OHDSI Home | Forums | Wiki | Github

Defining Cardiovascular related death in RWD

(Kristin Kostka, MPH) #1

@BridgetWang and I are revisiting the Population Level Estimation study on Xeljanz that we started at the F2F at Columbia last May – which is an attempt at recreating parts of this study. We are focusing on Major Adverse Cardiovascular Events (MACE). Since it’s a composite, we’ve broken it into 3 separate Os: acute myocardial infarction, stroke and cardiovascular-related death. We have 2 of these from LEGEND but we’re struggling with cardiovascular related death and could use some support from the community.

So we’re asking the community: does anyone have an existing O cohort definition for cardiovascular related death? We think this endpoint is valuable in recreating a MACE definition but we’re not entirely sure if this is well captured in real world data. Any advice on how others have approached this would be super helpful. We’d like to pull from the knowledge of the crowd here (cough cough: @jon_duke @hripcsa @rchen @Patrick_Ryan @Gowtham_Rao etc etc :smiley:)

(Seng Chan You) #2

I usually use MACCE (Major Adverse Cardio-Cerebrovascular Events) for the primary outcome, including heart failure, MI, and stroke. And I released my definitions of stroke in the OHDSI phenotype library git repo.

In NHIS-NSC (National Sample Cohort), the cause of death is recorded.
So I thought to make prediction model to choose CV-related death by using this labeled data. Again, due to my laziness, I’ve never tried this project, yet.
If you need this, we can try.

(Kristin Kostka, MPH) #3

Welcome back, @SCYou! We missed you!! :smile:

Thanks for sharing these resources. Very helpful!

Unfortunately, we don’t have explicit labels in a lot of US data sets for CV-related death. This is our unique challenge. Trying to figure out if there are reliable ways to define this cohort in the absence of a formal label.

(Seng Chan You) #4

Thank you for warm message, @krfeeney
I admit that the post I’ve written is somewhat confusing.

Actually, what I meant was a kind of simple PLP model for CV death, which is externally usable.

For example, if the patient has myocardial infarction during short-term period before death, and does not have any history of malignancy or other life-threatening condition, we can assume that his/her mortality was caused by cardiovascular disease (It would be more definite if he/she had a history of ECMO during short-term period before death).

(Kristin Kostka, MPH) #5

Ahhh… that makes sense! Well this is definitely an intriguing option and something that seems like it could be reusable to the community. I’ll chat with @BridgetWang.

Our biggest constraint in this particular study is completing the analysis and submitting our paper before the Phase IV trial results are published. May not have enough time to do a nested PLP in here but I love the idea and think this endpoint is something that comes up quite a bit. It’s a really tough thing to create in RWD.

For the record, you are the least lazy person I know! :wink:

(Seng Chan You) #6

We’ve started to make the PLP model to predict mortality and its cause when patient’s observation period was stopped (or based on his/her last visit_end_date) on the claim data.