OHDSI Home | Forums | Wiki | Github

Chohort definition: What is the difference between the first time of history and first_occurrence per person

dear all,
when I use the atlas to construct a cohort, I am confused about the difference between the first time in the person’s history(picture 1) and the first_occurrence per person(picture 2).
picture 1:image

thank you very much!

As long as I understand,

Let’s say you have entry criteria A and inclusion criteria B

  1. Option one (first time): When the patient have A event for the first time, he/she has to satisfy B event to be included in the cohort.

  2. Option two (initial event): The first event A will be captured when the patient has B event.

So usually, option two cohort has more patients than option one.

Please correct me, if I’m wrong @Chris_Knoll

Hi @pandamiao,

Although @SCYou’s explanation is pretty straightforward, I recommend you check out this topic as well.

Thank you very much @SCYOU @gyeol99, I think I understand the difference. At the same time, I have another question, how the researcher choose the two options? For example, which kind of researches or topics is better for the researcher to choose Option one?

Thank you very much @SCYOU @gyeol99, I think I understand the difference.But at the same time, I have another question, how the researcher choose the two options? For example, which kind of researches or topics is better for the researcher to choose Option one?

Thank you very much @SCYOU @gyeol99, I think I get it.But at the same time, I have another question, how the researcher choose the two options? For example, which kind of researches or topics is better for the researcher to choose Option One or Two?

I don’t think this is quite correct, because for Option 1: you are associating some relationship between cohort entry event A with an inclusion criteria B. But, ‘first time in history’ doesn’t depend on any inclusion rule. ‘First time in person history’ means: if you order the observations found by the concept set (in this case from the condition_occurrence table, using the concept set ‘Diabetes’) by start date, only the first of all those observations will be considered as a cohort entry event, even before applying other criteria (like age). For example, if you said ‘Diabetes, using the first time in patient history, age >= 18’, if the person’s first diagnosis was at 17 and then later had a diagnosis at 20, only the first event (at age 17) will be considered for cohort entry, and so this person would not make it into the cohort. Conversely, if you use the ‘earliest event per person’ without the ‘first time in history’ criteria, then the observations that appear when a person is >= 18 would be considered for cohort entry, but then only the ‘earliest event per person’ will be used in the cohort. This is a very subtle difference, but there are situations where you want to use one over another.

After these cohort entry events are identified, any inclusion rules that are defined in the cohort definition are applied, and the remaining cohort entry events are used to create the cohort episodes.

Yes, this is correct, because ‘first time in person’s history’ is very restrictive…you only can have one ‘first’ occurrence of an observation per person. However, when using ‘limit to earliest event’, you are taking potentially many entry events and limiting it down to the earliest of the set. Sometimes, the ‘first’ and ‘earliest’ result in exactly the same thing. but other times (as I described above) it does not.


1 Like

We see ‘first time in patient history’ for those cases where you want first ever exposure to a drug. In that case, using ‘first time in history’ is very common. Other times, like 'earliest diagnosis of Diabetes (entry event) where they are hypertensive (Inclusion 1) and Kidney damage (Inclusion 2). The reason for the second option is that you want to allow multiple diabetes diagnosis to be considered but only include those dates of diabetes diagnosis that have prior hypertension and kidney damage. If you limited to ‘first time in history’ then you’d be restricting your population to those people who presented with hypertension and kidney damage at their first diabetes diagnosis. It depends on your study design if that’s how you want to define your study population.

@Chris_Knoll I think we say the exactly same thing, but I know I said it so simply :wink:

@pandamiao my two cents added to @Chris_Knoll 's comment, is that if you want to do some ‘first-user’ cohort setting, then I recommend you to use ‘first time’ restriction.

Hey, @SCYou , yeah, I’m sorry, I reread it again, and you’re right! Apologies!

Your example does make sense because you show how the inclusion rules would react if you specified ‘first event in history’ vs. if you didn’t. So, good job there, you said it very well.


The difference is that the first picture specifies a FIXED time 0 while the second picture defines a baseline that flows (shifting back and forth depending on your other eligibility criteria, which should be considered simutaneously and can be met at mutiple times)

Super useful discussion here! And perhaps a good use case for the Tooltip feature suggested on yesterday’s community call :slight_smile:

If that helps you think about it, that’s fine, but I’m not sure I’d use terminology like ‘fixed time 0’ in the definition of what ‘first event in history’ means.

Tooltips are good suggestions, but it’s incomplete. We need actual content for these tootips to make them useful. In this thread alone, we have 3 different ways to describe this criterion. We can’t fit all 3 into a tooltip (we couldn’t even put my detailed description in a tooltip).

If I had to put it into a tooltip, I’d describe it as:
‘Use only the first condition occurrence event found using the specified concept set.’

This has all the implications that @SCYou, @yzhang and I describe, but I don’t think we could put that into a tooltip, but the implications of these decisions are very necessary to make the ‘correct’ cohort definition for the clinical question. This type of ‘implication question’ I think is best answers on a forum like this one (much like Stack Overflow hosts Q&A type of functions for a community of developers) but I do think we can incorporate some of this information into better documentation.

1 Like

For what it is worth, we use the terms “first ever event” and “first qualifying event”. For the latter, we use the term “recurrent index event study” meaning that we are studying something that has many possible index events and we are evaluating all possible ones to “enroll” people. Examples would be “first ever myocardial infarction” versus “first qualifying statin prescription”.

Got it, thank you very much for your help!

Thank you very much to everyone, and I think many more people will benefit from your answers!
Best regards