- We are looking for which SNOMED codes associate with X (recommendation of yoga, diagnosis of lyme disease)
- CHALLENGES:
- In this case we don't have a 1 to 1 map from ICD to SNOMED
- Also, SNOMED is roughly hierarchical
- This OHDSI example explains this issue and a solution... BUT the solution works for querying patients, not for association studies
- assume complete ignorance about which ICD-SNOMEDCT map is most meaningful
- when we have an instance of a person who has been recommended yoga...
- dole out weights to each snomed code to which the ICD code is mapped
- then, for each of the snomed codes who were doled out weights, dole out weights to the parents of that code
- continue recursively up the tree
- this is the naiive probability that SNOMED code X applies to this instance