Wednesday, Feb. 1
How budgetary restrictions affect post-mammography management decisions: a constrained MDP approach
featuring Mehmet U.S. Ayvaci, Ph.D. candidate, industrial and systems engineering, University of Wisconsin
Early diagnosis through screening mammography is the most effective means of reducing the death rate from breast cancer. While mammography is inexpensive, failures in the post-mammography management decisions in the form of false positives and overdiagnosis are not rare and therefore are important determinants of the total cost. One study reports that for every $100 dollars spent on screening, an additional cost of $33 dollars is incurred due to false positives. In this research, we demonstrate how budgetary constraints may alter a radiologist’s diagnostic decisions in the pursuit of optimal breast cancer diagnosis as measured by quality adjusted life years (QALYs) , i.e., given mammography features, demographic factors and a limited budget, what is the optimal course of action: routine screening, short-term follow-up or biopsy?
Ayvaci and others developed a finite-horizon discrete-time constrained Markov decision process (MDP) to model breast cancer diagnostic decisions after mammography to maximize the total expected quality adjusted life years (QALYs) of a patient under resource constraints. The researchers prove that the optimal value function is concave in the allocated budget. Model parameters are derived from using 62,219 consecutive mammography records reported in Breast Imaging Reporting and Data System (BI-RADS) format and the medical literature and solve the MDP model as a mixed-integer program.
Bio: Mehmet Ayvaci is currently a Ph.D. candidate in the Department of Industrial and Systems Engineering at the University of Wisconsin (UW)-Madison. He received his B.S. from the Department of Industrial Engineering at Texas A&M University and M.S. from the Department of Management Science and Engineering at Stanford University. He is a member of INFORMS, SMDM, RSNA and Tau Beta Pi Engineering Honor Society. His research interests include stochastic optimization, medical decision-making, Markov decision processes, simulation, risk-prediction modeling and health technology assessment.