I use the atlas to construct the target cohort, which entry event is the target drug. the analysis cohorts contain seven cohorts, one cohort only containing the target drug. then I run the cohort pathway, find that the portion of persons in pathways is 91.6%, but I think it should be 100%, isn’t it?
thank you very much
Usually this is a result of a mismatch between the index of your target cohort and your index within your 'even’t cohort (it sounds like you have 1 Target and 7 Event cohorts, one of the event cohorts is a cohort with a single target drug that is part of your target cohort).
The treatment pathway shows event sequences between your Target’s start and end date. If your event cohort was ‘first exposure to target drug’, and that event cohort’s date is outside of the Target’s start/end dates, then it will not appear in the pathway analysis, and you will have a ‘persons in pathway’ that is less than 100%.
I suggest you generate each of the cohorts via cohort generation in Atlas, then query the results schema looking for anyone who exists in your Target cohort ID but does not have any start dates from your event cohorts that fall between the target’s start and end date. This will tell you the people that are being dropped. Then you can look for those people in your event cohorts to understand the reason that they are being included in your event cohorts, but excluded from the analysis due to ‘out of bounds’ of the Target’s start/end date.
thank you very much @Chris_Knoll. Are there methods to make the results equal to 100%?
- Make sure that your target cohort’s start/end date spans a timeframe that will capture all possible events defined in your event cohorts. (a Target cohort which selects earliest event per person and ends 1 day after that event will only have 1 day to find the pathway events in the event cohorts. Usually your target cohort will end ‘at end of observation’.
- Use ‘all events per person’ in your event cohorts. If you use only ‘earliest event per person’ in your event cohorts, and the target cohort begins after the earliest event cohort event, that event cohort will not apepar as a step in the pathway analysis (potentially not finding a person in the pathway and leading to a < 100% number). Contrast to point 1. above, sometimes you don’t have a target cohort defined that incorporates the events of all of your event cohorts…you just have a long running timeframe (such as first occurrence of procedure X, ending 3 years after). In that case, you want to use all events per person in your event cohorts because you want to try to find any of those event cohort episodes within the timeframe of your target cohort.