Acceptance of Machine Learning for Medication Selection in Epilepsy to Inform Clinical Trial Design: Co-Design Survey Study.

Background: Antiseizure medications (ASMs) are the mainstay of epilepsy treatment; however, there is currently no reliable way to predict which medication will be most effective for an individual patient. Machine learning (ML) approaches are increasingly being explored to support personalized ASM selection, but successful implementation will depend on acceptance by both people with epilepsy and treating neurologists. Understanding factors influencing acceptability of ML in clinical decision-making is therefore critical to support engagement, trust, and adherence. The extended unified theory of acceptance and use of technology (UTAUT2) framework has previously been applied to evaluate acceptance of health care technologies, but its suitability for ML-supported ASM selection has not been established. Objective: The objective of this study was to co-design a UTAUT2-based questionnaire to measure ML technology acceptability that is suitable for use in a clinical trial. Methods: Adults living with epilepsy and prescribing neurologists were recruited using purposive sampling to participate in a co-design process evaluating the relevance, comprehensiveness, and clarity of the UTAUT2 framework in this clinical context. Participants completed an online survey that collected structured feedback on existing UTAUT2 constructs and identified additional factors influencing ML acceptance. Quantitative responses were analyzed descriptively, and qualitative responses were analyzed thematically to inform adaptation of the framework and development of a context-specific ML acceptability questionnaire. Results: A total of 32 participants completed the survey, including 22 (68.8%) adults living with epilepsy and 10 (31.2%) neurologists. While participants considered core UTAUT2 constructs relevant, qualitative feedback identified additional domains influencing ML acceptance, including emotional attitudes, perceived risks, knowledge enhancement, conflicts in shared decision-making, and contextual factors such as workplace policy and regulation. On the basis of this feedback, the ML acceptability questionnaire for ASM selection retained relevant UTAUT2 domains and incorporated additional constructs addressing trust and perceived risk. Other themes provided contextual insights for interpretation of ML acceptability and future implementation. Conclusions: This study highlights the importance of co-design when adapting existing frameworks to the specific clinical context. The co-designed ML acceptance questionnaire for ASM selection developed in this study can be used to evaluate the acceptability of ML technology in the field of epilepsy, strengthening future trials and supporting ongoing technology development.