FROM PROMISE TO PRACTICE: EVALUATING THE HEALTH AND ECONOMIC IMPACT OF AI IN BREAST CANCER IMAGING SURVEILLANCE
Madelon Voets is a PhD student in the Department of Health Technology & Services Research. (Co)Promotors are prof.dr. S. Siesling and prof.dr.ir. H. Koffijberg from the Faculty of Behavioural, Management and Social Sciences, dr. J. Veltman from the Faculty of Science & Technology and prof.dr.ir. C.H. Slump from the Faculty of Electrical Engineering, Mathematics and Computer Science.
Medical imaging systems generate vast amounts of data, demanding advanced methods for efficient analysis and interpretation. Artificial intelligence (AI) has demonstrated remarkable potential in this domain, enabling more accurate diagnostics, earlier disease detection, and personalized treatment planning. As AI technologies continue to transform healthcare, their integration into clinical workflows and evaluation by health technology assessment (HTA) agencies require rigorous evidence of clinical validity, reliability, and cost-effectiveness.
The introduction of AI-driven tools into healthcare presents new methodological and practical challenges for HTA bodies, particularly in assessing their broader economic and clinical impact. Traditional health economic modelling approaches may not fully capture the dynamic and adaptive nature of AI systems, necessitating innovative frameworks that can evaluate both their short- and long-term value. Oncology provides an ideal setting to explore these challenges, given its complex treatment pathways and rapid technological innovation. Within oncology, breast cancer remains a major focus due to its global prevalence and the potential for AI-enhanced imaging to improve outcomes through more tailored surveillance strategies.
This thesis examines how AI can add value to medical imaging and how its health and economic impact can be systematically assessed. A review of recent health economic evaluations of AI applications revealed that most studies have focused narrowly on short-term cost savings rather than broader health outcomes, with limited transparency and insufficient assessment of uncertainty. These findings underscore the need for more comprehensive and methodologically robust evaluations to guide the responsible adoption of AI in healthcare.
Using extensive real-world data from Dutch national cancer registries and hospital records, this research explored variations in breast cancer surveillance practices and their adherence to clinical guidelines. The analyses revealed substantial inconsistencies in follow-up intensity, with some low-risk patients receiving more imaging than necessary. These patterns highlight the potential of AI-based risk prediction tools to support more personalized, risk-adapted surveillance strategies, optimizing resource use while maintaining quality of care.
To further investigate AI’s potential, a dynamic discrete event simulation (DES) model was developed to assess how AI-enhanced mammography sensitivity and alternative surveillance intervals could influence patient outcomes and healthcare costs. The model, calibrated with empirical data and expert input, demonstrated that AI could reduce missed recurrences and improve early detection. However, threshold analyses indicated that, under current assumptions, AI would be cost-effective only if its incremental health benefits were sufficiently large or its implementation costs relatively low.
Overall, the findings highlight both the promise and the limitations of AI in breast cancer imaging. While AI has the potential to enhance diagnostic accuracy and support more efficient, individualized care, current economic evaluations often fail to capture its full value, particularly non-traditional benefits such as improved patient autonomy, decision-making, and well-being. Future research should expand health economic frameworks to include these dimensions, ensuring that AI technologies are assessed not only for their efficiency but also for their contribution to more equitable, patient-centered healthcare.
