Authored by Jessica Santos, Konovo Chief Compliance and Privacy Officer, Elise Berliner and Michael R. Fronstin
This article explores the growing use of agentic AI in healthcare analytics and the critical question of whether these systems can be trusted to generate valid real-world evidence. Agentic AI platforms allow users to run complex analyses without coding, autonomously determining how to structure studies and interpret results. While this technology creates significant efficiency and scalability, it also raises concerns around transparency, reliability, and the ability to validate outputs in high-stakes healthcare environments.
To address these concerns, the authors apply the ELEVATE-GenAI framework to evaluate an agentic AI system, testing its performance across multiple research scenarios. Their findings suggest that, contrary to the “black box” perception, these systems can be transparent when designed with human oversight. Researchers remain actively involved through structured planning steps such as query validation, data selection, and statistical design, ensuring that the AI’s decisions are reviewed and guided. The system also documents each step of the analysis and references external scientific sources, supporting reproducibility and trust.
The article concludes that while agentic AI holds strong potential to accelerate research, its outputs are only as reliable as both the evaluation framework and the underlying data. Validity still depends on fit-for-purpose data, appropriate study design, and human accountability. Frameworks like ELEVATE-GenAI provide a strong starting point for assessing these systems, but continued refinement and oversight are necessary as the technology evolves. Ultimately, trust in agentic AI will come from a combination of rigorous validation, transparency, and responsible human involvement.
If you like to read the full article click the link below.
