Digital health technologies present assurance challenges that existing regulatory frameworks were not designed to address. Unlike traditional medical devices with predictable failure modes, AI and software-based systems exhibit emergent behaviours, data dependency, and human-AI interaction effects that compliance checklists cannot capture. Multiple recent failures demonstrate these critical gaps. For instance, published trials have reported that AI-assisted colonoscopies paradoxically reduced cancer detection rates, as clinicians' checking behaviours changed over time with continuous AI assistance. 

Building on foundational statistical principles of risk, this presentation introduces a quantitative approach to digital health assurance, drawing on hard-won lessons learned from other sectors, including engineering and finance. Attendees will learn how to incorporate Failure Modes and Effects Analysis (FMEA), uncertainty quantification using computational statistics, risk and reliability analysis, and value of information methods for risk-proportionate monitoring. 

A worked example will show why "human in the loop" risk mitigation measures may provide less protection than assumed, using conceptual and statistical arguments. Attendees will leave with a principled approach for determining whether a digital health technology meets acceptable safety thresholds in a given clinical setting.