Abstract:
|
A dynamic treatment regime (DTR) formalizes personalized medicine as a series of functions over decision points. At each decision point, it takes the available information of a patient as input and outputs a recommended treatment for that patient. A high-quality DTR tailors treatment decisions to individual patient as illness evolves, and thus improves patient outcomes while reducing cost and treatment burden. In the development of DTRs, statisticians must integrate clinical science into the models and clinicians must scrutinize the estimated DTR for scientific validity. To facilitate meaningful information exchange, it is important that the estimated DTR be interpretable in a subject-matter context. We propose a simple, yet flexible class of DTRs whose members are representable as a short list of if-then statements. DTRs in this class are immediately interpretable and are therefore appealing choices for broad application in practice. We develop a robust Q-learning procedure to estimate the optimal DTR within this class and establish its consistency and rate of convergence. We demonstrate the performance of the proposed method using simulation experiments and a clinical dataset.
|