INCORPORATING INDIVIDUAL PATIENT PREFERENCES IN TREATMENT DECISION MODELING FOR BREAST CANCER

Purpose: Technological advances, dramatic increase in costs, and more patient involvement have been posing challenges to healthcare providers in medical decision making.  This paper proposes a dynamic decision modeling framework to determine optimal breast cancer treatment decisions that incorporate individual patient preferences.

Method: A Markov decision process (MDP) model is formulated to identify optimal treatment decisions when breast cancer is diagnosed at an annual mammogram exam.  The underlying Markov chain represents a patient’s health status at each screening.  We assume follow-up tests (e.g., biopsy) are used to confirm the presence of breast cancer after a positive mammogram.  Type II error of mammography and the possibility of spontaneous regression of early stage cancers are considered in the model.  While the patient can choose to wait on treatment, there are various types of treatments including different combinations of surgeries and adjuvant treatments considered in this study: breast conserving surgery (BCS), mastectomy, radiation therapy, chemotherapy, and hormone therapy  The goal of the decision model is to maximize a patient’s life score, which is a metric that seeks to take into account the patient’s expected remaining life years as well as her personal preferences.  Life scores depend on various factors: age, cancer stage, hormone receptor status, type of treatment, and the patient’s personal opinion (as measured by emotional weights) on treatment side effects including cancer recurrence, change in appearance, figure and general pain.  

Result: The results show that the optimal treatment policies vary with the patient’s personal preferences.  No treatment decision is found to be optimum for elderly patients, when side effects outweigh the gains in life years.  The threshold of no treatment decisions depends on individual patient preferences.  The average life score of the invasive cancer patients is around one-fourth of the average life score of in situ patients.  When emotional weights of side effects are high, the treatment combination of a BCS plus radiotherapy and chemotherapy is often recommended as optimal.

Conclusion: This study provides a dynamic decision framework to evaluate and identify optimal treatment decisions based on individual patient preferences. It shows how patient preferences can be incorporated as an outcome measure in the MDP model.