Background: The European Medicines Agency (EMA) provides a list of designated medical events (DMEs), MedDRA Preferred Terms (PTs) for medical conditions that are serious and often medicine-related. [1] Developing phenotype algorithms to capture these DMEs in observational data (OHD) is necessary for generating real-world evidence to inform safety decisions.
Objectives: To develop and evaluate phenotype algorithms to identify DMEs in real world data such as electronic health records (EHRs) and claims.
Methods: A team of safety clinicians and epidemiologists followed a 7-step iterative process to develop phenotype algorithms for 62 identified DMEs. Step 1: Refined the 62 MedDRA terms of the DMEs into 31 clearly defined clinical conditions and developed a clinical description for each. Step 2: Performed a systematic literature review to identify existing phenotype algorithms from past studies. Step 3: Developed one or more phenotype algorithms for each DME using the OHDSI ATLAS tool [3]. Step 4: Implemented candidate algorithms across a network of 7 data sources from 4 countries (3 EHRs and 5 claims databases) and characterized the demographic and clinical profiles of identified individuals using the OHDSI CohortDiagnostics R package. [4] Step 5: Reviewed the group level characterization to assess consistency with expectations and identify possible misclassification errors. Step 6: A sample of cases were randomly selected by two clinicians to review the available structured data in the EHR data sources. When potential misclassification was identified, the phenotype algorithms were modified when possible. Step 7: Executed PheValuator R package to estimate the sensitivity, specificity, and positive predictive value (PPV) of the final algorithms.[2]
Results: The full list of DMEs and the final phenotype algorithms may be viewed on-line (https://data.ohdsi.org/icpe_2023_designated_medical_events/). Of the 31 DMEs, 4 were not feasible to implement in OHD. Estimated sensitivity and PPV varied by DME and ranged from 0.3 to 0.9. For most DMEs, sensitivity and PPV estimates were consistent across data sources.
Conclusions: Phenotype algorithms for 27 DMEs were developed and are publicly available for future studies. Measurement error estimates for DMEs should be considered for bias analysis.
