| Breast cancer is the most common non-preventable cancer among women. Although it has been demonstrated in randomized trials that mammography screening reduces the breast cancer mortality rate, the optimal screening policy is not known. When screening should start and stop, and the optimal interval between screening sessions are controversial issues. In this thesis, we present dynamic programming algorithms that find optimal variable-interval screening policies that can either minimize lifetime cancer mortality risk or maximize life expectancy. We evaluate these policies using a simulation based on the MISCAN-Fadia breast cancer model. By applying the optimal policies, we can typically either increase life expectancy by 4.0 days or reduce the lifetime cancer mortality risk by 0.16% , which is equivalent to saving 3200 women annually from breast cancer death, compared to the standard constant-interval screening guidelines, without increasing the number of screenings. We also find that increasing life expectancy and decreasing cancer mortality risk can be contradictory goals. We demonstrate that variable screening intervals can increase the effectiveness of screening. We show that the benefits of optimizing screenings policies vary according to the cancer incidence risk of the women; but also that optimizing policies over each risk subgroups does not give promising results. |
