# Does artificial intelligence provide value for money in oncology care? A personal perspective Author: Lastname F, Lastname F Pages: 7 Source: /home/steve/Documents/Books/Does_artificial_intelligence_provide_val.pdf ---

Page 1

Perspective Published: 2 April 2026 https://doi.org/10.20935/AcadOnco8235 1Health Economics and Management Unit, Department of Public Health and Primary Care, Ghent University, Ghent, Belgium. ∗email: lieven.annemans@ugent.be Does artificial intelligence provide value for money in oncology care? A personal perspective Lieven Annemans1,* Academic Editor: Godefridus J. Peters Abstract Artificial intelligence (AI) in healthcare holds the promise of more accurate, efficient, and accessible healthcare. In the field of oncology, we experience rapidly evolving possibilities with AI. This paper describes recent insights into the potential benefits of AI at four levels of the health production model: 1. health promotion; 2. early detection and screening; 3. treatment and follow-up; and 4. patient support. The scientific literature points to the (potential) benefits of AI at all of these levels, but also to its pitfalls. It is thereby striking that there is still an important shortage of real-world studies demonstrating the added value of AI, in terms of effectiveness, as well as the way it is embedded in healthcare processes and its impact on human interaction and human–machine interaction. In addition, the evidence regarding the value for money of AI in oncology is very limited. Hence, the potential of AI should be leveraged, and more efforts should be made to systematically allow for assessing its real added value and value for money in the field of oncology. Keywords: artificial intelligence, cost-effectiveness, real-world evidence, cancer Citation: Annemans L. Does artificial intelligence provide value for money in oncology care? A personal perspective. Academia Oncology 2026;3. https://doi.org/10.20935/AcadOnco8235

• Introduction The reported use of artificial intelligence (AI) in healthcare has been growing exponentially in recent decades [1]. AI is expected to be (and is already) reshaping healthcare, with the promise of more accurate, efficient, and accessible healthcare [2]. It has the poten- tial to contribute to Europe’s guiding principles of high-quality, equitable, and sustainable healthcare [3]. The field of oncology is, herein, obviously not an exception. Given the rapidly evolving possibilities with AI on the one hand and the trend towards the digitization of a wide variety of cancer-related data on the other, AI applications increasingly find their way into different aspects of oncology, such as screening, diagnosis, and treatment, hence covering the entire cancer care continuum [4, 5]. This evolution is generally welcomed, given the expected increase in incidence and prevalence of cancer [6]. More concretely, AI has the potential to address cancer and im- prove oncology patients’ health at all four levels of the recently published health production model [7]:
• Health promotion, i.e., keeping people healthy and therefore preventing cancer [8];
• Early detection and diagnosis before the presence of clinical signs, via screening and biomarker-based technologies [9];
• Treatment of cancer and its follow-up [9];
• Support if cancer is no longer curable [10]. Yet, AI in cancer care is also associated with pitfalls, such as lack of transparency regarding the underlying data and algorithms, biases in the data used to train AI systems, which can result in inaccurate outcomes, concerns about accountability and liability, in case things go wrong, and concerns about privacy violations arising from the collection of large data sets [11]. These challenges have likely prevented the full-scale adoption of AI technologies in healthcare and in oncology accordingly [12]. Given the limited healthcare resources, the fundamental objective of healthcare policy is to make the best use of these resources in order to promote health and provide healthcare. The underlying principle can be seen as maximizing value for money by selecting the optimal mix of services subject to the budget constraints faced by the system [13]. This means that the question about value for money also needs to be addressed when considering the use of AI in cancer care. This personal perspective first provides a brief explanation of added value and value for money, followed by a recent overview of the potential added value of AI in cancer care and its potential value for money at each of the four levels of the health production model. Rather than a systematic review, which would be out of the scope of this personal perspective, a selection of best cases and typical cases was made to illustrate the potential added value and the currently limited evidence on the value for money of AI in oncology care, thereby also pointing to AI’s pitfalls. The paper concludes with future challenges and the proposed ways forward.
• Added value and value for money—a primer The major aim of innovative treatments in cancer care is to add ACADEMIA ONCOLOGY 2026, 3 1 of 7

  Page 2

  https://doi.org/10.20935/AcadOnco8235 benefit to patients in terms of quality of life and life expectancy. In the field of oncology, this added benefit has been proposed to be expressed in the magnitude of the clinical benefit scale [14]. Given the limited healthcare resources, there is a need for health eco- nomic evaluations to address the question of the value for money of innovations [15]. Health economic evaluations are comparative analyses whereby both the costs and the health consequences of a new intervention are compared to the standard of care (SoC). Often, the innovative treatment comes at a higher cost than the SoC, but this might be partially compensated for by cost offsets elsewhere in the healthcare system. The resulting net cost (called the incremental cost) is then divided by the difference in health effects between the new and the current treatment (the incremen- tal effect). The latter is mostly expressed in quality-adjusted life years (QALYs). QALYs are calculated by multiplying the time that patients are in a given condition by the quality of life (in terms of jargon, this is called ‘utility’) associated with that condition, whereby the quality of life is expressed on a scale between 0 and

• One QALY is equal to living one year in perfect health (score = 1). For instance, a patient with a life expectancy of 5 years at a quality level of 0.6 will have 5 × 0.6 = 3 remaining QALYs. The incremental cost-effectiveness ratio (ICER) is then the ratio between the incremental costs and the incremental effects [15]. For instance, a treatment with an incremental cost of 60,000€ and leading to two additional QALYs will cost 30,000€ per QALY gained. It is then up to policymakers to decide what is an accept- able cost per QALY gained. For instance, in the UK cancer drug fund, the threshold of societal acceptance = 50,000£ per QALY gained [16]. Innovations with an ICER exceeding such a threshold are considered not cost-effective. Of note, when the incremental costs are negative, i.e., the cost offsets are larger than the upfront investment, and QALYs are gained, then the new treatment is considered to be ‘dominant’. A key challenge in health economic evaluations is that at the time of the evaluation, evidence from the real-world use of health technology is still lacking. This leads to uncertainties about the added value and value for money. In the overview included in this paper, this limitation will receive special attention.
• Artificial intelligence (AI) and health promotion to reduce the risk of cancer Health promotion can be defined as “the process of enabling people to increase control over, and to improve, their health”. It goes beyond individual behaviour change to include a wide range of social and environmental interventions aimed at addressing the broader determinants of health such as income, housing, employ- ment, and living conditions [17]. Measures for smoking cessation, reducing UV exposure, increasing physical activity, and improving healthy nutrition have been shown to significantly reduce the risk of several cancers, as has been confirmed and summarized in a recent review [18]. AI has the potential to make health promotion more effective and, as such, add value. A recent review by Yousefi et al. (2025) points to the benefits of AI-driven apps, chatbots, and interactive websites to facilitate patient access to health promotion activities and motivate people to adopt healthy behaviours and adhere to lifestyle programmes, via improved personalized coaching and targeted health communication, thereby also reducing providers’ workloads [19]. Also, the use of AI-generated influencers has been shown to contribute to addressing modifiable cancer risk factors, such as tobacco consumption, unhealthy diet, sun ex- posure, alcohol consumption, and Human Papillomavirus (HPV) infection [20]. Amil et al. [21] describe the potential effectiveness of AI-driven conversational agents in the promotion of healthy food patterns. Although the results were mixed, it was found that features such as goal setting, frequent feedback, and tailored recommendations were linked to better results, thereby contributing to a reduction in the incidence of unhealthy nutrition-related cancers. Never- theless, the authors reported challenges such as the unnatural conversation style of the AI tools, the simplistic content, and limited perceived usefulness by several participants. Also, none of the involved studies investigated the real value for money of the AI tools. Yousefi et al. identified some additional remaining challenges with the use of AI for health promotion purposes, namely the lack of long-term evidence, the need for human oversight to avoid mistakes, AI’s access to underserved populations (despite its potential), and the lack of engagement from participants in the development of AI tools [19]. From the scarce evidence available, it seems that AI has clear po- tential to add value in health promotion in general and specifically with the aim to reduce cancer incidence, but also, this potential needs to be confirmed in real-life settings. Moreover, research on the value for money of AI applications in the setting of health pro- motion is still an untapped domain. The WHO has called health promotion ‘the best buy’ in healthcare policy [22], but whether AI will make it an even better buy still needs to be investigated.
• AI and early cancer detection Current evidence supports the use of AI in the early detection of cancer, particularly via improved imaging or the identifica- tion of prognostic and predictive biomarkers. In a recent large Swedish study, 105,000 women who were eligible for mammogra- phy screening were randomly allocated to AI-supported screening or standard double reading. In the AI-supported group, screening resulted in a 24% increased detection of invasive cancers (mainly small lymph-node-negative cancers) and 51% increased detection of in situ cancers. The recall and false-positive rates were not significantly higher in the AI-screened group. The authors also pointed to a 44% reduced screen-reading workload, supporting the potential cost-effectiveness of AI [23]. AI also has a transformative potential in the detection of other cancers, such as prostate cancer. A systematic review by Rajih et al. [24] describes how AI tools enhance the interpretation of multiparametric MRI (mpMRI) by improving lesion detection, segmentation, and risk stratification, thereby reducing unnec- essary biopsies, pointing to possible efficiency gains. Progress is expected from novel biomarkers based on multi-omics data (imaging, genomics, transcriptomics, etc.), as, for instance, shown by Khalili-Tanha et al. in the field of breast cancer [25]. The emerging evidence of using AI in cancer screening has also resulted in evidence on its cost-effectiveness in this domain. In a systematic review of health economic evaluations of AI, of the 19 included papers, two were related to the early detection of breast and lung cancer, both with positive (i.e., cost-effective) ACADEMIA ONCOLOGY 2026, 3 2 of 7

  Page 3

  https://doi.org/10.20935/AcadOnco8235 outcomes [26]. Yet the authors of this review point to several methodological weaknesses in the economic evaluations of AI tools overall. The cost of AI (upfront capital expenditures, soft- ware integration costs, and ongoing maintenance expenses) was often omitted. Moreover, the results are generally very sensitive to the diagnostic accuracy. The (in)efficiency of the integration of AI in existing workflows and its impact on cost and health is rarely considered. The authors conclude that more real-world evidence is needed to support AI’s value for money. In lung cancer screening, cost-effectiveness was investigated based on a recent systematic review assessing 27 papers [27]. The authors conducted a de novo cost-effectiveness analysis and found that for symptomatic and incidental populations, AI-assisted CT image analysis was cheaper than the AI-unaided radiologist and delivered more correct detections. However, when the full clinical pathway was considered, including the impact on quality-adjusted life years (QALYs), the AI-assisted approach was no longer better; on the contrary, the worse cost-effectiveness was explained by the costs and loss of quality of life associated with false-positive results and increased CT surveillance. The challenges with AI applications clearly affect their potential value for money. Rajih et al. [26] refer—among other challenges— to the need for more prospective validation of the potential and to integrating AI into existing workflows. It seems that the use of AI in cancer screening has clear potential to add value, which has already been shown in some cancer types in high-quality studies. Also, the cost-effectiveness has been largely studied, often with good results. But as was the case with its use in health promotion, the use of AI in screening also requires more real-world confirmation, thereby also accounting for all related aspects, such as embedding AI in workflows and the training of health professionals.

• AI and cancer treatment Once cancer is identified, AI plays an increasing role in aiming to improve its treatment, via clinical decision support, predictive and precision medicine, treatment adherence programmes, and monitoring of adverse events [5]. Ample evidence is already avail- able regarding the added value of AI in several of these treatment- related aspects, albeit often focusing on surrogate and short-term endpoints. AI-driven clinical decision support systems have been trained to conduct a thorough analysis of diverse patient data such as electronic health records, imaging, pathology, genomics, and biomarkers [28]. For instance, in prostate cancer, such multi- modal AI models, synthesizing imaging, biomarker, and clinical data, create robust predictive tools for superior clinical decision support [26]. AI’s predictive capacity also enables a more tailored approach in terms of risk assessment. This was illustrated in a recent study in renal cell cancer (RCC), where AI enhanced the prediction of the survival time of RCC patients undergoing targeted drug therapy. The authors concluded that their AI-based prediction model offers