The first step is to clearly define the clinical feature of interest and establish inclusion and exclusion criteria. This requires clinical input, particularly from general practitioners (GPs) who are best placed to understand how clinical features are coded in a primary care setting. For rare conditions, which GPs encounter infrequently, it may also be important to get clinical input from hospital specialist doctors. Reliable sources of clinical information should be used, for example:
International Classification of Primary Care (ICPC), which defines symptoms and diagnoses, provides synonyms for them and, importantly, lists what should be excluded from the definition.10
The BMJ Best Practice guidelines (http://bestpractice.bmj.com/best-practice/welcome.html).
National Institute for Health and Care Excellence (NICE) Clinical Knowledge Summaries (http://cks.nice.org.uk/).
ICD-10 (http://apps.who.int/classifications/icd10/browse/2016/en) – this is less useful for symptoms, as it focuses on diseases.
Medical Subject Headings (MeSH) (https://www.nlm.nih.gov/mesh/2016/mesh_browser/MBrowser.html).
National Health Service (NHS) Digital Technology Reference data Update Distribution: https://isd.digital.nhs.uk/trud3/user/guest/group/0/home. Downloadable technology reference files including READ Code Browers with cross-map files.
Other potential resources include patient support groups, online discussion forums and already published codelists (eg,
https://clinicalcodes.org). Hierarchical classifications such as Read, Systematized Nomenclature of Medicine (SNOMED) or ICD-10 may be useful for identifying additional search terms and synonyms.
For some symptoms, it is necessary to tailor the definition to the context of the disease under investigation. Abdominal pain is a good example, where pancreatic disease may cause pain in the epigastrium and left hypochondrium, whereas disorders in the sigmoid colon generate pain in the left iliac fossa.
