Sepsis is characterised by a dysregulated host response to infection that leads to life-threatening organ dysfunction. Although immune dysregulation is fundamental to its definition, most immunomodulation trials enrol patients based on clinical severity rather than the degree of immune disturbance, potentially contributing to variable treatment effects.
A practical method to quantify immune dysregulation could enhance prognostication, support assessment of treatment response, and help identify patients most likely to benefit from immunomodulatory therapies. A recent study aimed to develop a parsimonious machine-learning tool to define and quantify immune dysregulation, enabling a more biologically informed approach to treatment.
In this multicohort analysis, including a reanalysis of a randomised controlled trial, the primary objective was to derive and validate both categorical and continuous measures of immune dysregulation that are independent of clinical presentation and outcomes. The investigators measured 35 plasma biomarkers representing key domains of the host response in patients with community-acquired pneumonia (CAP) across a range of care settings (emergency department, general ward, and intensive care unit) and disease severities, using data from three independent cohorts.
Unsupervised trajectory inference analysis was applied to identify a gradient of immune dysregulation, represented as discrete stages Dysregulated Immune Profiles (DIP1–3) and a continuous score (cDIP; range 0–1). Two parsimonious machine-learning models were then developed to predict DIP stage and cDIP values from the biomarker data. These models were subsequently validated in five independent cohorts to assess their ability to capture immune dysregulation and their association with clinical outcomes.
To explore whether treatment effects vary according to the degree of immune dysregulation, a post-hoc analysis of the CAPE COD randomised trial (evaluating hydrocortisone in severe CAP) was conducted. This analysis examined outcomes across DIP stages and the cDIP continuum, as well as the impact of hydrocortisone on immune dysregulation trajectories over time.
A total of 398 patients with CAP were positioned along a continuum of immune dysregulation based on 35 measured biomarkers, yielding three distinct stages (DIP1–3) and a continuous cDIP score. Clinical severity was found to be a poor surrogate for immune dysregulation.
A simplified machine-learning model incorporating three biomarkers, procalcitonin, soluble TREM-1, and interleukin-6, demonstrated high accuracy in predicting the degree of dysregulation defined by the full biomarker panel.
Although not designed primarily as a prognostic tool, increasing levels of immune dysregulation (as measured by DIP stage and cDIP) were independently associated with worse outcomes. Each 10% increase in cDIP was linked to higher mortality and a greater risk of secondary infections, regardless of clinical severity. The three-biomarker model was successfully validated in five external cohorts comprising 1,191 patients with varying infections, illness severities, and care environments.
Reanalysis of the CAPE COD trial revealed that hydrocortisone improved survival only in patients with severe immune dysregulation. Specifically, 30-day mortality was reduced in those classified as DIP3 or with cDIP ≥0.63. This benefit was accompanied by more rapid recovery from immune dysregulation over time. No similar treatment effect modification was observed when patients were stratified by clinical severity alone.
This study presents a publicly available, three-biomarker machine-learning framework capable of quantifying host immune dysregulation in sepsis. By moving beyond traditional severity-based stratification, this approach offers a potential pathway towards precision-guided immunomodulatory therapy, enabling more targeted treatment of patients most likely to benefit.
Source: The Lancet
Image Credit: iStock
