OHDSI Home | Forums | Wiki | Github

Patient medical history/Events outside observation period

(Olga Osintseva) #1

We have source tables such as ‘patient medical history’ or ‘patient familial history’

The old approach:
Before the THEMIS issue #23, when START_DATE and END_DATE must be between OBSERVATION_PERIOD_START_DATE and OBSERVATION_PERIOD_END_DATE we populated SOURCE_VALUE with the name of source table and CONCEPT_ID populate with concept like ‘HISTORY OF…’.
And diagnosis/procedures from patient history was kept in VALUE_AS_STRING and VALUE_AS_CONCEPT_ID fields
field DATE contains OBSERBVATION_PERIOD_START_DATE or date when information about patient history diagnosis/procedures was inputted

The new approach:
After the THEMIS issue above, according to that events can fall outside of an observation period we may use new approach:
keep diagnosis/procedures from patient history in SOURCE_VALUE and CONCEPT_ID fields and use date of diagnosis/procedures from patient history as DATE.

Disadvantage of this approach is that information about data provenance (medical history) would be lost.
Advantage is that we keep real dates of diagnosis/procedures of patient medical history.

So questions are:

  1. For the new approach is it important that diagnosis/procedures are patient history?
  2. Do we need change the approach from the old to new?
  3. Or we may choose approach depending on the percent of records?
  4. Or any other way?

Thank you in advance!

(Christian Reich) #2


  1. Yes. Look. It’s important to know when the heart attack happened. Not when the doctor asked the question.
  2. Yes, please. THEMIS is law. Not a moral rule.
  3. No. Even if everybody crosses the street at the red light it’s still wrong and illegal.
  4. You want another way? :slight_smile:

(Olga Osintseva) #3

Thank you for the answer!

(Alexandra Orlova) #4


ETLers should define the date from which patient medical history could be recorded as real events in different domains (drugs/conditions/procedures/etc.). This date could be provided by a data expert/data dictionary (custom date). If there is no such date, then 1990-01-01 could be used as a default date.

So the rule will be:
IF patient medical history is about events that happened earlier than 1990-01-01 (or custom date)
THEN patient medical history will go to the OBSERVATION table:
‘History of’ concept as OBSERVATION_CONCEPT_ID
‘event itself’ as VALUE_AS_CONCEPT_ID
‘date when this info was asked by provider’ as OBSERVATION_DATE
‘Patient reported’ concept as OBSERVATION_TYPE_CONCEPT_ID

IF patient medical history is about events that happened later than 1990-01-01 (or custom date)
THEN patient medical history will go to drugs/procedures/conditions/etc. based on the domain of ‘event itself’:
‘event itself’ as EVENT_CONCEPT_ID
‘date of event’ as EVENT_DATE
‘Patient reported’ concept as EVENT_TYPE_CONCEPT_ID

Tagging @Christian_Reich, @Natalia_Karataeva

(Christian Reich) #5


Sounds like a good default rule set for me.

Usually, these records are patient-reported. And there is a reverse proportionality between distance in time and need for precision.

  • We need precise dates for all things during an Observation Period. We should make events “History of” only if we really don’t know the date.
  • Before the beginning of the Observation Period the precise date becomes less and less relevant the further we go back. We should never use it as an index date anyway, and whether that broken wrist was in 1984 or 1985 makes no difference. In fact, at some point it should become “History of” to adequately reflect the precision of the data.

Anybody objecting to the 1990?

@clairblacketer: Want to put this into your DQD?

(Clair Blacketer) #6

Yes, I have been working with something similar for CPRD. 1990 Seems like a fine date to use as a default, though I will probably add a custom option to change the date if necessary. I wonder if it should be a sliding scale, like > 10 years prior to the OBSERVATION_START_DATE should become “history of”?

(Christian Reich) #7

I like that better.

(Vojtech Huser) #8

Well. In a situation where there is a block of data in obs. period 1 (e.g., 5 years). Then some darkness (e.g., 10 years) (patient had no insurance) and than again block of data (e.g., 3 years).

Patient history can be about the time of 10 years between the OPs. So delegating some “-x years of time time” for history-of-events may not work in all cases.

(Christian Reich) #9

True, @Vojtech_Huser. Let’s apply the rule to the first Observation Period.

(Clair Blacketer) #10

That makes sense! I linked to this conversation as a github issue so I won’t forget to add it.