(C-117) Feasibility assessments in federated networks of disease-specific harmonized data assets: best-practice recommendations based on the PHederation experience

Background: Federated Data Networks [FDNs] can enhance Real-World Evidence [RWE] generation across various interconnected data assets which are usually standardized to a Common Data Model [CDM], protecting patient-level data through firewalls, and only sharing aggregate information. Key pre-requisites for a reliable, transparent, and reproducible FDN study are scientific alignment on the research question, and a concise feasibility and distributed analytics workflow.

Objectives: To share best practices for study feasibility assessments in disease-specific FDNs, based on learnings from a use case in PHederation, the Pulmonary Hypertension FDN.

Methods: The use case describes a feasibility assessment for a treatment pattern study in newly diagnosed Pulmonary Arterial Hypertension [PAH] patients, which aimed to better understand the patients’ initial treatment regimens.
The feasibility workflow included the following steps:
• Definition of feasibility objectives
• FDN stakeholder engagement: roles and responsibilities
• Qualitative feasibility assessment:
o IT and data infrastructure readiness assessment
o Detailed meta data transfer and evaluation
• Quantitative feasibility assessment:
o Preparation of feasibility protocol, analysis plan, and programming scripts
o Sharing these documents with the FDN partners for review, testing, and execution
o Return and consolidation of aggregate results
• Evaluation and reporting of feasibility results

Results: Four PHederation data assets with a total of 5939 patients participated in the quantitative feasibility assessment. Reasons for non-participation were staff shortages and exclusion of data assets that did not pass initial meta data evaluation.
The feasibility plan and R programming script (using RMarkdown for stepwise accuracy control) allowed assessing rates of PAH diagnoses, patient enrollment, and treatment initiation over time, and obtaining an overview of patient characteristics and initial treatment patterns. It demonstrated that there were enough patients for temporal stratification according to changes in treatment guidelines. It also allowed for agreement on important study eligibility criteria such as maximum time between diagnosis and database enrollment (6 months), maximum time between diagnosis and start of first treatment (3 months), and minimum follow-up time (2 months).
All four databases passed the feasibility assessment and contributed a total of ~2000 PAH patients from North America and Europe to the study.
Major challenges included differences in data provenance, treatment start and end dates, and database-specific custom source-to-CDM conversions.

Conclusions: The PHederation use case creates a blueprint for future feasibility assessments in disease-specific FDNs.