Work

Clinical phenotyping of septic shock with latent profile analysis: A retrospective multicenter study

Year: 2024

Type: article

Abstract: Septic shock (SS) is a highly fatal and heterogeneous syndrome. Identifying distinct clinical phenotypes provides valuable insights into the underlying pathophysiological mechanisms and may help to propose precise clinical management strategies. Latent profile analysis (LPA), a model-based unsupervised method, was used for phenotyping in the MIMIC cohort, and the model was externally independently validated in the eICU and AUMC cohorts. Three phenotypes, labeled phenotype I, II, and III, were derived. These phenotypes varied in demographics, clinical features, comorbidities, patterns of organ dysfunction, organ support, and prognosis. Phenotype I, characterized by the most severe organ dysfunction (especially liver), the youngest age, and the highest BMI, had the highest mortality ( p < 0.001). Phenotype II, with moderate mortality, was characterized by severe renal injury. In contrast, phenotype III, associated with the oldest age and the fewest comorbidities, exhibited significantly lower mortality. Phenotype I patients had the steepest survival curves and demonstrated an ultra-high risk of death, particularly within the first few days after SS onset. The individualized identification of phenotypes is well suited to clinical practice. The three SS phenotypes differed significantly in pathophysiological and clinical outcomes, which are crucial for informing management decisions and prognosis. • Three phenotypes were identified, with significantly different pathophysiological characteristics and clinical outcomes. • High reproducibility of characterization of the phenotypes in the three cohorts. • Phenotype I had a severe liver injury and the worst outcome; II had a severe renal injury; and III had the best outcome. • The model-based unsupervised approach, LPA, enables individualized phenotype identification, closer to clinical practice.

Source: Journal of Critical Care

DOI: 10.1016/j.jcrc.2024.154932