Patient-reported outcomes based on voice-based artificial intelligence (VAAIPRO): A fast and innovative QOL data collection application.

Abstract

e13658 Background: Health-related quality of life (QoL) is a key endpoint in oncology, typically assessed using validated paper or electronic questionnaires. Data collection faces major methodological and logistical challenges, including declining patient adherence, reporting burden, barriers to electronic PROs, staff workload, adherence to assessment schedules and data reformatting. As a result, QoL outcomes are often unavailable in the primary analysis. Therefore, we evaluated the feasibility of a mobile phone–based, artificial intelligence–driven patient-reported outcome (PRO) application that directly captures data in an executable format, automates assessment scheduling, and minimizes human intervention in data acquisition and processing. Methods: Health-related QoL instruments (EORTC QLQ-C30 and EQ-5D/EuroQol-5D) were simulated as voice-based questionnaires using a dedicated software application. Each participant first received a written questionnaire with instructions and then called a designated telephone number. An AI-driven voice agent conducted an automated interview, reading the items aloud and prompting numerical responses. The agent used an Automatic Speech Recognition system trained for multi-accent speech optimized for North Indian dialects. It analyses the voice signal in real time to detect fatigue or exhaustion and automatically logs all responses into structured Excel spreadsheets. Analysing exhaustion in voice, the agent was capable of rescheduling a follow-up call at the patient’s convenience, preserving the executed questionnaire context across interruptions. It also controlled the timing and frequency of longitudinal data collection by automatically initiating outbound calls at pre-specified dates and times. Results: The EORTC questionnaire was administered to 50 patients. The average time spent on completion was.12 minutes. On the first attempt, only 25/50 completely answered the questionnaire, 10 answered only a part, and 15 did not answer at all. A second attempt to continue questionnaire in patients who interrupted answering, was unsuccessful. At the four week follow up, only 11/25 (44%) completed the questionnaire. The most common reason for not answering was no call pick-up. A shorter Euro-QL was administered to 40 patients with an average completion time of 2.5 minutes. In this series 37 / 40 (96%) responded completely. The reason for not answering was non-mobile connectivity. Conclusions: Voice-assisted, AI-based PRO assessment is a novel approach that enables efficient, user-friendly, and timely data collection while reducing the need for human resources. The questionnaire length may hinder completion; however, this can be mitigated with better patient counselling and user support. If widely adopted and rigorously validated, VAAIPRO could substantially transform QoL data capture and reporting.