Anmerkungen:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Beschreibung:
Personalized medicine has the potential to revolutionize how healthcare is provided. The aim of personalized medicine is to tailor treatment decisions to the individual patient. To this end, we must estimate what effect a particular treatment has on the individual patient. However, estimating this individualized treatment effect is a challenging undertaking, since conventional medicine uses clinical trials to estimate treatment effects. These clinical trials are usually too small to capture all possible variations of patients. As such, clinical trials can only provide treatment effects for the average patient, but not for the individual patient. In order to estimate individualized treatment effects, so-called real-world data (RWD) bear great potential. Since RWD are generated outside of clinical trials, they are available in large quantities and, hence, can capture more variations among patients. However, it is precisely because RWD are generated outside of clinical trials that standard analytics tools may yield incorrect estimates of treatment effects. This is because treatment effects are causal effects and, hence, analytics tools for RWD must be able distinguish causality from correlation. Therefore, leveraging RWD for personalized medicine requires the development of advanced analytics tools that can reliably estimate individualized treatment effects. In this thesis, we approach personalized medicine by leveraging RWD. For this, we develop novel methods in the field of causal machine learning. These methods are able to estimate causal effects and, hence, can be used to reliably estimate treatment effects from RWD. In the first three chapters of this thesis, we develop causal machine learning methods that leverage RWD in order to: (I) estimate average treatment effects with high accuracy, (ii) learn personalized treatment policies that can be robustly generalized to the entire population, and (iii) estimate individualized treatment effects. While we show the potential of causal machine learning for leveraging ...