Date of Graduation


Document Type


Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level



Industrial Engineering


Shengfan Zhang

Committee Member

Richard Cassady

Second Committee Member

Edward A. Pohl

Third Committee Member

Lisa M. Maillart


Pure sciences, Applied sciences, Maintenance decisions, Operations research, Preventative healthcare, Stochastic decision modeling


This dissertation focuses on the preventive maintenance decision modeling in healthcare and service systems. In the first part of this dissertation, some issues in preventive health decisions for breast cancer are addressed, and in the second part, the required characteristics for preventive maintenance of an unreliable queuing system are derived.

Adherence to cancer screening is the first issue that is addressed in this dissertation. Women’s adherence or compliance with mammography screening remained low in the recent years. In this dissertation, we first develop a design-based logistic regression model to quantify the probability of adherence to screening schedules based on women’s characteristics. In Chapter 3, we develop a randomized finite-horizon partially observable Markov chain to evaluate and compare different mammography screening strategies for women with different adherence behaviors in terms of quality adjusted life years (QALYs) and lifetime breast cancer mortality risk. The results imply that for the general population, the American Cancer Society (ACS) policy is an efficient frontier policy. In Chapter 4, the problem of overdiagnosis in cancer screening is addressed. Overdiagnosis is a side effect of screening and is defined as the diagnosis of a disease that will never cause symptoms or death during a patient's lifetime. We develop a mathematical framework to quantify the lifetime overdiagnosis and mortality risk for different screening policies, and derive the (near) optimal policies with minimum overdiagnosis risk.

In the second part, we consider an unreliable queuing system with servers stored in a shared stack. In such a system, servers have heterogeneous transient usage since servers on the top of the stack are more likely to be used. We develop a continuous-time Markov chain model to derive the utilization and usage time of servers in the system. These quantities are critical for the decision maker for deriving a maintenance policy.