Evaluating Cancer Patients’ Experience in Health Services using Bayesian Belief Network Models

Abstract

Cancer patients lead more difficult lives than other types of patients. The cancer journey starting from the moment of suspicion of the disease, going through diagnosis, GP and hospital visits, in-hospital and home treatments, and ending with the post-treatment phase is very difficult to handle for the cancer patients and their care providers. Multiple studies have showed that to ensure high healthcare quality is being delivered to the cancer patients, healthcare institutions have to take cancer patients’ experience as a key performance index. Therefore, lately, many questionnaires and surveys have been widely used to capture the overall satisfaction cancer patients experience throughout their tough journey with cancer. Nonetheless, despite the commonality of types of research involving predictive models developed to predict the patents’ satisfaction as a function of some critical-to-quality healthcare factors, there is very limited research reporting how the probabilistic interdependencies among these factors are driving the overall cancer patients’ satisfaction. Hence, in this study, Bayesian Belief Network (BBN), a probabilistic graphical modelling technique, has been developed to address this issue. The study used data from 2016 to 2019 of the cancer patients’ satisfaction survey, designed and managed by National Health Services (NHS) in England. The original survey contains 59 questions categorized under 12 sections: Seeing your GP, diagnostic tests, finding out what was wrong, treatment, clinical nurse specialist, support for people with cancer, operations, hospital care as inpatient, hospital care as outpatient, home care and support, care from GP, and overall NHS care. The survey questions were recategorized under only seven new sections (factors) and a target variable to increase the internal reliability of the survey. The dataset was then discretized into three schemes: two-state, three-state, and mix-state. Four structural leaning algorithms, Bayesian Search (BS), Greedy Thick Thinning (GTT), and Tree Augmented Naïve Bayes (TAN), and Augmented-Naïve Bayes (ANB), were applied to create a Bayesian Belief Network. Results showed that the three-state ANB outperformed the other models, yielding the highest predictive accuracy and area under the Operating Characteristics Curve (ROC). Sensitivity analysis and scenario-based analysis were then applied to better understand the interactions in the created network.

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