Combination of clinical symptoms and blood biomarkers can improve discrimination between bacterial or viral community-acquired pneumonia in children

Globally, pneumonia is the leading cause of hospitalizations and death among children with nearly 120 million new cases and one million deaths each year [1]. In Australia, pneumonia is associated with 5–8 hospitalisations per 1000 child-years among children < 5 years old, with deaths being rare [2]. Australian Aboriginal children are 14 times more at risk of infectious diseases than non-Aboriginal children [3].

Respiratory bacteria and viruses are frequently detected in specimens collected from children with pneumonia [4]. Identifying the infectious agents associated with illness can guide management of the infection and facilitate judicious use of antibiotics. Differentiating bacterial from viral pneumonia based on clinical characteristics is challenging as the clinical signs and symptoms overlap [5, 6].

Despite the growing availability of molecular techniques for pathogen detection, including quantitative and qualitative pathogen detection, laboratory results are usually only available after treatment decisions have been made. Several studies have assessed the utility of non-specific inflammatory biomarkers such as C-reactive protein (CRP), an acute-phase reactant released in response to cytokine interleukin-6, white cell count (WCC) and absolute neutrophil count (ANC) to discriminate probable bacterial infections from non-bacterial infections and also to assess the severity of illness [7, 8]. Bacterial pneumonia has been associated with higher CRP, WCC and ANC than viral pneumonia [8‐11], while some studies have found no difference in biomarkers among bacterial and viral cases of pneumonia [12, 13]. Even studies that report differences in these biomarkers cannot determine any reliable thresholds for differentiating bacterial pneumonia from viral pneumonia [10, 14]. Previous studies were also affected by small sample size, use of less sensitive methods for pathogen detection which could have resulted an inaccurate categorization of bacterial and viral cases [12, 14]. Furthermore, studies have been conducted in low-income settings where the contribution of bacterial infection may be higher and where children are at higher risk of other infectious diseases that could influence the biomarker levels and confound the analysis [10].

The aetiology of childhood pneumonia has changed in high-income settings with routine pneumococcal conjugate vaccination program – decrease in bacterial aetiology and increase in viral aetiology have been reported [15, 16]. However, the utility of inflammatory biomarkers in a highly-vaccinated paediatric population with pneumonia to differentiate bacterial from viral pneumonia has rarely been assessed [11]. We therefore assessed the distribution of inflammatory biomarkers (CRP, WCC and ANC) in blood samples from a prospective case-control study of Western Australian children with radiologically-confirmed pneumonia [17]. We compared the clinical characteristics and biomarkers level among pneumonia cases detected with bacteria and viruses. We assessed the cut-off threshold of these biomarkers for discriminating bacterial pneumonia from viral pneumonia. Findings from this study could help in developing a rapid point-of-care diagnostic tool or algorithm to predict the likely causative pathogen and to assist clinicians to target management of childhood pneumonia.

A total of 230 children with radiologically confirmed community-acquired pneumonia (cases) were enrolled during the study period. Of these, 120 (52%) were male and 147 (64%) aged ≤ 5 years; the median age was 38 months (IQR: 19, 81). Twenty-one (9%) cases were Aboriginal. Of the 230 cases, 210 (91%) had received at least 2 doses of the 13-valent pneumococcal conjugate vaccine. Demographics and existing or ever diagnosed with co-morbidities of the enrolled children are summarized in Table 1.

The median length of hospitalization was 2 days (IQR: 1, 3) and all children were discharged with no deaths. At hospital presentation, 38 (17%) cases had blood oxygen saturation (SPO₂) level ≤ 92% and tachypnoea observed in 88 (38%). Nearly half of cases (109/230) received antibiotics in the 7 days prior to hospital presentation and all but three (227/230, 99%) received antibiotics during hospital stay. Twenty-four cases were diagnosed with pleural effusion and of these, 21 (88%) had pleural fluid drained: all 21 had microscopic purulence consistent with empyema.

There were 30 (13%) cases of definite bacterial pneumonia: 9 with bacteraemia, 15 with empyema, and 6 with both bacteraemia and empyema. Of the 21 pleural fluid samples from empyema cases, 1 was culture and PCR positive, and 9 were PCR positive (only) for Streptococcus pneumoniae, 1 each cultured methicillin-resistant Staphylococcus aureus, methicillin-sensitive Staphylococcus aureus, and Streptococcus pyogenes; and 1 sample was PCR positive for Mycoplasma pneumoniae. At least one respiratory virus was detected in NPS of 12 of the 30 cases with definite bacterial pneumonia. Of the remaining 200 cases, at least one virus was detected in nasopharyngeal swab from 118 (59%) cases including 98 with co-detection of respiratory bacteria. Among these 118 presumed viral pneumonia cases, 43 (36%) had RSV detected, 31 (26%) had rhinovirus, 21 (18%) had HMPV, 15 (13%) had influenza and 9 (8%) each had adenovirus and parainfluenza. No virus was detected in the nasopharyngeal swabs of 82 (41%) cases (other pneumonia) including 51 had detectable respiratory bacteria on NPS. The distribution of bacteria detected in NPS in three case groups of pneumonia are presented in Additional file 1: Table S1. The distribution of bacteria in NPS in three case groups were similar. Among cases in three groups, 53% of definite bacterial pneumonia cases, 40% of presumed viral pneumonia cases and 56% of other pneumonia cases received antibiotics before hospitalization.

The clinical features of cases and medical interventions are summarized in Table 2. Few differences in clinical symptoms and signs were observed across the different groups (Table 2). Fever (defined as temperature ≥ 38.0 °C) was more frequently observed in definite bacterial pneumonia than in presumed viral (p = 0.002) or other pneumonia cases (p < 0.001). Rhinorrhoea was more frequent in presumed viral pneumonia than in either definite bacterial pneumonia (p < 0.001) or other pneumonia cases (p < 0.001). Age-specific tachypnoea was also more common in presumed viral pneumonia than in definite bacterial (p = 0.08) or other pneumonia cases (p = 0.003). More definite bacterial pneumonia cases required intravenous fluid therapy than presumed viral (p = 0.02) or other pneumonia cases (p = 0.001). Furthermore, more than half (53%) of definite bacterial cases required both supplemental O₂ and intravenous fluid, substantially higher than presumed viral (32%; p = 0.03) or other pneumonia cases (23%, p = 0.002) (Table 2). Definite bacterial pneumonia cases also had a greater median length of hospital stay (6.5 days, 9 days with empyema and 2 days without) than presumed viral (2 days) or other pneumonia (2 days) (p < 0.001 for each). The mean RISC severity score was 1.2 (range: 0, 5) for definite bacterial cases, compared to 1.0 (− 2, 5) for presumed viral and 0.8 (− 2, 4) for other pneumonia cases, respectively.

The median blood CRP concentration was more than 6 times higher in definite bacterial cases than in presumed viral (174 versus 24 mg/L; p < 0.001) and other pneumonia cases (174 versus 27 mg/L; p < 0.001). The CRP, WCC and ANC did not vary significantly between presumed viral and other pneumonia cases and between empyema and bacteraemia cases (Table 2, Additional file 1: Table S2). The blood biomarker values did not vary significantly between presumed viral cases with different viral pathogens detected (data not shown).

The AUC for CRP was 0.82 (95% CI: 0.73, 0.91) for discriminating definite bacterial from presumed viral pneumonia (Fig. 1a) and 0.81 (95% CI: 0.72, 0.89) for discriminating definite bacterial from presumed viral plus other pneumonias (Fig. 1b). For WCC, the AUCs were 0.63 (95% CI: 0.53, 0.74) and 0.65 (95% CI: 0.53, 0.76) for discriminating definite bacterial from presumed viral plus others pneumonias, and definite bacterial from presumed viral pneumonia, respectively (Additional file 2: Figure S1a, 1b). For ANC, the AUCs were 0.68 (95% CI: 0.58, 0.78) and 0.69 (95% CI: 0.58–0.79) for discriminating definite bacterial from presumed viral plus other pneumonias, and definite bacterial from presumed viral pneumonias, respectively (Additional file 2: Figure S2a, 2b). Based on these AUCs, we further assessed the sensitivity, specificity, PPV and NPV at CRP cut-offs of ≥ 40 mg/L, ≥ 60 mg/L and ≥ 100 mg/L, for differentiating bacterial from presumed viral plus other pneumonias, and definite bacterial from presumed viral pneumonias (Additional file 1: Table S3). CRP ≥ 40 mg/L, CRP ≥ 60 mg/L and CRP ≥ 100 mg/L cut-off had sensitivity of 83, 75 and 67%, respectively for differentiating definite bacterial pneumonia from presumed viral pneumonias. From the Youden index, the optimal CRP threshold of ≥72 mg/L was found to discriminate definite bacterial from presumed viral plus other pneumonias with sensitivity 75% (95% CI: 55, 89), specificity 82% (95% CI: 76, 87), PPV 38% and NPV 96%; for discriminating definite bacterial from presumed viral pneumonias the sensitivity was 75% (95% CI: 55, 89), specificity 84% (95% CI: 76, 90), PPV 53% and NPV 93% (Table 3).

Significant differences in symptoms and signs were then included to assess the impact on the diagnostic performance of the algorithm (Table 3; Additional file 1: Table S3). The combination of CRP (≥72 mg/L) with either the presence of fever or absence of rhinorrhea, improved the specificity and PPV for differentiating bacterial pneumonia from presumed viral plus other pneumonias compared to the CRP alone, with little loss of sensitivity and NPV. While 73% of cases with definite bacterial pneumonia had both fever and elevated CRP (≥72 mg/L), only 11% of presumed viral plus other pneumonias and 10% of presumed viral pneumonias did. Similarly, 65% of definite bacterial pneumonia had elevated CRP (≥72 mg/L) and the absence of rhinorrhea, compared to 7% of presumed viral and other pneumonias, and 2% of presumed viral pneumonias. The PPV for the combination (CRP ≥72 mg/L + fever) and (CRP ≥72 mg/L + absence of rhinorrhea) was 48 and 59% for discriminating definite bacterial from presumed viral plus other pneumonia, and was 63 and 87%, respectively, for discriminating definite bacterial from presumed viral pneumonia, respectively (Table 3).

This study describes the clinical characteristics of children with radiologically confirmed pneumonia, and assesses the utility of serum biomarkers and clinical signs and symptoms for differentiating definite bacterial from other pneumonias in a highly vaccinated population. There were few differences between children with definite bacterial pneumonia, and those with presumed viral pneumonia, and other pneumonias which were neither definite bacterial nor presumed viral. CRP, WCC and ANC were higher in definite bacterial pneumonia and the CRP had value for distinguishing these from presumed viral and other pneumonias. The combination of high CRP with either fever ≥38.0 °C or with absence of rhinorrhoea increased the specificity and PPV compared to elevated CRP alone with little loss in sensitivity, suggesting that combining biomarkers with clinical features is of diagnostic value.

Timely identification of pneumonia aetiology could improve clinical management including decisions about the use of antibiotics. In line with previous studies, the clinical signs and symptoms among cases of definite bacterial and presumed viral pneumonia overlapped and were insufficiently specific in themselves to reliably differentiate one from the other [5, 24]. Our findings are mostly consistent with previous studies which have associated viral pneumonia with low grade fever, tachypnoea, rhinorrhoea and wheezing [4, 25‐27]. We did not attempt to distinguish cases based on specific radiographic features, but others have not found any radiographic feature that can reliably distinguish bacterial from viral pneumonia [14, 28, 29].

There are conflicting reports on the utility of serum inflammatory biomarkers such as CRP, WCC, and ANC for differentiating bacterial from viral pneumonia in children [9‐12, 14]. In UK children, Elemraid et al. found a higher CRP, WCC and ANC in those with bacterial compared to viral pneumonia, with a CRP level > 80 mg/L reportedly distinguishing severe bacterial pneumonia from viral pneumonia [9]. The sensitivity and specificity for CRP > 80 mg/L for differentiating bacterial versus viral/atypical pneumonia in children with community-acquired pneumonia was 71 and 52%, respectively, in Norway [11] and was 52 and 72%, respectively, in Finland [14]. We found that CRP had better discriminatory utility (AUC > 0.8) than the WCC or ANC, consistent with findings similar settings [9, 11, 14]. The optimum CRP threshold for differentiating definite bacterial from presumed viral and other pneumonia had only moderate predictive value which could limit its use as a sole criterion for antibiotic decision-making in clinical settings. Serum procalcitonin is not routinely used in our study setting, so we could not assess its value for distinguishing bacterial pneumonia from presumed viral pneumonias. Serum procalcitonin (PCT) has been described as a better biomarker for bacterial sepsis than CRP [30], while other found serum CRP as better predictor for lobar consolidation and pleural effusion than PCT [31].

Other studies have also attempted to discriminate bacterial from viral pneumonia cases using either blood biomarkers [8‐10, 12, 13, 32], clinical characteristics [5, 24] or radiological differences [14, 28] with modest success. To our knowledge, however, combining inflammatory biomarkers with clinical symptoms has rarely been explored. We found that compared to high CRP (≥72 mg/L) alone, high CRP and either fever or absence of rhinorrhoea improved the specificity and PPV with little loss in sensitivity, thereby improving the diagnostic accuracy.

Respiratory bacteria are often detected in the nasopharynx of healthy asymptomatic children, but are rarely detected in normally sterile sites like blood and pleural fluid. Furthermore, purulent pleural effusions are generally considered to be of bacterial origin even if bacteria cannot be detected. Respiratory viruses are usually detectable in the nasopharynx of children with viral pneumonias, but may also be detectable in healthy children [16, 33, 34]. In the absence of any gold standard microbiological assays for distinguishing bacterial from viral and mixed infections, we categorised cases based on whether they were highly likely to be bacterial (definite), most likely viral (presumed), or possibly either based on whether bacteria were detected in sterile sites, respiratory viruses were detected in the nasopharynx, or neither. This pragmatic approach to classification reflects the limitations of current methods, and some misclassification is likely to have occurred. Previous studies categorised cases as either bacterial or viral, are were more limited due to lower pathogen detection rates and by the use of less sensitive assays [8, 9, 12, 14, 26]. The strength of this study is that the range of pathogens tested covered virtually all important bacteria and viruses associated with childhood pneumonia using sensitive molecular techniques.

Our study had some limitations. While there were clear differences between the definite bacterial and the presumed viral and other pneumonia cases, it is likely that a number of the latter groups also had bacterial pneumonia or mixed infections and we note that many had detectable respiratory bacteria in their nasopharyngeal swabs (but not in their blood or pleural fluid). Of note, there were differences between presumed viral pneumonia and other pneumonia cases; the latter were less likely to have rhinorrhoea and tachypnoea suggesting they may comprise a higher proportion of bacterial infections than the presumed viral group. Half of all cases had prior exposure to antibiotics at enrolment which might have impacted on the natural progression of signs and symptoms and biomarkers, and the sensitivity of bacterial culture. Sputum samples are often used for diagnosis of community-acquired pneumonia in adults but collection of sputum from children is challenging and studies found limited utility of sputum as diagnostic tool for pneumonia [35].

Empiric use of antibiotics remains as cornerstone of treating pneumonia in the absence of effective point-of-care diagnostics for differentiating bacterial from viral infection. Many children who have viral pneumonia will continue to receive antibiotics without benefit. Early reliable detection of viral pneumonia, or early exclusion of bacterial pneumonia, could reduce unnecessary antibiotic therapy, thereby mitigating the risk of emerging antibiotic resistance [36]. While we have been unable to identify a single biomarker or clinical feature that could be used to confidently distinguish bacterial from viral pneumonia, our findings suggest there may be utility in more sophisticated algorithms that integrate a number of clinical, microbiological, inflammatory biomarker, or radiological factors to improve pneumonia diagnostics and better targeting therapies.