TITLE: Optimal Design of Prostate Cancer Screening Policies

SPEAKER: Brian Denton

ABSTRACT:

Prostate cancer is the most common solid tumor that affects American men. Screening typically involves the use of prostate specific antigen (PSA) tests. However, the imperfect nature of PSA tests, and the potential for subsequent harm from unnecessary biopsies and treatment, has raised debate about whether and when to screen. In this talk I will provide some background on prostate cancer, current screening guidelines, and a summary of the recent controversy over PSA testing. Next, I will discuss a partially observable Markov decision process (POMDP) model to investigate the optimal design of screening policies. Screening policies are defined by the patient’s probability of having prostate cancer which is estimated from their history of PSA tests results using Bayesian updating. The core states are the patients’ prostate cancer related health states. Transition probabilities among health states are estimated using a large longitudinal dataset from Olmsted County, the Mayo Clinic Radical Prostatectomy Registry (MCRPR) and the medical literature.  Reward functions that are considered include quality adjusted survival (patient perspective) and costs (third party payer perspective).

Some theoretical properties that define the optimal policy will be discussed, and a new approximation method suited to solving finite horizon non-stationary POMDPs will be presented.  The results of computational experiments will be used to illustrate the use of the model for making screening decisions, such as if and when to recommend a patient for a PSA test, and when to refer patients for biopsy and subsequent treatment.  Sensitivity analysis will be presented to demonstrate the relative importance of factors that define patient specific preferences and risk factors. Finally, future research directions in the area will be discussed.