Continuous observation 365 days before - Monitor the patient for 365 days or look at his past 365 days records (in retrospective data analysis) and if it’s clean, consider only those patients for analysis. Right?
Continuous observation 1095 days after - Monitor the patient for 1095 days or follow-this patient for another 1095 days to know/record/fetch information on what other “Diabetic ingredients” he had, so that we can know his treatment pathways (different medications/drug he took). Am I right? If any patient who doesn’t have 1095 days of records, he will be dropped from the cohort I guess
But why “earliest event” in the file that you shared. Both for Target and Event cohorts? Then is there any use in we having follow up period of 1095 days when we are going to get his earliest (1st record)? I might totally be wrong.
When we pick “Diabetic Ingredients” in cohort entry/Qualifying criteria, you have chosen “Exactly 0” drug era of Diabetes ingredients because you don’t want ingredient items to be collapsed? can you please help me understand why this condition? I know that drug era collapses two drug exposures into one single episode based on ‘n’ interval days. what does setting “exactly 0 drug era of diabetic ingredients” do?
when you set “All days before” and “1 days before”, what does it do? Meaning doesn’t “1 day before” come under “All days before” ?
I understand we briefly touched upon this discussion in another post, but just had a look at your doc in detail while trying to replicate it, so curious to understand your thought process behind cohort definition?
Yes, we’re requiring 365d of prior observation in order to determine if we are ‘clean’. However, remember that finding the earliest observation after 365d of continuous observation is not their ‘first’ (they may have an observation on day 5). See the other threads that I’ve responded to related to ‘earliest’ vs. ‘first’
Correct: we want to require that a patient has 1095 days of continuous observation after their cohort entry so that we can be sure that everyone in our pathway analysis swill have at least 3 years of follow up. If they don’t have the follow up, then they won’t be in the cohort.
For the Target cohort, we choose earliest event because we want the Target cohort’s start date to just begin at the earliest of any T2DM medication specified. We don’t need each exposure record because we’ll specify that the people in T continue in the cohort until end of observation. Any additional drug exposures included in cohort entry event are just redundant.
For the event cohorts, it may be a mistake that the event cohorts should use ‘all events’, however, it depends on what you want the pathway analysis to show: if you want the pathway to show only ‘new’ exposures experienced any time in a patient’s history, then you want your event cohorts to only include those ‘first’ events. If you want to just know about any activity during the Target cohort’s episode, then those event cohorts should specify ‘all events per person’.
This is the rule that’s making sure that this person is ‘clean’. Since we want our people in T to be those newly exposed to diabetic ingredient, we want to throw people out if we find out that the exposure post-365d was not their first exposure. To do this, we assert that they musth have 0 exposures to these ingredients between 1d before and all days before the cohort entry event (which is also a ingredient exposure event).
Way say all days before and 1 day before in the exactly 0 criteria because we do not want to include the day of the exposure when looking for disqualifying exposures. To put this in data terms:
Person 1 has exposure on Day 5 and day 367
Person 2 has exposure on day 369.
IF you make the lookback window `all days before to 1 day before’
Person 1 does have a exposure between day 0 and day 366 (day 367 - 1 = 366). This person is out.
Person 2 does NOT have an exposure between day 0 and day 368 (day 369 - 1 = 368). This person is in.
If you make the lookback window ‘all days before and 0 days before’
Person 1 does have an exposure between day 0 and day 367 (day 367 - 0 = 367). This person is still out
Person 2 DOES have an exposure between day 0 and day 369 (day 369 - 0 = 369). This person is now OUT!
We don’t want the case of person 2 is out, so, when looking for the prior exposure (to throw away people who are not clean) we can’t use the same date as the entry event when looking for the disqualifying, prior exposure. So, to exclude the index day, you say all days before to 1 day before index.
I think there may be another post where I went into this into quite fine detail, so you may want to search for that again.
Thanks for the response as always. Might be a basic question.
When I set “continuous observation 0 days before and 0 days after index date”, I get the same count of patients (3861) which is same as patient count that I get with 365d look back and 1095d follow-up. And they all have data for more than 3 years.
Does this mean all subjects in the data, had more than 365 days history before getting their diabetic ingredients.
“Continuous observation of 365 days before and 1095 days after” - look back 365 days and follow-him for 1095 days atleast
“Continuous observation of 0 days before and 0 days after” - don’t look back and don’t follow-him up.
Even though I don’t look back/follow-up, when I get the same count, it means that our subjects in the data (IN OUR DATASOURCE), had more than 365 days history before getting their diabetic ingredients. And they all have data for more than 3 years. Am I interpreting this correct?