Acute respiratory distress syndrome: phenotyping and precision medicine

Systematic review of ~4400 studies plus empirical dataset of 269 ICU patients evaluated by a panel of 18 experts using Delphi consensus.

• Goal: replace the 1994 AECC definition (acute lung injury vs ARDS terminology, PaO2/FiO2 < 300 irrespective of PEEP, four-category scheme)
• Key changes: dropped 'acute lung injury' (ALI) term; unified under ARDS with 3 severity bands; mandated PEEP ≥ 5 cmH2O; required timing within 1 week; specified imaging and origin of oedema
• Predictive validity (empirical cohort): as severity increased (mild → moderate → severe), mortality rose (27% → 32% → 45%), median ventilator-free days fell (16 → 12 → 9 days), and median ICU-free days fell (15 → 11 → 7 days)
• CONCLUSION: Berlin provided a reproducible, internationally agreed definition that standardised enrolment into trials and epidemiology — but it is anatomically/physiologically descriptive, NOT biologically mechanistic. Heterogeneity within 'severe ARDS' is why trials must additionally stratify by subphenotype.[1]

Prospective, multicentre, 4-week inception cohort in winter 2014. Detected ARDS using Berlin criteria; clinical recognition recorded prospectively.

• Incidence: 10.4% of ICU admissions fulfilled ARDS criteria (3022/29144)
• Severity at onset: mild 30%, moderate 46%, severe 24%
• Under-recognition: ARDS was recognised clinically in only 60-65% of cases — about 1 in 3 patients with ARDS was never clinically diagnosed as having it (mild ARDS least recognised)
• Under-treatment: lung-protective ventilation (Vt 6 mL/kg PBW) used in only ~two-thirds; prone positioning used in only 16% of severe ARDS despite PROSEVA-level evidence
• Mortality (hospital): mild 34.9%, moderate 40.3%, severe 46.1%
• CONCLUSION: ARDS is common, frequently unrecognised, and undertreated — even in ICUs in high-income countries. Recognition gap is the first barrier to delivering evidence-based care and to phenotyping-driven precision medicine.[7]

Applied latent class analysis (LCA) to 1022 and 944 patients from two NHLBI ARDS Network trials (ALVEOLI higher-vs-lower PEEP, and FACTT conservative-vs-liberal fluid) using 25 clinical and biomarker variables (IL-6, IL-8, sTNFr-1, SP-D, angiopoietin-2, ICAM-1, protein C, vWF, bicarbonate, creatinine, platelets, vasopressor use, etc.).

• Two classes consistently identified in both cohorts: a hyperinflammatory class (~30%) and a hypoinflammatory class (~70%)
• Mortality difference (FACTT): hyperinflammatory mortality higher than hypoinflammatory at 90 days
• PEEP response (ALVEOLI re-analysis): differential treatment effect by class — higher PEEP favoured the hyperinflammatory class, lower PEEP favoured the hypoinflammatory class (interaction test significant)
• Fluid response (FACTT): conservative fluid strategy improved ventilator-free and organ-failure-free days preferentially in the hyperinflammatory class
• CONCLUSION: ARDS is biologically heterogeneous; two reproducible subphenotypes respond DIFFERENTLY to the same therapy. This is the foundational evidence that precision medicine in ARDS is feasible and necessary.[2]

Retrospective, multicohort analysis. Trained machine-learning classifiers (random forest, logistic regression) using ONLY readily available clinical variables (no specialised biomarkers) in 4 cohorts: the St Paul's Hospital cohort, VALID, SJTRI, and the LUNG SAFE multinational cohort.

• Two subphenotypes reproduced using clinical data alone (PaO2/FiO2, ventilator settings, vasopressor use, bicarbonate, bilirubin, creatinine, platelets, age, comorbidities)
• Prevalence: hyperinflammatory ~30%, hypoinflammatory ~70% — consistent with biomarker-based LCA
• Mortality (LUNG SAFE): hyperinflammatory phenotype had significantly higher ICU and hospital mortality than hypoinflammatory, replicating the biomarker-defined gradient
• Clinical feasibility: because the model uses routine EHR data, bedside phenotyping is operationally feasible WITHOUT waiting for research biomarker panels
• CONCLUSION: ARDS subphenotypes can be identified in real-world ICU populations using routinely collected data, moving precision medicine from research to the bedside.[5]

Individual-patient-data meta-analysis of 1115 ARDS patients across 7 cohorts measuring plasma sRAGE.

• sRAGE independently associated with increased 28-day mortality after adjustment for severity
• Higher sRAGE correlated with fewer ventilator-free and organ-failure-free days
• sRAGE tracks alveolar type I epithelial cell injury and is one of the most reproducible ARDS biomarkers across cohorts
• CONCLUSION: sRAGE is the most validated single epithelial-injury biomarker in ARDS; useful for phenotyping and as a candidate enrichment biomarker for future precision-medicine trials.[12]

Secondary latent class analysis of HARP-2 (540 patients with ARDS randomised to simvastatin 80 mg vs placebo).

• Hyperinflammatory phenotype + simvastatin: improved 28-day survival (HR for death ~0.41) and more ventilator-free days
• Hypoinflammatory phenotype + simvastatin: trend toward harm
• Significant phenotype × treatment interaction for the primary outcome
• CONCLUSION: statin therapy in ARDS may benefit only the hyperinflammatory subphenotype — a precision-medicine rationale. The original HARP-2 trial (McAuley 2014) was neutral overall, illustrating how mixing phenotypes dilutes a real treatment effect.[3][4]

Latent class analysis and machine-learning validation across NHLBI ARDS Network trials (ALVEOLI, FACTT) and LUNG SAFE:

• Hyperinflammatory (Type 1): mortality ~40-50% (worse prognosis); higher vasopressor, acidosis, thrombocytopenia, low protein C
• Hypoinflammatory (Type 2): mortality ~20-25% (better prognosis); lung-predominant disease
• PEEP response (ALVEOLI re-analysis): higher PEEP BENEFITED hyperinflammatory (lower mortality), HARMED hypoinflammatory (overdistension)
• Fluid strategy (FACTT re-analysis): conservative fluid BENEFITED hyperinflammatory (more ventilator-free days), neutral in hypoinflammatory
• Statin (HARP-2 re-analysis): possible benefit in hyperinflammatory (improved 28-day survival), possible harm in hypoinflammatory
• Stability (Delucchi 2018): subphenotypes are reasonably stable over the first week but transitions occur — serial reassessment is rational [1]

CONCLUSION: ARDS subphenotypes determine treatment response and prognosis. Precision medicine (stratify by subphenotype) may improve outcomes. Future trials should stratify.[1][2][5][6]

Analysis of 549 ALVEOLI patients combining clinical variables (age, APACHE) with biomarkers (IL-6, IL-8, sTNFr-1, SP-D, angiopoietin-2).

• Combining clinical + biochemical indices outperformed either alone for predicting mortality and ventilator-free days
• sTNFr-1, IL-6, IL-8 and SP-D were among the strongest individual predictors
• Established the principle that ARDS biology is multidimensional and that biomarker panels (later formalised by latent class analysis) carry prognostic information beyond severity scores
• CONCLUSION: the rationale for biomarker-based phenotyping — biology adds information that severity scoring (Berlin, APACHE) cannot capture.[11]