Phenoflow: A Microservice Architecture for Portable Workflow-based Phenotype Definitions.

Area of Science:

• Biomedical Informatics
• Clinical Research Informatics
• Data Science

Background:

• Phenotyping is crucial for identifying patient cohorts but definitions often lack clarity, hindering cross-institutional portability.
• Abstract phenotype definitions require computational implementation, which is frequently impeded by unclear practical guidance.

Purpose of the Study:

• To introduce a novel multi-layer, workflow-based model for phenotype definition.
• To present Phenoflow, an authoring architecture for developing structured definitions and realizing them as computable phenotypes.
• To evaluate the impact of the proposed model on the portability of phenotype definitions.

Main Methods:

• Development of a multi-layer, workflow-based model for phenotype definition.
• Creation of the Phenoflow authoring architecture to support structured definition development.
• Evaluation of the model's impact on code-based (COVID-19) and logic-based (diabetes) phenotype definition portability using real-world datasets.

Main Results:

• The proposed model and Phenoflow architecture significantly enhance the portability of phenotype definitions across different institutions.
• Demonstrated successful implementation and evaluation on a dataset of 26,406 patients at North-western University.
• The approach ensures the transparency of studies utilizing computable phenotypes.

Conclusions:

• The workflow-based model and Phenoflow architecture effectively address the challenges in sharing phenotype definitions.
• This work contributes to more transparent and reproducible clinical research by improving phenotype definition portability.
• Facilitates the reliable identification of patient cohorts for diverse research applications.