Parsimonious Subphenotyping Algorithms Perform Differently in Patients With Sepsis and Hematologic Malignancy

Area of Science:

• Critical Care Medicine
• Immunology
• Oncology

Background:

• Latent class assignment algorithms can identify sepsis subgroups for targeted treatment.
• Generalizability of these algorithms to critically ill patients with comorbid malignancy is uncertain.
• Malignancy may alter inflammatory biomarkers, potentially affecting subphenotype accuracy.

Purpose of the Study:

• To evaluate if malignancy or neutropenia modifies subphenotype assignment by two algorithms.
• To assess algorithm performance in a prospective cohort of ICU patients with active malignancy.

Main Methods:

• Prospective cohort study at a single U.S. quaternary referral center.
• Included 930 ICU patients with sepsis; 396 had active malignancy.
• Applied two subphenotyping algorithms using IL-6 or IL-8, TNFR1, and bicarbonate.

Main Results:

• Hematologic malignancy patients showed higher "hyperinflammatory" assignment with the IL-8 algorithm (58%) vs. IL-6 (32%).
• Leukemia and neutropenia were overrepresented in the IL-8 hyperinflammatory group.
• Hematologic malignancy attenuated IL-6 hyperinflammatory subphenotype mortality risk (p=0.037) but not IL-8 (p=0.260).

Conclusions:

• Subphenotyping algorithms require validation in specific populations like critically ill cancer patients.
• Differential algorithm behavior in hematologic malignancy necessitates independent derivation and validation.
• Understanding generalizability is crucial as these tools are considered for point-of-care use.