Exploring tuberculosis physicians’ preferences for AI explainability in China: a protocol for a discrete choice experiment
Jiale Zhang, Qian Fu, Xiaojun Wang, Luo XuIntroduction
Artificial intelligence (AI) is increasingly used in tuberculosis (TB) diagnosis, but its clinical adoption depends not only on accuracy but also on physicians’ trust and understanding of AI outputs. Explainable AI (XAI) has been proposed to address this challenge, but limited evidence exists on which explanation features are most valued by physicians in TB diagnostic settings. Discrete choice experiments (DCEs) offer a structured method to elicit physicians’ preferences and quantify trade-offs among explainability attributes. This protocol describes a DCE designed to examine TB physicians’ preferences for AI explainability to inform user-centred AI development and implementation.
Methods and analysis
Six attributes were identified through a comprehensive literature review, one-on-one semi-structured interviews with TB physicians and expert consultations. A D-efficient experimental design was used to construct choice sets. TB physicians in Hubei Province, China, will be recruited using a stratified random sampling approach. Preference data will be analysed using multinomial logit and mixed logit models to estimate the relative importance of attributes and explore preference heterogeneity across physician subgroups.
Ethics and dissemination
Ethical approval has been obtained from the Ethics Committee of Wuhan Pulmonary Hospital. All participants will provide written informed consent prior to participation. Study findings will be disseminated through peer-reviewed journal publications, conference presentations and academic forums, with the aim of informing the design and implementation of XAI systems for TB diagnosis.