Abstract Rationale In ARDS, investigators have identified two sub-phenotypes (hyper and hypo-inflammatory). There is increasing data that these sub-phenotypes are also present in ICU patients with sepsis and may predict differential treatment responses. Currently, these sub-phenotypes are classified retrospectively using ELISA-based research biomarkers which take 6-8 hours to measure in batches using 96-well plates with variable reproducibility. We completed a pilot study using the Roche Cobas e 411 platform to rapidly (20 mins) measure three plasma biomarkers, interleukin-6 (IL-6), angiopoietin-2 (Ang-2) and soluble triggering receptor expressed on myeloid cell-1 (sTREM-1) and evaluate their relative effectiveness in classifying sub-phenotypes in two observational cohorts. These assays are labeled Research Use Only, and only the Elecsys IL-6 assay is available commercially in the US under Emergency Use Authorization. Methods We enrolled two prospective cohorts of patients with sepsis (CITRC-1 and 2) and measured IL-6, Ang-2, and sTREM-1 in plasma collected within 24 hours of ICU admission using the Roche platform. Sub-phenotypes were separately derived based on research biomarkers using published validated models that included IL-6, sTNFR-1, and serum bicarbonate (PMID: 31948926). We then applied LASSO regression to fit models for predicting sub-phenotypes using Roche biomarker measurements. Areas under the receiver operator curves (AUCs) were constructed using 5-fold cross-validation, and 95% confidence intervals were derived from a nonparametric bootstrap with 2000 replicates, shown below. Results Both cohorts included ICU patients with sepsis (CITRC-1: n = 332 and CITRC-2: n = 249). In CITRC-1, 223 (67%) were male, 171 (52%) were mechanically ventilated at baseline and hospital mortality was 8%. In CITRC-2, 196 (79%) were male, 121 (49%) were mechanically ventilated and mortality was 11%. Roche biomarkers had low (10%) inter-assay CVs. In CITRC-1, 307 were classified as hypo-inflammatory and 25 as hyper-inflammatory, and the Roche IL-6-based model showed strong predictive performance for classification (AUC: 0.92; 95% CI: 0.90-0.93), without significant improvement with the addition of Ang-2 or sTREM-1. Roche-derived hyperinflammatory sub-phenotype was associated with persistent day 3 mechanical ventilation (sHR: 2.43 (95% CI: 1.52-3.89), p-value: 0.0002)). In CITRC-2, we found similar predictive performance for Roche IL-6 based model to predict sub-phenotypes (AUC 0.87 (95% CI: 0.86-0.89) and associated with day 3 mechanical ventilation (sHR: 2.7 (95% CI: 1.7-4.4)). Conclusions In two ICU sepsis cohorts, IL-6, which is clinically available, was found to effectively predict hyper and hypo-inflammatory sub-phenotypes. IL-6 guided sub-phenotype prediction could be used as enrichment tools in prospective trials to identify treatment response within heterogeneous ICU populations. This abstract is funded by: Collaborative Research Agreement with Roche Diagnostics
Nacharaju et al. (Fri,) studied this question.