of non-codable items.
• Frequency and timing of coding runs.

^{36, 37}

MedDRA has two scheduled releases annually: X.0 on 1 March, with simple and complex changes at all levels, and X.1 on 1 September, with simple changes at LLT and PT levels. ^{37, 38} The version used in an analysis should be documented because a PT's primary SOC can change between versions, complicating comparisons and interim analyses in long studies. ^{38, 39} A clinical-coding training recommendation retains the version chosen at study start until database lock and upgrades at study end only if the coding specification permits; safety databases instead update twice yearly and code with the latest version. ²⁴ A 2014 regulatory-writing recommendation favoured the most recent version for each analysis. ³⁸

New conditions can prompt additional updates: COVID-19 terms added in the first half of 2020 included infection, symptoms, immunisation and testing. ⁴⁰ For studies whose adverse events were already coded, the decision whether to recode with the new COVID-19 terms depended on the trial. ⁴¹ The MedDRA Version Analysis Tool compares any two versions and supports transitions between versions; a mapping tool supports interoperability with other terminologies. ³⁵

Browsing and non-current terms

MedDRA provides desktop and web browsers for searching and browsing, alongside guides, training and queries. ²² The desktop browser allows selection of version and language, display of the SOC hierarchy and queries, multilingual searching and inspection of term details and query membership. ^{34, 42} Support documents accompany each version, and the term-selection guideline is version-specific. ^{42, 43} The official MedDRA website supplies information on use, subscriptions, news, events and training, with free training and webinars for subscribers. ⁷

Non-current terms are flagged at LLT level, are not recommended for continued coding and remain available for historical retrieval and analysis rather than being deleted. ^{43, 44, 45, 27} They include vague, ambiguous, outdated, truncated or misspelt terms and terms from other terminologies that do not fit MedDRA rules. ⁴⁴

Standardised MedDRA Queries

Standardised MedDRA Queries (SMQs) are validated sets of terms representing safety topics of regulatory interest for signal detection and monitoring. ^{46, 35} Some are simple PT sets and others have subordinate queries, such as hearing impairment and vestibular disorders at a second level; searches can operate at PT or LLT level. ^{46, 47}

Most SMQs have narrow and broad terms, while some have only one scope. ⁴⁷ Narrow terms are highly likely to identify the condition: pancreatitis acute and pancreatitis haemorrhagic are narrow terms in the acute pancreatitis SMQ. ⁴⁸ Broad terms are compatible but less specific: blood bilirubin increased does not always indicate acute pancreatitis. ⁴⁸ Narrow-only retrieval is an option, and some SMQs use algorithms in which multiple PTs within one case improve case identification. ^{48, 10}

An SMQ and its subordinate terms have active or inactive status; an active SMQ is maintained by the MSSO, while an inactive SMQ is no longer maintained. ^{47, 48} Inactivated PTs and LLTs remain in the query rather than being deleted. ⁴⁸ Standardised queries are predefined and should not be tailored by removing individual terms considered unhelpful. ^{49, 23}

SMQs replaced special search categories and are developed collaboratively by CIOMS and ICH, with updates accompanying the six-monthly terminology releases. ²³ Users may request new queries, for example for a novel adverse effect, but the committees determine their content rather than individual companies. ²³ Queries retrieve matching events across classes, including related concepts that secondary-SOC analysis cannot combine; clinical judgement is then needed to determine whether the retrieved cases constitute a significant signal. ²³

Coding reliability and granularity

Coding detail can support precise categorisation while also splitting a broader safety concept among terms. ^{50, 39, 51} A 2014 overview compared approximately 1,200 COSTART terms with about 18,000 MedDRA PTs and 66,000 LLTs. ⁵⁰ If a frequency table includes only events with incidence of at least 5%, several related granular terms can each fall below the threshold even when their broader concept would qualify. ⁵⁰ Secondary-SOC analyses and SMQs can partly address this fragmentation. ^{50, 23}

Published validation evidence was limited in the systematic review discussed in that 2014 overview; one study comparing two blinded coders found different coding for 12% of verbatim terms, and investigator training in recording terms could improve coding quality. ⁵⁰ A regulatory writer's assessment was that independent coding before analysis and documentation of coding queries limited opportunities to influence outcomes through coding. ⁵⁰ The same assessment described little available information in 2014 on how well coding detected unsafe drugs before approval or how often potentially useful drugs were discontinued. ³⁹ It proposed comparing development discontinuations for safety with post-approval withdrawals, and examining pre-approval data for withdrawn products, as disclosure improved. ⁵⁰

Medical writers should document adverse-event coding and provide a glossary mapping investigator verbatim terms to the PTs selected. ⁵⁰

Languages and report wording

Shared numerical codes allow MedDRA concepts to be matched across languages, and coding in a user's native language can improve accuracy and facilitate multinational data sharing. ^{52, 7, 53} LLTs include both American and British spellings, while PTs generally use British spelling. ^{54, 13} In narrative text, American spelling can follow the document's language, and catalogue word order can be adjusted for readability, such as “abnormal acid base balance” or “right bundle branch block”. ^{13, 55} Where statistical tables use British spellings and the report uses American English, manually changing every table is unnecessary; a general note should explain the spelling difference and its effect on alphabetisation, as with oesophageal and esophageal cancer. ⁵⁵

Development and maintenance

Before MedDRA, COSTART, the International Classification of Diseases, the WHO Adverse Reaction Terminology and company-specific dictionaries hindered comparison and pooling of safety data and could require recoding for different regulatory regions. ⁵⁶ A working group of regulators from the United Kingdom, Spain and France and industry representatives developed its precursor; CIOMS proposed global adoption in 1994, and ICH took it into the M1 multidisciplinary initiative. ⁵⁶

The ICH established the MedDRA Management Board in May 1998, and the Japanese Maintenance Organization was established in January 1999. ^{2, 57}

The Maintenance and Support Services Organization (MSSO) maintains, develops and distributes MedDRA. ⁵⁸ MSSO user-group meetings gather feedback, and Blue Ribbon Panels bring industry and regulatory experts together to recommend solutions to difficult terminology issues. ⁵⁹ An ICH advisory panel on SMQs works with CIOMS on support for pharmacovigilance signal detection and monitoring. ^{59, 52}