New AI model predicts sepsis mortality in the ICU with high accuracy

Septicemia is one of the deadliest conditions of intensive care units (USI), triggered by the body -controlled response to infection. Despite medical progress, its death mortality rate still oscillates between 20% and 50%. The challenge lies in early identification – SEPSIS is very dynamic, and current rating systems like Apache -II and SOFA are not specifically designed to follow its rapid progression. Although automatic learning is promising, most models are struggling to take into account real -time fluctuations in patient data. Given these challenges, an advanced predictive system capable of constantly learning incoming clinical data is required to improve early detection and patient results.

On February 8, 2025, researchers from the University of Sichuan, the University of a Coruña, and their collaborators published their results (DOI: 10.1093 / PCMEDI / PBAF003) Precision clinical medicinePresentation of a model based on a transformer into two steps designed to predict mortality between septicemia in USI. Trained on data from the EICU collaborative research database, which includes more than 200,000 patients, the model dynamically deals with timetable and daily health indicators. In admission of the fifth day of the USI, he made an impressive ASC of 0.92, considerably surpassing traditional rating systems like Apache-II.

This model powered by AI marks a significant leap in the prediction of sepsis. It operates in two stages: the first step analyzes hourly data, identifying intra-day criticism fluctuations in vital signs and laboratory results, while the second step includes daily data to enter longer-term trends. This layer approach allows the model to adapt to the rapidly evolving nature of sepsis.

The main predictors of mortality – such as lactate levels, respiratory frequencies and coagulation markers – were identified with high precision. A major breakthrough lies in the ability of the model to generate real -time risk alerts, equipping intensive care teams with usable information when it is most necessary. The inclusion of formalizations of the form (additive explanations of Shapley) guarantees interpretability, allowing clinicians to understand what factors stimulate predictions. In addition, the model has demonstrated exceptional robustness when validated on external data sets, including patient cohorts from China and the Mimic-IV database.

“This model based on the transformer represents a paradigm shift in the way we tackle the prognosis of septicemia in the USI,” said Dr. Bairong Shen, one of the corresponding authors of the study. “By integrating real -time data from the chronological series, we can now provide clinicians with more precise and more timely risk assessments, ultimately improving patient results and reducing mortality rates.”

The impact of this research could be transformer for intensive care management. By integrating the AI ​​model into hospital information systems, clinicians could receive daily risk alerts, allowing previous and more targeted interventions. Its adaptability between different populations of patients and resilience to missing data make it a precious asset in various health care establishments worldwide. Future developments could see the model integrated into real -time surveillance systems, continuous risk scores and minimization of diagnostic delays.

Beyond immediate clinical applications, the interpretability of the model by analysis of the form offers more in-depth information on the progression of sepsis, potentially guiding the development of precision therapies. This innovation not only improves patient care, but also establishes a new reference for predictive modeling focused on AI in intensive care medicine.

With its ability to use large amounts of data in real time and translate it into vital ideas, this AI -fed tool could redefine the standard of care for septicemia – turning early warnings into timely interventions and improving survival rates on a global scale.