Presentation (CTI)

OBJECTIVES: Regulatory review of PRO data has identified completeness, missingness, and interpretability as barriers to decision-making. We evaluated whether eCOA implementation failures create systematic missingness affecting compliance thresholds, endpoint estimability, and statistical confidence.
METHODS: We conducted a combined evaluation of cross-study compliance and training data, statistical/submission considerations, and desk research on regulatory expectations for PRO data. Key variables captured behaviors required of sites and patients, including training timing, setup accuracy, and assessment completion, evaluated across eCOA and complementary study data sources. Statistical considerations included missing data assumptions, sensitivity analysis requirements, and implications for treatment effect estimation.
RESULTS: Across studies, compliance patterns reflected failures at points where site and patient actions were required, rather than random variation. Late/missing training and incomplete/incorrect setup were associated with lower compliance, including failure to meet commonly used thresholds (e.g., 80%). These failures created structured patterns of missing/incomplete PRO data most apparent when integrating eCOA, EDC, and operational sources, and not visible within eCOA alone. Statistically, these patterns create additional complexity, potential bias, and increased uncertainty in treatment effect estimates by reducing usable data, increasing reliance on missingness assumptions, and requiring sensitivity analyses. Desk research reinforced the relevance of these findings, as regulatory and HTA review experience continues to identify data quality, missing data, and interpretability as barriers to PRO evidence use.
CONCLUSIONS: eCOA compliance problems are often managed operationally, but their consequences are statistical. Implementation failures can create non-random missingness that affects endpoint interpretability and introduces additional complexity, potential bias, and increased uncertainty in treatment effect estimates, and may therefore impact approval and reimbursement decisions based on those estimates. Protecting PRO evidence requires linking implementation, cross-system monitoring, and statistical planning. These findings suggest eCOA systems should be designed as behavioral systems to support site and patient actions, with predictive analytics used to identify high-risk implementation patterns before missingness accumulates.