Clinical operations
Real-world data, AI and digital twins: when does data become clinical evidence?
Explore China's 2026 investigator-initiated trial safeguards, rare-disease access dilemmas and operational implications for sponsors and CROs.
By Lampgen Clinical Research

2 min read · estimated
What the source reported
FDA’s October 22, 2026 CDRH Grand Rounds focused on its real-world evidence program and recent medical-device examples, including digital health. FDA’s device framework evaluates the relevance and reliability of real-world data for the proposed regulatory use. Data volume alone does not establish that a study can answer its clinical question.
A separate FDA January 2025 draft guidance for drugs and biological products proposes a risk-based framework for assessing AI model credibility within a particular context of use. It remains labeled draft, not for implementation. These are distinct regulatory documents: the CDRH session should not be described as blanket acceptance of digital twins or virtual efficacy endpoints in drug development.
What this means for trial teams
Lampgen’s interpretation is that trial teams should define the decision before selecting a dataset or model. Specify the population, treatment context, endpoint and intended role of the analysis. Then assess data provenance, missingness, outcome classification, follow-up completeness and whether the source population reflects the people the decision will affect.
Digital-twin applications also need to be distinguished. Simulating a protocol before enrollment, using a prognostic score as a prespecified baseline covariate within a randomized trial, and using an external comparator create different evidentiary questions. A simulated outcome is a prediction, not an observed patient outcome. Any claimed gain in efficiency needs support for the specific design and context.
For oncology operations, incomplete treatment histories, inconsistent progression assessments and missing survival follow-up are examples of problems that can undermine downstream analysis. CRAs, coordinators, data managers and study leaders help prevent these weaknesses through clear definitions, source traceability, appropriate monitoring and timely reconciliation.
Representation belongs in validation. Teams should examine performance and missingness across clinically relevant subgroups, care settings and access patterns. A model can reproduce gaps in the data from which it learns.
The potential impact is better-informed protocol choices and more efficient evidence generation when the approach is appropriately validated. These benefits should be demonstrated rather than assumed. Early regulatory engagement can clarify whether a proposed use is suitable. Lampgen supports oncology trial execution through CRA staffing and clinical-operations support; credible innovation depends on dependable work at the site level.