Patient, carer and clinical stakeholder views on using artificial intelligence and machine learning for the histopathological diagnosis of duodenal biopsies
En palabras de los autores
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 attitudes1 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.
Apareció: miércoles, 18 de noviembre. medRxiv. Preprint, todavía sin revisión por pares.