患者の白内障手術教育における会話型人工知能チャットボットの利用：パイロット無作為化比較試験

To evaluate the effectiveness and usability of a safety-first, clinician-validated conversational AI chatbot for cataract surgery education compared with standard brochures.

University Hospital (UZ Brussel), Brussels, Belgium.

Prospective, single-center, randomized controlled trial.

Adults scheduled for cataract operation were randomized to receive either standard information brochures alone (control group) or brochures plus access to a hospital-specific chatbot ("Mina"). Primary outcomes were knowledge gain, change in pre- to post-information anxiety, and satisfaction. Those outcomes were measured with questionnaires. Secondary outcomes included chatbot usability (measured with the System Usability Scale, SUS) and engagement with the chatbot.

Sixty-four patients were randomized (chatbot group 33, control group 31). Postoperative questionnaires were completed by 35 patients (14/33 chatbot, 21/31 control). No significant differences were detected in knowledge gain, anxiety change, or satisfaction (p>0.05). Knowledge increased in both groups after receiving information (p<0.001). In the chatbot group, 17/33 (52%) did not engage with the chatbot. Participants engaging with the chatbot tended to be younger (mean age: 64.1 ± 10.9 years) than those who did not (mean age: 74.1± 10.5 years). Among users, 63% of submitted questions matched validated answers. The SUS mean score indicated high usability (83.1±12.1).

A custom-built chatbot with only clinician-validated responses showed high usability but did not improve knowledge, reduce anxiety, or increase satisfaction compared with brochures. Chatbot engagement barriers, particularly among older adults, and limits of validated-only content indicate the need for a hybrid approach of those models, to balance safety and flexibility in digital patient education.