Immune patterns predict tuberculosis and death in sepsis with high HIV in East Africa

Sepsis is a leading cause of death among people living with HIV in Africa, but detailed information about how the immune system behaves in these patients has been limited. In work led by corresponding author Christopher C. Moore, researchers set out to identify immune subphenotypes — distinct patterns of immune response — among adults admitted with sepsis in East Africa where HIV prevalence is high. The team measured serum cytokine and antibody concentrations and looked for links between those measurements and clinical markers, including CD4+ T-cell count and blood lactate concentration, as well as the presence of tuberculosis and death within 30 days. To make sense of complex immune data they applied machine learning and statistical tools used in many modern biomarker studies, including K-means clustering, principal component analysis (PCA), and logistic regression. The investigators also tested whether the patterns they found in one group of patients could be reproduced in a separate validation cohort of adults with sepsis from the same region, strengthening the reliability of their findings.

The study used PCA to find which immune molecules most strongly associated with tuberculosis and with 30-day mortality. For tuberculosis, the main contributing signals were G-CSF, IL-5, IL-6, IL-8, and IL-13. For mortality, the principal contributors were IL-4, IL-6, IL-8, IL-12, and IL-13. By combining biomarker measurements with clinical information in multivariable models, the researchers were able to predict tuberculosis with an area under the receiver operating characteristic curve (AUC) of 0.84, and to predict 30-day mortality with an AUC of 0.78. Those predictive performances were replicated in the separate validation cohort. Cross-testing of the two models suggested a functional difference: the tuberculosis model captured pathogen-specific immune activation, while the mortality model reflected non-pathogen-specific immune dysregulation. The report also notes conclusions related to antibody responses.

These findings point to measurable immune signatures that distinguish between infection with tuberculosis and broader immune dysfunction that predicts death in people with sepsis and high HIV prevalence. If the patterns identified here hold up in further work, they could help clinicians and researchers to sort patients into groups with different risks or likely causes of illness — for example, highlighting those whose immune profile points toward tuberculosis versus those with a more generalized dysregulated response. The use of standard analytical tools such as K-means clustering, principal component analysis (PCA), and logistic regression, together with replication in a validation cohort, strengthens confidence that these are real patterns and not artifacts of a single patient group. The study emphasizes the importance of measuring both cytokine and antibody responses alongside routine clinical markers like CD4+ T-cell count and blood lactate, and suggests avenues for future research to translate immune subphenotypes into diagnostic or prognostic tests for sepsis in settings with high HIV prevalence.