Advertisement

Performance of host-response biomarkers to risk-stratify children with pneumonia in Bhutan

  • Sophie Jullien
    Correspondence
    Corresponding author at: Carrer Rosselló 132, 08036 Barcelona, Spain.
    Affiliations
    Institut de Salut Global de Barcelona (ISGlobal), Universitat de Barcelona (UB), Barcelona, Spain

    Departament de Medicina, Facultat de Medicina i Ciències de la Salut, Universitat de Barcelona (UB), Barcelona, Spain

    Jigme Dorji Wangchuck National Referral Hospital, Thimphu, Bhutan
    Search for articles by this author
  • Melissa Richard-Greenblatt
    Affiliations
    Hospital of the University of Pennsylvania, Philadelphia, PA, USA

    Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, USA

    University of Pennsylvania, Philadelphia, Pennsylvania, USA
    Search for articles by this author
  • Michelle Ngai
    Affiliations
    Sandra-Rotman Centre for Global Health, Toronto General Hospital Research Institute, University Health Network-Toronto General Hospital, Toronto, Ontario, Canada
    Search for articles by this author
  • Tenzin Lhadon
    Affiliations
    Jigme Dorji Wangchuck National Referral Hospital, Thimphu, Bhutan

    Khesar Gyalpo University of Medical Sciences of Bhutan (KGUMSB), Thimphu, Bhutan
    Search for articles by this author
  • Ragunath Sharma
    Affiliations
    Jigme Dorji Wangchuck National Referral Hospital, Thimphu, Bhutan
    Search for articles by this author
  • Kumbu Dema
    Affiliations
    Jigme Dorji Wangchuck National Referral Hospital, Thimphu, Bhutan
    Search for articles by this author
  • Author Footnotes
    1 These authors contributed equally to this work and should share senior authorship.
    Kevin C. Kain
    Footnotes
    1 These authors contributed equally to this work and should share senior authorship.
    Affiliations
    Sandra-Rotman Centre for Global Health, Toronto General Hospital Research Institute, University Health Network-Toronto General Hospital, Toronto, Ontario, Canada

    Tropical Disease Unit, Division of Infectious Diseases, Department of Medicine, University of Toronto, Toronto, Ontario, Canada
    Search for articles by this author
  • Author Footnotes
    1 These authors contributed equally to this work and should share senior authorship.
    Quique Bassat
    Footnotes
    1 These authors contributed equally to this work and should share senior authorship.
    Affiliations
    Institut de Salut Global de Barcelona (ISGlobal), Universitat de Barcelona (UB), Barcelona, Spain

    ICREA, Pg. Lluís Companys 23, 08010 Barcelona, Spain

    Centro de Investigação em Saúde de Manhiça (CISM), Maputo, Mozambique

    Pediatric Infectious Diseases Unit, Pediatrics Department, Hospital Sant Joan de Déu (University of Barcelona), Barcelona, Spain

    Consorcio de Investigación Biomédica en Red de Epidemiología y Salud Pública (CIBERESP), Madrid, Spain
    Search for articles by this author
  • Author Footnotes
    1 These authors contributed equally to this work and should share senior authorship.
Open AccessPublished:October 12, 2022DOI:https://doi.org/10.1016/j.jinf.2022.10.010

      Highlights

      • Inflammatory and endothelial activation markers, whose plasma concentrations can be readily measured upon first encounter, can predict poor prognosis in children with pneumonia.
      • sTREM-1 performed significantly better than commonly used acute phase markers (e.g., CRP, PCT).
      • The addition of sTREM-1 significantly improved the predictive accuracy of lower chest wall indrawing as a clinical sign.

      Summary

      Pneumonia is the leading cause of post-neonatal death amongst children under five years of age; however, there is no simple triage tool to identify children at risk of progressing to severe and fatal disease. Such a tool could assist for early referral and prioritization of care to improve outcomes and enhance allocation of scarce resources. We compared the performance of inflammatory and endothelial activation markers in addition to clinical signs or scoring scales to risk-stratify children hospitalized with pneumonia at the national referral hospital of Bhutan with the goal of predicting clinical outcome. Of 118 children, 31 evolved to a poor prognosis, defined as either mortality, admission in the paediatric intensive care unit, requirement of chest drainage or requirement of more than five days of oxygen therapy. Soluble triggering receptor expressed on myeloid cells 1 (sTREM-1) was the best performing biomarker and performed better than clinical parameters. sTREM-1 levels upon admission had good predictive accuracy to identify children with pneumonia at risk of poor prognosis. Our findings confirm that immune and endothelial activation markers could be proactively used at first encounter as risk-stratification and clinical decision-making tools in children with pneumonia; however, further external validation is needed.

      Keywords

      Background

      Pneumonia is the leading infectious cause of preventable deaths amongst children under five years of age,

      World Health Organization. Pneumonia. Fact sheets. (2019). Available at: https://www.who.int/news-room/fact-sheets/detail/pneumonia. (Accessed: 27th January 2020)

      causing an estimated 740,000 deaths annually, or 13.9% of all global deaths in this age group.
      • Perin J.
      • 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.
      Every year, up to 226 million children in this age group are diagnosed with pneumonia.
      • McAllister D.A.
      • et al.
      Global, regional, and national estimates of pneumonia morbidity and mortality in children younger than 5 years between 2000 and 2015: a systematic analysis.
      While most children will have self-limited disease, a small proportion of them will progress to severe disease and fatal outcome.
      • Ginsburg A.S.
      • et al.
      Placebo vs amoxicillin for nonsevere fast-breathing pneumonia in Malawian children aged 2 to 59 months. A double-blind, randomized clinical noninferiority trial.
      Early recognition of children with severe pneumonia enables more aggressive referral and treatment, leading to reduced mortality.
      • Sibila O.
      • Restrepo M.I.
      Biomarkers in community-acquired pneumonia: still searching for the one.
      Thus, there is a need for early identification of children at risk of progressing to severe disease, particularly at the moment of first contact with the healthcare system. At a primary health care level, a simple triage tool that would discriminate children at risk of severe pneumonia from those with self-limited disease could assist decision-making for early referral to a higher healthcare level, particularly in resource-limited settings. In addition, the identification of high-risk children will aid prioritization of care in busy healthcare centres and guide rational allocation of scarce resources.
      Clinical signs and simple laboratory testing have been combined to generate clinical severity scores to improve early detection of children with fever or respiratory symptoms at risk of poor outcomes.
      • Hooli S.
      • et al.
      Predicting hospitalised paediatric pneumonia mortality risk: an external validation of RISC and mRISC, and local tool development (RISC-Malawi) from Malawi.
      • Conroy A.L.
      • et al.
      Prospective validation of pediatric disease severity scores to predict mortality in Ugandan children presenting with malaria and non-malaria febrile illness.
      • Deardorff K.V.
      • Mccollum E.D.
      • Ginsburg A.S
      Pneumonia risk stratification scores for children in low-resource settings. A systematic literature review.
      However, most of these severity scores involve the measurement of vital signs (e.g., temperature, respiratory rate, or blood pressure), the assessment of clinical signs (e.g., recognising chest indrawing), or the interpretation of laboratory parameters that require trained healthcare workers. Furthermore, risk scores are validated and routinely used in adults with pneumonia, but none have been widely implemented for childhood pneumonia.
      • Deardorff K.V.
      • Mccollum E.D.
      • Ginsburg A.S
      Pneumonia risk stratification scores for children in low-resource settings. A systematic literature review.
      Therefore, the unresolved need for a simple severity assessment for children with respiratory symptoms may require an innovative approach to currently proposed clinical strategies.
      • Waterer G.
      Severity scores and community-acquired pneumonia. Time to move forward.
      Specific markers of host response including those associated with immune and endothelial activation, have been previously implicated in the pathogenesis of severe infections, irrespective of their underlying aetiology (“pathogen-agnostic”).
      • Sibila O.
      • Restrepo M.I.
      Biomarkers in community-acquired pneumonia: still searching for the one.
      ,
      • Thomas J.
      • Pociute A.
      • Kevalas R.
      • Malinauskas M.
      • Jankauskaite L.
      Blood biomarkers differentiating viral versus bacterial pneumonia aetiology: a literature review.
      ,
      • Leligdowicz A.
      • et al.
      Risk-stratification of febrile African children at risk of sepsis using sTREM-1 as basis for a rapid triage test.
      The quantification of such biomarkers may enable risk stratification and guide clinical decision-making regarding the need for early triage, referral, hospitalization, and admission to intensive care units.
      • Uwaezuoke S.N.
      • Ayuk A.C.
      Prognostic scores and biomarkers for pediatric community-acquired pneumonia: how far have we come?.
      Quantifying these markers at clinical presentation has been shown to be useful in predicting severity and outcomes in adults and children with life-threatening infections, including pneumonia, severe malaria, COVID-19, haemorrhagic fevers, or sepsis.
      • Leligdowicz A.
      • Richard-Greenblatt M.
      • Wright J.
      • Crowley V.M.
      • Kain K.C.
      Endothelial activation: the Ang/Tie axis in sepsis.
      • Xing K.
      • Murthy S.
      • Liles W.C.
      • Singh J.M
      Clinical utility of biomarkers of endothelial activation in sepsis-a systematic review.
      • Richard-Greenblatt M.
      • et al.
      Prognostic accuracy of soluble triggering receptor expressed on myeloid cells (sTREM-1)-based algorithms in febrile adults presenting to tanzanian outpatient clinics.
      • Wright J.K.
      • et al.
      Biomarkers of endothelial dysfunction predict sepsis mortality in young infants: a matched case-control study.
      • Wright S.W.
      • et al.
      sTREM-1 predicts mortality in hospitalized patients with infection in a tropical, middle-income country.
      • Balanza N.
      • et al.
      Host-based prognostic biomarkers to improve risk stratification and outcome of febrile children in low- and middle-income countries.
      • Van Singer M.
      • et al.
      COVID-19 risk stratification algorithms based on sTREM-1 and IL-6 in emergency department.
      • Conroy A.L.
      • et al.
      Host biomarkers are associated with progression to dengue haemorrhagic fever: a nested case-control study.
      However, they have not been widely evaluated in low- and middle-income countries, and their prognostic utility in childhood pneumonia has not been validated vis-à-vis standard risk-stratification clinical algorithms.
      • Uwaezuoke S.N.
      • Ayuk A.C.
      Prognostic scores and biomarkers for pediatric community-acquired pneumonia: how far have we come?.
      ,
      • Balanza N.
      • et al.
      Host-based prognostic biomarkers to improve risk stratification and outcome of febrile children in low- and middle-income countries.
      The Respiratory Infections in Bhutanese Children (RIBhuC) study recruited Bhutanese children aged 2 to 59 months hospitalized with clinical pneumonia. Here we assessed the performance of inflammatory, immune, and endothelial activation markers alone or in addition to clinical signs or scoring scales to risk-stratify children hospitalized with pneumonia and predict their outcome.

      Methods

      Study design

      The RIBhuC study was prospectively conducted during 12 consecutive months at the Jigme Dorji Wangchuck National Referral Hospital (JDWNRH) in Thimphu, Bhutan.
      • Jullien S.
      • et al.
      Pneumonia in children admitted to the national referral hospital in Bhutan: a prospective cohort study.
      Briefly, the paediatric department at JDWNRH consists of 38 beds, including five beds in the paediatric intensive care unit. All children aged 2 to 59 months admitted at JDWNRH and fulfilling the World Health Organization (WHO) criteria for pneumonia or severe pneumonia were recruited.

      World Health Organization. Revised WHO classification and treatment of childhood pneumonia at health facilities. Evidence summaries. (2014).

      Pneumonia was defined as history of cough or reported breathing difficulty and increased respiratory rate (≥ 50 breaths per minute in children aged 2–11 months or ≥ 40 breaths per minute in children aged 12–59 months) or chest indrawing (subcostal and/or intercostal retractions defined as lower chest wall indrawing and supraclavicular and/or suprasternal retractions defined as very severe chest indrawing). Severe pneumonia was defined as history of cough or reported breathing difficulty, and at least one of the following: oxygen saturation < 90%, central cyanosis, severe respiratory distress, or any danger sign (inability to breastfeed or drink, lethargy or reduced level of consciousness, convulsions). We excluded children when the principal reason for admission was a non-respiratory illness or a condition that was not caused by respiratory illness, those admitted in the previous seven days in order to exclude hospital-acquired infection, and children with evidence of a foreign body in the respiratory tract.
      For all eligible patients whose parents provided written consent for study participation, we collected demographic and clinical data, biological samples, and a chest radiography on admission. Radiographical endpoints were defined as per WHO radiological criteria.
      • Cherian T.
      • et al.
      Standardized interpretation of paediatric chest radiographs for the diagnosis of pneumonia in epidemiological studies.
      The study protocol was approved by the Research Ethics Board of Health, Ministry of Health, in Thimphu, Bhutan (PO/2016/086) and by the research ethics committee from the Hospital Clínic in Barcelona, Spain (HCB/2017/0741). All methods were performed in accordance with the relevant guidelines and regulations.

      Clinical scoring scales and outcomes

      We used three simple clinical scoring scales developed for predicting disease severity and mortality in low-resource settings (Table 1). Clinical parameters were assessed upon admission. The Respiratory Index of Severity in Children (RISC) score was developed amongst children 0–24 months hospitalized with respiratory infections.
      • Reed C.
      • et al.
      Development of the Respiratory Index of Severity in Children (RISC) score among young children with respiratory infections in South Africa.
      The RISC-Malawi is a modified version, which was developed amongst children < 59 months hospitalized with WHO-defined pneumonia.
      • Hooli S.
      • et al.
      Predicting hospitalised paediatric pneumonia mortality risk: an external validation of RISC and mRISC, and local tool development (RISC-Malawi) from Malawi.
      The Lambaréné Organ Dysfunction Score (LODS) was developed amongst children with severe malaria for identifying those needing referral or close monitoring.
      • Helbok R.
      • et al.
      The Lambaréné Organ Dysfunction Score (LODS) is a simple clinical predictor of fatal malaria in African children.
      Although LODS was not specifically developed for pneumonia, it is a promising prognostic tool used in childhood diseases other than malaria.
      • Conroy A.L.
      • et al.
      Prospective validation of pediatric disease severity scores to predict mortality in Ugandan children presenting with malaria and non-malaria febrile illness.
      Table 1Clinical scoring scales.
      RISC
      For non-HIV infected children.
      score
      RISC-Malawi scoreLODS
      Severity of respiratory signs

      SpO2 ≤90%

      OR

      Chest indrawing

      Wheezing

      Refusal to feed

      Growth standards

      WAZ ≤ −3 SD

      −2 ≤ WAZ < −3 SD


      3 points



      2 points

      −2 points

      1 point



      2 points

      1 point


      −2 ≤ WAZ < −3 SD
      Moderate and severe malnutrition were originally assessed with middle-upper arm circumference (MUAC). We substituted these measurements by using WAZ as we did not collect MUAC in our study.


      WAZ ≤ −3 SD
      Moderate and severe malnutrition were originally assessed with middle-upper arm circumference (MUAC). We substituted these measurements by using WAZ as we did not collect MUAC in our study.


      SpO2 90–92%

      SpO2 <90%

      Wheezing

      Unconscious at exam

      Female gender


      3 points

      7 points

      2 points

      7 points

      −2 points

      8 points

      1 point


      Prostration
      Prostration was defined by not being able to breastfeed, sit, stand, or walk, depending on the age of the child.


      Blantyre coma score <3

      Deep breathing
      Deep breathing is also known as Kussmaul's respiration or ‘acidotic’ breathing.


      1 point

      1 point

      1 point

      Abbreviations: LODS: Lambaréné Organ Dysfunction Score; RISC: respiratory index of severity in children; SD: standard deviations; WAZ: weight-for-age Z-score.
      a For non-HIV infected children.
      b Moderate and severe malnutrition were originally assessed with middle-upper arm circumference (MUAC). We substituted these measurements by using WAZ as we did not collect MUAC in our study.
      c Prostration was defined by not being able to breastfeed, sit, stand, or walk, depending on the age of the child.
      d Deep breathing is also known as Kussmaul's respiration or ‘acidotic’ breathing.
      The primary outcome was prognosis, defined as “good” if the child survived, did not require admission in the paediatric intensive care unit, did not require supplemental oxygen or only received oxygen therapy for five days or less, and did not present with pleural effusion that required chest drainage; and “poor” if the child died, required care in the paediatric intensive care unit, received oxygen for more than five days, or presented pleural effusion that requested chest drainage. The usual time of duration of hypoxaemia (oxygen saturation < 90% in room air) is 2 to 5 days, therefore we considered longer duration as poor prognosis.
      • Bradley B.D.
      • Howie S.R.C.
      • Chan T.C.Y.
      • Cheng Y.L.
      Estimating oxygen needs for childhood pneumonia in developing country health systems: a new model for expecting the unexpected.
      ,

      World Health Organization. The clinical use of oxygen in hospitals with limited resources. Guidelines for health-care workers, hospital engineers and programme managers. (2009).

      Clinical decisions such as weaning oxygen and time of discharge were taken by any treating paediatrician working at JDWNRH, unaware of the study outcomes for analysis, and therefore at low risk of introducing performance bias.

      Laboratory testing

      Blood samples were collected from each participant at time of enrolment and were processed following local standard of care.
      • Jullien S.
      • et al.
      Pneumonia in children admitted to the national referral hospital in Bhutan: a prospective cohort study.
      For measurement of immune and endothelial activation markers, blood (2 mL) was collected in EDTA tube and centrifugated (3000 g for three minutes). Plasma was separated and stored at −80 °C until shipment to the University of Toronto, Canada, for analyte testing. Plasma concentration of interleukin-6 (IL-6), interleukin-8 (IL-8), soluble triggering receptor expressed on myeloid cells 1 (sTREM-1), soluble tumour necrosis factor receptor 1 (sTNFR1), angiopoietin-2 (Angpt-2), soluble fms-like tyrosine kinase-1 (sFlt1), and procalcitonin (PCT) were quantified using a multiplex Luminex platform with reagents from R&D Systems (Minneapolis, MN) as described.
      • Leligdowicz A.
      • et al.
      Validation of two multiplex platforms to quantify circulating markers of inflammation and endothelial injury in severe infection.
      C-reactive protein (CRP) was quantified by enzyme-linked immunosorbent assay (R&D DuoSet, Minneapolis, MN). Biomarker concentrations outside of the detection limits were assigned a value of one third below or above the lowest or highest limit in the standard curve, respectively. Erythrocyte sedimentation rate (ESR), and CRP were measured at the study site (JDWNRH). We refer to CRP-study and CRP-ref for differentiating CRP analysed at the study and reference laboratories, respectively. Biomarkers were measured blinded to children clinical characteristics and outcome.

      Data management and statistical analysis

      The statistical associations were assessed using Chi-square, Fisher exact, and Mann-Whitney U tests, as appropriate. Univariable logistic regression models were used to estimate odds ratios of biomarker levels as predictors of prognosis, and multivariable logistic regression models to estimate the degree of association after adjusting for observed confounders. All continuous variables with non-parametric distribution were log transformed for inclusion in logistic regression models. Area under the receiver operating characteristics (AUROC) curve and other performance characteristics (sensitivity, specificity, and likelihood ratios) were calculated to assess the predictive capability, based on each univariable logistic regression model and using cut-off points defined with the Youden's index method (J=max[sensitivity+specificity-1]). AUROCs were compared using the algorithm suggested by DeLong et al. (1988).
      • DeLong E.R.
      • DeLong D.M.
      • Clarke-Pearson D.L.
      Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach.
      Classification and regression tree analyses were performed to create simple algorithms based on risk-stratification. We established the settings of a minimum of 10 cases for parent node and 5 for child node, pruning set with a maximum difference in risk to 0 to prevent over fitting and a maximum level of tree depth of 2.
      • Richard-Greenblatt M.
      • et al.
      Prognostic accuracy of soluble triggering receptor expressed on myeloid cells (sTREM-1)-based algorithms in febrile adults presenting to tanzanian outpatient clinics.
      ,
      • Van Singer M.
      • et al.
      COVID-19 risk stratification algorithms based on sTREM-1 and IL-6 in emergency department.
      We performed subgroup analysis by age groups, as age is a potentially relevant cofounder for clinical signs and biomarker levels. Data analyses and figures were performed with Stata™ v.16.0 (StataCorp, College Station, Texas, USA), SPSS Statistics version 23, and RStudio.

      StataCorp. Stata v.16. (2019).

      ,

      RStudio Team. RStudio: Integrated Development for R. (2020).

      Statistical significance was set at 0.05.

      Results

      Of 189 children with clinical pneumonia recruited to the RIBhuC study, 118 (62.4%) had biomarker quantification and were included in the analysis (Supplementary Fig. S1). Our study did not perform additional blood draws outside of clinical care, and therefore children that did not have blood collected at first presentation did not have biomarker analysis performed. The characteristics of children included and excluded from the analysis are summarized in Supplementary Table S1. Except for hypoxaemia, which was more common amongst children included in the analysis (p = 0.048), there were no significant differences between children included and excluded from the analysis.

      Association of demographic characteristics, clinical signs, and scoring scales with prognosis

      Of the 118 children included, 31 evolved to a poor prognosis. Tables 2 and 3 present demographic, clinical, radiological, and laboratory findings collected upon admission, according to prognosis. One-quarter of children with poor prognosis were referred from another healthcare centre. Parental education, employment, and access to care were not associated with prognosis. Amongst children with a poor prognosis, 39.1% (9/23) presented with a normal chest radiograph, while amongst those with a good prognosis, 21.9% (16/73) presented radiological endpoint pneumonia. A positive (and not considered contaminated) blood culture was not associated with prognosis. Hypoxaemia, prolonged capillary refill time, increased respiratory rate, lower chest wall indrawing, very severe chest indrawing, nasal flaring, grunting, rhonchi, prostration, and decreased level of consciousness at presentation were all associated with poor prognosis. The oxygen saturation upon admission was significantly lower amongst children with poor prognosis. An elevated score in any of the four clinical scoring scales (WHO, RISC, RISC-Malawi, and LODS) was also associated with poor prognosis.
      Table 2Demographic characteristics of participants according to prognosis.
      CharacteristicsGood prognosis (N = 87)Poor prognosis (N = 31)p-value
      Comparison of proportions using the chi-square or fisher tests.
      Infants (<12 months)43 (49.4)19 (61.3)0.256
      Gender, female44 (50.6)10 (32.3)0.079
      Immunization status

      Fully

      Partially

      None


      66/84 (78.6)

      18/84 (21.4)

      0/84 (0)


      22 (71.0)

      9 (29.0)

      0 (0)
      0.393
      Wasting (WAZ ≤ −2SD)
      Nutritional status was based on the WAZ score generated using the 2000 Centers for Disease Control and Prevention Growth Reference48,49.
      6/87 (6.9)5/30 (16.7)0.147
      Known case of HIV infection0 (0)0 (0)NA
      Exposure to tobacco smoke12/84 (14.3)4/30 (13.3)1.000
      Exposure to betel nut (doma)51/84 (60.7)23/30 (76.7)0.116
      Exposure to heater with kerosene7/75 (9.3)1/29 (3.5)0.438
      Parental education

      Both parents are illiterate

      Only one parent has primary education

      Both parents have primary education

      At least one parent has university education


      12/84 (14.3)

      11/84 (13.1)

      39/84 (46.4)

      22/84 (26.2)


      9/30 (30.0)

      4/30 (13.3)

      11/30 (36.7)

      6/30 (20.0)
      0.317
      Both parents unemployed

      Both parents are unemployed

      Only one parent is employed

      Both parents are employed


      1/81 (1.2)

      50/81 (61.7)

      30/81 (37.0)


      1/28 (3.6)

      20/28 (71.4)

      7/28 (25.0)
      0.291
      Time to access health care facility > 30 min5/84 (6.0)3/29 (10.3)0.421
      Abbreviations: NA: not applicable; SD: standard deviations; WAZ: weight-for-age Z-score.
      a Comparison of proportions using the chi-square or fisher tests.
      b Nutritional status was based on the WAZ score generated using the 2000 Centers for Disease Control and Prevention Growth Reference
      • Vidmar S.I.
      • Cole T.J.
      • Pan H.
      Standardizing anthropometric measures in children and adolescents with functions for egen: update.

      Centers for Disease Control and Prevention. CDC Growth charts. Available at: https://www.cdc.gov/growthcharts/. (Accessed: 5th February 2019)

      .
      Table 3Clinical characteristics of participants according to prognosis.
      CharacteristicsGood prognosis (N = 87)Poor prognosis (N = 31)p-value
      Comparison of proportions using the chi-square or fisher tests.
      Clinical history for current illness
      Reported duration of illness prior to admission ≥ 5 days40 (46.0)16 (51.6)0.589
      Reported duration of fever prior to admission

      No fever

      < 5 days

      ≥ 5 days


      14/86 (16.3)

      51/86 (59.3)

      21/86 (24.4)


      6/30 (20.0)

      15/30 (50.0)

      9/30 (30.0)
      0.675
      Referred from another healthcare centre6 (6.9)8 (25.8)0.005
      Started on antibiotics prior to admission17/85 (20.0)8 (25.8)0.501
      Clinical characteristics at admission
      Capillary refill > 3 s0 (0)4 (12.9)0.004
      Tachycardia for age
      Tachycardia was defined as heart rate > 160/minute for infants 〈12 months, and 〉 150/minute for children ≥ 12 months of age7.
      25/86 (29.1)8 (25.8)0.729
      Increased respiratory rate
      Increased respiratory rate was defined as >50breaths per minute in children aged 2 to 12 months and >40 breaths per minute in children aged ≥ 12 months.
      40 (46.0)22/30 (73.3)0.010
      SpO2 (median, IQR)
      Peripheral capillary oxygen saturation was measured in room air using Mindray VS-800 Vital Sign Monitor or Biolight BLT M800 Handheld pulse oximeter. Eight children (all of them with poor prognosis including two children with fatal outcome) had SpO2 <90% and were put on oxygen before or at arrival, with missing exact SpO2 value.
      86 (80 to 90)77 (70 to 84)0.0002
      Hypoxaemia (SpO2 < 90%)64 (73.6)30 (96.8)0.004
      Fever (≥37.5 °C, axillar)38 (43.7)11 (35.5)0.427
      High fever (> 39 °C, axillar)2 (2.3)3 (9.7)0.113
      Lower chest wall indrawing
      Lower chest wall indrawing was defined as subcostal and/or lower intercostal retractions, and very severe chest indrawing was defined as supraclavicular and/or suprasternal retractions.
      37/86 (43.0)26 (83.9)<0.0001
      Very severe chest indrawing
      Lower chest wall indrawing was defined as subcostal and/or lower intercostal retractions, and very severe chest indrawing was defined as supraclavicular and/or suprasternal retractions.
      3/86 (3.5)11 (35.5)<0.0001
      Nasal flaring12/86 (14.0)14 (45.2)<0.0001
      Grunting2 (2.3)5 (16.1)0.013
      Crackles46/86 (53.5)22 (71.0)0.091
      Ronchi37/86 (43.0)22 (71.0)0.008
      Wheezing27/83 (32.5)5 (16.1)0.103
      Decreased level of consciousness0 (0)4 (12.9)0.004
      Prostration7 (8.1)12 (38.7)<0.0001
      Seizure0 (0)0 (0)NA
      Clinical scoring scales at admission
      Severe WHO pneumonia65 (74.7)30 (96.8)0.007
      RISC score (median, IQR)2 (1 to 3)3 (3 to 4)0.0003
      RISC-Malawi score (median, IQR)6 (3 to 8)7 (7 to 8)0.0012
      LODS (median, IQR)0 (0 to 0)0 (0 to 1)0.0001
      Radiological findings
      Endpoint pneumonia

      Other infiltrates

      Normal
      16/73 (21.9)

      15/73 (20.6)

      42/73 (57.5)
      9/23 (39.1)

      5/23 (21.7)

      9/23 (39.1)
      0.195
      Laboratory findings at admission
      Anaemia (Haemoglobin < 11 g/dL)23 (26.4)19 (61.3)0.001
      Leucocytosis
      Leucocytosis was defined as white blood cells greater than 15 × 109 cells/L for children aged between 2 and 11 months and greater than 13 × 109 cells/L for children aged between 12 and 59 months.
      30 (34.5)11 (35.5)0.920
      Thrombocytosis (> 450 × 109 platelets/L)20 (23.0)10/29 (34.5)0.221
      High ESR (≥ 50 mm)12/78 (15.4)6/30 (20.0)0.564
      High CRP-study (> 4 mg/dL)11/83 (13.3)7 (22.6)0.224
      High CRP-ref (> 4 mg/dL)66 (75.9)24 (77.4)0.861
      High PCT (≥ 250 pg/mL)23 (26.4)14 (45.2)0.054
      Non-contaminated positive bacterial blood culture5/73 (6.9)2/27 (7.4)1.000
      Hospital management
      Antibiotic therapy57 (65.5)27 (87.1)0.023
      Oxygen therapy58 (66.7)31 (100)<0.001
      Hospital stay ≥ 7 days5 (5.8)17 (54.8)<0.001
      Abbreviations: CRP: C-reactive protein; ESR: erythrocyte sedimentation rate; IQR: interquartile range; LODS: Lambaréné Organ Dysfunction Score; NA: not applicable; PCT: procalcitonin; RISC: respiratory index of severity in children; WHO: World Health Organization.
      a Comparison of proportions using the chi-square or fisher tests.
      b Tachycardia was defined as heart rate > 160/minute for infants 〈12 months, and 〉 150/minute for children ≥ 12 months of age
      • Conroy A.L.
      • et al.
      Prospective validation of pediatric disease severity scores to predict mortality in Ugandan children presenting with malaria and non-malaria febrile illness.
      .
      c Increased respiratory rate was defined as >50breaths per minute in children aged 2 to 12 months and >40 breaths per minute in children aged ≥ 12 months.
      d Peripheral capillary oxygen saturation was measured in room air using Mindray VS-800 Vital Sign Monitor or Biolight BLT M800 Handheld pulse oximeter. Eight children (all of them with poor prognosis including two children with fatal outcome) had SpO2 <90% and were put on oxygen before or at arrival, with missing exact SpO2 value.
      e Lower chest wall indrawing was defined as subcostal and/or lower intercostal retractions, and very severe chest indrawing was defined as supraclavicular and/or suprasternal retractions.
      f Leucocytosis was defined as white blood cells greater than 15 × 109 cells/L for children aged between 2 and 11 months and greater than 13 × 109 cells/L for children aged between 12 and 59 months.

      Association of host-response biomarkers with prognosis

      Overall, results of the routinely ordered laboratory testing, including white blood cells (WBC), platelets, ESR, PCT, and CRP, were not associated with outcome when evaluated using common clinical thresholds (Table 3). Similar results were observed when these laboratory parameters were assessed as continuous variables, with the exception of PCT, which was associated with prognosis (Supplementary Table S2). In contrast, plasma levels of all the immune and endothelial activation factors, except for IL-6, were significantly higher at presentation in children that progressed to severe and fatal infections (Fig. 1 and Supplementary Table S2). After adjusting for selected potential confounding factors, differences remained significant for sTREM-1, sTNFR1, IL-8 and PCT (Supplementary Fig. S2).
      Fig 1
      Fig. 1Performance of host-biomarkers levels according to prognosis.Abbreviations: Angpt-2: angiopoietin-2; CRP: C-reactive protein; ESR: erythrocyte sedimentation rate; IL6: interleukin-6; IL8: interleukin-8; PCT: procalcitonin; sFlt1: soluble fms-like tyrosine kinase-1; sTNFR1: soluble tumour necrosis factor receptor 1; sTREM-1: soluble triggering receptor expressed on myeloid cells 1; WBC: white blood cells.Levels of each biomarker is summarized graphically through the median (red dot) and interquartile range (lower and upper side of the box). Good prognosis was defined as survival, no admission in the paediatric intensive care unit, no requirement of oxygen or oxygen therapy for ≤ 5 days, and no requirement of chest drainage; while poor prognosis was defined as death and/or admission in the paediatric intensive care unit and/or oxygen therapy for > 5 days and/or required chest drainage. Statistical significance of differences between good and poor outcome for each biomarker level was calculated using the Mann-Whitney U tests, with p-value shown at the top of each biomarker comparison.

      Performance of clinical characteristics, scoring scales, and biomarkers at predicting poor prognosis

      Of single clinical characteristics, oxygen saturation (AUROC 0.75, 95% confidence interval [CI] 0.63–0.86) and lower chest wall indrawing (AUROC 0.70, 95% CI 0.62–0.79) on admission displayed the best predictive accuracy for prognosis (Fig. 2). Of the clinical scoring scales, RISC presented the best predictive performance (AUROC 0.71, 95% CI 0.61–0.80) and was significantly higher than the WHO severity score at predicting prognosis (AUROC 0.61, 95% CI 0.55–0.67; P < 0.05).
      Fig 2
      Fig. 2Prognostic accuracy of clinical characteristics, scoring scales and host-response biomarkers in children with pneumonia.Abbreviations: Angpt-2: angiopoietin-2; AUROC: area under the receiver operating characteristics; CRP: C-reactive protein; ESR: erythrocyte sedimentation rate; IL6: interleukin-6; IL8: interleukin-8; LODS: Lambaréné Organ Dysfunction Score; PCT: procalcitonin; RISC: respiratory index of severity in children; sFlt1: soluble fms-like tyrosine kinase-1; sTNFR1: soluble tumour necrosis factor receptor 1; sTREM-1: soluble triggering receptor expressed on myeloid cells 1; WBC: white blood cells; WHO: World health Organization.Nonparametric ROC curves were generated. AUROC was plotted for each variable to illustrate its ability to discriminate between good and poor prognosis. For each variable, AUROC value with the 95% confidence interval in parenthesis are displayed to the right of its plot.
      The best host-response biomarker for predicting poor prognosis was sTREM-1 (AUROC 0.74, 95%CI 0.63–0.88) (Fig. 2). sTREM-1 performed significantly better than the commonly used inflammatory markers (WBC, ESR, and CRP) and IL-6, but not significantly better than the other immune and endothelial activation markers (AUROC 0.61–0.67) (Supplementary Table S3).
      Supplementary Table S4 summarizes additional performance characteristics (sensitivity, specificity and likelihood ratios) of clinical scoring scales and biomarkers.

      Top performing biomarkers improve the prognostic performance of clinical characteristics

      We assessed the performance of combinations of the best performing clinical signs, scales, and biomarkers. The addition of sTREM-1 significantly improved the prognostic performance of lower chest wall indrawing or the RISC score, but these combinations did not perform better than sTREM-1 alone (Table 4). Taking into consideration that RISC is a clinical scoring scale that includes the assessment of chest indrawing, we concluded that sTREM-1 combined with assessment of lower chest wall indrawing was the most parsimonious prognostic model.
      Table 4Performance of clinical parameters associated with top predicting biomarker sTREM-1.
      AUROC
      Clinical parameter+ sTREM-1
      Oxygen saturation0.75 (0.63 to 0.86)0.81
      p < 0.10 for comparison of AUROC of the clinical parameter alone versus AUROC of the combination of the clinical parameter with sTREM-1.
      Lower chest wall indrawing0.70 (0.62 to 0.79)0.84
      p < 0.05 for comparison of AUROC of the clinical parameter alone versus AUROC of the combination of the clinical parameter with sTREM-1.
      RISC score0.71 (0.61 to 0.80)0.82
      p < 0.05 for comparison of AUROC of the clinical parameter alone versus AUROC of the combination of the clinical parameter with sTREM-1.
      sTREM10.74 (0.63 to 0.88)
      Abbreviations: AUROC: area under the receiver operating characteristics; RISC: respiratory index of severity in children; sTREM-1: soluble triggering receptor expressed on myeloid cells 1.
      Differences in AUROCs were assessed using the algorithm suggested by DeLong et al. (1988)
      • DeLong E.R.
      • DeLong D.M.
      • Clarke-Pearson D.L.
      Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach.
      .
      low asterisk p < 0.10 for comparison of AUROC of the clinical parameter alone versus AUROC of the combination of the clinical parameter with sTREM-1.
      low asterisklow asterisk p < 0.05 for comparison of AUROC of the clinical parameter alone versus AUROC of the combination of the clinical parameter with sTREM-1.

      Prognosis performance of clinical characteristics, scoring scales, and biomarkers differ by age groups

      We investigated the performance of biomarkers by age groups since inflammatory response varies by age.
      • Haugen J.
      • et al.
      Cytokine concentrations in plasma from children with severe and non-severe community acquired pneumonia.
      ,
      • Saghafian-Hedengren S.
      • et al.
      Assessment of cytokine and chemokine signatures as potential biomarkers of childhood community-acquired pneumonia severity.
      We performed subgroup analyses amongst infants (< 12 months) and older children (≥ 12 months) (Table 5). PCT and IL-6 performed better at predicting poor outcome in children ≥ 12 months compared to infants. The performance of the clinical characteristics and scoring scales did not significantly differ between infants and older children.
      Table 5Performance of clinical characteristics and biomarkers for identifying children at risk of poor prognosis.
      AUROC (95% CI)
      All<12 months≥12 months
      Clinical characteristics
      Increased respiratory rate0.64 (0.54 to 0.73)0.63 (0.49 to 0.77)0.67 (0.56 to 0.78)
      Oxygen saturation0.75 (0.63 to 0.86)0.79 (0.65 to 0.93)0.70 (0.50 to 0.89)
      Lower chest wall indrawing0.70 (0.62 to 0.79)0.74 (0.64 to 0.84)0.65 (0.51 to 0.80)
      Very severe chest indrawing0.66 (0.57 to 0.75)0.65 (0.54 to 0.76)0.69 (0.54 to 0.83)
      Clinical scoring scales
      WHO severity score0.61 (0.55 to 0.67)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.66 (0.57 to 0.75)0.57 (0.52 to 0.62)
      p < 0.10 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      RISC score0.71 (0.61 to 0.80)0.74 (0.62 to 0.87)0.69 (0.56 to 0.82)
      RISC-Malawi score0.69 (0.60 to 0.79)0.76 (0.65 to 0.88)0.65 (0.48 to 0.81)
      LODS0.66 (0.56 to 0.75)0.67 (0.55 to 0.79)0.63 (0.49 to 0.78)
      Acute phase proteins and inflammatory markers
      WBC0.52 (0.40 to 0.65)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.57 (0.40 to 0.74)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.62 (0.43 to 0.81)
      Platelets0.58 (0.46 to 0.71)
      p < 0.10 for comparison of AUROCs between age groups for each clinical characteristic, clinical scoring scale or biomarker.
      0.53 (0.36 to 0.70)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.70 (0.50 to 0.89)
      ESR0.54 (0.40 to 0.68)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.60 (0.43 to 0.78)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.76 (0.59 to 0.93)
      CRP-study0.60 (0.48 to 0.72)0.57 (0.41 to 0.74)0.65 (0.49 to 0.82)
      CRP-ref0.59 (0.47 to 0.71)
      p < 0.10 for comparison of AUROCs between age groups for each clinical characteristic, clinical scoring scale or biomarker.
      0.51 (0.35 to 0.68)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.73 (0.56 to 0.89)
      PCT0.61 (0.49 to 0.72)
      p < 0.05 for comparison of AUROCs between age groups for each clinical characteristic, clinical scoring scale or biomarker.
      0.52 (0.37 to 0.67)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.72 (0.55 to 0.90)
      Immune activation factors
      IL-60.61 (0.48 to 0.74)
      p < 0.05 for comparison of AUROCs between age groups for each clinical characteristic, clinical scoring scale or biomarker.
      0.50 (0.32 to 0.68)
      p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      0.76 (0.61 to 0.91)
      IL-80.62 (0.50 to 0.74)0.62 (0.46 to 0.78)0.59 (0.39 to 0.79)
      sTREM-10.74 (0.63 to 0.88)0.69 (0.52 to 0.85)0.77 (0.61 to 0.93)
      sTNFR10.67 (0.55 to 0.79)0.58 (0.40 to 0.75)0.77 (0.60 to 0.93)
      Endothelial activation factors
      Angpt-20.61 (0.48 to 0.74)0.66 (0.50 to 0.83)0.54 (0.33 to 0.75)
      sFlt10.60 (0.48 to 0.73)0.62 (0.45 to 0.79)0.55 (0.35 to 0.75)
      Abbreviations: Angpt-2: angiopoietin-2; AUROC: area under the receiver operating characteristics; CI: confidence interval; CRP: C-reactive protein; ESR: erythrocyte sedimentation rate; IL-6: interleukin-6; IL-8: interleukin-8; LODS: Lambaréné Organ Dysfunction Score; NA: not applicable; PCT: procalcitonin; RISC: respiratory index of severity in children; sFlt1: soluble fms-like tyrosine kinase-1; sTNFR1: soluble tumour necrosis factor receptor 1; sTREM-1: soluble triggering receptor expressed on myeloid cells 1; WBC: white blood cells; WHO: World Health Organization.
      Differences in AUROCs were assessed using the algorithm suggested by DeLong et al. (1988)
      • DeLong E.R.
      • DeLong D.M.
      • Clarke-Pearson D.L.
      Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach.
      .
      low asterisk p < 0.10 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      low asterisklow asterisk p < 0.05 for comparison of AUROCs of the RISC score versus each of the other scoring scales and biomarkers.
      # p < 0.10 for comparison of AUROCs between age groups for each clinical characteristic, clinical scoring scale or biomarker.
      ## p < 0.05 for comparison of AUROCs between age groups for each clinical characteristic, clinical scoring scale or biomarker.
      The RISC score in infants (AUROC 0.74, 95%CI 0.62–0.87) performed significantly better than WBC, platelets, ESR, CRP-study, CRP-ref, PCT and IL-6 in predicting poor prognosis. In older children, the RISC score (AUROC 0.69, 95%CI 0.56–0.82) performance was similar to all biomarkers. Biomarker levels by age group are reported in Supplementary Table S2.

      sTREM-1-based algorithms predict poor prognosis in children with pneumonia

      As sTREM-1 demonstrated good prognostic accuracy for children with pneumonia, we examined this marker with top performing clinical characteristics to generate simple algorithms for risk-stratification in community and hospital settings. We performed classification and regression tree analyses to identify optimal cut-off points. We forced the clinical variable to be included in the model first for clinical relevance. Alone, sTREM-1 presented a sensitivity of 35.5% (95% CI 19.2–54.6) and specificity of 98.9% (95% CI 93.8–99.9), and the positive and negative likelihood ratios were 32.27 and 0.65, respectively (Fig. 3). The combination of very severe chest indrawing with sTREM-1 was found to be the best performing combination of sTREM-1 with any clinical characteristics, and increased sensitivity to 61.3% (95% CI 42.2–78.2) with a small decrease in specificity (95.4%; 95% CI 88.6–98.7).
      Fig 3
      Fig. 3Classification and regression tree analysis algorithms to predict poor outcome in children with pneumonia.Abbreviations: LR: likelihood ratio; sTREM-1: soluble triggering receptor expressed on myeloid cells 1.The algorithms were generated for sTREM-1 (left) and very severe chest indrawing and sTREM-1 (right). Good prognosis was defined as survival, no admission in the paediatric intensive care unit, no requirement of oxygen or oxygen therapy for ≤ 5 days, and no requirement of chest drainage; while poor prognosis was defined as death and/or admission in the paediatric intensive care unit and/or oxygen therapy for > 5 days and/or required chest drainage. For all models, the cost of misclassifying a child with poor prognosis was designated as 10 times the cost of misclassifying a child with good prognosis. Classification and regression tree analysis selected the optimal cut-off points. We forced the clinical variable to be included in the model first for clinical relevance. The performance of each of the three algorithms are presented in the table below them.

      Discussion

      Prognostic tools that enable the early identification of children with pneumonia that will progress to severe and potentially fatal disease are currently lacking. Early risk-stratification of children with respiratory symptoms could facilitate triage, early referral, and prioritization of care, and improve outcomes. In the following study, we assessed potential prognostic factors in children hospitalized with WHO-defined pneumonia, including clinical characteristics and a wide range of host-response biomarkers.
      We found that several clinical signs upon admission were associated with poor prognosis, including typical clinical indicators of pneumonia such as increased respiratory rate or grunting. However, and in agreement with previous studies, the prognostic performance of single clinical signs would not support clinical decision-making in the field on their own.
      • Uwaezuoke S.N.
      • Ayuk A.C.
      Prognostic scores and biomarkers for pediatric community-acquired pneumonia: how far have we come?.
      ,
      • Wright J.K.
      • et al.
      Biomarkers of endothelial dysfunction predict sepsis mortality in young infants: a matched case-control study.
      ,
      • Dean P.
      • Florin T.A.
      Factors associated with pneumonia severity in children: a systematic review.
      ,
      • Esposito S.
      • Principi N
      Unsolved problems in the approach to pediatric community-acquired pneumonia.
      In addition, the detection of clinical signs depends on the health worker ability to correctly assess them, leading to applicability limitations due to interobserver variability and the need of trained health workers.
      • Dean P.
      • Florin T.A.
      Factors associated with pneumonia severity in children: a systematic review.
      ,
      • Fernandes C.D.
      • et al.
      Host inflammatory biomarkers of disease severity in pediatric community-acquired pneumonia: a systematic review and meta-analysis.
      To improve prognostic performance of single clinical signs, several scoring scales have been developed, combining clinical signs, risk factors, and simple laboratory testing. LODS was initially developed for the risk assessment of children with malaria but was then found to yield good discrimination to predict in-hospital mortality (AUROC 0.86) amongst febrile Ugandan children aged 2–59 months with no malaria.
      • Conroy A.L.
      • et al.
      Prospective validation of pediatric disease severity scores to predict mortality in Ugandan children presenting with malaria and non-malaria febrile illness.
      ,
      • Helbok R.
      • et al.
      The Lambaréné Organ Dysfunction Score (LODS) is a simple clinical predictor of fatal malaria in African children.
      In the Bhutanese cohort, LODS was associated with poor prognosis but showed a low sensitivity (38.7; 95% CI 21.8–57.8) and low prognostic performance (AUROC 0.66). Each of the three components of LODS (coma, prostration, and deep breathing) is indicative of severe disease and therefore may have limited utility in the early stages of severe disease.
      • Waterer G.
      Severity scores and community-acquired pneumonia. Time to move forward.
      The RISC score was developed specifically for children with respiratory infections and include clinical signs, several of which may have greater utility in early identification of severe pneumonia.
      • Reed C.
      • et al.
      Development of the Respiratory Index of Severity in Children (RISC) score among young children with respiratory infections in South Africa.
      RISC demonstrated higher sensitivity (87.1%; 95% CI 70.2–96.4), which is an essential characteristic for a community-based triage tool. However, the RISC score is difficult to determine in low resource settings as it requires anthropometric measurement to assess weight-for-age, a pulse oximeter, the ability to recognize chest indrawing, and auscultation for wheezing. The WHO severity criteria are widely used and rely on their high sensitivity to detect most cases for antibiotic therapy and hospital management.

      World Health Organization. Revised WHO classification and treatment of childhood pneumonia at health facilities. Evidence summaries. (2014).

      We observed similar findings in our cohort, where all the children progressing to poor prognosis except one were classified as severe pneumonia. In conclusion, we found that clinical scoring scales were significantly associated with poor prognosis and presented high sensitivity at established cut-off points, but specificity was low, leading to a high number of false-positive cases. In addition, they rely on clinical signs, which does not solve the problem of subjectivity and interobserver variability in their assessment.
      Biomarker concentrations can be measured in the blood, with the benefits of objectivity, accuracy, and reproducibility. Currently, their measurement require training in blood collection and specialized equipment, although the development of rapid diagnostic testing with the best performing biomarkers to conduct with blood drops collected by finger prick could easily overcome these difficulties. WBC, platelets, ESR, and CRP are commonly used in clinical practice as aetiological and prognostic markers. However, studies have consistently shown that these biomarkers are poor prognostic predictors for childhood pneumonia.
      • Reed C.
      • et al.
      Development of the Respiratory Index of Severity in Children (RISC) score among young children with respiratory infections in South Africa.
      ,
      • Dean P.
      • Florin T.A.
      Factors associated with pneumonia severity in children: a systematic review.
      ,
      • Wu J.
      • et al.
      Evaluation and significance of C-reactive protein in the clinical diagnosis of severe pneumonia.
      In previous studies, levels of immune and endothelial activation markers were associated with disease severity and fatal outcome in life-threatening infections, including pneumonia, sepsis, severe malaria, haemorrhagic fevers, or COVID-19.
      • Leligdowicz A.
      • et al.
      Risk-stratification of febrile African children at risk of sepsis using sTREM-1 as basis for a rapid triage test.
      ,
      • Leligdowicz A.
      • Richard-Greenblatt M.
      • Wright J.
      • Crowley V.M.
      • Kain K.C.
      Endothelial activation: the Ang/Tie axis in sepsis.
      ,
      • Erdman L.K.
      • et al.
      Combinations of host biomarkers predict mortality among Ugandan children with severe malaria: a retrospective case-control study.
      ,
      • Xing K.
      • Murthy S.
      • Liles W.C.
      • Singh J.M
      Clinical utility of biomarkers of endothelial activation in sepsis-a systematic review.
      • Richard-Greenblatt M.
      • et al.
      Prognostic accuracy of soluble triggering receptor expressed on myeloid cells (sTREM-1)-based algorithms in febrile adults presenting to tanzanian outpatient clinics.
      • Wright J.K.
      • et al.
      Biomarkers of endothelial dysfunction predict sepsis mortality in young infants: a matched case-control study.
      • Wright S.W.
      • et al.
      sTREM-1 predicts mortality in hospitalized patients with infection in a tropical, middle-income country.
      • Balanza N.
      • et al.
      Host-based prognostic biomarkers to improve risk stratification and outcome of febrile children in low- and middle-income countries.
      • Van Singer M.
      • et al.
      COVID-19 risk stratification algorithms based on sTREM-1 and IL-6 in emergency department.
      • Conroy A.L.
      • et al.
      Host biomarkers are associated with progression to dengue haemorrhagic fever: a nested case-control study.
      ,
      • Zhang R.
      • et al.
      Dysregulation of angiopoietin-Tie-2 axis in ugandan children hospitalized with pneumonia.
      In the Bhutanese cohort, IL-8, sTREM-1, sTNFR1, Angpt-2, and sFlt1 were all significantly associated with poor prognosis despite the moderately small size of the cohort and few children with fatal outcome.
      Amongst the immune and endothelial markers analysed in this study for children with pneumonia, sTREM-1 exhibited the highest AUROC (0.74, 95% CI 0.63–0.88). The addition of sTREM-1 significantly improved the prognostic performance of the best performing clinical characteristics such as lower chest wall indrawing. These findings suggest that simple sTREM-1-based algorithms for pneumonia management may represent a strategy to improve care and outcome in children, particularly in resource-limited settings.
      • Silva-neto P.V.
      • et al.
      sTREM-1 predicts disease severity and mortality in COVID-19 patients: involvement of peripheral blood leukocytes and MMP-8 activity.
      ,
      • Hogendoorn S.
      • et al.
      Clinical sign and biomarker-based algorithm to identify bacterial pneumonia among outpatients with lower respiratory tract infection in Tanzania.
      Selection of which biomarker-based model to apply clinically will depend on the primary goal of the triage tool. For example, highly sensitive algorithms with associated low negative likelihood ratio would perform well to correctly classify children at low risk of evolving to poor prognosis. These children could be sent home confidently, while those classified at high risk might require close monitoring to ensure early detection of deterioration of the child. On the other hand, algorithms aiming for higher specificity with associated higher positive likelihood ratio, such as the one based on sTREM-1, perform better at correctly classifying children at risk of poor prognosis and as such, could assist care prioritization decisions. This approach is also useful in the context of the COVID-19 pandemic in any setting, to help improve rationale allocation of resources and decision on patient triage in overburdened hospitals.
      This study has several limitations. As there were only three deaths in the cohort, we used a composite primary outcome, which limits direct comparison with other studies using mortality as the primary outcome. We did not include other factors known to impact circulating biomarker concentrations such as duration of illness, prior administration of antibiotics, malnutrition, and other comorbidities, which are important considerations in biomarker discovery and validation studies.
      • Méndez R.
      • et al.
      Initial inflammatory profile in community-acquired pneumonia depends on time since onset of symptoms.
      • Rytter M.J.H.
      • Kolte L.
      • Briend A.
      • Friis H.
      • Christensen V.B.
      The immune system in children with malnutrition - a systematic review.
      • Krüger S.
      • Welte T
      Biomarkers in community acquired pneumonia.
      Since pneumonia progresses rapidly, increases or decreases between two measurements of the same biomarker over time (dynamic monitoring) might further help in the risk-stratification of children with this disease.
      • Savvateeva E.N.
      • Rubina A.Y.
      • Gryadunov D.A.
      Biomarkers of community-acquired pneumonia: a key to disease diagnosis and management.
      Also, since excess mortality can be observed up to three months after discharge from severe infections, we encourage assessing outcomes including post-discharge mortality to avoid missing late events.
      • Wiens M.O.
      • et al.
      Postdischarge mortality in children with acute infectious diseases: derivation of postdischarge mortality prediction models.
      ,
      • Madrid L.
      • et al.
      Postdischarge mortality prediction in sub-Saharan Africa.
      We did not adjust the statistical analyses for multiple comparisons, due to the exploratory nature of the analysis. Thus, the differences with statistical significance need to be evaluated as such. The size of the clinical cohort and number of outcomes is small, and therefore a potential risk of over fitting models containing more than one predictive variable exists. Further research from additional cohorts is needed to elucidate the best biomarker or combination of biomarkers and which cut-off points to use for risk-stratification of children with pneumonia.
      Nonetheless, our study confirms that immune and endothelial activation markers have the potential to become objective risk-stratification tools of children with pneumonia. A biomarker point-of-care tool alone or integrated into a simple clinical algorithm is likely to enhance clinical decision-making (such as triage and prioritization of care) and improve outcomes, in addition to optimizing resource allocation, especially in low- and middle-income countries, where mortality associated with childhood pneumonia remains greatest.

      Data availability

      The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

      Author contributions

      SJ, QB, MRG, and KCK conceptualized the work. SJ, TL, and KD collected the data. MN and RS performed the laboratory investigations. SJ run the analyses and drafted the main text. QB, MRG, KCK, and MN substantively revised the manuscript. MRG and MN prepared Fig. 1, Fig. 2, Fig. 3 and S2. All authors reviewed the manuscript.

      Declaration of Competing Interest

      The authors declare no competing interests.

      Acknowledgements

      We are grateful to the children and families who participated and made this study possible. We thank all the nurses who contributed to the sample collection, and the paediatric and microbiology departments of JDWNRH for their support. We also thank Gaurav Kwatra and Laura Puyol who helped us with samples shipment, and Kathleen Zhong for the measurement of host-response biomarkers. 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. Canadian Institutes of Health Research (CIHR Foundation grant no. FDN-148439 to K.C.K.); Tesari Foundation (to K.C.K); and the Canada Research Chair program (to K.C.K.). The funding bodies had no role in the design of the study and collection, analysis and interpretation of data, and writing the manuscript.

      Appendix. Supplementary materials

      References

      1. World Health Organization. Pneumonia. Fact sheets. (2019). Available at: https://www.who.int/news-room/fact-sheets/detail/pneumonia. (Accessed: 27th January 2020)

        • Perin J.
        • 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. Heal. 2022; 6: 106-115
        • McAllister D.A.
        • et al.
        Global, regional, and national estimates of pneumonia morbidity and mortality in children younger than 5 years between 2000 and 2015: a systematic analysis.
        Lancet Glob. Heal. 2019; 7: e47-e57
        • Ginsburg A.S.
        • et al.
        Placebo vs amoxicillin for nonsevere fast-breathing pneumonia in Malawian children aged 2 to 59 months. A double-blind, randomized clinical noninferiority trial.
        JAMA Pediatr. 2019; 173: 21-28
        • Sibila O.
        • Restrepo M.I.
        Biomarkers in community-acquired pneumonia: still searching for the one.
        Eur. Respir. J. 2019; 531802469
        • Hooli 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: 1-13
        • Conroy A.L.
        • 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: 1-11
        • Deardorff K.V.
        • Mccollum E.D.
        • Ginsburg A.S
        Pneumonia risk stratification scores for children in low-resource settings. A systematic literature review.
        Pediatr. Infect. Dis. J. 2018; 37: 743-748
        • Waterer G.
        Severity scores and community-acquired pneumonia. Time to move forward.
        Am. J. Respir. Crit. Care Med. 2017; 196: 1236-1238
        • Thomas J.
        • Pociute A.
        • Kevalas R.
        • Malinauskas M.
        • Jankauskaite L.
        Blood biomarkers differentiating viral versus bacterial pneumonia aetiology: a literature review.
        Ital. J. Pediatr. 2020; 46: 4
        • Leligdowicz A.
        • 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: 1-9
        • Uwaezuoke S.N.
        • Ayuk A.C.
        Prognostic scores and biomarkers for pediatric community-acquired pneumonia: how far have we come?.
        Pediatr. Heal. Med. Ther. 2017; 8: 9-18
        • Leligdowicz A.
        • Richard-Greenblatt M.
        • Wright J.
        • Crowley V.M.
        • Kain K.C.
        Endothelial activation: the Ang/Tie axis in sepsis.
        Front. Immunol. 2018; 9: 838
        • Xing K.
        • Murthy S.
        • Liles W.C.
        • Singh J.M
        Clinical utility of biomarkers of endothelial activation in sepsis-a systematic review.
        Crit. Care. 2012; 16: R7
        • Richard-Greenblatt M.
        • 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: 1304-1312
        • Wright J.K.
        • et al.
        Biomarkers of endothelial dysfunction predict sepsis mortality in young infants: a matched case-control study.
        BMC Pediatics. 2018; : 1-12
        • Wright S.W.
        • et al.
        sTREM-1 predicts mortality in hospitalized patients with infection in a tropical, middle-income country.
        BMC Med. 2020; 18: 159
        • Balanza N.
        • et al.
        Host-based prognostic biomarkers to improve risk stratification and outcome of febrile children in low- and middle-income countries.
        Front. Pediatr. 2020; 8552083
        • Van Singer M.
        • et al.
        COVID-19 risk stratification algorithms based on sTREM-1 and IL-6 in emergency department.
        J Allergy Clin. Immunol. 2020; (S0091-6749): 31401-31409
        • Conroy A.L.
        • et al.
        Host biomarkers are associated with progression to dengue haemorrhagic fever: a nested case-control study.
        Int. J. Infect. Dis. 2015; 40: 45-53
        • Jullien S.
        • et al.
        Pneumonia in children admitted to the national referral hospital in Bhutan: a prospective cohort study.
        Int. J. Infect. Dis. 2020; 95: 74-83
      2. World Health Organization. Revised WHO classification and treatment of childhood pneumonia at health facilities. Evidence summaries. (2014).

        • Cherian T.
        • et al.
        Standardized interpretation of paediatric chest radiographs for the diagnosis of pneumonia in epidemiological studies.
        Bull. World Health Organ. 2005; 83: 353-359
        • Reed C.
        • 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: e27793
        • Helbok R.
        • et al.
        The Lambaréné Organ Dysfunction Score (LODS) is a simple clinical predictor of fatal malaria in African children.
        J. Infect. Dis. 2009; 200: 1834-1841
        • Bradley B.D.
        • Howie S.R.C.
        • Chan T.C.Y.
        • Cheng Y.L.
        Estimating oxygen needs for childhood pneumonia in developing country health systems: a new model for expecting the unexpected.
        PLoS ONE. 2014; 9: e89872
      3. World Health Organization. The clinical use of oxygen in hospitals with limited resources. Guidelines for health-care workers, hospital engineers and programme managers. (2009).

        • Leligdowicz A.
        • et al.
        Validation of two multiplex platforms to quantify circulating markers of inflammation and endothelial injury in severe infection.
        PLoS ONE. 2017; 12e0175130
        • DeLong E.R.
        • DeLong D.M.
        • Clarke-Pearson D.L.
        Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach.
        Biometrics. 1988; 44: 837-845
      4. StataCorp. Stata v.16. (2019).

      5. RStudio Team. RStudio: Integrated Development for R. (2020).

        • Haugen J.
        • et al.
        Cytokine concentrations in plasma from children with severe and non-severe community acquired pneumonia.
        PLoS ONE. 2015; 10e0138978
        • Saghafian-Hedengren S.
        • et al.
        Assessment of cytokine and chemokine signatures as potential biomarkers of childhood community-acquired pneumonia severity.
        Pediatr. Infect. Dis. J. 2017; 36: 102-108
        • Dean P.
        • Florin T.A.
        Factors associated with pneumonia severity in children: a systematic review.
        J. Pediatric Infect. Dis. Soc. 2018; 7: 323-334
        • Esposito S.
        • Principi N
        Unsolved problems in the approach to pediatric community-acquired pneumonia.
        Curr. Opin. Infect. Dis. 2012; 25: 286-291
        • Fernandes C.D.
        • 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: ofz520
        • Wu J.
        • et al.
        Evaluation and significance of C-reactive protein in the clinical diagnosis of severe pneumonia.
        Exp. Ther. Med. 2015; 10: 175-180
        • Zhang R.
        • et al.
        Dysregulation of angiopoietin-Tie-2 axis in ugandan children hospitalized with pneumonia.
        Cytokine. 2020; 133155175
        • Erdman L.K.
        • et al.
        Combinations of host biomarkers predict mortality among Ugandan children with severe malaria: a retrospective case-control study.
        PLoS ONE. 2011; 6: e17440
        • Silva-neto P.V.
        • 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: 2521
        • Hogendoorn S.
        • et al.
        Clinical sign and biomarker-based algorithm to identify bacterial pneumonia among outpatients with lower respiratory tract infection in Tanzania.
        BMC Infect Dis. 2022; 39https://doi.org/10.1186/s12879-021-06994-9
        • Méndez R.
        • et al.
        Initial inflammatory profile in community-acquired pneumonia depends on time since onset of symptoms.
        Am. J. Respir. Crit. Care Med. 2018; 198: 370-378
        • Rytter M.J.H.
        • Kolte L.
        • Briend A.
        • Friis H.
        • Christensen V.B.
        The immune system in children with malnutrition - a systematic review.
        PLoS ONE. 2014; 9e105017
        • Krüger S.
        • Welte T
        Biomarkers in community acquired pneumonia.
        Expert Rev. Respir. Med. 2012; 6: 203-214
        • Savvateeva E.N.
        • Rubina A.Y.
        • Gryadunov D.A.
        Biomarkers of community-acquired pneumonia: a key to disease diagnosis and management.
        Biomed Res. Int. 2019; 1701276https://doi.org/10.1155/2019/1701276
        • Wiens M.O.
        • et al.
        Postdischarge mortality in children with acute infectious diseases: derivation of postdischarge mortality prediction models.
        BMJ Open. 2015; 5e009449
        • Madrid L.
        • et al.
        Postdischarge mortality prediction in sub-Saharan Africa.
        Pediatrics. 2019; 143e20180606
        • Vidmar S.I.
        • Cole T.J.
        • Pan H.
        Standardizing anthropometric measures in children and adolescents with functions for egen: update.
        Stata J. 2013; 13: 366-378
      6. Centers for Disease Control and Prevention. CDC Growth charts. Available at: https://www.cdc.gov/growthcharts/. (Accessed: 5th February 2019)