人工知能および機械学習支援による敗血症の個別化免疫療法のためのサブフェノタイピング

Background/Objectives: Sepsis heterogeneity limits advances in immunotherapy. Increasing use of artificial intelligence (AI) and machine learning (ML) attempts to turn multi-dimensional data into meaningful clusters, indicating biological mechanisms. We provide an overview of the existing evidence on AI-derived sepsis subtyping, exploring treatment response to available immune modulating therapies. Methods: On 1 October 2025, we conducted a structured search on all relative publications on MEDLINE and undertook a narrative review. Results: Multiple subphenotyping algorithms were identified, using clinical, biological, and omics data, across different cohorts, mainly through secondary analyses of randomized trials. The main classification was between hyper- and hypoinflammatory subphenotypes. Statins, corticosteroids, activated protein C, or thrombomodulin displayed differential effects on the outcome of these subphenotypes. Conclusions: Further research is required to prospectively validate findings and to offer pragmatic solutions to patients who need them the most. Issues of validity, equity, ethics, and feasibility are discussed.