Evaluation of spoken dialogue technology for real-time health data collection

Background: A real-time assessment of patients' experiences is an important methodology for studies in health care, quality of life, behavioral sciences, and new drug and treatment development. Ecological momentary assessment is a methodology that allows for real-time assessment of experience and behavior in a subject's natural environment. Recently, electronic data collection techniques have been introduced, including systems utilizing interactive voice response. Objective: The objective of this project was evaluation of spoken dialogue methodology for real-time data collection of information from patients for health, behavioral, and lifestyle studies and monitoring. While the management of the data collection process was Internet-based, this additional eHealth communication channel was based on over-the-phone natural language conversation with a dialogue system utilizing automated speech recognition technology. For this study we implemented a dialogue system for patients' assessment and monitoring of chronic pain. Methods: Experimental evaluation of usability of the Pain Monitoring Voice Diary was performed with 24 volunteers. The volunteers were asked to contribute 10 sessions with the system over a period of 2 weeks; in practice, the number of sessions per subject ranged from 1 to 20. The subjects. were asked to either relate to pain episodes in their past while answering the system's questions, or use as a guidance one of nine provided medical scenarios compiled by a pain specialist, ranging from migraines and back pain to post-surgical pain (knee injury) and cancer- and chemotherapy-related afflictions. Results: From 24 volunteers, we collected a total of 177 dialogue sessions: 171 sessions were completed, while the caller hung up in the other 6 sessions. There were a total of 2437 dialogue turns, where a dialogue turn corresponds to one system prompt and one user utterance. The data capture rate, measuring the percentage of slots filled automatically, was 98%, while the other 2% were flagged for transcription, Among the utterances sent to transcription, where the user had opted for the none of those option, 70% corresponded to the type of pain slot, 20% to the symptoms slot, and 10% to the body part slot, indicating that those are the grammars with the highest out-of-vocabulary rate. Conclusions: The results of this feasibility study indicated that desired accuracy of data can be achieved with a high degree of automation (98% in the study) and that the users were indeed capable of utilizing the flexible interface, the sessions becoming more and more efficient as users' experience increased, both in terms of session duration and avoidance of troublesome dialogue situations.