Finding the Maximum Safe Treatment Interval for Patients With Chronic Conditions
针对慢性病（如年龄相关性黄斑变性）患者，提出一种基于马尔可夫决策过程的序贯决策方法，帮助医生快速安全地确定个体化的最大安全治疗间隔，相比现有临床方案可减少38%的症状暴露并加快7%的优化速度。
For patients diagnosed with a chronic condition, appropriately timing treatment is critically important. Many conditions, such as age-related macular degeneration (AMD), have a maximum safe treatment interval (MSTI) in which treatment is required to prevent disease progression. We introduce a Markov decision process (MDP) with an ordinal action space as a way to quickly and safely identify the MSTI for an individual patient. The MDP’s ordinal action space allows users to update the expected outcomes of multiple actions with only a single decision. We solve the MDP and show how the expected outcomes of each action can be presented to clinicians as a menu of treatment options. We describe conditions under which the MDP is guaranteed to find the MSTI under uncertainty. We illustrate the model’s effectiveness through an application to AMD, and show that following the MDP can reduce patient exposure to symptoms by 38% and find the optimal interval 7% faster when compared to current clinical protocols.