Identification and longitudinal assessment of sepsis phenotypes derived from routine clinical data in critically ill patients: a retrospective repeated measures latent profile analysis

Background: The definition of sepsis as an organ dysfunction with dysregulated host response leads to a considerable heterogeneity in this cohort of patients. Research is ongoing to identify subgroups of septic patients who share a common, potentially treatable pathomechanism. There are now several examples of reproducible subgroups, but they often rely on complex biomarker panels and data on their longitudinal stability are scarce, which limits their translation to the bedside. The objective of this study was to identify sepsis subgroups using routinely available clinical data and to assess the temporal stability of these subgroups.

Methods: We retrospectively collected data on all adult patients treated for sepsis according to sepsis-3 criteria in our intensive care unit at a university hospital in Germany between 2013 and 2021. Subgroup identification was performed using latent profile analysis, based on data collected within 48 h of the onset of their sepsis, and was repeated with data collected 120-144 h after onset. We assessed the stability of subgroup assignment over time and the in-hospital mortality of these subgroups.

Results: The analysis included 637 patients with sepsis, 83% of whom were in septic shock. Latent profile analysis of clinical data from the first period identified four subgroups with a high median probability of class membership in all subgroups and distinct characteristics. Subgroup 1 included 76 patients (12%) and was characterized by a hepatobiliary and cardiovascular dysfunction. Subgroup 2, which included 242 patients (38%), showed the least inflammation and organ dysfunction. Subgroup 3 included 236 patients (37%) and was characterized by hyperinflammation. Subgroup 4 included 83 patients (13%) who were older and had more comorbidities. They tended to have high procalcitonin and INR levels. In-hospital mortality was excessive in Subgroup 1 and lowest in Subgroups 4 and 2. In the longitudinal assessment conducted at 120-144 h post-inclusion, subgroup 1 demonstrated the greatest stability over time.

Conclusion: Analysis of clinical routine data identified four distinct clinical subgroups. In the longitudinal analysis, subgroup 1, which was characterized by hepatobiliary dysfunction and high mortality, demonstrated stability over the course of the illness.