OHDSI Home | Forums | Wiki | Github

ATLAS Incidence Rate Analysis - Denominator Cohort Numbers

Dear OHDSI Community,

My name is Saptha from Gilead Sciences, I am an Associate Data Scientist working on the Real World Evidence team.

I am working on the incidence rate analysis using ATLAS, where I have one target cohort and a few outcome cohorts. I am interested in calculating the overall incidence proportion of an outcome as well as by time period post index (0-3, 0-6, 0-9, 0-12, 0-24 months).

The overall incidence proportion is calculated with the following settings

- Time at risk starts with start date plus 0 days
- Time at risk ends with end date plus 0 days
- Study window begins on 2017-01-01 and ends on 2022-12-31

The incidence proportion for time period post index is calculated with the following settings. lets say for example 0-3 months

- Time at risk starts with start date plus 0 days
- Time at risk ends with start date plus 90 days
- No Study window is defined

I am noticing the overall incidence proportion is lower when compared to the time period post index. Can you please advise why the numbers are lower and also suggest how to define the IR analysis to have the same denominator for both overall as well as by time period post index.

Thanks a lot for all your help in advance.


For one, your top report is showing per 1k people, and your bottom one is showing per 100 people. You can adjust the per X by clicking the +/- buttons in the header.

That’s a bit of a paradox: overall means you’ll get more people in your denominator than if you restrict those people to a study window.

Hi @Chris_Knoll,
My understanding is that the denominator is the number of people who did not experience the outcome prior to the index. Is that correct?

In my target cohort, the index date cannot be earlier than 1/1/2017 (one of the cohort exclusion criteria). So, if add a study window that starts on 1/1/2017, wouldn’t the number of people who experience the outcome prior to index remain the same as the one without a study window?

If not, can you explain how that could be different in this case?

It’s more accurate to say denominator is the people who are at risk for the outcome. There are multiple ways that would exclude a person, the first being that they had the prior outcome Another is that they have no time at risk (for example, if your Target cohort has a person with a cohort record has a start date = end date, and your TAR is specified as start + 0, end + 0, you may have some people that have 0 TAR (and will be excluded) vs. start + 0, start + 90, everyone is guaranteed to have 90d (provided their observation period will allow 90d of TAR).,

This wasn’t information you provided initially, so given that new information, if you require your Target cohort to only start on or after 1/1/2017, then all your T will satisfy the left-side of the study window, but anyone after 2022-12-31 will be dropped (unless it is another piece of undisclosed info that you have no one in your cohort that will appear after 2022-12-31). Anyone who has a cohort_start of 2022-12-31 does satisfy the study window range, but won’t be in the analysis becuase they have 0 TAR.

Thank you Chris for the explanation. It all makes sense now