OptAB2: Causal AI for Clinical Decision Support in Sepsis

The OptAB2 PhD project develops a clinical decision support system based on causal artificial intelligence to improve antibiotic selection in sepsis treatment. Early and appropriate antibiotic therapy is critical for survival, yet difficult to determine. Existing AI approaches such as reinforcement learning often suffer from spurious correlations and misleading treatment strategies.

OptAB2 uses causal neural ordinary differential equations (causal NODEs) to model disease progression in continuous time, even from irregular and incomplete patient data. This allows the system to simulate and compare treatment trajectories under different antibiotics and to generate counterfactual predictions — enabling data-driven, individualized treatment recommendations.

The model is trained on the MIMIC-IV intensive care database and externally validated on AmsterdamUMCdb. In the next phase, OptAB2 will be extended to additional therapies, validated on German patient data, and prepared for regulatory approval as a medical device (MDR). The project is conducted in collaboration with the University Hospitals of Mainz and Bonn and the industry partner Qurasoft GmbH, which also supports the regulatory pathway.