A metadata framework for electronic phenotypes

As many phenotyping algorithms are being created to support precision medicine or observational studies using electronic patient data, it is getting increasingly difficult to identify the right algorithm for the right task. A metadata framework promises to help curate phenotyping algorithms to facilitate more efficient and accurate retrieval. We recruited 20 researchers from two phenotyping communities, the eMERGE and the OHDSI communities, and used a mixed-methods approach to develop the metadata framework. Once we achieved a consensus of 39 metadata elements, we surveyed 47 new researchers from these communities to evaluate the utility of the metadata framework. Two researchers were also asked to use it to annotate eight type 2 diabetes mellitus phenotypes. The survey consisted of a series of multiple-choice questions, which allowed rating of the utility of each element on a scale of 1-5, and open-ended questions, which allowed for narrative responses. More than 90% of respondents rated metadata elements concerning phenotype definition and validation methods and metrics with a score of 4 or 5. Our thematic analysis of the respondents’ feedback indicates that the strengths of the metadata framework were its ability to capture rich descriptions, explicitness, compliance with data standards, comprehensiveness in validation metrics, and ability to enable cross-phenotype searches. Limitations were its complexity for data collection and entailed costs.