A conversation between Chifundo Kanjala (UNICEF) and Jay Greenfield (CODATA) on profiles

Chifundo: The cholera analytics work that we have been talking about, would it involve creating a profile from the upper model or the profiles currently under development are generic enough to handle multiple diseases?

Jay: Neither the upper model nor the profiles are disease specific. The upper model is able to specify person-specific and place-based (environmental, SDOH including SDGs) factors that figure into one or more outcomes, depending on the research design(s) in play in a study.

Three research designs are supported — descriptive analysis, causal analysis (population health field experiments, emulated clinical trials) and predictive analysis that uses machine learning to predict concept-specific outcomes using user-defined and/or “standard” covariates. The upper model grows the schema.org MedicalObservationalStudyDesign to include protocols/schemas for each type of analysis. In the upper model these schemas are largely empty.

The profile will give each schema specificity. Not each data analysis workbench is the same when it comes to analysis specifics. Within each analysis category, we may encounter different subtypes and/or each subtype may have its own workbench-specific UI. For example, in predictive analysis OHDSI and I-DAIR don’t support the same algorithms, and, even when they do, they each have their own way of specifying hyperparameters.

Using profiles, we can take a layered approach through which we can navigate the differences between various workbenches. Of course, this is easier said than done.