Developing a Novel Methodological Paradigm for Live Clinical Evaluation (Silent Trials) of AI Models to Support Accountable Translation and Governance

Humphrey's doctoral research investigates how artificial intelligence systems can be ethically evaluated and governed before their integration into routine clinical practice.

Although AI technologies demonstrate considerable potential to improve healthcare delivery, many models are evaluated primarily using retrospective datasets and conventional performance metrics. Such evaluations may not adequately capture the ethical, organisational and operational challenges that emerge when AI systems are introduced into real-world clinical environments.

His research focuses on silent trials, an approach in which AI models operate prospectively using live clinical data without influencing patient care or clinical decision-making. This enables researchers and healthcare institutions to evaluate model performance and operational behaviour under real-world conditions before active clinical deployment.

The research aims to:

1. Examine existing approaches to clinical AI evaluation and identify ethical, organisational and governance gaps.
2. Develop an operational, ethics-informed framework for institutional evaluation, monitoring and accountability throughout the clinical AI lifecycle.
3. Establish a normative justification for responsible clinical AI evaluation, with particular attention to fairness, accountability, transparency and trust.

Using a multiphase mixed-methods approach, the research integrates qualitative investigation of institutional perspectives, quantitative assessment of operational evaluation indicators, and expert-informed framework development and validation.

The intended contribution is a practical methodological and governance framework that supports healthcare organisations in generating meaningful evidence about AI systems before clinical implementation, thereby strengthening patient safety, institutional accountability and responsible AI translation.