OHDSI Home | Forums | Wiki | Github

OMOP CDM and Clinical Trials


(Vojtech Huser) #1

I was asked to organize a group around this past proposal in the CDM workgroup

fixed link:
http://www.ohdsi.org/web/wiki/doku.php?id=documentation:next_cdm:clinical_trial

In addition to CDM, we also have “conventions” (see here Per CDM tutorial slides 5 visit types?).

If you are interested in creating conventions (or CDM changes) for clinical trials data, please reply to this thread.


Non-EHR data transformation to OMOP CDM
Requirements for Clinical Study/Trial data in OHDSI
(Jon Duke) #2

Make sure to delete the edit from the link so you can see it properly:

http://www.ohdsi.org/web/wiki/doku.php?id=documentation:next_cdm:clinical_trial


(Christian Reich) #3

@Vojtech_Huser:

Conventions? You don’t think we need some data structures? What about the arms?


(Joshua F Ransom) #4

This is an area that I and a couple of my colleagues would like to get involved with. Has the WG met yet?


(Vojtech Huser) #5

@christian - I agree that trial arms are also in scope.

I also want to document that there is communication with Columbia U on how to unify some clinical trial data conventions. (possibly to be used by the phenotypeDataExtraction package) (find similar patients given an a trial NCT)


(Vojtech Huser) #6

For representing enrollment and withdrawing for a trial identified by an NCT, possible parameters would be SNOMED codes from here

OMOP vocabulary seems to have also some SNOMED national extension (for example, scottish study is represented under this concept)


(Vojtech Huser) #7

to keep the conversation going:

Here is possible representation of trial enrollment using OBSERVATION table.

entering a trial - Enrollement (start)
Observation_concept_id: http://www.ohdsi.org/web/atlas/#/concept/4163733 (Patient consented to clinical trial)
value_as_string: NCT000035135

leaving the trial early (due to a decision of a patient (no longer interested) or researcher (non compliance) (end)

*Observation_concept_id:*http://www.ohdsi.org/web/atlas/#/concept/4087907 (Patient withdrawn from trial)
value_as_string: NCT000035135

fully completing the trial (end)
Observation_concept_id: http://www.ohdsi.org/web/atlas/#/concept/40482840 Completion of clinical trial
value_as_string: NCT000035135


(Eldar Allakhverdiiev) #8

Please, sign me up


(Vojtech Huser) #9

I was discussing study data transformation with someone (think SDTM data transformed to ADAM data).

Here is another issue for the group to discuss:

To capture raw data as well as “for analysis” transformed data - I propose to use the same table (e.g., measurement) and use type columns to indicate raw vs. transformed. (just to put a proposal to the discussion)

Not capturing raw data is probably not an option. This would make OMOP inferior to current standards.

To capture transformation steps (captured in Define.XML), OMOP METADATA table could be used for that purpose.


(Christian Reich) #10

I would start with the analysis. Because that is what the OMOP CDM is. And that’s were the use cases live. Raw data are pre-ETL in this world.


(Andrew Williams) #11

I am interested.


(Joshua F Ransom) #12

@Vojtech_Huser - Since there were a few posters at the Symposium on using OMOP and RWE for Clinical Trial analysis, thought it might make sense to try and revive this thread.

I see a few levels of items to consider for being able to fully leverage the power of OMOP CDM with RCTs. I don’t think this is exhaustive per se, but thought it might help spur the discussion.

A: Data Model Interoperability:

  1. CDISC SDTM conversion to OMOP CDM
  2. CDISC ADaM conversion to OMOP CDM
  3. OMOP conversion back to CDISC

B: RCT Administrative Records:

  1. Patient Screening (not the recorded determination of eligibility, but that the screening took place)
  2. Patient Randomization

C: RCT Visit Type (e.g., was the visit for screening, baseline, follow-up, etc.)

D: Non-standard Interventions and Outcomes (e.g., how to record and deal with investigational therapies, unvalidated PROs, etc.)

@Vojtech_Huser - I’m wondering where you’d recommend is the best place to capture the issues and lessons learned from our respective CDISC conversion efforts?


(Vojtech Huser) #13

I like your proposed topics. I focused mostly on how trial data can co-exist next to EHR data when both are ETLed into OMOP.

Some of my observations are summarized here:https://github.com/vojtechhuser/project/tree/master/sdtm-omop#sample-data-conversion-results

One community effort could be to produce “SOURCE_TO_CONCEPT” map for the lab test below: (that we saw in the sample data)

(Also tagging @Craig_Mayer in our clinical research informatics team)

(observations also pasted for those who do not click the link)

Sample data conversion results

  • sample CDISC data differes in format from trial RWD (3 trials)
    • person identifier used (SUBJID, RSUBJID,RUSUBJID))
      • large sample needed but not available; must also be recent (trial started and finished in the last 5 years)
  • SDTM manual is very long, in PDF format (this should change with CDISC SHARE initaitive)
  • Demographic (DM) domain does not contain year of birth. It is not clear at what time age is calculated. (at first visit of the subject?; day 1 of the trial) (convention)
  • trial data may be organized by visit number (not absolute date)
  • Demographic (DM) domain data on screen failure or arm: missing suitable concept_id (issue) (tentatively using LOINC 68839-0| Research Note) (todo new concept)
  • LOINC codes will be used in SDTM (in 2018, FDA clarified a LOINC strategy) so detailed mapping is not pursued.
  • Assignment into arms can be in COHORT table or as OBSERVATION or other table (convention).
  • Vital Signs (VS) domain is very similar to Laboratory (LB) domain. Both have standard units and original value and numeric value. LB has exact datetime, whereas VS has only date (issue).
  • _source_value columns were sometimes overloaded with multiple elements separated by delimiters (issue)
  • Medical History (MH) domain captures a condition at a given visit (date1) but
    records onset date (date2). Date2 can be null (not known) (convention) (issue)

update: the trial arm term will be in SNOMED CT in Jan 2019 :slight_smile:!
(I submitted it)


(Vojtech Huser) #14

Please vote/comment on this proposed high priority convention for study data https://github.com/OHDSI/Themis/issues/37

Also, we would need row in METADATA for date.


t