I know that we can get the number of observations and thus the number of patients for a particular time period, but something that is often necessary is determining how many patients were enrolled (regardless of whether or not they saw their doctor) and I don’t know if it is possible to get at that. Is is not possible to get that number? I’m not seeing anything obvious.
Have you considered using observation_period as the enrolment period? Observations and other events don’t have to line up with the observation period, but those that are outside of it are ignored.
I should have specified more in my original post, but yes what I normally do to count patients in effect is use observation period. That, however, is predicated on someone having a claim during the period in question. I have been asked if it is possible to get the total number of eligible patients in the population which would include those who are enrolled, but they may not have had a doctors visit or some other kind of claim. I’m just not seeing a way to capture that with Atlas. I wanted to confirm here that that is, in fact, something the tool cannot do.
You can create a cohort definition out of observation periods, and use the cohort counts as a ‘count of people’. There are a couple of filters you can use to identify those people who have a observation period that starts before and after a specific time (which would return those people who are in your data as-of a certain date).
But you seem to indicate an issue with your data where you depend on a doctor visit or some kind of claim in order to produce your observation period table, which isn’t a limitation of the tool, but rather your data. Is my understanding correct on this point?
What kind of claims are you processing? If they are open claims you don’t have enrollment, and you need to infer from the visits or other interactions. Just like in EHR data. But if you have adjudicated claims from a payer you get the enrollment, and even if the patient is completely healthy you know he counts as denominator.
Not exactly. We know the data has the information since one of our programmers queried it and she got a higher patient number than I did using the same observation period in the tool. I know the data has the information, I’m just not sure how to count healthy people who may not have an observation during the time period or if I even can within an Atlas cohort.
The Atlas tool does generate the cohort query for you so you can see where the counts from your programmer diverges from the tool. The cohort generation sql isn’t the simplest, but it’s in steps so it should be possible to run each step at a time and try to understand where the patient counts may be different.
Atlas relies on the observation_period for cohort definitions. You would have to adjust your ETL to ensure all enrolled patients have an entry in observation_period table, not just those who have a claims-based interaction.
Are you confident your ETL does this?
We found the solution! I shouldn’t say we. I should say @Brad_Rechkemmer found it. Instead of using Obs period if you want to just capture patients on the basis of enrollment so that those without an observation during that period get counted too you have to use Payer period which I had never had occasion to use so it wasn’t obvious to me, but that is the solution in case anyone else has this question.
Glad you found something that works, but it sounds off:
obseration_period should span the periods of time that anything that happens to a person, including periods of payer plans.
When a cohort definition is generated, anyone included in the cohort must have an observation period, because we need to know when the cohort_end_date is, which defaults to the end of the observation period that encloses the entry event. So, if you specify that persons enter at a payer plan period, but there’s no observation period during that payer plan period, then the person won’t be included in the cohort.
I’d suggest that if you have payer plan periods, you should incorporate that information when building your observation periods.
To add to @Chris_Knoll reply, per the CDM v5.4 ETL Conventions for the Observation Period:
“OBSERVATION_PERIOD record, which should represent time intervals with a high capture rate of Clinical Events. Some source data have very similar concepts, such as enrollment periods in insurance claims data”
With general insurance claims data, if a person was to have a clinical event (visit, condition, drug, etc.) your data would capture this information. Your Observation Period should match your enrollment data.