Economic evaluation of artificial intelligence in oncology: a scoping review of literature

Artificial intelligence (AI) has significant potential to address pressing challenges in healthcare, particularly in improving access to, and quality, and efficiency of cancer care. An increasing number of AI-based applications are being developed and implemented across various cancers, and in domains such as diagnostics, immunotherapy, radiation therapy, and research. This scoping review assesses the current literature on health economic evaluations of AI in oncology, examines methodological approaches used, and explores potential economic benefits and risks associated with integrating AI into cancer care. Literature on health economic evaluation (EE) of AI in oncology was reviewed from 1 January 2019 to April 2026. Papers were included if they included data on clinical effectiveness, resource utilisation, costs, or other relevant parameters related to AI systems. The key information from the papers was extracted, including the author, year, country, study design, population, care pathway phase (e.g. screening, diagnostic, planning, delivery). A total of 29 health economic evaluations were reviewed. Of these studies, 9 (31%) focused on breast cancer, 6 (21%) on colorectal cancer and 5 (17%) on lung cancer. AI-based technologies were most commonly evaluated in cancer screening, accounting for 22 (76%) studies. The studies were conducted primarily in Western developed countries, with 8 (28%) studies from Asia and 1 (3%) from Middle East. No studies were identified from Africa, Latin America, or other regions. Cost-utility analysis was the predominant evaluation approach, and Markov modelling was the most frequently used modelling technique. Most studies adopted a healthcare perspective. Overall, the evaluations reported economic benefits associated with AI implementation due to improved accuracy, labour cost reduction, and improved workflow. Results suggest that AI has potential to be both cost-effective and cost-saving, particularly in screening and diagnostic phases of cancer care. However, uncertainty remains regarding applicability of these results to real-world settings. The predominance of model-based evaluations using cost-utility and cost-effectiveness frameworks indicates that AI interventions are largely being assessed within traditional health technology assessment paradigms. However, this approach often overlooks critical aspects unique to AI, such as algorithmic adaptability, data dependency, and product lifecycle costs. Future evaluations should incorporate real-world performance data, dynamic learning costs, and implementation outcomes.