Russian Researchers Build National ICU Patient Dataset for AI Development

Researchers at Sechenov University are creating a national harmonized dataset of intensive care patients. It will include detailed information on 5,300 clinical cases and is expected to serve as a foundation for developing AI solutions in critical care.

A key feature of the dataset is the use of “clinical phenotypes,” which algorithmically identify pathophysiological conditions based on objective criteria such as vital signs, laboratory data, and patient condition dynamics. This approach allows AI systems to be trained on real-world clinical scenarios rather than theoretical documentation.

The method has already demonstrated effectiveness. Based on it, researchers have developed a machine learning model capable of predicting the onset of sepsis several hours before its first clinical symptoms appear.