Algorithmic approaches to measuring and visualising risk of falling by older adults

Abstract

Falls are the main cause of serious injuries and accidental deaths in older adults. Adults over the age of 65 years suffer from cognitive decline, which in turn makes them more vulnerable and prone to falls. Unfortunately, falls and their consequences have a huge impact on older adults, their families, and caregivers. When a fall occurs, health and social care professionals, individuals and their families need to work together to ensure the best care possible is given to the vulnerable older person. The aim of this research was to assess the feasibility of a computer-based risk visualisation tool for decision making to investigate risk factors for older adults and aim to help and understand the contributing factors to predict harm levels for older adults and hence provide decision support for health and social care professionals.

This PhD comprises several component parts including:

- a review of literature related to social science perspectives on risk and decision making.
- a review of literature related to computational machine-learning models.
- a theoretical exploration of the potential of types of decision trees to distinguish accuracy scores of risk probabilities and machine learning approaches.
- a series of statistical and computational modelling techniques using the Irish Longitudinal Study on Ageing (TILDA) dataset was used to examine the use of various statistical and machine learning algorithms to predict the risk of falls, the risk of recurrent falls, and gender differences; and
- a study of visualisation approaches, whereby health and social care professionals (including Social Workers, Nurses and Occupational Therapists) completed a questionnaire on what visualisations they currently used in their daily work; their views on presented visualisations that may be of benefit; and how such visualisations might help in decisions about the care of older adults.

Statistical and machine learning approaches were utilised to predict falls in older adults using health and social care risk factors whereby the results reflect the risk factors contributing to falls in older adults. Four different decision tree algorithms (J48 Decision Tree, Fast and Frugal Tree, Conditional Inference Tree and Classification and Regression Tree) were used with health and social care risk factor cues with the highest performing tree having an accuracy of 69%. Machine learning results provided similar accuracy results to decision trees when looking at results with accuracy scores between 56-69%. A further experiment using machine learning classified whether an individual was likely to fall in the next year based on the previous number of falls which resulted in higher accuracy results ranging between 88-90%.

Both statistical and machine learning algorithms have demonstrated promising results suitable for supporting decision making by health and social care professionals. Suitable methods of visualising this data in terms of potential harm levels allows machine learning algorithms to become more understandable to professionals, enabling their use in delivering services delivery for the benefit of older people and their families.

Date of Award	Apr 2023
Original language	English
Supervisor	Anne Moorhead (Supervisor), Sonya Coleman (Supervisor), Dermot Kerr (Supervisor) & Brian Taylor (Supervisor)