https://orcid.org/0000-0002-3581-0383
Figures
Abstract
Pneumonia is a leading cause of child mortality. However, currently we lack simple, objective, and accurate risk-stratification tools for pediatric pneumonia. Here we test the hypothesis that measuring biomarkers of immune and endothelial activation in children with pneumonia may facilitate the identification of those at risk of death. We recruited children <10 years old fulfilling WHO criteria for pneumonia and admitted to the Manhiça District Hospital (Mozambique) from 2010 to 2014. We measured plasma levels of IL-6, IL-8, Angpt-2, sTREM-1, sFlt-1, sTNFR1, PCT, and CRP at admission, and assessed their prognostic accuracy for in-hospital, 28-day, and 90-day mortality. Healthy community controls, within same age strata and location, were also assessed. All biomarkers were significantly elevated in 472 pneumonia cases versus 80 controls (p<0.001). IL-8, sFlt-1, and sTREM-1 were associated with in-hospital mortality (p<0.001) and showed the best discrimination with AUROCs of 0.877 (95% CI: 0.782 to 0.972), 0.832 (95% CI: 0.729 to 0.935) and 0.822 (95% CI: 0.735 to 0.908), respectively. Their performance was superior to CRP, PCT, oxygen saturation, and clinical severity scores. IL-8, sFlt-1, and sTREM-1 remained good predictors of 28-day and 90-day mortality. These findings suggest that measuring IL-8, sFlt-1, or sTREM-1 at hospital presentation can guide risk-stratification of children with pneumonia, which could enable prioritized care to improve survival and resource allocation.
Citation: Balanza N, Erice C, Ngai M, McDonald CR, Weckman AM, Wright J, et al. (2023) Prognostic accuracy of biomarkers of immune and endothelial activation in Mozambican children hospitalized with pneumonia. PLOS Glob Public Health 3(2): e0001553. https://doi.org/10.1371/journal.pgph.0001553
Editor: Sadia Shakoor, Aga Khan University, PAKISTAN
Received: November 1, 2022; Accepted: January 24, 2023; Published: February 23, 2023
Copyright: © 2023 Balanza et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: De-identified data analyzed as part of this study is available at CORA research data repository through the following link https://doi.org/10.34810/data595.
Funding: We acknowledge funding support from the Bill and Melinda Gates Foundation (OPP50092), the Canadian Institutes of Health Research (CIHR) Foundation grant FDN-148439 (KCK), the Canada Research Chair Program (KCK), and donations from Kim Kertland and the Tesari Foundation. ISGlobal receives support from the Spanish Ministry of Science and Innovation through the “Centro de Excelencia Severo Ochoa 2019-2023” Program (CEX2018-000806-S), and support from the Generalitat de Catalunya through the CERCA Program. CISM is supported by the Government of Mozambique and the Spanish Agency for International Development (AECID). NB is supported by an FPU predoctoral fellowship from the Spanish Ministry of Universities (FPU18/04260). BB is a Beatriu de Pinos postdoctoral fellow granted by the Government of Catalonia’s Secretariat for Universities and Research, and by Marie Sklodowska-Curie Actions COFUND Program (BP3, 801370). EL-V has received funding from the European Respiratory Society and the European Union’s H2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 847462. CRM, AMW, and JW are supported by CIHR Doctoral Awards. Funding bodies played no role in study design; collection, analysis, or interpretation of the data; writing of the report, or in submission of this article for publication.
Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: KCK is a named inventor on a patent “Biomarkers for early determination of a critical or life-threatening response to illness and/or treatment response” held by the University Health Network. EL-V is an editorial board member of PLOS Global Public Health. The remaining authors declare no competing interests.
1. Introduction
Pneumonia is a leading cause of illness and death in children. In 2019, pneumonia resulted in an estimated 740,000 deaths in children under the age of five, accounting for nearly 14% of all deaths in this age group [1]. Although childhood pneumonia is a global problem, it disproportionately affects low and middle-income countries.
Identifying children with pneumonia who are at high risk of severe or fatal disease is essential to reduce mortality, as such cases need early and adequate clinical management. Nevertheless, this is challenging in settings without radiographic imaging, basic or inexistent laboratory facilities, and limited qualified medical personnel. The World Health Organization (WHO) has set clinical algorithms for resource-constrained settings to diagnose pneumonia as a clinical syndrome and identify severe cases to guide patient case management [2–4]. Yet, these guidelines have limitations including subjective signs recognition. Clinical signs and laboratory measures have also been combined to generate prognostic scores for children with pneumonia or, more broadly, febrile illness [5–8]. However, these scores have not been well validated nor implemented for childhood pneumonia. New simple, accurate, objective, and inexpensive triage and risk-stratification tools are needed.
Immune and endothelial activation are implicated in the pathogenesis of severe and fatal infections. Previous studies have shown that circulating analytes of these pathways, measured at clinical presentation, have good prognostic accuracy in a range of infections, including pneumonia. They have been associated with mortality and other adverse outcomes, and they may enable improved childhood pneumonia risk-stratification [9–13]. However, only a few studies have assessed some of these biomarkers in children with pneumonia in sub-Saharan Africa [14–21].
Here we evaluated the prognostic accuracy of eight biomarkers for in-hospital and longer-term (28-day and 90-day) mortality in Mozambican children hospitalized with WHO-defined clinical pneumonia versus clinical signs and severity scores. Their biomarker levels were also compared to those in healthy community controls.
2. Methods
2.1 Study site
This study was conducted in Manhiça district, a rural area located in Maputo province, southern Mozambique. Studies conducted in the district of Manhiça have consistently described significant percentages of child hospital admissions and mortality due to pneumonia [22–24], which are exacerbated by suboptimal pneumonia management and referral practices [25]. A full description of the Manhiça population and study area can be found in previous publications [26, 27].
2.2 Study design
This analysis is part of a larger study designed to identify etiological biomarkers in children with respiratory symptoms [28–30]. Between 2010 and 2014, children presenting to the Manhiça District Hospital with symptoms of clinical pneumonia and fulfilling criteria for hospital admission were assessed for recruitment. Children were invited to participate in the study if they were <10 years old, had fever (≥37.5°C axillary temperature) or reported a history of fever in the preceding 24 hours, and met the 2005 WHO case definition for clinical pneumonia (cough or/and difficulty breathing plus increased respiratory rate for age) [31]. Exclusion criteria included use of antimalarials in the preceding two weeks, established or suspected tuberculosis (based on history of cough of >2 weeks or close contact with a documented case), marked hypoxemia (SpO2 ≤85%, as a proxy of Pneumocystis jirovecii infection), and participation in conflicting studies. Patients were managed in accordance with Mozambique’s national standard of care [22].
Community controls <10 years old and living in the district of Manhiça were recruited from 2010 to 2012. The Health and Demographic Surveillance System (HDSS) running in the district allowed the selection of a random sample of children within the same age group and geographic area than study patients. Controls were not considered healthy if they presented with fever, had taken any medication during the preceding 30 days, had a positive malaria microscopy exam, or showed any other important sign of disease. Human immunodeficiency virus (HIV) testing was not performed in this group.
2.3 Clinical, laboratory, and radiological data
Standard clinical questionnaires were completed for all subjects. Upon recruitment and prior to initiating treatment, venous blood and a nasopharyngeal aspirate were collected from pneumonia patients. Briefly, blood was used on-site for hematocrit, full blood count, blood smears for malaria diagnosis, blood culture, and HIV-1 testing. Dried blood spots were collected and used for molecular screening for Streptococcus pneumoniae and Haemophilus influenzae type B. The nasopharyngeal aspirate was used for PCR-based determination of influenza virus (A, B and C), respiratory syncytial virus (A and B), adenovirus, parainfluenza (1, 2, 3 and 4), coronavirus (229E and OC43), rhinovirus, enterovirus, human metapneumovirus, and bocavirus. Cerebrospinal fluid and pleural effusion were collected and cultured if clinically indicated. The detection of a non-contaminating bacterial organism in cerebrospinal fluid, pleural effusion, and/or blood was defined as an invasive bacterial disease. Pneumonia patients underwent anteroposterior chest radiography and X-rays were interpreted by two experienced clinicians following a WHO-designed protocol [32]. From community controls, venous blood was collected and used for blood smears for malaria diagnosis. Further details on all procedures can be found elsewhere [30].
2.4 Biomarker quantification
A volume of ~2 mL of the venous blood was collected in EDTA coated tubes, centrifugated, and the derived plasma was isolated and stored at -80°C until shipment to University Health Network, Canada. Plasma concentrations of angiopoietin-2 (Angpt-2), interleukin-6 (IL-6), interleukin-8 (IL-8), procalcitonin (PCT), soluble fms-like tyrosine kinase-1 (sFlt-1, also known as soluble vascular endothelial growth factor receptor-1 or sVEGFR-1), soluble tumor necrosis factor receptor 1 (sTNFR1), and soluble triggering receptor expressed on myeloid cells 1 (sTREM-1) were quantified using the multiplex Luminex platform with custom-developed reagents from R&D Systems [33]. Individual Luminex plates included 10% duplicates. C-reactive protein (CRP) was quantified by enzyme-linked immunosorbent assay (ELISA) (R&D DuoSet) and all assays were run in duplicate. Procedures were performed according to manufacturers’ instructions and blinded to clinical data. Biomarker concentrations outside the dynamic range were assigned a value of one-third of the lowest limit or the highest limit of the standard curve (S1 Table). If volume was insufficient for both assays, Luminex was prioritized over ELISA.
2.5 Post-discharge mortality data
Post-discharge mortality data were retrieved from the Manhiça HDSS. Each person living in the surveillance area is issued a unique permanent identification number, allowing HDSS data to be linked to the study database. HDSS mortality data were obtained through semestral household visits, weekly updates from key community informants, and daily hospital visits [26, 27].
2.6 Data analysis
All statistical analyses and graphs were performed using Stata 16.0 (Stata Corp, College Station, TX, USA). Our primary outcome was in-hospital mortality during initial admission. Secondary outcomes were 28-day and 90-day mortality. Children who were transferred to a higher-level hospital or absconded (i.e., left against medical advice) were excluded when evaluating the primary outcome, and children without information on vital status at 28 or 90 days were excluded when assessing secondary outcomes. A sensitivity analysis was conducted to evaluate 28-day and 90-day mortality excluding children who died during initial hospital admission. Biomarker concentrations in two groups were compared using Mann-Whitney U tests. Areas under the receiver operating characteristic curve (AUROCs) for one parameter were analyzed nonparametrically. Logistic models were used to estimate odds ratios (ORs) and AUROCs for multivariable models. Biomarker data were log-transformed for inclusion in logistic regression models. Models and variables were compared in terms of discrimination, by comparing AUROCs using the algorithm suggested by DeLong et al. [34]. Best two-biomarker combination was determined based on highest AUROC, excluding CRP from this analysis. Biomarker correlations were explored using Spearman’s rank correlation coefficients. Two-sided p-values <0.05 were considered statistically significant.
Severe pneumonia was defined, according to the WHO 2013 pocket book, as the presence of at least one of the following: oxygen saturation <90% or central cyanosis, severe respiratory distress (i.e., grunting, very severe chest indrawing), or any danger sign (i.e., inability to breastfeed or drink, lethargy or reduced level of consciousness, convulsions) [3]. This WHO severity definition, SpO2 [35], the Respiratory Index of Severity in Children-Malawi (RISC-Malawi) score [7, 36], and the Lambaréné Organ Dysfunction Score (LODS) [37–39] were pre-specified as important mortality prognostic factors. The RISC-Malawi score takes into account degree of hypoxemia, severity of malnutrition (here using mid-upper arm circumference [MUAC]), sex, presence of wheeze, and level of consciousness; while the LODS considers deep breathing, prostration, and coma. Biomarker cut-offs were derived for in-hospital mortality using the Youden’s index method (J = max[sensitivity+specificity-1]) and pre-specified likelihood ratios. A negative likelihood ratio (LR-) of 0.1 and a positive likelihood ratio (LR+) of 10 were used as they indicate large changes from pre-test to post-test probabilities. Lastly, we applied sTREM-1 cut-offs derived using same likelihood ratios for 7-day in-hospital mortality by Leligdowicz et al. [13] in a cohort of Ugandan febrile children. Corresponding sensitivity, specificity, LR-, LR+, positive predictive value (PPV), and negative predictive value (NPV) were calculated for each cut-off.
2.7 Ethical approval
This study was approved by the Mozambique National Bioethics Committee (reference No. CNBS-IRB00002657) and the Ethics Committee at the Hospital Clinic in Barcelona, Spain (reference No. 2010/5590), and performed in accordance with the principles of the Declaration of Helsinki. Parents or guardians provided written informed consent after a detailed explanation of the study. HIV counseling was provided to families of participating children undergoing HIV testing.
3. Results
3.1 Study population characteristics
A total of 576 pneumonia cases were recruited. Of these, 472 had a plasma sample for biomarker measurement and were included in this analysis. In 11 out of 472 pneumonia cases, CRP could not be assessed due to insufficient volume for all assays. Their clinical and demographic characteristics are presented in Table 1. Half (50.0%; 236/472) were females, median age was 14.1 months (interquartile range [IQR]: 6.1 to 27.1), 17.6% (83/472) had severe malnutrition (weight-for-age z-score (WAZ) <-3 SD), 18.6% (88/472) were HIV positive, 18.6% (88/472) had malaria, 13.1% (62/472) had an invasive bacterial disease, 65.1% (298/458) had a viral respiratory infection, and 37.4% (151/404) had a normal chest X-ray. A total of 15 (3.2%) died during hospital stay, 20 (4.2%) absconded, 20 (4.2%) were transferred to another hospital, and 417 (88.3%) were discharged alive.
Of 117 community controls examined, 80 fulfilled the definition of healthy community control and had a plasma sample for biomarker measurement. Of these, 56.3% (45/80) were female and median age was 39.0 months (IQR: 24.7 to 55.8) (Table 1).
3.2 Biomarkers as indicators of disease and severity
All biomarker concentrations were significantly higher in pneumonia cases versus healthy community controls (p<0.001). Equivalent results were observed when comparing pneumonia cases discharged alive with healthy community controls (Fig 1).
Boxplots with median and interquartile range. Whiskers represent one and a half times the interquartile range, and outliers are plotted as individual points. Concentrations are in pg/mL, except CRP in μg/mL. p-values were computed using the Mann-Whitney U test. *p<0.05, **<0.01; ***p<0.001; ns: non-significant. Abbreviations: Angpt-2 (angiopoietin-2), CRP (C-reactive protein), IL-6 (interleukin-6), IL-8 (interleukin-8), PCT (procalcitonin), sFlt-1 (soluble fms-like tyrosine kinase-1), sTNFR1 (soluble tumor necrosis factor receptor), sTREM-1 (soluble triggering receptor expressed on myeloid cells 1).
Forty-six percent of pneumonia cases fulfilled the WHO definition of severe pneumonia. IL-8 was the only biomarker significantly increased in severe cases compared to non-severe cases (p = 0.046) (S2 Table).
3.3 Biomarkers as indicators of in-hospital death
There were 15 in-hospital deaths in children with pneumonia during initial hospital admission. These occurred between the day of admission and 16 days afterwards, with 40.0% (6/15) happening within the first 48 hours of hospitalization. IL-8, sFlt-1, and sTREM-1 were higher in in-hospital deaths compared to children discharged alive (p<0.001), as well as sTNFR1 (p = 0.002) (Fig 1). The odds of in-hospital mortality increased significantly for a two-fold increase in IL-8, sFlt-1, sTREM-1, or sTNFR1 concentration (Table 2).
IL-8 had the best discriminatory capacity for in-hospital deaths, with an AUROC of 0.877 (95% confidence interval [CI]: 0.782 to 0.972) (Fig 2A and 2B). However, IL-8 was not significantly better than the next two most discriminative biomarkers, sFlt-1 and sTREM-1, with AUROCs of 0.832 (95% CI: 0.729 to 0.935) and 0.822 (95% CI: 0.735 to 0.908), respectively. These three biomarkers individually were significantly superior to CRP and PCT (S3 Table). IL-8 combined with sTREM-1 was the best two-biomarker combination (AUROC: 0.904 [95% CI: 0.820 to 0.988]), which performed slightly better than IL-8 alone (p = 0.024). Correlations between biomarkers can be found in S1 Fig.
(A) IL-8, sFlt-1, sTREM-1, sTNFR1, Angpt-2, PCT, and IL-6 AUROCs for in-hospital deaths (n = 432). (B) IL-8, sFlt-1, sTREM-1, and CRP AUROCs for in-hospital deaths (n = 421). (C) IL-8, sFlt-1, and sTREM-1 AUROCs for 28-day and 90-day mortality (in-hospital & post-discharge) (n = 309). (D) IL-8 and clinical parameters AUROCs for in-hospital deaths (n = 405). Abbreviations: Angpt-2 (angiopoietin-2), AUROC (area under the receiver operating characteristic curve), CRP (C-reactive protein), IL-6 (interleukin-6), IL-8 (interleukin-8), LODS (Lambaréné Organ Dysfunction Score), PCT (procalcitonin), RISC-Malawi (Respiratory Index of Severity in Children-Malawi), sFlt-1 (soluble fms-like tyrosine kinase-1), sTNFR1 (soluble tumor necrosis factor receptor), sTREM-1 (soluble triggering receptor expressed on myeloid cells 1), WHO (World Health Organization).
3.4 Biomarkers as indicators of 28-day and 90-day mortality
HDSS data provided long-term vital status information for 65.5% (309/472) of pneumonia cases. There were 22 deaths within 28 days of hospital admission, and 28 deaths within 90 days. These correspond to the 15 children dying during the initial hospital admission and 7 or 13 children dying after hospital discharge. Deaths happening after hospital discharge were from children either discharged alive (10/13) or transferred to another hospital (3/13). As per HDSS data, 5/13 post-discharge deaths occurred at home, 3/13 in a health facility, and 5/13 in an unspecified location.
IL-8, sFlt-1, sTREM-1 (p<0.001) and sTNFR1 (p = 0.004) were significantly elevated in children who subsequently died at 28 days. IL-8, sFlt-1, and sTREM-1 remained most significantly elevated in those dying at 90 days (p<0.001), as was sTNFR1 (p = 0.011) (S4 Table). ORs are provided in Table 2. IL-8, sFlt-1, and sTREM-1 were the best-performing biomarkers for discriminating deaths at 28 or 90 days, with AUROCs ranging from 0.742 (95% CI: 0.645 to 0.839) to 0.812 (95% CI: 0.716 to 0.908) (Fig 2C, S3 Table). Best two-biomarker combination for 28-day mortality was IL-8 and sFlt-1 (AUROC: 0.847 [95% CI: 0.770 to 0.924]), while for 90-day mortality was IL-8 and sTREM-1 (AUROC: 0.828 [95% CI: 0.742 to 0.914]). However, two-biomarker models did not outperform models with only IL-8 (p≥0.05).
The sensitivity analysis evaluating secondary outcomes only for children not dying during initial hospital admission showed similar results. Best AUROCs for 28-day mortality also were from sFlt-1 (0.744 [95% CI: 0.523 to 0.965]), sTREM-1 (0.686 [95% CI: 0.447 to 0.925]), and IL-8 (0.659 [95% CI: 0.473 to 0.845]); and for 90-day mortality the best ones were from IL-8 (0.700 [95% CI: 0.584 to 0.816]), sTREM-1 (0.672 [95% CI: 0.516 to 0.833]), and sFlt-1 (0.628 [95% CI: 0.478 to 0.777]) (S5 Table).
3.5 Biomarkers compared to clinical parameters to predict in-hospital death
Of the four clinical variables selected a priori, the RISC-Malawi score (OR: 1.38 [95% CI: 1.18 to 1.61], p<0.001) and SpO2 (OR: 1.19 [95% CI: 1.02 to 1.41] per unit decreased, p = 0.049) were associated with in-hospital mortality in univariable analysis (S6 Table). IL-8, sFlt-1, and sTREM-1 ORs for in-hospital mortality remained similar when adjusting for one of the pre-specified clinical variables at a time.
IL-8 performed significantly better in discriminating in-hospital mortality compared to the WHO severity definition, LODS, and SpO2 (p<0.001) (Fig 2D). sFlt-1 and sTREM-1 also performed significantly better than these three clinical parameters (p<0.009). IL-8, sFlt-1, and sTREM-1 were better than the RISC-Malawi score but differences were non-significant (p = 0.087, p = 0.466, and p = 0.429, respectively). When adding IL-8, sFlt-1, or sTREM-1 to a model with one of these clinical parameters, AUROCs improved on all occasions (non-significantly for the addition of sTREM-1 to RISC-Malawi score) (Table 3). Of note, models with one of these three biomarkers alone performed as well as models combining the same biomarker with SpO2 or a severity score/definition (p>0.05).
3.6 Using biomarkers for risk-stratification
Cut-offs for in-hospital mortality maximizing sensitivity and specificity (Youden’s index method) are reported in S7 Table. IL-8 was the biomarker with a cut-off associated with best performance estimators, as a 59.6 pg/mL threshold gave a sensitivity of 80.0 (95% CI: 51.9 to 95.7) and a specificity of 86.8 (95% CI: 83.2 to 89.9).
We also used a pre-specified LR+ of 10 and a LR- of 0.1 to define cut-offs and place children with pneumonia into low, intermediate and high-risk groups. We could only obtain these cut-offs for IL-8, sFlt-1, and sTREM-1, as other biomarkers were not good enough. Table 4 shows these values and their corresponding performance estimators. When we applied the cut-offs of <239 pg/mL and ≥629 pg/mL for sTREM-1 based on Leligdowicz et al. [13], this resulted in 1.0% (3/290) in-hospital deaths occurring among children placed in the low-risk group, 6.4% (8/125) among those in the intermediate-risk group, and 23.5% (4/17) among those in the high-risk group (Table 4).
4. Discussion
In this cohort of hospitalized Mozambican children fulfilling WHO diagnostic criteria for pneumonia, IL-8, sFlt-1, and sTREM-1 concentration determined at hospital admission showed good individual prognostic accuracy for in-hospital and longer-term mortality. They outperformed other commonly used clinical parameters and were unaltered in healthy community controls.
IL-8 had the highest discriminatory capacity for severity and in-hospital deaths. IL-8 is a chemokine produced by different blood cells upon stimulation with inflammatory stimuli, which attracts and activates neutrophils and other inflammatory cells. sFlt-1 and sTREM-1 were the next best prognostic biomarkers for in-hospital mortality. sFlt-1 regulates angiogenesis and inflammation-induced vascular permeability, and sTREM-1 is thought to contribute to immunosuppression via inhibition and apoptosis of neutrophils and monocytes. In accordance with our results, a previous study in 2,502 hospitalized Ugandan febrile children reported sTREM-1, followed by sFlt-1 and IL-8, as best prognostic biomarkers for 7-day in-hospital mortality [13]. When limiting the cohort to those children who met the IMCI diagnostic criteria for pneumonia, these three biomarkers remained the best ones for 48-hour in-hospital mortality [21]. Another study in Bhutanese children hospitalized with clinical pneumonia showed that sTREM-1 was the best prognostic biomarker for a composite outcome of poor prognosis [41]. In a study in India, IL-8 levels were also higher in fatal cases [42]. Moreover, IL-8 and sTREM-1 have been reported to be increased in severe childhood pneumonia cases [11, 43].
In this study, CRP and PCT were significantly elevated in pneumonia cases compared to controls, but they were neither good severity indicators nor prognosticators of death. A systematic review and meta-analysis found them to be significantly increased in children with severe pneumonia compared to those with non-severe pneumonia [11]. However, almost all included studies were conducted in high-income countries, and PCT and CRP have been suggested to be influenced by HIV, malaria, and malnutrition [14, 44, 45]. Other reviews reported limited or conflicting results [35, 46]. Studies conducted in sub-Saharan Africa have shown that they are poor indicators of childhood pneumonia severity and outcome [15–21], including a previous study in Manhiça [14].
Studies in children have found that biomarkers of immune and endothelial activation are better prognostic indicators than common clinical markers of pneumonia disease severity [21, 41]. Hypoxemia is considered a hallmark of pneumonia severity [35], but SpO2 had poor discriminating performance for in-hospital mortality in our study cohort (with an exclusion criterion of SpO2 ≤85%). The severity dichotomization based on the WHO 2013 pocket book [3] did not have good prognostic accuracy either. In fact, we did not find clear biomarker differences between these severity groups. Nonetheless, it is important to note that we applied this definition to children older than five years old. RISC-Malawi [7, 36] and LODS [37–39] were selected as relevant scores developed with data from African children to predict in-hospital mortality. The RISC-Malawi score demonstrated the best discrimination for in-hospital deaths among pre-specific clinical parameters and was the only one not significantly different from best-performing biomarkers. Remarkably, the addition of IL-8, sFlt-1, or sTREM-1 to a model with one of these clinical parameters improved AUROCs. Hyperlactatemia has been identified as a promising prognostic indicator in children with pneumonia [47, 48], although inferior to sTREM-1 [21]. Unfortunately, we lacked lactate data, and further studies are needed to evaluate its prognostic accuracy in comparison to the studied biomarkers.
Children at high risk of adverse outcomes could be prioritized for resources and higher level of care as they are more likely to benefit, while those at low risk could be confidently discharged and iatrogenic harm prevented [12, 49]. Biomarker cut-offs were derived and tested to exemplify biomarker potential, but a larger sample size is needed to provide definitive values. Moreover, cut-offs would need to be adapted to the specific use-case and consider both benefits and costs based on decision theory. Interestingly, when calculating different biomarker cut-offs, optimal cut-offs of sTREM-1 based on pre-specified likelihood ratios were similar to those derived by Leligdowicz et al. in a cohort of hospitalized febrile children [13]. The potential of these biomarkers is not limited to pediatric pneumonia, but could be used in a myriad of situations where children are at risk of severe and fatal infections. In fact, in this study we included children with WHO-defined clinical pneumonia, a heterogeneous syndromic group in which we detected diverse pathogens (bacteria, viruses, and malaria parasites). Our study population does not include SARS-CoV-2 infections, but recent studies have also highlighted the utility of IL-8, sFlt-1, and sTREM-1 to risk-stratify adult patients with COVID-19 pneumonia [50–53].
IL-8, sFlt-1, and sTREM-1 had also the best predicting value for 28-day and 90-day mortality after initial hospital admission. When limiting analyses to deaths occurring after hospital discharge (sensitivity analysis), we obtained the same best-performing biomarkers. As in similar settings, the burden of post-discharge mortality in Manhiça is not trivial [54, 55]. Children discharged after treatment for severe pneumonia are at increased risk of relapse, morbidities and mortality [56, 57], but adequate follow-up of all children is often unfeasible and unaffordable in low-income settings. Host biomarkers could improve discharge follow-up by enabling the early identification of children at higher risk of post-discharge complications, therefore improving survival after hospitalization [12, 49].
Measuring a panel of biomarkers from different pathophysiological pathways could improve outcome prognostication compared to measuring individual biomarkers [58]. However, two-biomarker combinational approaches did not importantly increase mortality discrimination compared to a more parsimonious approach with only IL-8. Although the statistical power of this analysis is limited due to the low numbers of events in this cohort, these results are in agreement with previous studies evaluating similar biomarkers in children with pneumonia or febrile adult populations, where one biomarker provided comparable mortality discrimination to few biomarkers combined [21, 59, 60].
This study benefits from a prospective design and exhaustive clinical data collection. Major strengths were the addition of a healthy community control group and having post-discharge mortality data. Nevertheless, the number of deaths limited our statistical power. For this reason, we constructed models with only two independent variables, pre-specified few relevant clinical parameters for comparison with biomarkers, did not attempt to derive a full risk prediction model nor validate one, and did not explore if prognostic performance interacted with other factors (e.g., HIV status, malnutrition, age, or pathogens detected). This limitation, common in other studies, highlights the need for individual patient meta-analyses. Besides, HDSS mortality data were not specifically collected for the purpose of this study and this could have impacted the accuracy of the dates of post-discharge deaths.
In this study, we provide evidence that IL-8, sFlt-1, and sTREM-1 are promising candidates to be used in the risk-stratification of childhood pneumonia. Nonetheless, further studies are necessary to establish with assurance which biomarkers and cut-offs are most useful in different use-cases and in combination with clinical parameters. Another important step is the incorporation of these promising biomarkers into rapid, easy-to-use, and inexpensive point-of-care triage tests to be used in low-resource settings. Their hypothetical implementation into clinical practice could transform risk-stratification and contribute in efforts to reduce pneumonia mortality among the most vulnerable children.
Supporting information
S1 Table. Luminex and ELISA biomarker dynamic ranges and distribution of values outside the dynamic range.
https://doi.org/10.1371/journal.pgph.0001553.s001
(DOCX)
S2 Table. Biomarker concentrations in pneumonia cases by severity group.
https://doi.org/10.1371/journal.pgph.0001553.s002
(DOCX)
S3 Table. Biomarker AUROCs for primary and secondary study outcomes.
https://doi.org/10.1371/journal.pgph.0001553.s003
(DOCX)
S4 Table. Biomarker concentrations in 28-day or 90-day deaths and survivors.
https://doi.org/10.1371/journal.pgph.0001553.s004
(DOCX)
S5 Table. Sensitivity analysis: Biomarker AUROCs for 28-day or 90-day deaths occurring only at post-discharge.
https://doi.org/10.1371/journal.pgph.0001553.s005
(DOCX)
S6 Table. Demographic and clinical characteristics associated with in-hospital mortality in pneumonia cases.
https://doi.org/10.1371/journal.pgph.0001553.s006
(DOCX)
S7 Table. Biomarker performance estimators to identify in-hospital mortality using cut-offs based on the Younden index.
https://doi.org/10.1371/journal.pgph.0001553.s007
(DOCX)
S1 Fig. Biomarker Spearman’s correlation coefficients in pneumonia cases.
https://doi.org/10.1371/journal.pgph.0001553.s008
(DOCX)
Acknowledgments
We would like to express our gratitude to participant children and their families. We also thank Dyann Wirth and the HSPH and Broad Institute teams which supported this study, in addition to everyone else who contributed directly or indirectly through clinical, laboratory, or data entry support.
References
- 1. Perin J, Mulick A, Yeung D, Villavicencio F, Lopez G, Strong KL, et al. Global, regional, and national causes of under-5 mortality in 2000–19: an updated systematic analysis with implications for the Sustainable Development Goals. Lancet Child Adolesc Health. 2022;6(2):106–15. pmid:34800370
- 2.
World Health Organization. Integrated Management of Childhood Illness. Chart Booklet. Geneva: 2014. Available from: https://www.who.int/publications/m/item/integrated-management-of-childhood-illness—chart-booklet-(march-2014)
- 3.
World Health Organization. Pocket book of hospital care for children. Guidelines for the management of common illnesses with limited resources. Second edition. Geneva: 2013. Available from: https://www.who.int/publications/i/item/978-92-4-154837-3n
- 4.
World Health Organization & United Nations Children’s Fund. Caring for the sick child in the community: a training course for community health workers, adaptation for high HIV or TB settings. Chart booklet. Geneva: 2020. Available from: https://www.who.int/publications/i/item/9789240017368
- 5. Deardorff KV, McCollum ED, Ginsburg AS. Pneumonia risk stratification scores for children in low-resource settings: a systematic literature review. Pediatr Infect Dis J. 2018;37(8):743–8. pmid:29278608
- 6. Ogero M, Sarguta RJ, Malla L, Aluvaala J, Agweyu A, English M, et al. Prognostic models for predicting in-hospital paediatric mortality in resource-limited countries: a systematic review. BMJ Open. 2020;10(10):e035045. pmid:33077558
- 7. Rees CA, Hooli S, King C, McCollum ED, Colbourn T, Lufesi N, et al. External validation of the RISC, RISC-Malawi, and PERCH clinical prediction rules to identify risk of death in children hospitalized with pneumonia. J Glob Health. 2021;11:04062. pmid:34737862
- 8. Rees CA, Colbourn T, Hooli S, King C, Lufesi N, McCollum ED, et al. Derivation and validation of a novel risk assessment tool to identify children aged 2–59 months at risk of hospitalised pneumonia-related mortality in 20 countries. BMJ Glob Health. 2022;7(4):e008143. pmid:35428680
- 9. Bordon J, Aliberti S, Fernandez-Botran R, Uriarte SM, Rane MJ, Duvvuri P, et al. Understanding the roles of cytokines and neutrophil activity and neutrophil apoptosis in the protective versus deleterious inflammatory response in pneumonia. Int J Infect Dis. 2013;17(2):e76–83. pmid:23069683
- 10. Page AV, Liles WC. Biomarkers of endothelial activation/dysfunction in infectious diseases. Virulence. 2013;4(6):507–16. pmid:23669075
- 11. Fernandes CD, Arriaga MB, Costa MCM, Costa MCM, Costa MHM, Vinhaes CL, et al. Host inflammatory biomarkers of disease severity in pediatric community-acquired pneumonia: a systematic review and meta-analysis. Open Forum Infect Dis. 2019;6(12):ofz520. pmid:31867405
- 12. Balanza N, Erice C, Ngai M, Varo R, Kain KC, Bassat Q. Host-based prognostic biomarkers to improve risk stratification and outcome of febrile children in low- and middle-income countries. Front Pediatr. 2020;8:552083. pmid:33072673
- 13. Leligdowicz A, Conroy AL, Hawkes M, Richard-Greenblatt M, Zhong K, Opoka RO, et al. Risk-stratification of febrile African children at risk of sepsis using sTREM-1 as basis for a rapid triage test. Nat Commun. 2021;12(1):6832. pmid:34824252
- 14. Díez-Padrisa N, Bassat Q, Machevo S, Quintó L, Morais L, Nhampossa T, et al. Procalcitonin and C-Reactive protein for invasive bacterial pneumonia diagnosis among children in Mozambique, a malaria-endemic area. PLoS One. 2010;5(10):e13226. pmid:20976241
- 15. Reed C, Madhi SA, Klugman KP, Kuwanda L, Ortiz JR, Finelli L et al. Development of the Respiratory Index of Severity in Children (RISC) score among young children with respiratory infections in South Africa. PLoS One. 2012;7(1): e27793. pmid:22238570
- 16. Huang H, Ideh RC, Gitau E, Thézénas ML, Jallow M, Ebruke B, et al. Discovery and validation of biomarkers to guide clinical management of pneumonia in African children. Clin Infect Dis. 2014;58(12):1707–15. pmid:24696240
- 17. Sawatzky J, Soo J, Conroy AL, Bhargava R, Namasopo S, Opoka RO et al. Biomarkers of systemic inflammation in Ugandan infants and children hospitalized with respiratory syncytial virus infection. Pediatr Infect Dis J 2019; 38: 854–9. pmid:31306398
- 18. Zhang R, Rai U, Ibrahim NBM, Amazouz Y, Soo J, Conroy AL, et al. Dysregulation of angiopoietin-Tie-2 axis in Ugandan children hospitalized with pneumonia. Cytokine. 2020;133:155175. pmid:32599540
- 19. Ahmed U, Soo J, Conroy AL, Namasopo S, Opoka RO, Bhargava R et al. Soluble T cell immunoglobulin and mucin-domain containing protein 3 in children hospitalized with pneumonia in resource-limited settings. Cytokine 2022; 151: 155794. pmid:35030468
- 20. Konrad ER, Soo J, Conroy AL, Namasopo S, Opoka RO, Hawkes MT. Interleukin-18 binding protein in infants and children hospitalized with pneumonia in low-resource settings. Cytokine 2022; 150: 155775. pmid:34875584
- 21. McDonald CR, Leligdowicz A, Conroy AL, Weckman AM, Richard-Greenblatt M, Ngai M, et al. Immune and endothelial activation markers and risk stratification of childhood pneumonia in Uganda: a secondary analysis of a prospective cohort study. PLoS Med. 2022;19(7):e1004057. pmid:35830474
- 22. Sigaúque B, Roca A, Bassat Q, Morais L, Quintó L, Berenguera A, et al. Severe pneumonia in Mozambican young children: clinical and radiological characteristics and risk factors. J Trop Pediatr. 2009;55(6):379–87. pmid:19401405
- 23. Roca A, Quintó L, Saúte F, Thompson R, Aponte JJ, Alonso PL. Community incidences of respiratory infections in an actively followed cohort of children <1 year of age in Manhiça, a rural area of southern Mozambique. Trop Med Int Health. 2006;11(3):373–80. pmid:16553918
- 24. Sacarlal J, Nhacolo AQ, Sigaque B, Nhalungo DA, Abacassamo F, Sacoor CN, et al. A 10 year study of the cause of death in children under 15 years in Manhiça, Mozambique. BMC Public Health. 2009;9:67. pmid:19236726
- 25. Acácio S, Verani JR, Lanaspa M, Fairlie TA, Nhampossa T, Ruperez M, et al. Under treatment of pneumonia among children under 5 years of age in a malaria-endemic area: population-based surveillance study conducted in Manhica district- rural, Mozambique. Int J Infect Dis. 2019;36:39–45. pmid:25980619
- 26. Sacoor C, Nhacolo A, Nhalungo D, Aponte JJ, Bassat Q, Augusto O, et al. Profile: Manhiça health research centre (Manhiça HDSS). Int J Epidemiol. 2013;42(5):1309–18. pmid:24159076
- 27. Nhacolo A, Jamisse E, Augusto O, Matsena T, Hunguana A, Mandomando I, et al. Cohort profile update: Manhiça health and demographic surveillance system (HDSS) of the Manhiça health research centre (CISM). Int J Epidemiol. 2021;50(2):395. pmid:33452521
- 28. Valim C, Ahmad R, Lanaspa M, Tan Y, Acácio S, Gillette MA, et al. Responses to bacteria, virus, and malaria distinguish the etiology of pediatric clinical pneumonia. Am J Respir Crit Care Med. 2016;193(4):448–59. pmid:26469764
- 29. Silterra J, Gillette MA, Lanaspa M, Pellé KG, Valim C, Ahmad R, et al. Transcriptional categorization of the etiology of pneumonia syndrome in pediatric patients in malaria-endemic areas. J Infect Dis. 2017;215(2):312–20. pmid:27837008
- 30. Gillette MA, Mani DR, Uschnig C, Pellé KG, Madrid L, Acácio S, et al. Biomarkers to distinguish bacterial from viral pediatric clinical pneumonia in a malaria-endemic setting. Clin Infect Dis. 2021;73(11):e3939–e3948. pmid:33534888
- 31.
World Health Organization. Revised WHO classification and treatment of childhood pneumonia at health facilities. Evidence summaries. Geneva: 2014. Available from: https://www.who.int/publications/i/item/9789241507813
- 32. Cherian T, Mulholland EK, Carlin JB, Ostensen H, Amin R, de Campo M, et al. Standardized interpretation of paediatric chest radiographs for the diagnosis of pneumonia in epidemiological studies. Bull World Health Organ. 2005;83(5):353–9. pmid:15976876
- 33. Leligdowicz A, Conroy AL, Hawkes M, Zhong K, Lebovic G, Matthay MA, et al. Validation of two multiplex platforms to quantify circulating markers of inflammation and endothelial injury in severe infection. PLoS One. 2017;12(4):e0175130. pmid:28419100
- 34. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837–45. pmid:3203132
- 35. Dean P, Florin TA. Factors associated with pneumonia severity in children: a systematic review. J Pediatric Infect Dis Soc. 2018;7(4):323–34. pmid:29850828
- 36. Hooli S, Colbourn T, Lufesi N, Costello A, Nambiar B, Thammasitboon S, et al. Predicting hospitalised paediatric pneumonia mortality risk: an external validation of RISC and mRISC, and local tool development (RISC-Malawi) from Malawi. PLoS One. 2016;11(12):e0168126. pmid:28030608
- 37. Helbok R, Kendjo E, Issifou S, Lackner P, Newton CR, Kombila M, et al. The Lambaréné Organ Dysfunction Score (LODS) is a simple clinical predictor of fatal malaria in African children. J Infect Dis. 2009;200(12):1834–41. pmid:19911989
- 38. Conroy AL, Hawkes M, Hayford K, Namasopo S, Opoka RO, John CC, et al. Prospective validation of pediatric disease severity scores to predict mortality in Ugandan children presenting with malaria and non-malaria febrile illness. Crit Care. 2015;19(1):47. pmid:25879892
- 39. George EC, Walker AS, Kiguli S, Olupot-Olupot P, Opoka RO, Engoru C, et al. Predicting mortality in sick African children: the FEAST Paediatric Emergency Triage (PET) Score. BMC Med. 2015;13(1):174. pmid:26228245
- 40. Vidmar SI, Cole TJ, Pan H. Standardizing anthropometric measures in children and adolescents with functions for egen: Update. Stata J 2013;13(2):366–78.
- 41. Jullien S, Richard-Greenblatt M, Ngai M, Lhadon T, Sharma R, Dema K, et al. Performance of host-response biomarkers to risk-stratify children with pneumonia in Bhutan. J Infect. 2022;S0163–4453(22)00603-X. pmid:36243198
- 42. Saghafian-Hedengren S, Mathew JL, Hagel E, Singhi S, Ray P, Ygberg S, et al. Assessment of cytokine and chemokine signatures as potential biomarkers of childhood community-acquired pneumonia severity: a nested cohort study in India. Pediatr Infect Dis J. 2017;36(1):102–8. pmid:27956727
- 43. Esposito S, Di Gangi M, Cardinale F, Baraldi E, Corsini I, Da Dalt L, et al. Sensitivity and specificity of soluble triggering receptor expressed on myeloid cells-1, midregional proatrial natriuretic peptide and midregional proadrenomedullin for distinguishing etiology and to assess severity in community-acquired pneumonia. PLoS One. 2016;11(11):e0163262. pmid:27846213
- 44. Carrol ED, Mankhambo LA, Jeffers G, Parker D, Guiver M, Newland P, et al. The diagnostic and prognostic accuracy of five markers of serious bacterial infection in Malawian children with signs of severe infection. PLoS One. 2009;4(8):e6621. pmid:19675669
- 45. Page A-L, de Rekeneire N, Sayadi S, Aberrane S, Janssens A-C, Dehoux M, et al. Diagnostic and prognostic value of procalcitonin and C-reactive protein in malnourished children. Pediatrics. 2014;133(2):e363–70. pmid:24446443
- 46. Bivona G, Angello L, Scazzone C, Lo Sasso B, Bellia C, Ciaccio M. Procalcitonin and community-acquired pneumonia (CAP) in children. Clin Chim Acta 2015; 451(Pt B): 215–8. pmid:26434548
- 47. Ramakrishna B, Graham SM, Phiri A, Mankhambo L, Duke T. Lactate as a predictor of mortality in Malawian children with WHO-defined pneumonia. Arch Dis Child. 2012;97(4):336–42. pmid:22267369
- 48. Ma C, Gunaratnam LC, Ericson A, Conroy AL, Namasopo S, Opoka RO, et al. Handheld point-of-care lactate measurement at admission predicts mortality in Ugandan children hospitalized with pneumonia: a prospective cohort study. Am J Trop Med Hyg. 2019;100(1):37–42. pmid:30398141
- 49. Chandna A, Osborn J, Bassat Q, Bell D, Burza S, D’Acremont V, et al. Anticipating the future: prognostic tools as a complementary strategy to improve care for patients with febrile illnesses in resource-limited settings. BMJ Glob Health. 2021;6(7):e006057. pmid:34330761
- 50. Van Singer M, Brahier T, Ngai M, Wright J, Weckman AM, Erice C, et al. COVID-19 risk stratification algorithms based on sTREM-1 and IL-6 in emergency department. J Allergy Clin Immunol. 2021;147(1):99–106.e4. pmid:33045281
- 51. da Silva-Neto PV, de Carvalho JCS, Pimentel VE, Pérez MM, Toro DM, Fraga-Silva TFC, et al. sTREM-1 predicts disease severity and mortality in COVID-19 patients: involvement of peripheral blood leukocytes and MMP-8 activity. Viruses. 2021;13(12):2521. pmid:34960790
- 52. Dupont V, Kanagaratnam L, Goury A, Poitevin G, Bard M, Julien G, et al. Excess soluble fms-like tyrosine kinase 1 correlates with endothelial dysfunction and organ failure in critically ill coronavirus disease 2019 patients. Clin Infect Dis. 2021;72(10):1834–7. pmid:32672790
- 53. Cesta MC, Zippoli M, Marsiglia C, Gavioli EM, Mantelli F, Allegretti M, et al. The role of interleukin-8 in lung inflammation and injury: implications for the management of COVID-19 and hyperinflammatory acute respiratory distress syndrome. Front Pharmacol. 2022;12:808797. pmid:35095519
- 54. Nemetchek B, English L, Kissoon N, Ansermino JM, Moschovis PP, Kabakyenga J, et al. Paediatric postdischarge mortality in developing countries: a systematic review. BMJ Open. 2018;8(12):e023445. pmid:30593550
- 55. Madrid L, Casellas A, Sacoor C, Quintó L, Sitoe A, Varo R, et al. Postdischarge mortality prediction in sub-Saharan Africa. Pediatrics. 2019;143(1):e20180606. pmid:30552144
- 56. Ashraf H, Alam NH, Chisti MJ, Salam MA, Ahmed T, Gyr N. Observational follow-up study following two cohorts of children with severe pneumonia after discharge from day care clinic/hospital in Dhaka, Bangladesh. BMJ Open. 2012;2(4):e000961. pmid:22842561
- 57. Ngari MM, Fegan G, Mwangome MK, Ngama MJ, Mturi N, Scott JAG, et al. Mortality after inpatient treatment for severe pneumonia in children: a cohort study. Paediatr Perinat Epidemiol. 2017;31(3):233–42. pmid:28317139
- 58. Erdman LK, Dhabangi A, Musoke C, Conroy AL, Hawkes M, Higgins S, et al. Combinations of host biomarkers predict mortality among Ugandan children with severe malaria: a retrospective case-control study. PLoS One. 2011;6(2):e17440. pmid:21364762
- 59. Richard-Greenblatt M, Boillat-Blanco N, Zhong K, Mbarack Z, Samaka J, Mlaganile T, et al. Prognostic accuracy of soluble triggering receptor expressed on myeloid cells (sTREM-1)-based algorithms in febrile adults presenting to Tanzanian outpatient clinics. Clin Infect Dis. 2020;70(7):1304–12. pmid:31102510
- 60. Wright SW, Lovelace-Macon L, Hantrakun V, Rudd KE, Teparrukkul P, Kosamo S, et al. sTREM-1 predicts mortality in hospitalized patients with infection in a tropical, middle-income country. BMC Med. 2020;18(1):159. pmid:32605575