ADAPTIVE DECISION-MAKING OF BREAST CANCER MAMMOGRAPHY SCREENING: A HEURISTIC REGRESSION-BASED MODEL

Purpose: For early diagnosis of breast cancer, all U.S. women are recommended by the American Cancer Society (ACS) to undergo routine mammography screenings from age 40. However, due to the potential harms associated with frequent mammograms, such as over-diagnosis and unnecessary follow-ups, how to design an appropriate mammography screening schedule is a very controversial problem. This study presents a regression-based optimization model for personalized screening decisions that aims to minimize the loss of life expectancy caused by undergoing and skipping mammograms.

Method: The decision-making process consists of two sub-models: breast cancer risk estimation and mammography screening decision based on the estimated risk. The risk estimation model is an age-specific logistic regression model to predict a woman’s breast cancer probability at her current age based on a number of risk factors using the Breast Cancer Surveillance Consortium data. In order to find the optimal regression models for different ages, we use the H measure to perform model selection. Due to the high dimensionality of independent variables and the extremely large number of observations, a heuristic algorithm is developed to select the best combination of independent variables. The next sub-model determines whether a woman should undergo or skip the mammogram at her current age based on the estimated breast cancer probability. An age-specific optimal cut-off point of cancer probabilities, which is expected to minimize the woman’s loss of life expectancy, serves as a threshold of accepting a mammogram. The misclassification cost term criterion is used to calculate the optimal cut-off points.

Result: The optimal combinations of independent variables are not the same for different age groups. The interaction effects between different risk factors play a vital part in every age’s model. The optimal decisions always outperform the ACS and the USPSTF breast cancer screening guidelines in terms of the average loss of life expectancy.

Conclusion: While most of earlier studies attempted to offer optimal lifetime mammography screening schedules, this study provides an adaptive yearly screening decision aid—whether a woman should receive a mammogram is determined based on her risk level at her current age. Thus, our “on-line” screening policy is adjusted according to a woman’s latest health status, which is supposed to produce better screening decisions as compared with a rigid lifetime screening schedule.