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

Summary

Across 52,226 ICU sepsis cases, the SENECA subtypes exhibited geographic variability, but a simple three-variable model (AST, lactate, bicarbonate) robustly identified δ-type patients with high external validity. The work advances precision medicine by enabling bedside subtype adjudication with routine labs.

Key Findings

• Validated SENECA clinical sepsis subtypes across four large ICU cohorts (n=52,226), revealing geographic differences in subtype distribution.
• A three-variable classifier (AST, lactate, bicarbonate) predicted δ-type with high accuracy: derivation AUC 0.93 and external validation AUC 0.86 (accuracy ~83–86%).
• A parsimonious four-class model had only moderate-to-low accuracy (62.2%), indicating δ-type is most readily and reliably identified.

Clinical Implications

Routine AST, lactate, and bicarbonate could triage δ-type patients associated with higher mortality, informing escalation strategies and enrollment in subtype-tailored interventions.

Limitations

• Retrospective observational design with heterogeneity across cohorts.
• Performance for full four-class assignment was modest; subtype distributions differed by geography, impacting generalizability.

Future Directions

Prospective implementation studies to test clinical utility and impact on outcomes; subtype-enriched randomized trials; exploration of therapeutic responsiveness by subtype.