In this cohort of Mozambican children and adults with febrile illness presenting to hospital, sTREM-1 was the best biomarker for predicting 28-day mortality. sTREM-1 was superior to other biomarkers of immune and endothelial activation and provided additional discriminative ability when combined with clinical severity scores. Notably, sTREM-1 consistently demonstrated the best performance across different patient subgroups, though its performance was diminished in HIV-positive individuals. sTREM-1 remained the best biomarker if 7-day mortality was considered instead. sTREM-1 was positively correlated with length of hospital stay and was also the best biomarker to predict mortality among inpatients. Outpatients had lower plasma concentrations of all studied biomarkers compared to inpatients, yet their sTREM-1 levels were associated with seeking further care and subsequent hospitalisation.

These results provide additional evidence supporting the use of sTREM-1 as a predictor of mortality in children and adults with all-cause febrile illness. In hospitalised febrile children in Uganda, sTREM-1 had the best prognostic performance for 7-day mortality among a similar biomarker panel, with an AUROC of around 0.90¹⁰. Likewise, in Tanzanian febrile adults presenting to outpatient clinics, sTREM-1 was the best biomarker to predict 28-day mortality with an AUROC of 0.87 (95% CI: 0.81–0.92)¹¹. In diverse populations with specific syndromes or detected infectious agents (including malaria, pneumonia, COVID-19, or sepsis), high sTREM-1 levels have been consistently associated with subsequent mortality^{12,13,14,15,28,29,30,31}. When we applied previously defined sTREM-1 cut-offs for risk stratification, we observed a stepwise increase in mortality across categories, similar to that reported in other cohorts^10,13,15. Considering the challenges in determining the underlying cause of febrile illnesses, these data collectively suggest the broad applicability of sTREM-1 to risk-stratify febrile patients in a pathogen-agnostic manner.

sTREM-1 significantly outperformed CRP and PCT in predicting 28-day mortality, consistent with earlier findings from similar populations^10,11. sTREM-1 also outperformed lactate. Although prior research on pneumonia, malaria, and sepsis has documented an association between lactate levels and fatal outcome, the prognostic performance of lactate was usually inferior to sTREM-1^12,14,32,33. In addition, we applied various relevant clinical severity scores to our study population, similarly to previous publications using data from the FIEBRE study³⁴, and compared them to sTREM-1. Comprehensive clinical data collection occurred under our study conditions; however, it is worth noting that some variables may be measured with less accuracy or be unavailable in routine clinical practice. sTREM-1 alone had similar discriminative ability for 28-day mortality when compared to clinical severity scores, and the addition of sTREM-1 added discriminative ability to the scores. This result, coupled with similar previous findings, suggests that sTREM-1 could be integrated with relevant and context-specific clinical severity scores or decision algorithms to better risk-stratify patients with suspected acute infections^{11,13,15,28,35}. CHI3L1 and Angpt-2 are other candidate biomarkers that had comparable performance to sTREM-1 overall, but not across all patient subgroups. Combining biomarkers could enhance mortality prediction, although this may limit feasibility and practicality for clinical use. sTREM-1 and Angpt-2 was the best two-biomarker combination to predict 28-day mortality based on AUROC, but adding Angpt-2 only slightly increased the AUROC over sTREM-1 alone.

The prognostic performance of sTREM-1 for 28-day mortality differed by HIV status. While sTREM-1 remained a good predictor in HIV-positive participants, its performance was significantly inferior in this group compared to HIV-negative participants. This was more notable in individuals not receiving ART or unaware of their HIV-positive status. Active HIV infection causes a combination of immunosuppression and chronic inflammation that can result in immune system exhaustion and altered host responses to pathogens³⁶. Most of the deaths in the sTREM-1-based low-risk group were HIV-positive. Moreover, previous research has linked HIV infection with upregulation of TREM-1 in immune cells and increased plasma levels of sTREM-1, suggesting a possible contribution of TREM-1 to HIV-induced chronic inflammation, and hence potentially diminishing the prognostic performance of sTREM-1 in acute infections^37,38. One previous study in children with signs of severe infection explored sTREM-1 levels in relation to HIV status and mortality, but in this cohort sTREM-1 was not associated with mortality regardless of HIV status³⁹. Additional research is needed to elucidate the mechanistic impact of HIV infection on the TREM-1 pathway during febrile illnesses. Except for sTREM-1 and HIV status, selected biomarkers showed similar prognostic performance for 28-day mortality across patient subgroups. Nevertheless, almost all biomarkers tended to have lower AUROCs in HIV-positive participants. Existing limited evidence in febrile children found no sex-related differences in biomarker levels in relation to mortality⁴⁰. However, in paediatric severe malaria, sTREM-1 and sFlt-1 predicted mortality better in males¹⁵. PCT and IL-6 were inferior at predicting poor outcome in infants compared to older children with pneumonia in Bhutan³⁵. Future studies are needed to examine potential differences in biomarker prognostic performance across patient characteristics.

Death is often the primary outcome in prognostic studies with inpatients. Yet associations with less severe outcomes are also important for patient management, particularly in lower mortality risk clinical settings. In this cohort, mortality among outpatients was very low, and other adverse outcomes were assessed within this subgroup. Along with other biomarkers, we demonstrated an association between sTREM-1 levels and seeking additional care or subsequent hospitalisation in outpatients, suggesting that the studied biomarkers could play a future role in mitigating these adverse outcomes. Previous evidence on sTREM-1 and adverse outcomes in patients attending outpatient clinics and emergency departments is scarce. In these contexts, sTREM-1 has been associated with hospital admission and with mortality, in line with our results^11,30. However, in Spanish outpatients aged over 50 years with mild-to-moderate COVID-19, sTREM-1 poorly predicted hospitalization by day 28. This contrasts with our study findings, likely due to differences in demographics, disease severity, and healthcare settings⁴¹. In addition, sTREM-1 has been reported to be less suitable for predicting supplementary oxygen requirement in respiratory diseases^42,43,44. Further research is necessary to better understand whether sTREM-1 is sufficiently altered early in the disease course and confirm its prognostic utility for other adverse outcomes beyond mortality.

sTREM-1 has a pathophysiologic link with sepsis. TREM-1 is a cell-surface receptor expressed mostly by myeloid immune cells, which amplifies inflammation in infections⁴⁵. sTREM-1 acts as a decoy receptor and dampens TREM-1 activation⁴⁵. It is thought that sTREM-1 release depends on TREM-1 pathway activation and that it counteracts excessive inflammatory reactions, but high levels may reflect an underlying immune dysfunction^46,47. Interestingly, efforts have been made to modulate TREM-1 amplification in severe infections⁴⁷. Recent clinical trials in septic shock and COVID-19 have demonstrated that nangibotide, a TREM-1 specific inhibitor, was safe and holds potential to improve clinical status^48,49. Therefore, risk stratification strategies based on sTREM-1 could be coupled with targeted therapeutic strategies acting on the same pathways.

This study had several strengths, including its prospective enrolment of participants, a large and well-characterized cohort comprising both inpatients and outpatients, and a head-to-head comparison of several candidate biomarkers described in the literature. However, it also has limitations. The stratified enrolment strategy resulted in the inclusion of a high proportion of patients requiring hospitalisation, potentially limiting the representativeness with regard to all febrile cases presenting at the recruiting hospitals. There was a considerable loss to follow-up for the 28-day visit. Although this study had a significant sample size and number of primary outcome events, the statistical power for some subgroup or secondary analyses was limited. Additionally, we lacked detailed information on barriers to seeking healthcare and when and where outpatients sought further care or were hospitalised, which could have provided further insights and allowed us to restrict events stringently to 28 days as we did with mortality. Furthermore, information on viral load or CD4 cell count for HIV-positive participants was unavailable, which would have better indicated participants with HIV-associated immunosuppression. Conversely, we relied on self-reported data regarding ART use and awareness of a previous HIV diagnosis. We also lacked data on HIV status confirmation by molecular methods for seropositive children aged <18 months and they were considered positive cases. All studied biomarkers, except lactate, were measured retrospectively on stored samples, which might have given slightly different results than if all the testing had been done at the point of care.

In conclusion, this study supports sTREM-1 as a robust predictor of mortality in both children and adults with all-cause fever and provides evidence of its association with other adverse outcomes. Moreover, it shows that sTREM-1 prognostic performance may be diminished in HIV-positive individuals. While its application in different clinical scenarios warrants further investigation through interventional studies, measuring sTREM-1 in patients with febrile illness in resource-limited settings may facilitate timely risk stratification and management decisions with a positive impact on clinical outcomes.