Q-Learning Based Therapy Planning Decision Support System

This paper presents an approach for a Q-learning based decision support system for therapy planning. It focuses the consideration on a multilevel approach for constructing a data driven evidence based model for classification of different drug dosages and their effectiveness for clinical trials. We consider time-ordered sequences of patient data (critical patient events) called patient trials consisting of state, time and the medication ordered by clinicians. A patient state generalization function to provide generalized patient states as categories of patient observation vectors is presented which is based on a neural network. For classification of medications we introduce a medication generalization function based on similarity classes of medications and a similarity function between two drug dosages. Both generalization functions are used for generalizing patient trials and the life long quality function to synthesize the Q-Learning agent for the Decision Support System.