High plasma IL-18 identifies high-risk ARDS patients not identified by latent class analysis sub-phenotyping: a secondary analysis of the SAILS and HARP-2 studies

Background: Both latent class analysis (LCA) assignment based upon a panel of plasma biomarkers and interleukin-18 (IL-18) plasma level have been shown to predict prognosis and treatment response in Acute Respiratory Distress Syndrome (ARDS). Interleukin-18 is a measure of inflammasome activation and plays a distinct role in inflammation that is not captured by the biomarkers used in LCA assignments. We hypothesized that elevated IL-18 would provide additive prognostic and therapeutic information to previously published LCA assignments in ARDS, identifying additional “high-risk” patients not captured by LCA who could be eligible for inclusion in future precision medicine-focused trials. Methods: IL-18 and a panel of protein markers used for LCA had been previously measured in plasma from 683/745 patients in the Statins for Acutely Injured Lungs from Sepsis (SAILS) and 511/540 patients in the Hydroxymethylglutaryl-CoA reductase inhibition with simvastatin in Acute lung injury to Reduce Pulmonary dysfunction (HARP-2) trials. We tested the association between high IL-18 (>800 pg/mL) and LCA class assignment using McNemar’s test and evaluated the association of each subgrouping as well as treatment with 60-day mortality using Fisher’s exact test. We assessed 60-day mortality in each combination (high/low IL-18, hypo-/hyper-inflammatory LCA class, and treatment/placebo) using Kaplan-Meier survival analysis. We evaluated the correlation between the log2 transformed IL-18 level and LCA biomarkers using Pearson’s correlation coefficient. Results: 33% of patients in SAILS and HARP-2 were discordant by IL-18 level and LCA class. Elevated IL-18 identified a high-risk group of individuals previously classified as hypo-inflammatory by LCA in both SAILS (OR 3.3, 95% CI 1.8-6.1, p of 0.17-0.47. Conclusions: High Plasma IL-18 level provides additional prognostic information to LCA sub-phenotypes in two large ARDS cohorts.