Analysis of chronic pain descriptions for base-pathology prediction: the case of rheumatoid arthritis versus spondylitis pathology prediction based on pain descriptions

Abstract

The language of pain is a sub-language used to describe a subjective, private, and painful experience. In the clinical assessment and management of chronic pain, which is not as straightforward as acute pain, verbal communication is key to convey relevant information to health professionals that would otherwise not be accessible, namely, intrinsic qualities of the painful experience and that of the patient. We raise the hypothesis of applying Natural Language Processing techniques to transcribed verbal descriptions of chronic pain, to capture that information in the form of linguistic features that characterize and quantify the experience of pain of each patient. Furthermore, we demonstrate the application of these features for base-pathology prediction, specifically regarding the diagnosis of rheumatoid arthritis and spondyloarthritis. A dataset of verbal descriptions was collected for this work, considering 85 patients. The descriptions were obtained by having each patient freely answer to an interview of seven questions. The dataset was pre-processed, and features were extracted, which were then fed into binary classification machine learning models. We obtained an accuracy of 79%, in a Leave-One-Out cross-validation fashion. Based on an extensive experimental setup, we conclude that the computational analysis of the language of pain can potentially extract useful information to aid health professionals, in this case, focusing on base-pathology prediction. We also conclude on which semantic features provided more useful information for the task (distribution of pain on the body), and which did not.