Artificial intelligence to help healthcare professionals detect cancer in the UK breast screening programme
| ISRCTN | ISRCTN81384017 |
|---|---|
| DOI | https://doi.org/10.1186/ISRCTN81384017 |
| ClinicalTrials.gov (NCT) | Nil known |
| Clinical Trials Information System (CTIS) | Nil known |
| Integrated Research Application System (IRAS) | 336814 |
| Central Portfolio Management System (CPMS) | CPMS 59556 |
| National Institute for Health and Care Research (NIHR) | NIHR162840 |
| Sponsors | University of Warwick, Cambridge University Hospitals NHS Foundation Trust |
| Funder | National Institute for Health and Care Research |
- Submission date
- 03/11/2025
- Registration date
- 26/11/2025
- Last edited
- 26/11/2025
- Recruitment status
- Not yet recruiting
- Overall study status
- Ongoing
- Condition category
- Cancer
Plain English summary of protocol
Background and study aims
Artificial Intelligence (AI) is being used more commonly in the NHS. The EDITH study is looking to see how well AI could support the breast screening programme in the UK. It will test how AI can help NHS trained experts to look for cancer in mammograms.
Who can participate?
We plan to have 660,000 women take part in the study from approximately 30 breast screening centres in the UK. Women will be informed about the study when they get their screening (mammogram) invitation letter.
What does the study involve?
Within the study, some women will receive standard breast screening, where two NHS experts look at the breast images for signs of cancer. Other women will receive AI-assisted breast screening, where the AI helps the NHS expert(s) by also looking for signs of cancer. We will try out two different types of AI-assisted breast screening. The first type will involve the use of AI after the NHS expert to look for cancers that they might have missed. The second type will involve AI being used first to say whether there is a high or low risk of cancer. All images will be read by a human expert with final decisions about cancer detection made by standard human arbitration.
Women will be followed up until their screening episode is closed and then for a further four years through the screening programme and cancer registries to ascertain if they develop a cancer in the future.
Results from the three groups will be looked at to see how well humans combined with AI assistance works compared to the human readers alone and the costs and benefits of this.
What are the possible benefits and risks of participating?
There is a possible benefit in taking part in the EDITH study as some studies have shown increased cancer findings when AI is used. The information gained from this study should help improve the screening service offered to women in the future.
AI may highlight more areas as high likelihood of cancer than human experts. Some of these may not be cancer. This may mean more women are recalled for additional testing than if AI is not used. Some women may feel anxious and/or worried if they are asked to attend additional follow-up tests. The mammography and any further testing are part of your routine care and helps the experts examine the breast more thoroughly. In most cases there is no abnormality to be found.
These procedures use ionising radiation to form images of the breast and provide NHS trained experts with clinical information. Ionising radiation may cause cancer many years or decades after the exposure. The majority of participants in this study will not undergo any additional procedures. For those participants the chance of this happening is the same whether they take part in the study or not. Participants who may not have been recalled for additional testing, but are recalled due to participation in the study, will undergo additional exposure to ionising radiation. The risk arising from these additional exposures is extremely small.
Where is the study run from?
University of Warwick and Cambridge University Hospitals NHS Foundation Trust (UK)
When is the study starting and how long is it expected to run for?
August 2025 to July 2029
Who is funding the study?
National Institute for Health Research (NIHR) Health Technology Assessment (HTA) Programme (UK)
Who is the main contact?
University of Warwick study team, edith@warwick.ac.uk
Contact information
Public
Warwick Clinical Trials Unit
University of Warwick
Gibbet Hill Road
Coventry
CV4 7AL
United Kingdom
 ORCID ID |
0009-0001-6821-7543 |
|---|---|
| Phone | +44 (0)24 7615 1912 |
| edith@warwick.ac.uk |
Scientific
Division of Health Sciences
Warwick Medical School
University of Warwick
Coventry
CV4 7AL
United Kingdom
| ORCID ID |
0000-0002-1841-4346 |
|---|---|
| Phone | +44 (0)2476 575882 |
| s.taylor-phillips@warwick.ac.uk |
Scientific
Department of Radiology
University of Cambridge
Cambridge
CB2 0QQ
United Kingdom
| ORCID ID |
0000-0002-0124-9962 |
|---|---|
| Phone | +44 (0)12 2374 6439 |
| fjg28@medschl.cam.ac.uk |
Study information
| Primary study design | Interventional |
|---|---|
| Study design | Randomized; Interventional; Design type: Screening, Imaging |
| Secondary study design | Randomised controlled trial |
| Study type | Participant information sheet |
| Scientific title | Early Detection using Information Technology in Health (EDITH) |
| Study acronym | EDITH |
| Study objectives | 1. To integrate AI into the NHSBSP to measure the difference in clinical and cost-effectiveness between three study arms: replacing the second reader with AI (intervention 1); using AI for triage (intervention 2); and the current double reading pathway (control) 2. To understand differences in accuracy by AI vendor, mammographic acquisition system and by population subgroup (e.g. age, ethnicity, breast density, socioeconomic status) in a nested test accuracy study 3. To develop a comprehensive post-market surveillance methodology for national rollout 4. To monitor barriers to participation in women opting out 5. To calculate the cost of the current pathway using two readers compared with replacing the second reader with AI or using AI for triage from the perspective of the NHS |
| Ethics approval(s) |
Approved 13/11/2025, East of England - Cambridge East Research Ethics Committee (2 Redman Place, Stratford, London, EC20 1JQ, United Kingdom; +44 (0)207 104 8096; cambridgeeast.rec@hra.nhs.uk), ref: 25/EE/0197 |
| Health condition(s) or problem(s) studied | Breast cancer |
| Intervention | EDITH is a multi-centre, pragmatic, cluster, randomised controlled trial (RCT). Cluster will be at the clinic day level including 660,000 women across approximately 30 screening centres with opt-out consent. Each centre will undertake all three trial arms with one approved AI tool randomly allocated to be used at the centre. All women who attend participating screening centres will take part in the study, unless they opt-out. Cluster randomisation will be allocated 1:1:1 ratio across: Control arm: women will receive standard breast screening, where two NHS experts look at the breast images for signs of cancer. Two groups of women (intervention arms) will receive AI-assisted breast screening, where the AI helps the NHS expert(s) by also looking for signs of cancer. Intervention arm 1: AI is used after the NHS expert has looked for cancer and will look for cancers that they might have missed. Intervention arm 2: AI is used first before NHS experts look at the mammograms to say whether there is a high or low likelihood of cancer. This information is then used by the NHS experts, with knowledge of the AI decision available to them. If the likelihood is low one NHS expert will read the images and if the likelihood is high two experts will read the images, before a decision is made whether to recall women for further tests. In both cases, if there is disagreement on whether to recall women for further tests the case will go to arbitration (this is a process where a single third expert or panel of human experts review the images and make a final decision whether to recall the woman). Where the AI score shows a very high likelihood of cancer, the images will always go to arbitration. In all arms, all images will continue to be looked at by at least one human expert with final decisions about cancer detection made by at least two human experts. A nested test accuracy study comparing AI systems from participating vendors will be undertaken towards the end of the study to establish which AI tool performs best with each of the mammography vendors. This is a retrospective sub-study of mammograms from the main study taking place towards the end of the trial. This study has no impact on patient care pathway, as it does not involve any further contact with participants. A qualitative study exploring women’s reasons for opt-out and acceptability of AI in breast screening will be undertaken in a small number of women. A health economic evaluation will be undertaken to evaluate the cost-effectiveness of implementing different AI systems and AI pathways and includes a (a) micro-costing study to calculate the resource use and cost of using AI to read mammograms; and (b) a qualitative study to identify barriers and facilitators to the introduction of AI which will inform the health economic model. The trial is designed to answer whether each intervention is more clinically effective than current practice. This will be assessed using a randomised controlled trial design – the gold standard method for generating the strongest possible evidence. The inclusion of a control arm (double reading by two human experts) allows us to directly compare current cancer detection and recall rates with those achieved in the intervention arms, where AI is integrated into the screening pathway. We will use hierarchical models to assess the primary outcomes (cancer detection rate and recall rate). Hierarchical models are statistical methods used to account for clustering in data – in this trial, clustering occurs when women are grouped into clinic days contained within screening centres. In addition, an intention-to-treat analysis will be performed, meaning all participants will be analysed according to their randomised intervention, regardless of whether the intervention was delivered as intended. For instance, if AI is unable to read a mammogram – for example, due to breast implants or the mammogram being acquired via a method other than the standard 2D digital four-view mammography – the woman will receive standard care (double reading by two human experts) but will be treated as if she had received the intervention to which she was assigned. Each intervention arm will be compared separately with the control arm. For cancer detection, we will assess superiority. A difference of at least 1 additional cancer detected per 1000 women screened, compared to the control arm, will be considered clinically meaningful. This difference was chosen to align with the difference in cancer detection rate observed in MASAI, a Swedish trial of AI in breast screening [https://pubmed.ncbi.nlm.nih.gov/37541274/]. Thus, the null hypothesis is that the difference in cancer detection rate between the intervention arm and the control arm is less than 1 cancer per 1000 women screened (in either direction), and the alternative hypothesis is that this difference is greater than or equal to 1 cancer per 1000 women screened (in either direction). For recall, we will assess non-inferiority. If the intervention arm results in no more than 2 additional recalls per 1000 women screened, compared to the control arm, it will be considered non-inferior. This margin was established in consultation with PPIE representatives and researchers. Thus, the null hypothesis is that the intervention leads to more than 2 additional recalls per 1000 women screened (in the positive direction only), and the alternative hypothesis is that it results in 2 or fewer additional recalls per 1000 women screened (in the positive direction only). |
| Intervention type | Other |
| Primary outcome measure(s) |
1. Cancer detection rate (CDR) measured as cancers detected out of all women screened |
| Key secondary outcome measure(s) |
1. Arbitration rate measured as women requiring additional read by third person out of all women screened |
| Completion date | 31/12/2028 |
Eligibility
| Participant type(s) | Service user |
|---|---|
| Age group | Mixed |
| Lower age limit | 50 Years |
| Upper age limit | 70 Years |
| Sex | Female |
| Target sample size at registration | 660000 |
| Key inclusion criteria | 1. Women routinely invited for NHSBSP breast screening at participating centres during the trial recruitment period, typically screening target age 50-70 years but pragmatically include all screened women even if outside target age range (for example older women are allowed to self-refer every 3 years). 2. Women whose mammograms cannot be read by AI (for instance due to cosmetic breast implants, or not having standard 2D digital four-view mammograms acquired) will be included in the main trial intention to treat analysis comparing the trial arms. In practice they are effectively a test failure and will receive standard care (two human readers) regardless of trial arm. This analysis will deliver comparison of the AI pathway as it would actually be implemented for all women in comparison to current standard practice. |
| Key exclusion criteria | 1. Women being screened in the moderate and high-risk screening service (separate to NHSBSP) 2. Women being screened following the very high-risk protocol (separate protocol to population risk women: annual mammography/MRI) 3. Women who have opted out from participation in the study |
| Date of first enrolment | 01/05/2026 |
| Date of final enrolment | 30/11/2028 |
Locations
Countries of recruitment
- United Kingdom
- England
Study participating centres
80 Newark Street
London
E1 2ES
England
Hills Road
Cambridge
CB2 0QQ
England
Sandford Road
Cheltenham
GL53 7AN
England
London
SE5 9RS
England
Barrack Road
Exeter
EX2 5DW
England
Beckett Street
Leeds
LS9 7TF
England
Westcliff-on-sea
SS0 0RY
England
Southmead Road
Westbury-on-trym
Bristol
BS10 5NB
England
Holdforth Road
Hartlepool
TS24 9AH
England
Headley Way
Headington
Oxford
OX3 9DU
England
Grove Lane
Smethwick
B66 2QT
England
Tooting
London
SW17 0QT
England
Uttoxeter Road
Derby
DE22 3NE
England
Infirmary Square
Leicester
LE1 5WW
England
Stoke-on-trent
ST4 6QG
England
Colney
Norwich
NR4 7UY
England
Oxford Road
Manchester
M13 9WL
England
Hermitage Lane
Maidstone
ME16 9QQ
England
Clifford Bridge Road
Coventry
CV2 2DX
England
Derriford Road
Derriford
Plymouth
PL6 8DH
England
Wigginton Road
York
YO31 8HE
England
Results and Publications
| Individual participant data (IPD) Intention to share | Yes |
|---|---|
| IPD sharing plan summary | Available on request |
| IPD sharing plan | The datasets generated during and/or analysed during the current study will be available upon request from the University of Cambridge and the University of Warwick. At the end of the trial third parties may request access to the trial data and mammogram images via the Chief Investigators or the trial team at edith@warwick.ac.uk. The trial will form a Data Access Committee (which will include the Chief Investigator(s) and Trial Management Group members) to consider requests. No reasonable requests will be denied but all requests must have the approval of the Data Access Committee. Requests must be made according to Warwick Clinical Trials Unit data request Standard Operating Procedure. If any images or meta data is released to a third party, full anonymisation will take place. Participants are informed that their data may be shared anonymously with other researchers to support future research in the Patient Information Sheet. |
Study outputs
| Output type | Details | Date created | Date added | Peer reviewed? | Patient-facing? |
|---|---|---|---|---|---|
| Participant information sheet | Participant information sheet | 11/11/2025 | 11/11/2025 | No | Yes |
| Study website | Study website | 11/11/2025 | 11/11/2025 | No | Yes |
Editorial Notes
03/11/2025: Study's existence confirmed by the NIHR.
