AI Tool Maps Early Risk Patterns in Bloodstream Infections
Posted on 24 Mar 2026
Bloodstream infections can deteriorate rapidly and carry high mortality, especially in immunocompromised patients. Clinicians need reliable early risk stratification to prioritize monitoring and organ support. A new study shows that artificial intelligence (AI) can group patients into clinically meaningful risk patterns within the first 48 hours of diagnosis. The approach aims to help teams identify those most likely to require escalated care.
Researchers at Houston Methodist Research Institute (Houston, TX, USA) used an unsupervised machine learning approach to stratify bloodstream infection (BSI) patients using data from the first 48 hours after diagnosis. From more than 15,000 records, the model identified three patient clusters to inform early risk assessment. The analysis was designed to assist clinicians in recognizing previously unseen infection patterns.
Investigators extracted 27 characteristics from electronic health records (EHRs) after BSI was confirmed in microbiology databases. Uniform Manifold Approximation and Projection (UMAP) reduced dimensionality, and k-means++ clustering delineated distinct phenotypes. Each cluster was further categorized as solid organ transplant (SOT) or non-SOT, and clinical characteristics and outcomes were evaluated.
The highest‑risk cluster included older, predominantly male patients who more often required ventilator and vasopressor support and encompassed transplant recipients. These individuals are especially susceptible to infection, with one in 10 developing an infection in the first year after transplant and mortality reported as high as 60%. The other two clusters appeared clinically similar at baseline but diverged in outcomes, with one milder and the other associated with greater severity and mortality.
The research was published in the American Journal of Transplantation (2026). Next steps include validating the approach in external healthcare systems to confirm reproducibility and refining the methodology to inform clinical decision-making and patient outcomes.
"Our model turns routine early data into a risk map clinicians can use immediately. This gives us a new way to understand and predict how sick a patient might become. If we can identify high-risk patients sooner, we can act sooner," said Stefano Casarin, Ph.D., assistant professor in the Center for Precision Surgery at Houston Methodist Research Institute.
Related Links
Houston Methodist Research Institute