Patient, carer and clinical stakeholder views on using artificial intelligence and machine learning for the histopathological diagnosis of duodenal biopsies
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Objective
Biopsies are examined under the microscope by a pathologist, giving a diagnosis, which guides treatment and predicts prognosis. A shortage of pathologists and increasing workload leads to diagnostic delays, while for duodenal (small intestinal) biopsies, variable diagnostic concordance between pathologists limits accuracy. These challenges have motivated us to develop an artificial intelligence (AI)-mediated approach for the automated diagnosis of up to 80% of duodenal biopsies (i.e., most cases of the two most frequent diagnoses: coeliac disease and “normal”). We set out to understand patient, carer and stakeholder views regarding the adoption of such technology.
Methods and Analysis
We conducted mixed-methods stakeholder engagement (February 2024 to June 2026). We ran a national online patients’ and carers’ questionnaire (n=231 respondents) and four UK focus-group workshops, one of which additionally captured Likert-scale attitudes 1 immediately before and after a structured educational session (n=34 matched pairs). Healthcare professionals completed an online questionnaire (n=37) and took part in a structured workshop. Quantitative data were summarised descriptively; free-text and workshop data were analysed thematically and triangulated across all sources.
Results
Most patients and carers supported AI-reliant diagnosis (68.4%) and would strongly trust systems with formal regulatory approval (66.7%). Most wanted to be informed if AI was used (73.6%), but only 52.4% felt explicit permission was necessary. Qualitative themes centred on clear and available evidence of accuracy, human oversight, transparent communication and clear terminology, with the term ‘machine learning’ being preferred to ‘artificial intelligence’. Agreement (‘agree’ or ‘strongly agree’ combined) with statements relating to trust, acceptance and perceived benefit of AI was significantly higher after a structured educational session (38.2% to 67.6%, 85.3% to 97.1%, and 52.9% to 97.1% respectively; all p≤0.003). Healthcare professionals favoured an AI ‘assistant’ model over full automation, citing the need for validation, explainability, data governance and workflow integration. Patients also preferred some level of human oversight.
Conclusion
Patients and healthcare professionals are cautiously supportive of AI-assisted diagnosis of duodenal biopsies, conditional upon demonstration of accuracy, appropriate human oversight and transparent communication. Attitudes towards AI-assisted diagnosis were more favourable after structured explanation.
Key Messages
What is already known on this topic
Very little is known about patients’, carers’ or healthcare professionals’ views about the automation of diagnosis in pathology, using AI-based software solutions.
What this study adds
This is the first study to investigate patient, carer and healthcare professional attitudes specifically to AI-mediated duodenal biopsy diagnosis. Around two-thirds of respondents supported AI-assisted diagnosis, contingent upon clear and available evidence of accuracy, appropriate human oversight and transparent communication, together with appropriate regulatory approval of the technology.
Attitudes were markedly more favourable after a single structured educational session: agreement that AI could improve duodenal biopsy diagnosis nearly doubled and trust in healthcare AI rose substantially, suggesting that public views of AI in diagnosis can be influenced by clear explanation.
Patients and healthcare professionals both favour some ongoing level of human oversight of any AI-based diagnostic approach.
How this study might affect research, practice or policy
Our study suggests that a phased, AI-assistant-first deployment model with defined validation and audit checkpoints would be most readily acceptable when introducing diagnostic AI into histopathology.
It indicates that appropriate information and education, for example, a patient and public involvement and engagement (PPIE) session, can increase trust in AI in healthcare.
It shows that patients and the public are more trusting of technology for which the term machine learning is used, rather than artificial intelligence (AI).
It offers a stakeholder-engagement template for use by others introducing AI-based diagnostics.