Improving Pain Management in Patients with Sickle Cell Disease from Physiological Measures Using Machine Learning Techniques

Abstract

Pain management is a crucial part in Sickle Cell Disease treatment. Accurate pain assessment is the first stage in pain management. However, pain is a subjective response and hard to assess via objective approaches. In this paper, we proposed a system to map objective physiological measures to subjective self-reported pain scores using machine learning techniques. Using Multinomial Logistic Regression and data from 40 patients, we were able to predict patients' pain scores on an 11-point rating scale with an average accuracy of 0.578 at the intra-individual level, and an accuracy of 0.429 at the inter-individual level. With a condensed 4-point rating scale, the accuracy at the inter-individual level was further improved to 0.681. Overall, we presented a preliminary machine learning model that can predict pain scores in SCD patients with promising results. To our knowledge, such a system has not been proposed earlier within the SCD or pain domains by exploiting machine learning concepts within the clinical framework.

Keywords: decision support; health informatics; machine learning; physiological sensing.

Figures

Fig. 1.

Sample Electronic Health Record

Fig. 2.

Intra-individual pain prediction accuracy results using MLR, SVM, KNN and RF

Fig. 3.

Intra-individual pain prediction weighted average F1 score results using MLR, SVM, KNN and RF