Data driven decision making frameworks with probabilistic graphical models in healthcare
Advancements in data gathering and uncertainty surrounding health care and human behavior require advanced modelling techniques such as Bayesian belief networks (BBNs) and Markov networks (MNs) to analyze such a complex environment, and provide insights for decision makers. The main objective of this dissertation is to develop probabilistic graphical models (PGMs) to solve prediction and survival problems in health care domain. In this research, the aim is to expand the knowledge for solving problems in this area by proposing various approaches using BBNs, MNs and data mining models. In the first study, elastic net (EN) and BBN based noshow prediction model for a primary care clinic is proposed. This study predicts the "no-show probability of the patient(s)" using demographics, socio-economic status, and current appointment information, and appointment attendance history of the patient and the family. The findings of noshow prediction model is integrated into patient scheduling and overbooking policy. Using the prediction model, we analyze the impact of integrating no-show prediction model into scheduling for effective overbooking using discrete event simulation. Third, we propose a holistic Markovian framework that incorporates different aspects of continuity, such as density, dispersion and sequence and propose a framework for assessment of clinic/ provider CoC. Last study proposes a multi-level survival analysis to define risk levels as low, medium, and high for kidney transplants patients using deceased donor database. Our findings have showed that big data and uncertainty challenge can be addressed by designing comprehensive frameworks combining expert knowledge with data analytic models.
Thesis (Ph.D.)-- Wichita State University, College of Engineering, Dept. of Industrial and Manufacturing Engineering