Economic evaluation of artificial intelligence for breast cancer detection in the UK national breast screening programme
Economic evaluation of artificial intelligence for breast cancer detection in the UK national breast screening programme
Theme 4: Applied Economic Evaluation
There is a global shortage of breast radiologists (1) and in the UK, there is currently a 29% deficit in clinical radiologists (2). This shortfall is projected to increase to 40% within the next five years(2). One potential solution to address these challenges is the application of artificial intelligence (AI) (3). Several retrospective studies have indicated that AI can achieve diagnostic accuracy comparable to that of human radiologists when assessing breast screening mammograms(3). While these studies have yielded promising results, they are susceptible to biases due to their retrospective designs and do not evaluate how AI integrates into existing screening workflows. Recently, the initial findings from the first population-based randomised trial of AI-supported breast screening were published (4). This trial recruited 58,344 Swedish women aged 40–74 years, and compared three AI-supported mammography approaches (AI-only reading, one radiologist plus AI, two radiologists plus AI) with standard practice of two radiologists reading the mammogram screen. All AI alternatives detected at least as many cancers as standard practice, and the AI alternatives with fewer radiologists resulted in a significantly reduced workload for screen reading. Additionally, the approach involving two radiologists plus AI detected more cancers over the one year trial period (269 vs. 250) while maintaining a similar workload. Although this RCT provides encouraging results, the RCT design does not assess the integration of AI system into the UK screening programme, consider the effect of AI screening on the entire population of women eligible for screening, and consider its impact on important non-clinical outcomes such net population health (QALYs), NHS resources and workforce capacity.
1. Nightingale, J., Sevens, T., Appleyard, R., Campbell, S. and Burton, M., 2023. Retention of radiographers in the NHS: Influencing factors across the career trajectory. Radiography, 29(1), pp.76-83.
2. Royal college of Radiologists. Clinical Radiology Workforce Census.Aviliable at: https://www.rcr.ac.uk/sites/default/files/documents/rcr_clinical_radiology_workforce_census_2023.pdf
3. van Nijnatten, T.J.A., Payne, N.R., Hickman, S.E., Ashrafian, H. and Gilbert, F.J., 2023. Overview of trials on artificial intelligence algorithms in breast cancer screening–A roadmap for international evaluation and implementation. European Journal of Radiology, 167, p.111087.
4. Lång K, Josefsson V, Larsson A-M, et al. Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): a clinical safety analysis of a randomised, controlled, non-inferiority, single-blinded, screening accuracy study. Lancet Oncol 2023; 24: 936–44
This project would extend our existing modelling to assess the workforce capacity impact and cost-effectiveness of three alternative approaches for introducing AI-supported screen-reading into the UK national breast screening programme
Harry Hill, Jim Chilcott, Andrew Metry
Harry Hill harry.hill@sheffield.ac.uk
Background:
Artificial intelligence is changing medical practices, speeding up radiological workflows, and indicating the long-term chance of disease. There are early signs from trials, where images are viewed at random by either AI or experienced radiologists, that AI powered image recognition for breast screening works better than the experienced radiologists. This will make sure that things like fatigue, overwork etc do not lead to cancers being missed. This would also lessen the workload for radiologists. However, there are worries that AI might identify extra diseases which could lead to overdiagnosis and overtreatment.
Aims and objectives:
The researchers will extend their existing SCHARR breast cancer screening model to work out the cost comparison of three AI-supported breast cancer screening strategies in the UK. These strategies include AI-only reading, one radiologist plus AI, and two radiologists plus AI, compared to the existing practice involving two radiologists.
Methods:
The model will consider lifetime health outcomes, NHS costs, and changes to NHS staffing requirements. The research will use data from a randomised controlled trial already being carried out in Sweden and other possible sources of evidence where AI has been used for cancer detection in a national breast screening service.
Policy relevance & dissemination:
The policy implications of this research are shown by how it agrees with the NHS's aims to use AI within clinical practice. For example, the UK National Screening Committee has called for an in-depth economic analysis to see how AI influences clinical outcomes and the broader health of the population. A report will be produced, and the findings will be reported at an international academic conference.