OHDSI Home | Forums | Wiki | Github

Condition and/or observation Mapping Challenges

(ella) #1

we’re wondering if folks are mapping many-to-one or many-to-many and if one or other, to please share why/why not…
here’s the current situation - our source system contains ‘events’ with ‘results’. these events can be various kinds of things, and in some cases there are event/result pairs such as “primary pain location”/“head” or ‘skin colour’/‘pink’ …
we aren’t sure about how best to map these things:

  • find a combo condition map - ‘headache’ - and don’t map the actual event/result
  • do the combo condition map, and also map the event to observation
  • map everything to both condition and observation

thoughts/comments welcome - thanks !


(Don Torok) #2

Can you provide a better explanation of the difference between the second and third option?
How best to record events with results depends a lot on how your site expects to use the data and your level of confidence that your event/result mapping to condition is correct. Primary pain location/head mapped to headache seems reasonable, but what is the condition for skin color/pink? I think in the end you will need to map the event/result to Observation because there will be combinations where you cannot confidently map to a condition. I suggest that you look at the CDMv6 adoption of two additional fields to the Observation table, observation_event_id and obs_event_field_concept_id. Adding these columns will allow an easy way to create a relationship between the Observation event/result and the derived condition record. Adding these fields to a V5 CDM will not interfere with any ‘standard’ analysis run on your data.

(ella) #3

Thanks for your response. the main difference between 2 and 3 is that in 2, the result would not be mapped to observation so the linkage between the event/result would be lost. in 3, there would be data redundancy in the DB. the condition we found to map ‘skin colour/pink’ to is ‘pink skin’. i was wondering about others who have dealt with this kind of thing to know: what they did, what’s been good about it, what’s been bad about it, etc. any such experience? thanks so much !

(Christian Reich) #4


Yes, the current mapping assumes source and target are semantically equivalent (one-to-one “Maps to” relationships), need to be split into several parallel concepts (one-to-many “Maps to”) or need to be split into EAV (“Maps to” plus “Maps to value”). We don’t have aggregation maps (like in your case), maps to number/unit combinations or more complicated situations where mapping results in concepts of different Domains. We are working on a proposal that will allow all these.