Evaluation Of Patient Experience In Mental Health Services Using A Bayesian Belief Network Model

Abstract

In recent times, mental health has been on the rise, with a major contributor being COVID-19. With mental patients being more susceptible to negative feelings and atmospheres, their experience must be perfect to be more willing and comfortable when visiting mental health institutes and hospitals. Mental health patients pose unique challenges as they are not treated the same as regular patients, they require special care and long-term care as they can not see immediate effects from the doctors and hospitals. This study aims to understand what are the major variables that affect their experience when getting the help they need, by using Machine Learning algorithms we can understand how they are interrelated and how they affect each other. By doing that we can help health institutions aim their efforts on the most important variables to focus on. To do that, we will be utilizing Bayesian Belief Network, a popular modeling technique.

The data was acquired from 2018-2021 as they were the most recent data and followed the same survey format. The survey had 36 questions divided into 11 sections. They covered Health and social care, Planning Care, Reviewing care, crisis care, Therapies, Support and well-being, Organizing care, medicine Feedback, and Overall in the last 12 months. The data were converted from continuous to discrete data. Discretized data then were changed into three types: Two-state, Three-state, and mixed-state. A BBN only works with discretized data, they were then modeled into 4 different network types that fit our purposes and goals. The algorithms used were Tree Augmented Naïve Bayes (TANB), Naïve Bayes (NB), Bayesian Search (BS), and Augmented Naïve Bayes (ANB). They were used due to their ability to have a parent variable which we needed to use in this research. Two-state TAG provided us with the highest accuracy with 0.87671. To grasp a better understanding of the results, sensitivity analysis was conducted along with scenario-based analysis which will be discussed at the end of the paper.

Date of Award	21 Dec 2023
Original language	American English
Supervisor	Mecit Simsekler (Supervisor)