Exploring Chronic Pain Experiences: Leveraging Text and Audio Analysis to Infer Well-Being Metrics

Abstract

Current advancements in digital health offer the promise of novel insights into chronic pain patients by combining subjective data from questionnaires and objective measures that are broadly available. However, unstructured information from speech data, which captures patients expressing themselves in their own words, has not been thoroughly analyzed in this area. Recognizing this limitation, we have implemented an approach to analyze thousands of chronic pain patients' responses (text and voice) from a longitudinal spinal cord stimulation study, where patients were asked to answer different prompts about their day and the recommendations provided to them to improve their outcomes. We used a large language model and acoustic techniques to extract features and infer seven well-being metrics in a cross-validated approach, including an overall health status assessment and a disability test, achieving statistically significant correlations of up to Spearman r=0.46.