search

OHDSI / Characterization

Various types of characterizations of a target and outcome cohort.
https://ohdsi.github.io/Characterization/
3 stars 5 forks source link

Event breakdown for outcomes in incidence rate analysis #52

Open conovermitch opened 7 months ago

conovermitch commented 7 months ago

This is a request for new data / new functionality in Strategus that is sometimes of-interest; to be clear this is "nice to have" and likely low-priority relative to other submitted Strategus issues. A frequent request from those using evidence generated by Strategus is to describe the breakdown of clinical concepts that contribute to the outcome definition. More specifically, the analysis that would be of interest would be to provide the frequency of occurrence of any concept from a concept set used in the highest level (i.e. not nested) events that qualify for cohort entry. It would be of interest to see the breakdown: 1) by person and by record; and 2) by standard codes and source codes. To the extent possible, it would be nice for the display and statistics to align as closely as possible with the current "index event breakdown" functionality within Strategus.

It is straightforward to provide this breakdown for the entire outcome cohort (i.e. not subset to target or indication) if we enter the outcome cohort as a target in CohortDiagnostics and inspect the "index event breakdown" tab. However, the limitation of such an approach is that this breakdown will not subset to patients who are in our target (e.g. exposure) cohort or in our indication/age/calendar time subsets applied within Strategus. In particular, we can imagine drawbacks to that approach when the outcome is extremely common (e.g. infections occurring in the inpatient setting) outside of our population of interest and may have a notably different breakdown of qualifying events once we subset to the target/indication cohorts.

For the few occasions where this analysis is requested, the current the approach that we use is that we produce a derivation of the outcome cohort which subsets to patients with prior target/indication/continuous enrollment/1st in history requirements. With enough tweaking you can often get the numbers to align exactly with the incidence rate analysis in Strategus but doing so requires time-consuming iteration. This cohort, when indexed on the outcome occurrence, can be used in conjunction with a characterization analysis (with features for 0d to 0d) to get an approximate breakdown of codes included in key outcome concept sets.

So to be clear, there are approximate / time consuming approaches to getting this information with the current functionality. However, I think it would be a relatively low-lift to produce this automatically and include the information in the Characterization tab of Strategus. Please reach out to me (mconove1@its.jnj.com) if I can provide more clarity.

anthonysena commented 2 months ago

@conovermitch - this seems like functionality specific to the incidence rate and in order to have this function available we'd need to have the cohort definition(s) involved (since the concept set definitions are part of the Circe definition) along with the incidence rate settings for time at risk, etc. Do I have this right? This is likely out of scope for Strategus but we can potentially move this to CohortIncidence for review.

conovermitch commented 2 months ago

@anthonysena Thanks for reviewing the issue. I think you are right - this may be specific to the CohortIncidence package but I'm not totally sure (I would defer to you). It seems like the ideal place for it to live within the Shiny App Viewer would be the "Exposed Cases Summary". Clarifying question - is the "Exposed Cases Summary" tab within the Characterization module being managed by the CohortIncidence package? I would think this issue/feature request should be assigned to whichever module is responsible for generating the "Risk Factor" and "Case Series" analyses, since it would likely use a lot of the same machinery to breakdown features that occur on the day of the outcome occurrence (which would need to, as you mentioned, be connected back to the concept set used in the outcome cohort definition so the analysis could subset to those concepts).

Let me know if you would like me to re-post this issue elsewhere - happy to do that if helpful. Again, this is a "nice to have" request but definitely not urgent. Just wanted to share since it's an analysis I've been asked for a handful of times.

anthonysena commented 2 months ago

Let me move this over to the Characterization repo as I suspect that package will be the one that can help with this use case. Tagging @jreps @chrisknoll for awareness.