New paper accepted at the conference “Machine Learning for Healthcare 2022”
Our paper “Learning Optimal Dynamic Treatment Regimes using Causal Tree Methods in Medicine” has been accepted for presentation and publication in the proceedings of the Machine Learning for Healthcare 2022 conference (MLHC 2022). The paper is authored by PhD candidate Joel Persson, his former MSc student Theresa Blümlein (now Data Scientist at QuantCo), and his co-supervisor Prof. Dr. Stefan Feuerriegel (AI in Management at LMU Munich).
In the paper, we propose a way of adapting the widely used causal tree and causal forest methods from estimating heterogeneous treatment effects to learning optimal sequential treatment decisions. The methods learn non-linear relationships, adjust for time-varying confounding, and are data-driven, doubly robust, and explainable to medical practitioners. We evaluate our methods using simulation studies and then apply them to real-world data from intensive care units. We find that the methods outperform state-of-the-art baselines in terms of cumulative regret and percentage of optimal treatment decisions by a considerable margin. Overall, our research enable improved treatment recommendations from electronic health record and is thus of relevance for personalized medicine.
external pageLink to preprint.
external pageLink to MLHC 2022