Data-driven decisions Optimizing critical care through date integration and predictive decision support

This thesis investigates the development, validation, and clinical relevance of data-driven clinical decision support systems in critical and perioperative care. While artificial intelligence (AI) and machine learning offer clear potential to improve diagnostic accuracy, reduce clinician workload, and enable proactive patient management, real-world implementation remains limited by challenges related to data quality, interpretability, regulation, and clinical trust.
Using a structured pipeline, from data acquisition and annotation to feature engineering and model development, this work demonstrates how clinically meaningful algorithms can be designed and evaluated. A cloud-based infrastructure for high-frequency physiological data is presented as a scalable foundation for real-time analytics. The thesis further addresses the often overlooked problem of annotation quality by validating expert-labeled rotational thromboelastometry (ROTEM) data and translating these annotations into an explainable diagnostic decision support tool for coagulation management in cardiac surgery. This tool is explicitly designed to support, rather than replace, clinician judgment.
In parallel, the thesis explores predictive modeling using arterial blood pressure waveform morphology to anticipate post-induction hypotension, illustrating how subtle physiological patterns beyond human perception can enable proactive intervention. The work critically discusses trade-offs between predictive performance and explainability, as well as ethical concerns such as accountability, bias, and the distinction between associative and causal inference.
Overall, this dissertation provides a practical and conceptual framework for developing trustworthy, clinically integrated data-driven decision support systems, emphasizing that technical performance alone is insufficient without robust data infrastructure, clinical insight, and careful implementation strategies.