Remotely-captured, free-text responses track with patient health states in chronic pain

Abstract

Chronic pain (CP) is a complex, multidimensional condition characterized by physiological and psychological com- ponents. Advances in digital health have helped improve our understanding of CP by remotely tracking multiple symptoms on a more granular time scale, allowing for a more com- prehensive representation of patient experience. Concurrently, natural language processing (NLP) holds great promise in its ability to assess cognition, emotion, and acoustic properties to identify symptoms and disease. Here, we highlight the ability to track daily health and wellness in CP patients (n=206, over 15k samples) using sentiment features derived from free-text messages via smartphone applications. Using NLP, we quantified positive and negative sentiment and emotion in unstructured text messages using Sentence Transformers and Watson Natural Language Understanding. In both tools, sentiment and emotion track with a summary metric, termed as Pain Patient States, which have demonstrated prior success in describing a patient’s overall well-being. This helps validate language analysis as a snapshot representation of health, and indicates that Pain Patient States may track with emotion in CP.