Anesthesiology and critical care medicine generate a large amount of diverse data, including electronic medical records, textual documentation, and continuous measurement of physiological parameters.[1] Machine learning models can use these data to make individual risk predictions and stratify patients, but their practical use is complicated by data heterogeneity, missing values, and inconsistent definitions of clinical outcomes.[1] The paper describes three main strategies for fusing different types of data: early fusion, which combines data into a single table and offers simplicity, intermediate fusion with specialized encoders, which achieves the most complex results, and late decision-level fusion, which provides flexibility when data arrives asynchronously.[1] The development of multicenter databases and federated infrastructures can enable a better understanding of patient treatment courses and support personalized therapy in perioperative and intensive care.[1]