Phenoflow: Portable Workflow-based Phenotype Definitions

Clinical phenotyping is an effective way to identify patients with particular characteristics within a population. In order to enhance the portability of a phenotype, it is often defined abstractly, with users expected to realise the phenotype computationally before executing it against a local dataset. However, complex definitions, which also provide little information about how best to implement a phenotype in practice, mean that this process is often not easy. To address this issue, we propose a new multi-layer model for a phenotype definition, which is realised as a workflow, and can be combined with different implementation units in order to produce a computable form. A novel authoring architecture, Phenoflow, supports the generation of these structured definitions. To illustrate the utility of our approach, we re-author a diabetes phenotype definition, and then compare its portability to the original definition, in the context of a population of 26,406 patients at Northwestern University.