Feasibility of an AI-Based Interactive Voice Response (IVR) System for Patient-Reported Outcome Measures (PROMs) Collection: A Scalable Solution to Collect Patient-Centered Real-World Evidence

OBJECTIVES: Interactive Voice Response (IVR) technology allow patients to respond patient-reported outcome measures (PROMs) via phone calls, enhancing accessibility and engagement. Artificial intelligence (AI) and Natural Language Processing (NLP) technologies can improve transcription accuracy and IVR scalability. This study aimed to implement and pretest an IVR solution for PROMs collection, leveraging AI to collect, process and map voice open answers to PROMs answer options.

METHODS: The IVR PROMs collection pretest was conducted in a public mid-size hospital in northern Portugal using EQ-5D-3L. This PROM was adapted for IVR delivery, integrated with Promptly Collect software, to be delivered to 32 patients with cardiovascular disorders. After integration with an automated phone call provider, we employed an open-source neural network to perform NLP, converting speech-to-text and mapping open answers to PROMs answer options. Call status was analyzed using the Mixpanel. The users’ interaction with the IVR system and AI algorithm mapping accuracy was assessed by two reviewers.

RESULTS: Out of 32 patients, 2 (6.3%) had their phones off and 2 (6.3%) did not answered or rejected the call. From the remaining 28 patients who answered the call, 13 (46.4%) completed all the questions. The average call duration for completed surveys was 114 seconds (approximately 2 minutes). The AI algorithm correctly mapped 52.7% (29 of 55) of all answers. The main usability issues identified were: difficulty in understanding the chatbot, background noise and poor patient dictation. When considering only answers without usability issues, the accuracy was 80.0% (21 of 26 answers).

CONCLUSIONS: This pretest demonstrates the feasibility and utility of an AI-powered IVR system to scalably collect PROMs in real-world settings. Besides addressing challenges of background noise and clarity of speech, future steps include improving the technology with more recent AI models, applying it to more complex PROMs, and performing psychometric validation.