OHDSI Home | Forums | Wiki | Github

Setting Comorbidities in ATLAS

Hi all,

I have a question about Setting Comorbidities.
I tried to put comorbidities like this.

initial event 
a drug exposure of ACE inhibitor  

inclusion criteria
` a condition exposure of diabetes
    where event starts between 60 days Before and 0 days Before index start date`

But this one can’t catch the diabetes that occurred in 90 days before index start date and still going on.
Is there any better idea to get a comorbidities?

plz help me

@11123:

Why don’t you set the time when the event starts to 90 days, then? Or any time before? Usually, diabetes doesn’t go away.

What if you set it to ‘event ends between 60 days before and all days after index start date’? I think you’re saying you want the diabetes episode to overlap with the period of time 60d before and 0d before the index but at the same time it’s ok if the diabetes went past the index date, so you just want 60d before and all days after index (note: you want the event (diabetes) end date to do this). If you also want to say soemthing about the diabetes start date, you can click the ‘add additional’ next to the ‘before-after’ part of the criteria and specify an additional date check using the event start/end date and the index start/end date.

Edit! I made a logical msistake, if you look for an end date of 60d before to all days after, you may find diabetes diagnosis that occurred complete after index! no Good! So:

inclusion criteria
 a condition exposure of diabetes
    where event starts between all days Before and 0 days Before index start date
       and event ends between 60 days before and all days after index

The addition of ‘event starts before index’ means that you make sure that you don’t get any of those events that actually started after index.

Yes, on this note: I’ve seen cases where in CDM data, they dont’ actually capture the duration of their diabetes, just the event that they diagnosed it, so you may not have a reliable ‘condition_end_date’ for diabetes anyways. If you are just looking for diagnosis in past 90 days just look for the event start date in the past 90 days, or 60 days or whatever makes sense. In the case where you said ‘but they could have had it 90 days and it’s still going on), your cohort definition looks like you are loking for recent diagnosis…if you dont’ care when they were diagnosed then you can just say 'betwwen all days before index and 0 days before index)

We should think of abolishing it altogether. I haven’t seen a single data source or use case that used that. Everybody considers them one-off declarations of things.

Cause diabetes is just an example :cry: If I want to set comorbidities that could be completely cured, then I need to know how to do!

I was wonder that diabetes couldn’t be cured, then if I use

inclusion criteria
 a condition exposure of diabetes
    where event starts between all days Before and 0 days Before index start date
       and event ends between 60 days before and all days after index

does diabetes patients could be included? Cause they can’t be end.
–> so what I mean is

Event end = last diagnosis record? or Event End = cured?

Yeah. That’s what I am saying. The end date is useless except maybe in some explicit cases where people put that in. Normally, nobody does. Think about it yourself. Did you ever call your doctor telling him or her that you are fine? No. You just stop showing up complaining.

So, if you really want to make sure a disease is cured the only way to do is to add a criterion where you require “with exactly 0 using all occurrences” of the disease you want to have gone. So, if you want diabetes “healed” then you require a diabetes up to, say, 60 days before index, and then you require no diabetes 60-0 days.

In most data assets the sensitivity of Condition records after initial diagnosis is pretty thin. So, you cannot expect that you have a record every day during the disease, and for the records to stop when it is over. And the condition_end_date is practically non-existent. Which is why I think we might get rid of it.

t