Clinical subtypes in critically ill patients with sepsis: validation and parsimonious classifier model development.

Summary

Across four large ICU cohorts (n=52,226), sepsis clinical subtype distributions varied by region, and a 3-variable model (AST, lactate, bicarbonate) accurately identified δ-type patients with high AUCs in external validation. Method-based validation showed only moderate overlap with original SENECA subtypes, underscoring cohort heterogeneity.

Key Findings

• Subtype distributions differed substantially between European cohorts (MARS/MARS2/NICE) and the US (MIMIC-IV).
• Three-variable classifier (AST, lactate, bicarbonate) predicted δ-type with high accuracy: AUC 0.93 (MARS) and 0.86 (MIMIC-IV).
• Method-based validation revealed only moderate overlap with original SENECA subtypes, indicating heterogeneity.

Clinical Implications

Use AST, lactate, and bicarbonate to screen for δ-type sepsis at ICU admission to inform prognosis and potential enrollment into subtype-targeted interventions.

Limitations

• Observational design with potential residual confounding and missingness.
• Subtype stability and transportability across care systems and time periods remain uncertain.

Future Directions

Prospective validation and integration into adaptive trials to test subtype-targeted therapies; evaluation of δ-type dynamics and treatment responsiveness.