Predictive Modelling Acute Exacerbations of Chronic Obstructive Pulmonary Disease

Student thesis: Doctoral ThesisDoctor of Philosophy (PhD)

Abstract

Chronic obstructive pulmonary disease (COPD) is a highly prevalent chronic respiratory disease
and a leading cause of disability and death globally. People living with COPD are at risk of
periods of sustained worsening of respiratory symptoms beyond their normal stable state and
are acute in onset. Current evidence suggests timely and accurate intervention in exacerbations
can reduce recovery time, severity, and frequency of exacerbations with marked improvements
in health-related quality of life. This PhD thesis aims to describe work on the development
of a COPD exacerbation support tool which will be a service within the myCOPD app that
can accurately predict acute exacerbations of COPD (AECOPD). A dual literature review of
remote patient monitoring (RPM) in randomised controlled trials (RCT) and machine learning
for AECOPD prediction highlighted machine learning has the potential to improve patient
outcomes. Data analysis of myCOPD patient data led to the creation of labels for exacerbations
and stable health. Patient users who matched the labels were more likely to be in a GOLD
(Global Initiative for Chronic Obstructive Lung Disease) group greater than A and B, with
more frequent use of rescue packs and more engagement with using myCOPD. Modelling of the
myCOPD dataset demonstrated the potential exacerbation predictive value of patient-entered
data in a real-world digital therapeutic. AdaBoost and EasyEnsemble Classifier models achieved
a sensitivity of 67% and 35% and a specificity of 69.5% and 89%, respectively. The Sensing,
Predictions, and Alerts in COPD Exacerbations (SPACE) Study I, using thematic analysis,
identified that a limited number of patient users understood that the machine learning models
were impractical for real-world application. There is a need to further improve model accuracy,
and develop a framework to build patient trust and understanding in predictive models. SPACE
Study II, a mixed methods study revealed that a group of myCOPD patient users were willing
to engage in physiological and functional sensors regularly to support exacerbation prediction.
However, the study suggested that maintaining long-term adherence to digital spirometry
among these individuals may prove challenging. Analysis of myCOPD patient-entered data and
sensor data shows the sensors’ predictive capability warrants further study. Continued research
following this PhD includes SPACE Study III, using sensor data to develop exacerbation
prediction models with greater accuracy and SPACE Study IV an interventional study to
identify the safety and efficacy of the finalised intervention.
Date of Award7 May 2024
Original languageEnglish
Awarding Institution
  • The University of Bristol
SupervisorJames Dodd (Supervisor) & Ian Nabney (Supervisor)

Cite this

'