Sepsis insult generates different vascular response phenotypes: an unsupervised time-series clustering preclinical study.

In this study, the temporal evolution of the cardiovascular response to sepsis was investigated by combining conventional hemodynamic parameters with novel indices derived from arterial blood pressure (ABP) waveforms that reflect vascular properties. The well-established association between sepsis and vascular dysfunction suggests that waveform-based indices may be instrumental in sepsis early detection and quantitative assessment of sepsis severity. ABP was continuously recorded at aortic and femoral sites in 40 pigs from a baseline condition to the full development of septic shock which was induced by intraperitoneal instillation of autologous feces. Standard beat-to-beat indices, including mean ABP, heart rate, and pulse pressure (PP), were computed alongside advanced cardiovascular markers such as baroreflex sensitivity, characteristic time constant ([Formula: see text]), PP amplification, and harmonic distortion (HD). Time series clustering was performed using a K-mean clustering approach with soft-dynamic time warping distance metric using 4 features - PP amplification, femoral [Formula: see text], aortic systolic ABP, and femoral HD - selected by the First Integer Neighbor Clustering Hierarchy filtering method. According to the main physiological parameters typically evaluated by clinicians, all animals experienced severe cardiovascular decompensation under septic conditions. However, the time series cluster analysis identified two clusters with distinct temporal patterns of the cardiovascular variables. Shapelet analysis confirmed these findings, revealing consistent variable-specific trends. The proposed indices may assist clinicians in the early identification of sepsis and monitoring of sepsis severity, supporting more timely and personalized therapeutic strategies. Further studies are needed to explore their relationship with treatment response and clinical outcomes.