How to subphenotype ICU patients in a simple manner beyond ferritin and the HScore (Histocyte Society score)?

Simple ICU Patient Subphenotyping Beyond Ferritin and HScore

Use machine learning-derived clinical parameter models that incorporate routinely available vital signs, laboratory values, and organ dysfunction scores to identify distinct sepsis and ARDS subphenotypes without requiring specialized biomarkers. 1, 2

Practical Clinical Parameter-Based Approaches

The most accessible subphenotyping strategy leverages more than 25 routinely measured clinical parameters already present in standard electronic health records, including vital signs, basic laboratory values, organ dysfunction scores, and demographic data. 1, 2 This approach has successfully identified four distinct sepsis subphenotypes with differential mortality and treatment responses, requiring no specialized testing beyond what is universally available. 1, 2

Key Clinical Parameters to Collect

• Vital signs: Heart rate, respiratory rate, blood pressure, temperature, oxygen saturation 1, 3
• Basic laboratory values: White blood cell count, platelet count, blood urea nitrogen, creatinine, bilirubin, lactate 1, 3
• Organ dysfunction metrics: PaO2/FiO2 ratio, Glasgow Coma Scale score, vasopressor requirements 1, 4
• Demographic and clinical context: Age, comorbidities (especially COPD), infection site, multilobar infiltrates 1

Validated Scoring Systems for Subphenotyping

Sequential Organ Failure Assessment (SOFA) Score

SOFA score serves as the foundation for trajectory-based subphenotyping, quantifying organ dysfunction severity across six systems (respiratory, cardiovascular, hepatic, coagulation, renal, neurological). 2, 4, 5 The number of acquired organ system failures is the single most important prognostic indicator for ICU patients. 2, 4

Four distinct SOFA trajectory-based subphenotypes have been validated across multiple cohorts: 5

• Rapidly Worsening (13%): Higher comorbidity burden, acidosis, visceral organ dysfunction, highest mortality (28.3%) despite lower admission SOFA
• Delayed Worsening (21%): Progressive deterioration over 72 hours
• Rapidly Improving (41%): Vasopressor use without acidosis, lowest mortality (5.5%)
• Delayed Improving (25%): Gradual improvement trajectory

These subphenotypes can be predicted with 78% accuracy at 6 hours and 87% accuracy at 24 hours after ICU admission using readily available clinical data. 5

PIRO System for Sepsis Subphenotyping

The PIRO classification system provides a structured framework analogous to cancer staging, evaluating four domains using only clinical data: 1

• Predisposition: Age, nutritional status, sex, comorbidities (particularly COPD)
• Infection: Site, pathogen type, spread pattern
• Response: Host inflammatory response, coagulation status
• Organ dysfunction: Number and severity of failing organs

This system demonstrates excellent correlation between increasing PIRO score and mortality rate (p < 0.001) and requires no specialized biomarkers. 1

ARDS-Specific Subphenotyping

For ARDS patients specifically, validated 3-4 variable classifiers using clinical data and plasma inflammatory markers (excluding ferritin) reliably identify subphenotypes with differential mortality and treatment responses. 2 Key parameters include:

• Respiratory system compliance: Identifies subgroups with differential responses to low tidal volume ventilation 2
• PaO2/FiO2 ratio: Severity stratification (≤250 indicates severe disease) 1
• Multilobar infiltrates: Presence indicates higher risk subphenotype 1

Alternative Inflammatory Markers Beyond Ferritin

While avoiding ferritin and HScore as requested, several readily available inflammatory markers provide subphenotyping value:

• Neutrophil-to-lymphocyte ratio (NLR): AUC 0.857 for mortality prediction, easily calculated from routine complete blood count 6
• Interleukin-6 (IL-6): Correlates with hemoglobin levels and mortality, though less universally available 7
• Hepcidin: AUC 0.808 with 87.3% specificity for 28-day mortality, superior to other inflammatory anemia markers 7
• Soluble transferrin receptor/log ferritin ratio: Associated with mortality when avoiding absolute ferritin values 7

The combination of NLR with any single inflammatory marker achieves comparable predictive performance to SOFA score (AUC 0.901 vs. 0.955, p=0.085) while being more accessible. 6

Practical Implementation Algorithm

1. At ICU admission (0-6 hours): Calculate baseline SOFA score, collect vital signs, basic labs, and demographic data 5
2. At 24 hours: Recalculate SOFA to establish trajectory pattern (worsening vs. improving) 5
3. At 72 hours: Confirm trajectory-based subphenotype classification 5
4. Throughout: Apply PIRO framework to contextualize predisposition, infection characteristics, response patterns, and organ dysfunction 1

Critical Caveats

Clinical severity criteria alone achieve prognostic enrichment but generally fail to enrich for treatment response. 1 Subphenotyping based solely on single organ dysfunction severity (e.g., sepsis-associated coagulopathy) has not yielded successful targeted treatments in randomized trials. 1

Integration of clinical, biological, and physiologic data provides more robust subphenotyping than single modalities. 2 The trajectory-based approach is superior to static admission measurements because patients with identical admission SOFA scores can have vastly different outcomes based on their subsequent trajectory. 5

Subphenotyping approaches must be readily measurable with acceptable operating characteristics to allow global implementation. 2 Machine learning models requiring specialized biomarkers or complex genomic profiling are not practical for routine clinical use, particularly in resource-limited settings. 1, 2

The majority of sepsis patients (66%) have an improving phenotype with lower mortality risk, yet they still account for over 20% of all deaths due to their larger numbers, emphasizing the importance of identifying the high-risk worsening subphenotypes early. 5