Country / Region
EMEA
Tags
Data analytics, Expression constraint language, Research, Tooling
The Royal College of General Practitioners (RCGP) Research and Surveillance Centre (RSC) has been monitoring disease incidence for fifty years and produces a weekly report with an emphasis on infectious respiratory disease. We collect data from over 2000 practices, representing 20 million patients or around a third of the population of England. Data is supplied using the UK Edition of SNOMED CT. Concepts to be extracted are defined using intensional valuesets using the “is a” hierarchy and can be expressed using the Expression Constraint Language. This can speed up the creation of valuesets and reduce the maintenance required. Inactivated concepts are included using the association refset, although this is by no means infallible and requires manual review
Description
We provide a national sentinel surveillance system using routinely collected primary care data. The main priority is infectious respiratory diseases – particularly acute respiratory infection. We have created a hierarchy of clinical conditions, defined using SNOMED CT, to identify disease and produce a weekly report which will typically show changes in the pattern of disease more quickly that lab based measures. The hierarchy allows both a granular view of disease, e.g. rates of bronchiolitis, or influenza-like illness, or a more general picture of acute respiratory infection.
Individual primary care practices in England have agreed that anonymised data can be sent Data is received from practice computerised medical record (CMR) systems twice weekly to the database and the report is generated from this. Data is also sent to the United Kingdom Health Security Agency (UKHSA) and forms part of their public health function.
Scope
Snomed CT has been the standard for primary care data in England since 2018. At that time all previous data in Read 2 or CTV3 was mapped to Snomed, although one system continues to use CTV3 internally. Snomed was therefore essential to this programme.
An intensional approach, using the Snomed Expression Constraint Language was selected as there was a reduction in the amount of maintenance of the valuesets that would be required as new editions of Snomed were released. Surveillance is a continuous process so a definition could be used weekly for a number of years and several different versions of Snomed. Whilst we have found that some review of changes is essential this is less than would be required if extensional valueset were used.
The existence of the Historical Association Refset further allows the automatic selection of inactive concepts that may still appear in the CMR. This can be very useful, although, at times, this can also be difficult to manage.
How SNOMED CT will be used
Data from primary care is supplied using the UK edition of SNOMED CT. This is a combination of historic data from before 2018 which has been mapped from Read v2, current data mapped from CTV3 and data directly entered with SNOMED CT. There is a small amount of metadata about encounters with the practice and some patient demographic details.
We used valueset for conditions which use the “is a” hierarchy from SNOMED CT. These will automatically incorporate changes that are made in the UK edition (i.e. the International Edition with the UK Extension). These are authored using an internally developed helper tool which is hosted on the same server as our data.
The valuesets are used alongside clinical logic to define an episode of disease. If two values from a valueset have been used within seven days of each other we consider that a single episode, starting with the earliest code entry
We use these valuesets to create a hierarchy of conditions, from the most specific to the most general. A patient may appear in more than one category at the lower level so that the total across each level may be different.
We seek to avoid having inactive terms in the definition. However, we do want to include inactive terms in the final list. Whilst they may not be available for use (and with mapped terms even that is not certain) they will remain in the record and need to be included in the extraction. We use the Historical Association Refset to find concepts equivalent to the active concepts that have been selected. This is equivalent to the {{+HISTORY-MOD}} expression in ECL.
A selection of valuesets have been uploaded to the NHS Terminology Server, which are available (after signing up) for anyone to use.
Why SNOMED CT will be used
Contact
