Molecular phenotypes of critical illness confer prognostic and biological enrichment in sub-Saharan Africa: a prospective cohort study from Uganda

Abstract

The generalisability of critical illness molecular phenotypes to low- and middle-income countries (LMICs) is unknown. We show that molecular phenotypes derived in high-income countries (hyperinflammatory and hypoinflammatory, reactive and uninflamed) stratify sepsis patients in Uganda by physiological severity, mortality risk and dysregulation of key pathobiological domains. A classifier model including data available at the LMIC bedside modestly discriminated phenotype assignment (area under the receiver operating characteristic curve (AUROC) 0.80, 95% CI 0.71 to 0.90 for hyperinflammatory vs hypoinflammatory; AUROC 0.74, 95% CI 0.65 to 0.83 for reactive vs uninflamed). Our findings highlight the potential for a globally relevant, clinicomolecular classification of critical illness and may support the inclusion of diverse populations in phenotype-targeted critical care trials. Improved laboratory capacity and access to rapid biomarker assays are likely necessary to optimise phenotype stratification in LMIC settings.

Keywords: Critical Care; Cytokine Biology.

Figure 1:. Prognostic relevance, biological profiles, and classifier models for molecular phenotypes.

(a) Adjusted probability of 30-day mortality by the probability of assignment to the hyperinflammatory phenotype (as per IL-8, sTNFR1, and vasopressor use [hypotension] classifier) (N=226). Probabilities of 30-day mortality (dependent variable) generated in multivariable logistic regression model including probability of hyperinflammatory phenotype assignment, age, sex, illness duration prior to enrollment, and HIV co-infection as independent variables. Shading indicates 95% confidence interval. (b) Volcano plot of differential protein expression in patients assigned to hyperinflammatory vs. hypoinflammatory phenotypes (N=242). Red shading indicates proteins differentially expressed at |log₂-fold change| ≥0.5 and Benjamini-Hochberg (BH)-adjusted p-value ≤0.05. Blue shading indicates proteins differentially expressed at |log₂-fold change| <0.5 and BH-adjusted p-value ≤0.05. Grey and green shading indicates proteins without significantly different expression (BH-adjusted p-value >0.05). (c) Adjusted probability of 30-day mortality by the probability of assignment to the reactive phenotype (as per IL-6, IFN–γ, PAI-1, and Angiopoeitin-1/2 classifier) (N=226). Probabilities of 30-day mortality (dependent variable) generated in multivariable logistic regression model including probability of reactive phenotype assignment, age, sex, illness duration prior to enrollment, and HIV co-infection as independent variables. Shading indicates 95% confidence interval. (d) Volcano plot of differential protein expression in patients assigned to reactive vs. uninflamed phenotypes (N=242). Red shading indicates proteins differentially expressed at |log₂-fold change| ≥0.5 and Benjamini-Hochberg (BH)-adjusted p-value ≤0.05. Blue shading indicates proteins differentially expressed at |log₂-fold change| <0.5 and BH-adjusted p-value ≤0.05. Grey and green shading indicates proteins without significantly different expression (BH-adjusted p-value >0.05). (e-f) Performance of clinical classifier (age, sex, temperature, heart rate, respiratory rate, systolic blood pressure, oxygen saturation, AVPU mental status assessment) for discrimination of (e) hyperinflammatory vs. hypoinflammatory phenotype and (f) reactive vs. uninflamed phenotype (N=253). Area under the receiver operating characteristic curve (AUROC) presented with 10,000 bootstrap-derived 95% confidence intervals.