Modeling Concepts

Here are some notes/conversations related to concept modeling that may be helpful resources when confronting modeling issues.

Modeling diagnoses

Hamish wrote:

 If I create a concept "malaria" say a boolean and I want to also have 
 "Malaria diagnosis date" how do I link these two items so that it is 
 clear that they refer to the same item?

 Also as Darius and I discussed today what if Carole our epidemiologist 
 friend says she wants all the diagnoses for her data collection form 
 to have true, false, unknown and "no data available". Should we just 
 create hundreds of coded concepts for this or formalize is as a 
 standard construct? If we formalize it does that help us to handle such
 data efficiently in the same analysis and data extraction tools that 
 normally take a boolean?

I would use DIAGNOSIS or PATIENT REPORTED DIAGNOSIS or similar generically and then record the diagnoses as answers (e.g. MALARIA) for most cases. That gives you the choice of recording the date of diagnosis into the obs date (probably simpler) or creating a DATE OF DIAGNOSIS observation to link dates to diagnoses through an obs_group_id.

If someone wants multiple choice answers for 1..n diagnois, then you are collecting two observations in each case: a multiple choice answer and the diagnosis for which the answer applies. There are a several ways to skin this.

1. Flat by diagnosis (add coded answers for each diagnosis)

 MALARIA with coded answers TRUE, FALSE, UNKNOWN, NO DATA AVAILABLE

PRO 

simple, MALARIA/etc could still be used as answers

CON 

not scalable, ties you to one model of answers to diagnoses and these must be replicated/managed for *every* diagnosis in the system.

2. Fully abstract (one concept for diagnsosis and a 2nd for answer to multiple choice question)

 INQUIRED DIAGNOSIS as coded answered by diagnosis
 DIAGNOSIS STATUS as coded with coded answers TRUE, FALSE, UNKNOWN, NO DATA AVAILABLE
 Tied together with obs_group_id

PRO 

easily scalable

CON 

needs tools/knowledge to convert back to one complex data point, diagnoses stored in a questionnaire-specific manner

3. Abstract, but use existing DIAGNOSIS concept

 INQUIRED DIAGNOSIS as coded answered by diagnosis
 DIAGNOSIS STATUS as coded with coded answers FALSE, UNKNOWN, NO DATA AVAILABLE
 Linked with obs_group_id

 Store TRUE answers as DIAGNOSIS with coded answer MALARIA, ASTHMA, etc.

PRO 

re-use of DIAGNOSIS, so you can still search DIAGNOSIS concept for all known diagnoses

CON 

application must know to treat TRUE answers differently

4. Flat by answer (one concept per answer for questionnaire)

 DIAGNOSIS
 DIAGNOSIS DENIED
 DIAGNOSIS STATUS UNKNOWN
 DIAGNOSIS DATA NOT AVAILABLE
 each is coded and is answered with the diagnosis

 CAROLE1 QUESTIONNAIRE DIAGNOSES OPTIONS as a concept_set of the above concepts 

PRO 

scalable, could re-use DIAGNOSIS concept or have a separate concept for diagnosis = true on questionnaire

CON 

requires search for multiple concepts (could be facilitated by concept_set)