Do we need a schema for our data warehouse?
We have had several efforts to use a traditional data warehouse approach (e.g., star schema) over the past decade or more, using Birt, Pentaho, and other/custom approaches. This traditional data warehouse approach works well when the data model and the questions to answer are known. We have neither of these. Our data model for analytics within OpenMRS isn’t the relational database model, but the data model of clinical data represented by our concept dictionary. Both this model and the questions to be answered are in constant motion. For this reason, tools that don’t depend on a fixed schema – e.g., Apache Lucene-based tools for indexing documents like Elasticsearch – are appealing.
@bashir, does your approach depend on coercing content into a FHIR SQL schema? Or is it possible to use documents (e.g., JSON) at the heart of the engine, focusing on coercing data into FHIR format, but not requiring a specific relational database schema to describe our universe? I assume a FHIR schema would not be adequate for data used by OpenMRS that is out of FHIR’s scope and would force us to come up with an alternative approach for non-FHIR and implementation-specific data.