Data-adaptive targeted machine learning methods for analyzing dynamic treatment effects in observational healthcare data

Helene Charlotte Wiese Rytgaard says: “One of the key challenges in medical research consists in analyzing the effects of treatments administered over time using real-world data. Here traditional statistical methods and standalone machine learning approaches may either be inapplicable or fail to yield clinically meaningful results. The obstacles that a sound statistical approach needs to deal with are continuous-time dynamics, including irregular monitoring, and complex treatment decisions, changes of patient characteristics, and health outcomes. This research project aims to develop, extend and implement advanced statistical methods integrating machine learning techniques for analyzing treatment effects in observational healthcare data, to provide more reliable tools for informed medical decision-making by patients, clinicians, and drug developers. The project will expand and enhance modern statistical causal inference tools combined with machine learning techniques and continuous-time models, to data-adaptively model the dependence between life-course events and treatment decisions, while accurately and efficiently addressing essential medical questions regarding dynamic administering of treatment. The goal is to provide a toolbox containing methods and corresponding software implementations that can be used to gain valuable insights into how the administration of treatments over time impacts patient survival and disease progression, beyond what is possible with existing methods.”