Application of a conversational AI system for voice-interactive history taking in neurosurgical patients

Background ccurate and structured medical history taking is essential in neurosurgical practice, but repetitive inpatient interviews can be time-consuming for both clinicians and patients. Recent advances in conversational artificial intelligence (AI) and speech recognition have created new opportunities to support clinical documentation and patient communication. This pilot study evaluated the feasibility and patient satisfaction of an AI-based voice-interactive history-taking system in neurosurgical inpatients. Methods This single-center, non-interventional pilot study enrolled 16 hospitalized patients undergoing microvascular decompression for hemifacial spasm or trigeminal neuralgia. Participants completed a diagnosis-tailored, AI-based voice-interactive interview using a bedside tablet device. The system presented questions sequentially, transcribed responses in real time, and generated a structured summary in an electronic medical record-compatible format. After the interview, patients completed a four-item satisfaction questionnaire using a five-point Likert scale. Results All 16 patients completed the AI-assisted interview without adverse events. Median patient age was 59 years (range, 29–76 years); 11 patients had hemifacial spasm and 5 had trigeminal neuralgia. Satisfaction scores were favorable across all domains, with median scores of 5 (IQR, 4–5) for comprehensibility, 4 (IQR, 3.75–5) for conversational naturalness, 4 (IQR, 4–5) for perceived convenience, and 4 (IQR, 4–5) for recommendation intent. Conclusion In this single-center pilot cohort of selected neurosurgical inpatients, an AI-based voice-interactive history-taking system was feasible to implement and was associated with favorable patient-reported satisfaction. Such systems may help standardize data collection and reduce the burden of repetitive interviews, supporting the potential role of AI-driven medical interview systems as an adjunct to routine clinical practice. However, because this study was not designed as a formal technology acceptance study and did not use a validated acceptance framework, these findings should be interpreted as preliminary. Larger studies using validated acceptance models and objective workflow measures are needed before broader clinical implementation can be recommended.