Sunday, October 19, 2008
Columbus A-C (Hyatt Regency Penns Landing)
Jagpreet Chhatwal, MS1, Oguzhan Alagoz, PhD2 and Elizabeth S. Burnside, MD, MPH, MS1, (1)University of Wisconsin, Madison, WI, (2)University of Wisconsin, Madison, USA
Purpose: A 2% threshold has traditionally
being used for biopsy recommendation based on the available literature without
taking into consideration patient characteristics. Our objective was to create
a personalized sequential decision model that determines whether to recommend: biopsy,
short interval imaging follow-up, or routine annual
mammography based on patient's current breast cancer risk.Methods: We constructed a mathematical
model using Markov Decision Processes (MDP), a decision-analytical tool used
for sequential decision making under uncertainty, which maximizes a patient's
total quality-adjusted expected life years (QALYs). Our MDP model provides the
optimal recommendations that are based on patients' probability of breast
cancer, estimated from demographic risk factors (age, family history,
and hormone use) and mammographic findings (using BI-RADS) using a
logistic regression (LR) model. Both the LR and MDP model were constructed
using data from literature and prospectively collected 48,744 consecutive
mammograms from a breast imaging practice between 1999-2004. We checked the
robustness of the MDP model by performing sensitivity analyses on its
parameters.Results: The MDP model provides optimal decision
rules (policy) that maximize patients' total QALYs. Our preliminary results are
shown in the figure below. For example, the optimal decision rule for a 42-year
old woman would be: follow-up and biopsy, if her risk of breast
cancer is above 1% and 2%, respectively; otherwise, wait until next annual
mammogram. . For an 82-year old woman, the thresholds to recommend follow-up
and biopsy increase to 2% and 4%, respectively. Our model
recommended a higher biopsy threshold for older patients suggesting less
aggressive management in older women.Conclusions: Given a patient's mammographic
features and demographic risk factors, our model provides a threshold for biopsy,
short interval follow-up, or annual mammography that maximizes QALYs in order
to personalize these decision thresholds.