Pain Level Modeling of Intensive Care Unit patients with Machine Learning Methods: An Effective Congeneric Clustering-based Approach

Abstract

Recent studies showed that machine learning can assist to better evaluate the induced pain level(s) on healthy individuals in a controlled environment. However, the role of these methods in clinical settings remained unclear and there is an unmet need to develop machine learning assisted tools in pain. The aim of this paper is to develop an automatic pain level assessment model based on patients' physiologic measures from the Medical Information Mart for Intensive Care (MIMIC III). There were two study phases; 1) pilot study, tested three existing machine learning methods proposed recently for healthy individuals. However, these yield poor performances in MIMIC III patients. 2) group study, a novel congeneric clustering method which divided patients into eleven categories and trained a dedicated model for each one. The clustering effectiveness of the proposed congeneric clustering method by showing the highest classification accuracy of 82.86%.