| OHDSI Home | Forums | Wiki | Github |
dear all,
when I use the atlas to do IR analysis, I found a question, if one target person developed outcome twice, how do the atlas compute the num of the outcome? for the bookofohdsi, it is “An incidence proportion provides a measure of the new outcomes per person in the population during the time-at-risk”, so does it mean only compute once per person?
thank you very much
@pandamiao,
The IR analysis will calculate the follow up time to the first occurrence of the outcome. If they had multiple outcomes, on the first one will be counted.
In addition, anyone who had the outcome prior to their cohort start date in the Target cohort will be excluded from this analysis. This is the behavior of Atlas, and may change in the future, but for now you should know the rule is to drop anyone with a prior outcome.
thank you very much@Chris_Knoll, I get it. And there are also two questions.
1, if the target cohort has 200 people and 300 records, so in the IR result section, the person num will be equivalent to the difference value of person num in target cohort (200, not 300) and anyone num with a prior outcome, is it?
2, if I make the TAR begin 0 of the target start date, then the person with outcome in the 0 day，same with the initial event begins， will also be excluded from the person in the IR results, is it ？
Hi,
You can see the logic for the IR analysis here. The question about ‘what happens with a person with multiple episodes’ is a good one. Each episode is going to be treated as a unique case for IR calculation. So a person with 2 episodes will contribute to two distinct Time At Risks for the analysis. I can think of cases where this may not be what you intend, so for the IR analysis, I’d define a cohort which has 1 episode per person (using the ‘earliest event per person’ in the last setting of the cohort definition). This is an area I think Atlas IR analysis can be improved to handle this better.
This part of the IR analysis SQL:
where (o.cohort_start_date is null or o.cohort_start_date > t.adjusted_start_date)
and t.adjusted_start_date < t.adjusted_end_date
and t.adjusted_start_date between op.observation_period_start_date and op.observation_period_end_date
The first line of the where clause Is only including people if they did not have an outcome or if the outcome was after the adjusted start date (the start of the time at risk). This means that people that have the outcome on the same day as the TAR start will NOT be included in the outcome. The reason for this is that this would lead to divide-by-zero errors for those people who were considered a case but the datediff between the TAR start and outcome date = 0. So, Atlas throws away people that have the outcome on-or-before the TAR start.
Again, this could be an area for improvement in the Atlas IR analysis tool, but it is important you know how it is currently behaving so you can make the appropriate decisions to calculate an incidence rate.
-Chris
thank you very much @Chris_Knoll, the first question I get it. And for the second, if I set the TAR like the picture in the Redbox, then it means that people that have the outcome on the 1 day as the TAR start will NOT be included in the outcome. is it?
Another quesion about the cohort definitions. I construct cohort like the picture, then the person with the condition occurrence on the 0 day before the index event, will the person be included in the cohort?
thank you very much!
Correct, the outcome has to appear after the start, not on the start of TAR.
Saying ‘event starts between 0 days before and 1 days after index start date’ means that the index start date and the day after the index start date are the dates that you’ll look for the ‘any condition’ to start in order to be included in the cohort.
thank you very much @Chris_Knoll, I get it.
So if there is one condition occurrence on the 0 day before index start date, on the day of index start date, the condition will be included in the cohort, is it?
And if i set the inclusion criteria’ event starts between 1 day before and 1 day after index start date’, the condition occurrence on 1 day before index date will also be included in the cohort, is it?
thank you very much
Yes, if you say ‘0 days before’ or ‘0 days after’ you’re just saying ‘on the index date’. if you say ‘1 day before to 1 day after’ then you will allow the event on the day before, the day of, and the day after.
In your picture, you say ‘between 0 days before and 1 day after’ meaning the condition occurrence could be found on the day of the index or the day after the index.
