“A View into the CCU” – Investigating Visual Analysis and Machine Learning Approaches to Aid Clinicians in Critical Care Units

Abstract

Despite the vast quantities of data stored by critical care units, there is a resistance to applying this to technological innovation. This is limiting the potential to enable the best clinical decision making and increasing time pressures on clinical staff. This thesis examines the current state of data use in two critical care units at University Hospitals Plymouth. Through discussions with clinical staff, the main limitations of their current digital work environment are considered to develop a three-element system. A thorough literature review is then carried out into critical care practices, visualisation principles for multi-variate heterogeneous data, and supervised machine learning approaches. Using this background and the requirements from the clinical team, each of the three elements is developed to create a novel software suite. The first element is a cross-disciplinary visualisation tool, which combines streams of heterogeneous medical data into a unique dashboard. The second element is patient recovery prediction tool using LSTM recurrent neural networks to predict how a patient’s critical care metrics change over time. The final element is a bespoke natural language processing pipeline used to automatically classify clinical feedback into pre-defined categories. Each of these elements is applied to previously unused, real-world, medical data with state-of-the-art approaches. The software has been reviewed and positively received by the University Hospitals Plymouth and has innovated the clinical team to continue work in this area.

Date of Award	2024
Original language	English
Supervisor	Nathan Clarke (Director of Studies (First Supervisor)) & Marco Palomino (Other Supervisor)