Daniel J. Riskin1,2, Lou Brooks3, Keri Monda4, Rayna K. Matsuno1, Brian Bradbury4
Background
Real-world data (RWD) support a range of applications, including precision medicine and point-of-care predictive analytics. The generation of high-validity real-world evidence (RWE) depends on data that are both relevant and reliable. The recent proliferation of RWD sources is enabling the advancement of large-scale, high-validity RWD in the healthcare sector.
Objectives
To build a high-validity Pragmatic Registry of patients with moderate-to-severe asthma using integrated claims-electronic health record (EHR) data and artificial intelligence (AI) technologies.
Methods
Patients were eligible if they were aged 12 or older and identified as being either diagnosed with or treated for moderate-to-severe asthma between 2015 and 2024. Data were sourced from US-based health systems as well as the Optum Market Clarity dataset, incorporating both claims and structured and unstructured EHR data for all patients. Unstructured data were processed at scale using AI technologies and validated by manual clinical annotation. Curation and enrichment were implemented using the Verantos Evidence Platform resulting in high-validity de-identified RWD optimized on the basis of both disease and data source. Data reliability was measured as accuracy, completeness, and traceability and calculated according to peer-reviewed published approaches.
Results
Out of over 200 million patients in the respective catchment areas, 349,600 patients with moderate-to-severe asthma were identified, with median follow-up time 83 months (interquartile range [IQR], 45-118 months). Median age at index was 55 years (IQR, 40-66 years). Patients were primarily female (64%), white (71%), and non-Hispanic (82%). Data reliability was 91% accuracy, 72% completeness, and 64% traceability; all optimized variables had accuracy >80%. Phenotypically critical data elements available exclusively within unstructured data, such as pulmonary function test values, were systematically curated with details including test, date, result, and unit. The combination of extracted attributes achieved the required accuracy levels,
thereby enhancing the value of these data compared to those traditionally derived solely from structured data.
Conclusions
The ability to generate high-validity, clinically relevant RWD at scale is needed to effectively address complex questions and support informed decision-making. Evidence from a clinical use case indicates that this level of data quality is attainable through the application of AI technologies to linked, extensive healthcare data that included claims and structured and unstructured EHR data. This approach establishes a framework for expanding the use of advanced data and technologies to benefit patient care across therapeutic domains.
