Pneumonia Severity in Children: Utility of Procalcitonin in Risk Stratification

Data were obtained from 2 prospective cohorts of children with pneumonia presenting for emergency care. The first was a pilot study performed at a single tertiary children’s hospital in the southeastern United States from December 2014 to January 2017. These data were combined with data from the same southeastern children’s hospital and a second tertiary children’s hospital in the western United States from September 2017 to May 2019. This study represents a subset of both of these cohorts because subjects were still eligible for each cohort if they declined to provide blood samples. Study protocols were approved by the institutional review board of each institution.

Eligibility criteria were identical for both cohorts. Children 2 months to <18 years presenting to the ED with signs and symptoms of acute lower respiratory infection, radiographic evidence of pneumonia, and a provider diagnosis of pneumonia were considered for inclusion. Radiographic evidence of pneumonia was determined by the treating clinical provider at the time of enrollment, in concert with the clinician’s interpretation of available clinical radiology reports. A secondary review of all radiology reports was performed by the 2 site primary investigators to further classify pneumonia by infiltrate pattern. Children <2 months of age were excluded, as were those who were immunocompromised, had cystic fibrosis, or had a tracheostomy. In an effort to exclude hospital-acquired pneumonias, children transferred from other hospitals and those recently hospitalized or previously enrolled were also excluded (Supplemental Table 3).

Enrolled children and caregivers were interviewed to collect baseline sociodemographic, clinical, and historical data. Blood samples for procalcitonin measurement were collected at the time of enrollment (up to 72 hours from the time of ED triage) and frozen at −80°C. All other testing was at the discretion of the provider and not uniform among subjects. Additional data collected from the electronic health record included physical examination, radiology and laboratory findings, vital signs, ED disposition, LOS, need for intensive care, and development of respiratory failure and/or shock.

Procalcitonin was the primary exposure of interest, measured in nanograms per milliliter. Frozen samples were batched and analyzed by using the bioMérieux VIDAS 3 platform, according to standard protocols by an institutional research laboratory.¹⁶  Procalcitonin values below the detection limit of 0.05 ng/mL were set at 0.025 ng/mL. Procalcitonin assays were not conducted in real time and were not available to clinicians. Seven children had procalcitonin obtained during clinical care. Because clinical and study procalcitonin values were not statistically different, only the study procalcitonin values were used during analysis.

The primary outcome of interest was an ordinal pneumonia severity outcome measure, defined as very severe (intubation, shock requiring vasoactive medications, or death), severe (intensive care admission without very severe features and/or high-flow nasal cannula), moderate (hospitalization without severe or very severe features), and mild (discharge from the hospital).¹⁷  Use of continuous or bilevel positive pressure outside of normal home settings required intensive care at both institutions. The outcome was assigned according to the most severe outcome experienced at any point during the encounter. LOS in hours was a secondary outcome and was measured in hours from the time of ED triage to the time of discharge from either the ED or inpatient hospital unit.

Covariates chosen for analysis in the model were selected on the basis of a previous study evaluating the use of 20 candidate predictors, including demographics, comorbidities, vital signs and other clinical factors, radiographic characteristics, and laboratory results in the prediction of severe pediatric pneumonia.¹⁷  We selected covariates from the reduced 10-predictor model for inclusion in our analyses, excluding altered mental status, which was rarely observed. The covariates included were the ratio of Pao₂ to fraction of inspired oxygen (PF ratio) (estimated from the pulse oxygen saturation to fraction of inspired oxygen ratio), age, triage heart rate (HR), triage respiratory rate (RR), triage systolic blood pressure (SBP), comorbidities, chest indrawing, infiltrate pattern on chest radiograph, and presence of pleural effusion.^17,18 

Baseline characteristics were evaluated by severity classification by using Wilcoxon rank and Kruskal-Wallis tests and χ² tests for continuous and categorical variables, respectively. Associations between procalcitonin and severity outcomes were evaluated by using ordinal logistic regression, adjusting for the covariates listed above. Restricted cubic spline functions were applied on age, PF ratio, and procalcitonin with 3 knots to relax linearity assumptions. To account for known age-based differences in normal HR, RR, and blood pressure, interaction terms between each of these variables and age were included. The proportional odds assumption was confirmed by visual inspection of partial residual plots. Because a cubic spline outcome was used for the primary exposure, we estimated adjusted odds ratios (aORs) for a variety of cut points of procalcitonin. We also generated predicted probability plots to demonstrate how varying procalcitonin changes the estimated probabilities for severe or very severe outcomes and LOS >3 days, holding all other covariates constant. Several additional analyses were also conducted. Because a small number of children had procalcitonin collected >24 hours after triage, we repeated our primary analyses in those whose procalcitonin was collected within 24 hours of triage. Next, we repeated our main analysis after excluding those with very severe pneumonia to explore how procalcitonin performs in those with less severe disease. We also explored procalcitonin results in those with bacteremia and those with parapneumonic effusion requiring a drainage procedure. Lastly, we examined the timing of antibiotic administration and procalcitonin collection.

A total of 1116 children were enrolled in the combined cohorts, with 488 children (44%) with procalcitonin samples of sufficient quantity for analysis: 167 from the pilot cohort and 321 from the main study (Supplemental Fig 4). Compared with children without blood samples, those with procalcitonin samples were older (64.5 vs 33.5 months; P < .001), more likely to be admitted to the hospital (95% vs 78%; P < .001), more likely to require ICU care (26% vs 16%; P < .001), and had a longer average LOS (median 50.5 vs 40.5 hours; P < .001).

Among children included in the study, the distribution of outcomes was as follows: 30 children (6%) were classified as very severe, 106 as severe (22%), 327 (67%) as moderate, and 25 (5%) as mild. Median LOS was 50.5 hours (interquartile range [IQR] 32.0–96.0). Among the 30 subjects in the very severe group, 28 (6% of total cohort) required intubation, 13 (3%) developed shock, and 2 (0.05%) died. Baseline demographic characteristics were largely similar among the 4 severity groups, although those in the very severe group were younger than those in other groups. Baseline clinical characteristics varied between groups, with the most severely ill children demonstrating higher HRs and RRs, lower PF ratios, increased incidence of chest indrawing and pleural effusion, and more comorbidities (Table 1). The median LOS for all children was 50.5 hours (IQR 32.0–96.0). In total, 164 children (34%) had a LOS >3 days.

Overall, procalcitonin concentrations among subjects were low, with a procalcitonin of <0.25 ng/mL in 44% of subjects and <1.0 ng/mL in 65% of subjects. Median procalcitonin in the very severe group was 5.06 ng/mL (IQR 0.90–16.83), 0.38 ng/mL (IQR 0.11–2.11) in the severe group, 0.29 ng/mL (IQR 0.09–1.90) in the moderate group, and 0.21 ng/mL (IQR 0.12–1.20) in the mild group (Fig 1). In unadjusted comparison, procalcitonin was weakly correlated with LOS (r = 0.21; P < .001).

For the ordinal severity outcome, increasing procalcitonin was associated with increasing severity with aORs comparing multiple procalcitonin cut points ranging from 1.03 (95% confidence interval [CI] 1.01–1.06) for 0.25 ng/mL (reference) versus 0.5 ng/mL to 1.25 (95% CI 1.04–1.51) for 0.10 vs 2.0 ng/mL; all comparisons were statistically significant (Table 2, Fig 2). Increasing procalcitonin was also associated with longer LOS, with aOR ranging from 1.04 (95% CI 1.02–1.07) for 0.25 vs 0.50 ng/mL to 1.36 (95% CI 1.17–1.59) for 0.1 vs 2.0 ng/mL; all comparisons were statistically significant (Table 2, Fig 3). When limiting our analysis to procalcitonin samples obtained within 24 hours of ED triage, results for both the ordinal severity outcome and LOS outcome were similar and remained significant (results not shown). In the secondary analysis excluding the very severe group, aORs remained >1.0 for all comparisons, although point estimates were reduced in magnitude and comparisons were no longer statistically significant for the ordinal severity outcome, whereas associations remained significant for the LOS outcome (Supplemental Table 4).

Bacteremia was identified in only 5 (1.7%) of 290 patients for whom a blood culture result was obtained, all with very high procalcitonin levels (142.3, 21.25, 18.18, 13.32, and 11.13 ng/mL). Bacterial pathogens detected were Streptococcus pneumoniae (3), Staphylococcus aureus (1), and S pyogenes (1). Seventy-four children had parapneumonic effusion noted on imaging, with 17 (23%) requiring drainage. Median procalcitonin of those requiring drainage was 10.45 ng/mL (IQR 3.05–20.01). Of the 17 undergoing drainage procedures, 4 (24%) had positive bacterial growth results on pleural culture, all with high procalcitonin levels (36.09, 31.21, 17.49, and 8.58 ng/mL). Bacterial pathogens detected in pleural fluid were S pyogenes (2), S pneumoniae (1), and S intermedius with Parvimonas micra coinfection (1).

Eighty-eight children (18%) had procalcitonin levels obtained before the administration of any antibiotic (including outpatient oral antibiotics). Of those who had not received an antibiotic before presentation to the enrolling ED, but for whom an antibiotic was administered before procalcitonin collection in the ED or while admitted, the mean time from antibiotic administration to procalcitonin collection was 15.8 hours (95% CI 13.7–17.9). Procalcitonin concentration was not different (P = .65) between the no antibiotic pretreatment group (median 0.55 ng/mL; IQR 0.15–2.75) and the antibiotic pretreatment group (median 0.33 ng/mL; IQR 0.10–2.29).

In our study of 488 children with pneumonia presenting for emergency care, increased procalcitonin was associated with both our primary outcome of increased pneumonia severity and our secondary outcome of increased hospital LOS. In our secondary analysis excluding those with very severe pneumonia, procalcitonin was no longer significant for the ordinal severity outcome but remained significant for the LOS outcome.

In our adjusted models, procalcitonin was a significant predictor of severe pneumonia outcomes. This is in line with adult studies associating elevated procalcitonin levels with increased risk of mechanical ventilation or vasopressor support within 72 hours of presentation and smaller pediatric studies associating elevated procalcitonin with hospital admission, ICU admission, and LOS.^13,14,19  These findings are similar to those observed by Florin et al,¹⁵  who also demonstrated associations between procalcitonin and an ordinal severity outcome (most severe defined as ICU care, vasoactive infusions, effusion drainage, or severe sepsis) adjusted for age, antibiotic receipt before arrival, and length of fever.

Very severe outcomes were rare in our cohort, with only 6% requiring mechanical ventilation and 3% requiring use of vasoactive medications. This group demonstrated significantly higher median procalcitonin concentrations (median 5.06 ng/mL) compared with other groups, all of which had median procalcitonin concentrations <0.40 ng/mL. Although our secondary analysis excluding the very severe group was limited by a smaller sample size, it provides further context to our results and suggests that very severe outcomes are a key driver of the associations between procalcitonin and pneumonia severity. In the study by Florin et al,¹⁵  procalcitonin also did not discriminate between those with less severe outcomes. Thus, although procalcitonin is associated with increasing disease severity, its utility for differentiating among those with less severe pneumonia outcomes in the clinical setting may be more limited.

We hypothesize that the utility of procalcitonin as a predictor of severe pneumonia is attributable in part to procalcitonin’s association with bacterial pathogens. Procalcitonin is upregulated in bacterial infections through the release of tumor necrosis factor and interleukin-1 and interleukin-6 and inhibited in viral infections through the action of interferon.^10,20,21  These unique qualities make procalcitonin a leading candidate for the differentiation of bacterial and viral etiologies, with procalcitonin used to guide antibiotic administration in sepsis and other bacterial processes and revealing superior ability to differentiate these etiologies from other biomarkers such as C-reactive protein.^22–24 

Procalcitonin values >2.0 ng/mL have been associated with bacteremic pneumococcal pneumonia, and although our study was not designed to assess pneumonia etiology, the very high procalcitonin concentrations in our subjects with bacteremia supports bacterial etiology association with procalcitonin elevation.²⁵  This is further supported by the high incidence of pleural effusion (40%) in the very severe group. In those with parapneumonic effusion severe enough to require drainage, procalcitonin values were very high (median 10.45 ng/mL [IQR 3.05–20.01]). Florin et al¹⁵  similarly demonstrated associations between procalcitonin and those requiring chest drainage. Although only 4 subjects had bacterial growth on pleural culture (all with procalcitonin >8 ng/mL), we suspect that nearly all of these children had bacterial pneumonia and that bacterial growth was limited by antibiotic pretreatment. This latter hypothesis is supported by the Etiology of Pneumonia in the Community study, in which researchers reported a dramatic decrease in pleural culture yield when antibiotics were administered before drainage (71% vs 37%).² 

In contrast, low procalcitonin values (<0.25 ng/mL) are associated with low risk for bacterial etiology.²⁶  In our study, approximately half of the children had a procalcitonin concentration <0.25 ng/mL, suggesting bacterial disease was unlikely. This assumption aligns with findings from the Etiology of Pneumonia in the Community study, with the majority of pediatric pneumonias in the pneumococcal conjugate vaccine era caused by viral pathogens.²  Although viral pneumonias may also contribute to severe outcomes, especially in younger children,²⁷  whether procalcitonin predicts these more severe outcomes in the absence of bacterial disease is unclear. Recent investigations suggest an association between procalcitonin and severe coronavirus disease 2019 illness, but these studies are small and limited to adults.²⁸ 

Difficulties identifying causative pathogens in pediatric pneumonia highlight the importance of using procalcitonin in concert with other clinical decision-making tools. The wide variation in procalcitonin concentrations observed in our study, with some mildly ill children demonstrating high procalcitonin values and some severely ill children demonstrating low procalcitonin values, reveals that procalcitonin alone is imperfect in predicting disease outcomes. In addition, the independent association between procalcitonin and pneumonia disease severity in children appears modest. Nonetheless, in adult studies, researchers have found that adding procalcitonin to established adult pneumonia risk classification systems improves prognostic accuracy, with procalcitonin levels in the ED more useful than measurements in the ICU or primary care locations.^19,29  It will be important to test the utility of procalcitonin in similar pediatric risk stratification tools.

Our study has limitations to consider. We were not able to enroll every eligible child presenting with pneumonia. Few children discharged from the ED provided blood samples, suggesting our population likely represents a more severely ill cohort than the average child presenting to the ED with pneumonia. This is not unexpected and aligns with current clinical practice, with well-appearing children quickly discharged from the hospital without laboratory evaluations. Nonetheless, we acknowledge that our study may not fully characterize the range of procalcitonin values in mild illness. Enrollment occurred at 2 tertiary care children’s hospitals, with the majority (86%) coming from a single institution. This may limit the generalizability of the study’s findings. Our outcome groups also incorporated hospital location (floor versus ICU) into severity classification, which is useful in terms of how clinicians assess patient needs but may not fully reflect the physiology of illness severity. However, our findings support previous work revealing association between increased procalcitonin and need for ICU admission in pediatric community-acquired pneumonia.¹³  As previously discussed, our study was not designed to systematically assess etiology of community-acquired pneumonia through comprehensive microbiologic testing. As a result, we were not able to determine if pneumonia etiology modifies associations between procalcitonin and severe pneumonia outcomes. Finally, 82% of children received some form of antibiotic before procalcitonin collection. Although procalcitonin values did not differ by receipt of antibiotics, it is possible that antibiotic pretreatment could have falsely lowered procalcitonin. Antibiotic treatment would not be expected to influence procalcitonin values in those with viral pneumonia, and given that the majority of pneumonias in US children are caused by viral pathogens, the influence of antibiotics on procalcitonin in our study may be minimal.² 

Our findings indicate that increased procalcitonin concentration is associated with more severe outcomes and longer LOS among children with pneumonia. Procalcitonin may be useful in helping clinicians evaluate pneumonia severity, although further study is needed, particularly as it relates to procalcitonin in those with noncritical illness. In future studies, researchers should explore the utility of procalcitonin integration into risk stratification models and other decision support applications to determine if these tools improve clinical decision-making.