Identifying individual health events

[SRW612]

Our study requires counting the number of events satisfying a given codelist (e.g. codelist for RTIs). For example, if patient X attends their GP on two different occasions during the study period, and are coded as having an RTI both times, we need to count both events for that patient. However, the study_definition.py template is set up for patient-level data. Is there a way to derive counts of events rather than patients by either

1. defining the study population at event-level, where the two events for Patient X would be represented by two rows.
   OR
2. identifing multiple events for a given patient from patient-level data, where Patient X is represented by one row, but the two events for that patient are represented by other variables in the row, e.g. Event1='CTV3 code for RTI', Event2='CTV3 code for RTI' ?

[alexwalkerepi]

Currently there isn't a straightforward way to return all the individual events themselves. We hope to add this functionality eventually.

However, if you only need to count the individual events, you can return the count of recorded events within a given time period by using the returning="number_of_matches_in_period" option, which will return an integer count, as documented here:
https://docs.opensafely.org/study-def-variables/#cohortextractor.patients.with_these_clinical_events

Does that fulfil your use case?

[SRW612]

Many thanks Alex, I think the with_gp_consultations function will work for our needs.

[SRW612]

Just to clarify, can this approach be used to count consultations defined by a particular diagnosis, e.g. only consultations with a respiratory infection diagnosis? If not then I assume I have to look into Will's suggtions below?

[alexwalkerepi]

For that you'll need the patients.with_these_clinical_events function, which you use with a codelist to define the condition you're interested in. That also has the returning="number_of_matches_in_period" option.

[SRW612]

Brilliant, I think that will do it.

[wjchulme]

Option 1 is not currently possible -- see this issue and the issues linked therein for more info about how we're working to resolve this.

Option 2 is possible with a bit of guile. See section Extracting a series of consecutive events in this issue. WARNING I wouldn't recommend this approach if you want to return a large number of events. This would mean you need to create lots of variables (columns) which means more queries and slower extraction time. It might be useful for you to talk through your use-case in a bit more detail first.

[SRW612]

Thanks for the alternative appraoch. If the with_gp_consultations function isn't right for our study I will try your suggestions for option 2.