Figure 1.

Workflow data from an optimal medication regimen. We aimed to derive an optimal medication regimen that suggests the best combination of drugs given the current disease state to minimize motor symptoms of PD. (a) PPMI database consists of trajectories of visits of PD patients. Each visit has clinical assessments to measure the current disease status and the record of the medications used to address the motor symptoms. UPDRS III score measures the motor response to the medication. (b) Each visit was characterized as one of the discrete disease states that are defined using a decision tree regressor. (c) We formulated this medication therapy as an MDP. Based on the current disease state, a clinician makes a decision on a combination of medications, which in turn change motor symptoms and alters the disease state iteratively. (d) We computed the optional medication option on each state using policy iteration (in reinforcement learning).