Optimization-enhanced Gaussian mixture modeling for data-driven subphenotyping of septic shock

Septic shock is associated with high mortality and pronounced clinical heterogeneity, posing significant challenges for patient stratification and individualized clinical management. To address the limitations of empirically driven subtyping approaches, this study presents an optimization-enhanced Gaussian mixture modeling framework for data-driven clinical subphenotyping of septic shock. By integrating the Black-Winged Kite Algorithm (BKA) with a Gaussian Mixture Model (GMM), the proposed framework aims to improve parameter optimization stability, clustering robustness, and subtype separability. A retrospective cohort of 780 patients was analyzed using multidimensional clinical data collected at admission, including vital signs and laboratory parameters. After data preprocessing and standardization, the BKA was employed to optimize GMM parameters, enhancing clustering robustness and reducing sensitivity to initialization. The optimized model identified three clinically distinct subphenotypes. Subtype I (n = 213) exhibited more severe physiological and laboratory abnormalities; Subtype II (n = 330) showed intermediate clinical patterns; and Subtype III (n = 237) presented relatively milder clinical profiles. Compared with the conventional clustering methods, the proposed framework demonstrated improved clustering performance, achieving a silhouette coefficient of 0.8528 ± 0.0112 across repeated runs, indicating stable subtype separation. Overall, this study provides a data-driven methodological framework for septic shock subphenotyping, which may contribute to improved understanding of patient heterogeneity and support future research in clinical stratification.