Adoption and implementation of AI-driven Clinical Decision Support Systems in Cancer Care

Abstract

The Dutch healthcare system is facing increasing pressures on accessibility, quality, and affordability, making it unsustainable in its current form. This is particularly evident in cancer care, where rising incidence and treatment costs threaten timely and effective care. To maintain sustainable cancer care, innovations proven to improve outcomes must be quickly adopted. AI-driven Clinical Decision Support Systems (AI-CDSS) are often cited as promising tools to support adequate, patient-centered cancer care, reducing over-treatment and enhancing quality of life. Despite extensive research on AI-CDSS development, adoption and implementation remain limited, especially in the Netherlands. This thesis investigates why adoption of AI-CDSS in cancer care fails, using Mammaprint, a molecular diagnostic AI-CDSS for breast cancer, as a case study.

The research finds that the primary barrier to adoption is an institutional void in reimbursement by basic health insurance. Institutional voids refer to the absence or inadequacy of supportive structures, regulations, and frameworks. In the case of Mammaprint, the reimbursement process relies on the “state of science and practice” (SWP) criterion, which assesses whether clinical utility—demonstrated health benefits for the patient—is proven. Since no definitive requirements exist for diagnostic AI-CDSS, the SWP criterion is open to interpretation. Disagreements between policy analysts and medical specialists emerged regarding burden of proof, study design, and the trade-off between quality of life and survival, highlighting the institutional ambiguity that hinders adoption.

The study further reveals that the use of AI per se is not a determining factor in reimbursement. AI-CDSS adoption pathways vary depending on technology type (molecular diagnostics vs. image analysis) and use case (in-hospital vs. screening). Similar institutional voids were identified for other AI-CDSS, emphasizing that these challenges are not unique to Mammaprint. Broader contextual factors also affect adoption, including limited hospital-based development resources, regulatory hurdles for certification, and additional evidence requirements for marketing. Comparisons with other countries reveal significant differences between predominantly public European healthcare systems and private U.S. systems.

This thesis applies a case study methodology using interviews and grey literature. Eighteen interviews with nineteen stakeholders, including policy analysts and medical specialists, were analyzed through an institutional actor analysis framework. This framework maps formal and informal institutions, identifies key actors and interactions, and assesses their power and interests. Findings were categorized using a sequential framework of device adoption phases, allowing structured analysis of barriers across the innovation lifecycle.

The research identifies actionable recommendations for Mammaprint and similar AI-CDSS. Consensus is needed on the required burden of proof, appropriate study designs to establish clinical utility, and ethical trade-offs between quality of life and survival. Temporary admission policies can facilitate data collection while maintaining accessibility. Furthermore, clarifying reimbursement pathways for molecular diagnostics and image analysis-based AI-CDSS, alongside developing a strategic vision for AI-CDSS in Dutch cancer care, is essential to future-proof the system. Without a clear vision and strategy, AI-CDSS adoption risks being impeded, potentially undermining sustainability and quality of cancer care.

In conclusion, institutional voids in reimbursement are the main factor stalling adoption of AI-CDSS in Dutch cancer care. Addressing these gaps through consensus building, clearer frameworks, and strategic planning can enhance the uptake of AI innovations, contributing to a more sustainable, patient-centered healthcare system.