Thank you very much for joining out first Open Health Stack Community Call.
If you missed the session and you would like to catch up on the discussion, you can find the recording here.
For the questions that we did not manage to address on the call, @jing and @bashir will be addressing them in the respective categories on the forums, so please be sure to follow the conversations.
If you would like to make a presentation on your implementation in future community sessions, please reach out to us and we would be glad to schedule this with you.
We look forward to engaging with you in future sessions!
Thanks @Sylvia for the post, there was just one question that I remember for the OHS Analytics component, which I briefly answer here (hopefully it is okay to use this same thread):
The question was about the performance of the query engine and how it impacts the Text-to-SQL-on-FHIR framework. It is actually critical to have a performant query engine for Text-to-SQL because the framework runs SQL queries on the dataset multiple times to come up with the right query.
Here are the slides for my FHIR DevDays presentation. There is some data about performance, in particular on slides 21 and 24. The MIMIC-IV dataset with ~300K Patients and ~461M Observations is used and it is shown queries with JOINs on Observation views take less than 10 seconds (obviously it depends on the machine spec).
In general, one of the main goals of the fhir-data-pipes component of OHS is to convert FHIR resources into a format/structure for performant population level queries. For details, please see the documentation or the repo.