A discrete Bayesian network for analysing hospital discharge data

Exploratory research requires models that can explain the underlying phenomena of interest in new research areas. We present the design and application of discrete Bayesian networks (DBN) for knowledge discovery in a hospital discharge dataset. For the learning phase of the network, the automated learning methods are preceded by customising the initial network. Structural learning is done using three state-of-the-art algorithms and is inter-validated. A new method is suggested for drawing selective inferences from the posterior conditional probability tables (CPT). As an illustration, functional inferences are drawn on length of stay and treatment charges for three disease groups using the developed method. Our analysis shows that for longer hospital stays, hospital visits involving mental disorders cost less than visits with other types of health conditions. This study contributes to data science research by demonstrating the application of Bayesian networks, evaluating different structure learning methods for given contexts, and developing a measure for selective inference using the CPT of the network.