| OHDSI Home | Forums | Wiki | Github |
@Chris_Knoll and Atlas builders:
What’s the difference between the two? I can choose First Exposure in History as a Criteria attribute, or I can select from the pop-down below the Initial Event Cohort definition by clicking on “earliest event”. Is there a difference?
First exposure locks in the first event in your patient record.
If you had 3 records for criteria A that appears 5, 50 and 120 days after your observation period start, and you said that you require 30 days of continuous observation prior to the record, specifying ‘first’ means you’d only look at the record at appeared at day 5, and then you’d throw it out because it does not have 30d of prior observation.
On the other hand, if you specified ‘earliest’, record at day 50 would be selected because it’s the earliest record that has the required 30d of prior observatoin.
-Chris
@Chris_Knoll: Will put in the new Ux issue. Totally clear what you mean, but hard to guess for users.
Ok, but to be fair: the full description for ‘first’ is ‘for the first time in the person’s history’…so we can try to rephrase that to make it immediately understandable, but I acknowledge that this very specific terminology is hard to see exactly what it is doing there at first glance. On the other hand, this very specific terminology is used to identify a very specific event: the first event that is found in the patient history.
As part of the cohort definition tutorial, we’ll be making adjustments to the UI language to follow a lexicon for cohort definitions. The initial events will be referred to as a ‘pool’, and the events in the pool will be used to construct the ‘episodes’ of the cohort, etc. If you have any other concerns with some of the terminology in the current UI, this git issue is where to talk about it:
Before we make changes the UI, let’s please try to reach community consensus around ‘lexicon’. I am not familiar with the term ‘pool’ and ‘episode’ has a very specific meaning in epidemiologic contexts which may or may not be what we have in mind.
The current ATLAS UI applies a set of terms (initial events, conceptsets, inclusion criteria, exit criteria) that have definitions consistent with the current implementation. I am open to revising the set of terms we use if there are objections to the current vocabulary from the community, but first we should level set on the vocabulary and definitions we are currently use, and as necessary, evaluate alternatives.
Regardless of what vocabulary we ultimately settle on (keeping it the same or changing it), there is a education/training component that is needed to provide the users the vocabulary and ensure they understand its meaning, as well as a development task to ensure the vocabulary is consistently applied throughout our applications. I would argue we need to start with the design (what is our intention that we want to achieve, fully specified and documented), then propagated through our implementation (what do we see in the UI and resulting artifacts), then further propagated through our dissemination plan (what materials are provided as help/documentation in the tool, what artifacts are generated and shared as part of videos and tutorials, etc.).
Yes, the intention here is to have these decisions made so that we can use them in our Cohort Definition Tutorial session at OHDSI 2018, and then we will have the recorded session available for other people to learn the terminology. I do expect that this process will take some time to fully mature into the OHDSI ecosystem, and that this is just the beginning of that process.
In the Tutorial you want to do what? Introduce new terms? Test out new terms? Ask the crowd? Anything we need to do to prepare for that?
It would be pointless to have the tutorial with current terms and then change them later.
As it stands now, there’s already confusing/conflicting terms used in the language that we want to clean up before the tutorial (or else we’ll be left teaching on confusing terms). For example:
Intial Event Cohort: The ‘inital event cohort’ is not a cohort at all. Events selected here could have overlapping periods, which violates the rule that periods of time in a cohort must be non-overlapping for a given person.
The real idea in the ‘initial events’ is that it is a set (or pool) of events that the final cohort episodes will be derived from. You can’t say that these are the ‘cohort start dates’ because a series of events could be chained together A->B->C such that the cohort episode starts with Event A and ends at event C. These ideas are key to understanding how to use the cohort definition tool in Atlas.
We can’t expect the lexicon of CIRCE to translate 1:1 to all other contexts: CIRCE will talk about events differently than an effect estimation context will talk about events (the effect estimation might assume ‘events’ means ‘adverse events’ or outcomes). CIRCE defines things like ‘censoring’ to work a certain way, while other contexts will think of censoring differently.
Therefore, the aim is to clean up the terminology so that we can teach the tools using a specific lexicon. Note: this is a lexicon relative to CIRCE, not across all OHDSI or modern science. It would be up to the people leading the other groups, such as population level estimation or patient level prediction, to ‘map’ terms in CIRCE to their own lexicon. For example, a ‘Time At Risk’ in PLE may map over to a cohort ‘episode’.
In Chris’s explanation, I would also like to see a definition of ‘30 days of continuous observation’.
To be honest, I was not able to follow his explanation.
What events count as making my record “continuous”? (visit to doctor or just dispensation of a refill)?
Simply put: continuous observation means that there is no break in the observation period between 2 points in time. Specifically when talking about initial events with 'continuous prior observation", we are saying that the observation period that contains the initial event begins (observation_period_start_date) at least X number of days prior to the event date.
Example:
Event Date: 4/1/2010
Observation period Containing the event: 1/1/2010 - 1/1/2012
There are 4 months of continuous observation observation prior to the event.
There are 20 months (1 year + 8 months) of continuous observation post-event.
Mental exercise: Which events will not have 30 days of prior continuous observation?
Answer: Any events between 1/1/2010 and before 1/30/2010 will not have 30 days of continuous observation prior to event.