Evaluation of a national risk stratified breast cancer screening programme based on short term breast cancer risk measured by an automated system at mammogram screening

In the UK screening programme women are currently invited to mammography screening every three years. The detection rate of cancers in the screening programme will increase if clinicians are able to quickly assess, at a routine cancer screening appointment, women’s risk of developing cancer in the period until her next screening appointment (i.e. in the next three years). Several models to assess risk of breast cancer have been developed, which are based on information on family history, hormone and reproductive history and some include genetic testing to establish polygenic risk scores (PRS). These have proved accurate at predicting breast cancer incidence in the medium term (10 years) and long term (over a lifetime). However, of more relevance to clinicians is establishing immediate risk – risk of cancer incidence until she examined again at her next screening appointment. The variables that are important for predicting medium- and long-term risk of cancer are not as relevant for short term risk. For example, gene susceptibility cancer is the most important predictor of lifetime cancer risk while mammographic density, which is a measure of the amount of fibroglandular tissue in the breast taken from a woman’s screening mammogram, is the strongest predictor of cancer in the short term (3-6 years).  

Recently an artificial intelligence model based on digital screening mammograms called Mirai has been designed to identify early signs of cancer. Mirai is a computer algorithm that uses routine data from a woman’s mammogram such as mammographic density. It also draws from the mammogram data additional information relevant to short-term risk that has a high chance of being detected using a computer-aided analysis, such as the difference in the number of calcifications between left and right breasts. As a result, the predictive performance of Mirai over in assessing short term (3-6y) risk appears to be stronger than all other approaches. A further benefit is that information needed for Mirai assessment is already collected at scale worldwide including at regular NHS screening appointments, unlike other risk assessment approaches that for example require genetic testing, which is not used at scale within the NHS, or involvement from specialist staff (e.g. trained screen readers to interpret risk from negative screen mammograms). Mirai involves using only the mammogram, a resource which the NHS already has, so there is no burden of questionnaires or additional tests on the provider or on the women attending screening appointments.

Aims

The aim of this research is to evaluate the cost-effectiveness of introducing novel risk-stratified screening programmes into the UK National Breast Cancer Screening programme. The risk regimens we will consider are novel first because risk is uniquely considered as the chance of cancer incidence in the intervening period to their next mammogram screen (which is also their next risk assessment). Second, the risk assessment instrument  (an artificial intelligence tool, Mirai) is easier to introduce into the National Screening programme than other risk assessment instruments because it is based on information produced automatically from a routine mammogram screen, and third, it has superior performance in predicting short term risk.

Project team

Harry Hill, Adam R. Brentnall (Wolfson Institute of Population Health, Queen Mary University of London), Stephen Duffy (Cancer PRU), Cristina Roadevin (Nottingham Clinical Trials Unit, University of Nottingham)

Contact

Harry Hill
harry.hill@sheffield.ac.uk

Plain English Summary

Report

Hill H, Brentnall AR, Duffy S, Roadevin C (2024) Evaluation of risk stratified breast cancer screening regimens that employ artificial intelligence driven short-term risk assessment. Policy Research Unit in Economic Evaluation of Health and Care Interventions. Universities of Sheffield and York. Report 072. DOI: https://doi.org/10.15131/shef.data.25219358