Aetiology of sepsis in adults living with HIV in East Africa: a secondary analysis of an open-label, multicentre, randomised, controlled phase 3 trial

Abstract

Background: Sepsis in people living with HIV (PLWH) in East Africa has high mortality. Regionally, the etiology of sepsis is incompletely understood. We performed a planned analysis of the microbiological data obtained from a randomised clinical trial of early empiric anti-Mycobacterium tuberculosis (Mtb) therapy for sepsis (ATLAS) in Tanzania and Uganda. Methods: We present a prespecified, secondary analysis of a phase three, open-label, multicentre, randomised, controlled trial conducted at four regional referral hospitals in Tanzania and Uganda. Participants were adults living with HIV admitted with concern for infection and a modified quick sepsis-related organ failure assessment (qSOFA) ≥2. Participants were randomised to (1) immediate or diagnosis-dependent antituberculosis therapy and to (2) high-dose or conventional-dose antituberculosis therapy. Tests for sepsis etiology included bacterial blood and urine cultures, multi-pathogen qPCR from blood, GeneXpert MTB/RIF Ultra from sputum and urine, urine lipoarabinomannan (LF-LAM), and Mtb cultures from sputum and blood. We used multivariable logistic regression analysis and random forest analysis to determine variables that predicted Mtb as the sepsis etiology. The trial is registered with ClinicalTrials.gov, NCT04618198. Findings: From January 5, 2022 through December 9, 2024, we randomised 437 participants to receive immediate and/or high dose antituberculosis therapy. Mtb was the most common pathogen, detected in 229 (52%) of 437 participants, and in 54 (50%) of 108 participants with a bloodstream infection. Combined urine LF-LAM and sputum GeneXpert MTB/RIF testing missed 17 (32%) of 54 Mtb bloodstream infections. The most frequent non mycobacterial bacteria were Klebsiella species and Escherichia coli, which were identified in 39 (9%) and 33 (8%) of 437 participants, respectively. We detected ceftriaxone resistance in 21 (64%) of 33 bacterial isolates. In a random forest prediction model (accuracy: 0.6; precision: 0.5; recall: 0.6; F1-score: 0.5), the best indicators of Mtb as a sepsis pathogen were a greater number of ill-days before presentation (mean decrease in accuracy [MDA] 10.1), younger age (MDA 8.7), a longer duration of cough (MDA 7.7), and low CD4+ T-cell concentration (MDA 3.7). Interpretation: Mtb was the most common pathogen causing sepsis and bloodstream infection and was frequently missed by conventional rapid diagnostics. We also identified a high prevalence of non-mycobacterial pathogens resistant to ceftriaxone in blood and urine cultures. Limitations of our study included exclusion of cryptococcal antigen positive participants, non-systematic drug susceptibility testing, and potential regional differences in sepsis etiology and resistance patterns.