Prognostic peripheral blood biomarkers at ICU admission predict COVID-19 clinical outcomes

Melina Messing, Mypinder S. Sekhon, Michael R. Hughes, Sophie Stukas, Ryan L. Hoiland, Jennifer Cooper, Nyra Ahmed, Mark Hamer, Yicong Li, Samuel B. Shin, Lin Tung, Cheryl L. Wellington, Don D. Sin, Kevin B. Leslie, Kelly M. McNagny

TL;DR

Researchers have identified a powerful two-part biomarker "signature"—serum IL-10 levels combined with a specific subset of $CD11c^{low}$ classical monocytes—that predicts with 91% accuracy whether a COVID-19 patient admitted to the ICU will recover quickly or face a prolonged stay or death. This streamlined prognostic tool, measurable at the point of admission, could significantly improve clinical triage and resource allocation.

What Problem Does This Paper Address?

While the peak of the pandemic has passed, COVID-19 continues to strain Intensive Care Units (ICUs) globally. A major challenge for clinicians is "prognostic uncertainty": when a patient is admitted, it is difficult to tell who will recover within a few days and who is at high risk for multi-week stays or mortality.

Standard clinical markers like C-reactive protein (CRP), D-dimer, age, and BMI are useful for measuring current disease severity, but they often lack the statistical power to predict future clinical trajectories. Without reliable prognostic tools, clinicians often apply broad-spectrum interventions (like corticosteroids) to all patients, potentially over-treating some while failing to identify those needing more aggressive management.

Methods

The researchers employed a multi-omic approach to find subtle immune signals that standard blood tests miss:

  1. Cytokine Profiling: They used the Simoa HD-1 platform (an ultra-sensitive digital immunoassay) to measure four key inflammatory cytokines: IL-1β, IL-6, IL-10, and TNFα.
  2. Deep Immunophenotyping (CyTOF): For a subset of patients, they used Mass Cytometry (CyTOF). Unlike traditional flow cytometry, which uses fluorescent tags, CyTOF uses heavy metal isotopes and mass spectrometry. This allowed the team to simultaneously track 38 different cell-surface markers, providing a high-resolution map of the "immune landscape" at the single-cell level.
  3. Statistical Integration: The team used dimensionality reduction (UMAP) and clustering (Rphenograph) to identify specific cell populations that correlated with clinical outcomes, eventually distilling complex high-dimensional data into a simple, two-step screening logic.

Datasets

The study analyzed a total of 90 COVID-19 patients admitted to two ICUs in Vancouver, Canada, between November 2020 and February 2021.

  • Cytokine Cohort (n=90): Used for the initial cytokine screening.
  • Initial Cohort (n=14): A subset used for the first round of deep CyTOF immunophenotyping.
  • Replication Cohort (n=28): A second subset used to validate the findings from the initial cohort.
  • Healthy Controls (n=8): Age-matched volunteers used to establish baseline immune signatures.

Key Insights and Ideas

  • The "Short-Stay" vs. "Long-Stay/Died" Threshold: The authors defined a new outcome metric: recovery in <6 days vs. stay ≥6 days or death. This 6-day cutoff was determined empirically as the point where immune markers showed the greatest statistical divergence.
  • The Power of Specificity: Interestingly, broad immune shifts (like total T-cell lymphopenia or total monocyte counts) failed to distinguish between the two outcome groups. The predictive power was hidden in a very specific sub-population: $CD11c^{low}$ classical monocytes.
  • IL-10 as a "Pre-Screen": While several cytokines were elevated in severe cases, IL-10 emerged as the most reliable cytokine predictor of ICU stay length.
  • Synergy over Solitude: Neither IL-10 nor the monocyte subset was perfectly predictive on its own. However, when used in a tiered approach—checking IL-10 first, then the monocyte signature—the accuracy jumped significantly. This suggests that severe COVID-19 outcomes are driven by a specific combination of systemic inflammation and cellular dysregulation.

Key Results

The combined evaluation of serum IL-10 and $CD11c^{low}$ classical monocytes at the time of ICU admission yielded:

  • 91% Sensitivity: Correct identification of those who would die or have a long stay.
  • 91% Specificity: Correct identification of those who would recover quickly.
  • Positive Likelihood Ratio (LR+) of 10.1: In clinical terms, an LR+ over 10 is considered a "strong" indicator that a test can confirm a diagnosis or outcome.
  • Cut-off Values: The predictive thresholds were set at IL-10 ≥ 15pg/ml or $CD11c^{low}$ classical monocytes ≥ $2.7 \times 10^7/L$.

Limitations and Future Work

  • Sample Size: While the results were validated across two cohorts, the total number of patients for the cellular analysis (n=42) is relatively small. Larger, multi-center studies are needed for clinical adoption.
  • Treatment Confounding: All patients in this study received dexamethasone (a corticosteroid). The authors suggest these biomarkers might actually be identifying patients who are less responsive to steroid therapy, an avenue that requires further investigation.
  • Variant Specificity: The data was collected during the "second wave." It remains to be seen if these biomarkers hold the same weight for newer variants (like Omicron) or in highly vaccinated populations.
  • Clinical Translation: While CyTOF is a research tool, the authors demonstrated that the monocyte marker can be detected using standard 2D flow cytometry, which is available in most major hospitals.

Summary updated

Abstract

Abstract The COVID-19 pandemic continues to challenge the capacities of hospital ICUs which currently lack the ability to identify prospectively those patients who may require extended management. In this study of 90 ICU COVID-19 patients, we evaluated serum levels of four cytokines (IL-1β, IL-6, IL-10 and TNFα) as well as standard clinical and laboratory measurements. On 42 of these patients (binned into Initial and Replication Cohorts), we further performed CyTOF-based deep immunophenotyping of peripheral blood mononuclear cells with a panel of 38 antibodies. All measurements and patient samples were taken at time of ICU admission and retrospectively linked to patient clinical outcomes through statistical approaches. These analyses resulted in the definition of a new measure of patient clinical outcome: patients who will recover after short ICU stays (&lt; 6 days) and those who will subsequently die or recover after long ICU stays ( &gt; 6 days). Based on these clinical outcome categories, we identified blood prognostic biomarkers that, at time of ICU admission, prospectively distinguish, with 91% sensitivity and 91% specificity (positive likelihood ratio 10.1), patients in the two clinical outcome groups. This is achieved through a tiered evaluation of serum IL-10 and targeted immunophenotyping of monocyte subsets, specifically, CD11c low classical monocytes. Immunophenotyping revealed clear predictors of clinical outcome in COVID-19 providing a highly sensitive and specific prognostic test that could prove useful in guiding clinical resource allocation. Graphical Abstract
BibTeX
@inproceedings{messing2022prognostic,
  author = {Melina Messing and Mypinder S. Sekhon and Michael R. Hughes and Sophie Stukas and Ryan L. Hoiland and Jennifer Cooper and Nyra Ahmed and Mark Hamer and Yicong Li and Samuel B. Shin and Lin Tung and Cheryl L. Wellington and Don D. Sin and Kevin B. Leslie and Kelly M. McNagny},
  title = {{Prognostic peripheral blood biomarkers at ICU admission predict COVID-19 clinical outcomes}},
  booktitle = {Frontiers in Immunology},
  year = {2022},
  doi = {10.3389/fimmu.2022.1010216},
}
Figure