Please use this identifier to cite or link to this item:
Title: Analytics of patient flow patterns in a healthcare ecosystem : a blockchain approach
Authors: Abela, Stephen
Keywords: Medical care
Artificial intelligence -- Medical applications
Blockchains (Databases)
Machine learning
Issue Date: 2019
Citation: Abela, S. (2019). Analytics of patient flow patterns in a healthcare ecosystem : a blockchain approach (Master’s dissertation).
Abstract: Patient flow is a key concept in understanding the problem of ongoing shortage of hospital beds in healthcare. This lack of bed capacity has been extensively researched using various ICT techniques including data analysis, machine learning and information visualisation in order to improve operational efficiency, resource allocation and healthcare outcomes. In spite of these initiatives, there is a general tendency in healthcare, for collected data to remain in silos. This lack of sharing across organisations undermines any opportunities for data processing and the extraction of useful insights from this data. With the emergence of digital ledger technologies, however, new opportunities are being created that may help overcome limitations in existing systems, and advance the use of artificial intelligence for improving healthcare. The aim of this research was to explore a solution based on a design which facilitates understanding of patient flow by healthcare stakeholders using a combination of data analytics, machine learning and visualisation techniques. The system leverages on blockchain technology and its advantages, to promote data collection and to safeguard the security implications. Datasets used for our research were readily available from internet sources. A series of experiments were conducted to examine the characteristics, ascertain whether the data was fit-for-purpose and to explore the datasets using various machine learning techniques. The real de-identified dataset was found to be appropriate for the purpose of machine learning and visualisation, while the computer generated dataset was more suitable for exploring blockchain smart contracts. Previous research using machine learning on medical datasets showed many categorical attributes, imbalanced data and the need for partitioning data, with Decision trees and Random Forests being found to be particularly efficient. For our research, Random Forest was found to provide a better accuracy level of 0.719 than the Extra Tree and the Decision Tree machine learning tools. The implemented system incorporated a four tier system comprised of an application layer, a logical layer, a data storage layer and a blockchain layer. This architecture was used in developing a web based application that allowed access to different users and with specific views according to roles, and to provided machine learning and visualisation modules to analyse the datasets. Blockchain smart contracts were explored for sharing of aggregate and de-identified data over blockchain. Twelve experts with technical, academic and medical backgrounds participated in evaluating the system by providing usability feedback on their experience while carrying out a series of nine pre-determined tasks. As part of the evaluation, they were also asked to complete a survey reviewing the system as a whole and to make recommendations for further improvement. Participants acknowledged the machine learning and visualisation tools as the most commended aspects of the system.
Appears in Collections:Dissertations - FacICT - 2019
Dissertations - FacICTAI - 2019

Files in This Item:
File Description SizeFormat 
  Restricted Access
3.41 MBAdobe PDFView/Open Request a copy

Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.