Title: On the relationship among different statistical methods for dynamic treatment regimes
Authors: Woojoo Lee - Inha University (Korea, South)
Seung Jae Lee - Inha University (Korea, South) [presenting]
Abstract: For each patient, medical doctors need to determine the best treatment among available options based on patient's information. Especially, for chronic diseases such as hypertension and diabetes, they need to determine the best sequence of treatments using the updated information including the past outcomes affected by the past treatment. This topic has been studied under the name of optimal dynamic treatment regime. So far, various statistical methods haven been proposed for finding optimal dynamic treatment regime. Q-learning, A-learning and O-learning are those examples. In addition, in epidemiology literature, g-computation and inverse probability treatment weighting methods for time-varying treatment have been considered for a similar purpose. However, their relationship is not clearly recognized in literature. We examine the relationship among different statistical methods for discovering optimal dynamic treatment regime and discuss their advantages and disadvantages in terms of computation.