Fig. 7. Relationships between immune mediators and severe acute respiratory infection etiology.

a Violin plots showing immune mediator concentrations stratified by severe acute respiratory infection (SARI) etiology (N = 292; 66 patients with asymptomatic SARS-CoV-2 infection excluded). Concentrations across groups compared using Kruskal-Wallis H test followed by Dunn’s test for multiple comparisons with p values adjusted using Benjamini-Hochberg method; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. b Associations between COVID-19 and immune mediator concentrations (N = 292; 66 patients with asymptomatic SARS-CoV-2 infection excluded); coefficients with 95% confidence interval bars generated in multivariable linear regression models including log₁₀-transformed mediator concentration as dependent variable and SARI etiology (binary; COVID-19 vs. influenza/non-influenza SARI), age (continuous), self-reported sex (binary), HIV co-infection (binary; 4 patients not known to be living with HIV but with missing rapid diagnostic tests analyzed as HIV negative) and WHO clinical severity classification (categorical) as independent variables. c Importance of immune mediators in gradient-boosted machine classifier for prediction of COVID-19 (N = 240) vs non-COVID-19 (N = 52) SARI as per model split-gain values; top 20 variables presented in descending order of importance. d Shapley additive explanations (SHAP) values derived from gradient-boosted machine model for prediction of COVID-19 (N = 240) vs non-COVID-19 (N = 52) SARI; SHAP values > 0 indicate positive impact on prediction while values < 0 indicates negative impact (e.g., high concentrations of EGF have a strong positive contribution to prediction of COVID-19). Source data are provided in the Source Data file.