Monte Carlo Tree Search for optimal cancer intervention strategies among BRCA mutation carriers
Breast and ovarian cancers are the second and fifth leading causes of cancer-related death among women in the United States. Compared to non-carriers, BRCA mutation carriers are subject to a higher risk of developing breast and ovarian cancers. Prophylactic surgeries including prophylactic mastectomy and prophylactic bilateral salpingo-oophorectomy can significantly reduce the incidence risks of breast and ovarian cancers for BRCA mutation carriers. However, both prophylactic surgeries are permanent one-time interventions. Determining the optimal age for BRCA carriers to undergo such surgeries is critical. The optimal solution depends on sequential decision-making over long periods during a patient's life involving the cancer incidence risk and quality of life (QOL). Thus, there is an urgent need to develop an optimal cancer intervention strategy with the goal to reduce cancer risk and maintain a high-level QOL. This paper presents a novel sequential decision-making framework for BRCA mutation carriers by jointly considering the two objectives. We first propose to model the dynamic progression of cancer risk using a continuous-time Markov Decision Process. Second, we propose to solve for the optimal intervention strategies through Monte Carlo Tree Search (MCTS) by optimally balancing the exploitation of current knowledge and exploration of uncertainty factors about the cancer state. Experimental results show that the proposed MCTS planning method can effectively provide optimal sequential intervention strategies for BRCA mutation carriers, reducing the risk of cancer incidence while maintaining a high-level QOL.
