Beyond the Tick Box: How SNOMED Codes Turn a GP Appointment into Health Data
Photo: U.S. District Court for the District of Massachusetts. 1789, Public domain, via Wikimedia Commons
On a wet November morning in 2026, a patient in a suburban surgery describes a cough that has lasted three weeks. The GP listens, examines and types. By the end of the ten-minute appointment, the practice record contains a problem-list entry, a prescription, a set of observations and a few lines of free text. Much of that record will be structured with SNOMED CT codes. For anyone interested in how disease becomes digital data, the GP consultation is where the process begins.
The code as the hinge
SNOMED CT is a clinical terminology, not a classification. It contains concepts, synonyms and hierarchical relationships, allowing a GP to record ‘asthma’ at different levels of granularity and to attach it to a problem list, a prescription or a review. In England, GP systems such as EMIS Web, SystmOne and Vision use SNOMED CT; legacy Read codes have largely been replaced. When a clinician selects a code, the note becomes computable: it can be searched, extracted, aggregated and linked.
The Quality and Outcomes Framework adds another layer. QOF indicators reward practices for recording certain data—blood pressure in hypertension, smoking status, asthma reviews and so on. The template is a prompt, not a script. It improves completeness but can also produce a record that reflects the indicator set as much as the consultation. A blood pressure reading may be recorded because it is clinically needed and because it is incentivised; the data do not always distinguish between the two.
Coded and uncoded health data
Coded data are structured and extractable. Uncoded data include free text, scanned letters, attachments and images. Free text can carry the nuance that codes lose: ‘possible viral wheeze, not convinced asthma’, ‘patient declines statin’, ‘social circumstances affecting adherence’. A research extract built only from codes may miss a diagnosis mentioned in a letter or a risk factor described in a note. Natural language processing can help to classify free text, but it is not infallible. Uncoded does not mean unimportant; it means invisible to simple queries.
Where errors and gaps creep in
Errors are not exotic. A clinician may choose a broad code when a specific one exists, or a ‘rule-out’ code that later becomes ambiguous. Copy-forward can perpetuate an old problem. Historical diagnoses may remain on the list after resolution. Template prompts can produce entries that satisfy a target but do not capture severity. At practice level, coding habits vary; at system level, migration between software can create artefacts.
For research, code lists define cohorts. If a study’s definition of diabetes uses one set of codes and a practice uses another, patients can be misclassified. Data quality is not a single score: completeness, accuracy, timeliness and consistency all matter, and missingness is often patterned rather than random. Patients who do not attend reviews, or whose care is shared across settings, may be less completely represented in the record than the clinical reality warrants.
From consultation to secondary use
Secondary use is the term for using health records beyond direct care—for research, planning, safety monitoring and commissioning. A GP record can be extracted, pseudonymised and placed in a secure data environment. Linked datasets, such as those used by the Clinical Practice Research Datalink or OpenSAFELY, allow analysts to study medicines, diagnoses and outcomes at scale. But ‘research-ready’ is a process, not a property. Data must be cleaned, validated and interpreted with the original clinical context in mind.
Consent is important but not the only legal basis. The NHS handles confidential patient information under the common law duty of confidence, data-protection law and specific approvals. The NHS National Data Opt-out allows patients to register a preference that their confidential patient information is not used for research or planning, subject to exceptions. Research using confidential patient information may require approval from the Health Research Authority and, in some cases, the Confidentiality Advisory Group. In 2026, the debate is less about whether secondary use should happen than about how it is governed, explained and trusted.
What the record says about you
A patient looking at their GP record should understand that it is both a care document and a data source. The coded entries are not a complete biography; they are a series of decisions made by clinicians, administrators and software. Patients can ask what is coded, whether free text is used in research, who can access the data and how to opt out. Curious readers should treat GP data as evidence with a provenance, not as a neutral transcript of illness.
The coded consultation is one of the most consequential pieces of digital health infrastructure in the UK. It supports care, payment, audit and research at once. Its value depends on the quality of the coding and the honesty of the interpretation. That is the lesson of digitised disease: the data are made in the clinic, not found in the wild.
