The PERCH study sought to determine the cause of severe pneumonia in children aged 1–59 months living in developing areas [21]. The detailed methods of this multisite case-control study, including a description of case and control enrollment, have been published elsewhere [22, 23]. Briefly, in each of 9 sites in 7 countries (Dhaka and Matlab, Bangladesh; Basse, The Gambia; Kilifi, Kenya; Bamako, Mali; Soweto, South Africa; Sa Kaeo and Nakhon Phanom, Thailand; and Lusaka, Zambia), from August 2011 to January 2014, children presenting to study facilities with WHO-defined severe or very severe pneumonia were evaluated for enrollment using standardized criteria. Controls were randomly selected children from the community without severe or very severe pneumonia and were frequency matched for age (4 age strata) within 2 weeks of the matched case enrollment. Blood, NP/OP swab specimens, and urine were collected from both cases and controls; blood cultures and IS specimens were also collected from cases.

IS specimen collection was attempted within 24 hours of hospital admission for each PERCH case unless contraindicated. The methods for sputum induction are described elsewhere [24, 25]. In brief, participants were given a β2-agonist followed by nebulized hypertonic saline solution. A sterile closed-loop mucus extractor attached to a suction device was inserted through the nose into the nasopharynx. Suction was applied once the extractor was in the nasopharynx and was discontinued before the extractor was removed. Secretions were flushed into a collection container using 5 mL of sterile normal saline.

All laboratory methods were highly standardized across all study laboratories [25, 26]. Quantitative real-time PCR for respiratory pathogens was performed on IS and NP/OP specimens, as described elsewhere [27].

Chest radiographs from cases were classified as consolidation, other infiltrate, both consolidation and other infiltrate, normal, or uninterpretable by members of a panel of 14 radiologists and pediatricians, who were trained in the standardized interpretation of pediatric chest radiographs [28, 29].

Because the WHO definitions for severe and very severe pneumonia are not specific [30], a proportion of children enrolled as PERCH cases probably did not have true pneumonia. We therefore defined 2 subsets of PERCH cases: (1) children with radiographic evidence of pneumonia defined by the presence of consolidation and/or other infiltrate on chest radiographs, representing children who truly had infectious pneumonia, and (2) children with clinical and laboratory characteristics highly indicative of the absence of pneumonia, representing children enrolled into the case arm but with a very low likelihood of having infectious pneumonia (ie, nonpneumonia group; Text Box 1). Because IS specimens were not collected from controls, the nonpneumonia cases acted as “mock controls,” and this group was compared for similarity with PERCH control children.

Hypoxemia was defined as oxygen saturation <92% (or < 90% at elevations of >1200 meters) or receipt of supplemental oxygen if a room air oxygen saturation reading was not available. A high-quality IS specimen was defined as a sputum specimen with <10 squamous epithelial cells per low-power field [25].

To test the hypothesis that pathogens causing pneumonia will be detected more frequently in IS than in NP/OP specimens, we compared individual paired IS and NP/OP specimen results among cases with radiographic pneumonia, using McNemar’s χ² test. We calculated additional pathogen detection from the IS specimen, over and above the NP/OP specimen, as the ratio (number of infections detected in either specimen divided by number of infections detected in the NP/OP specimen) minus 1 and expressed as a percentage. The proportion of chest radiographic–positive cases with IS specimen density higher or lower than NP/OP specimen density (difference, ≥1 log₁₀ copies/mL) was evaluated using the 2-sided sign test.

To assess whether pathogen detection in IS specimens (using multiplex PCR) was associated with radiographic evidence of pneumonia, we compared pathogen prevalence in the IS specimens of children with radiographic pneumonia and those without evidence of pneumonia (nonpneumonia) and calculated odds ratios and 95% confidence intervals (CIs), with and without adjustment for the presence of the pathogen on the NP/OP specimen. We compared PCR cycle threshold (Ct) values (a semiquantitative measure of pathogen load) in IS specimens from radiographic pneumonia and nonpneumonia case groups, using the Wilcoxon rank sum test. To compare demographic and baseline characteristics of individuals between study groups, we calculated frequencies for categorical variables and median values with interquartile ranges (75th centile minus 25th centile) for continuous variables. To evaluate the similarity of our mock control group (ie, nonpneumonia cases) with true control children, we compared the prevalence of organisms detected in the NP/OP specimens and calculated odds ratios and 95% CIs. Odds ratios were adjusted for potential confounding variables (eg, age, sex, site, and human immunodeficiency virus [HIV] status) where appropriate.

The majority of analyses involving IS specimens were restricted to high-quality specimens, those more likely to have originated from the lower respiratory tract [25]. The significance of differences between specimens or groups for the 28 organisms evaluated was assessed using Bonferroni correction (0.05/28), so that were differences were considered statistically significant at P < .002.

The study protocol was approved by the institutional review board or ethical review committee at each of the study site institutions and at Johns Hopkins Bloomberg School of Public Health. Parents or guardians of participants provided written informed consent.

A total of 4232 cases with WHO-defined severe (n = 2862) or very severe (n = 1370) pneumonia and 5325 control children were enrolled during a 24-month period. IS specimens were collected from 3800 (89.8%) cases, including 94.8% (n = 2713) of those with severe pneumonia and 79.3% (n = 1087) with very severe pneumonia (Supplementary Table 1). Reasons for noncollection of IS specimens included presence of a contraindication (169 of 432; 39.1%), death (96 of 432; 22.2 %), intubation (85 of 432; 19.7%), parent refusal (14 of 432; 3.2%), and other or unknown reason (68 of 432; 15.7%). The IS specimen collection procedure was generally well tolerated, but adverse consequences were reported in 13 cases (0.34% of procedures); the most common was a transient drop in oxygen saturation [31]. Of the 4232 cases enrolled in PERCH, 1935 (45.7%) had radiographic pneumonia, 573 (13.5%) were classified as nonpneumonia, and 1724 (40.7%) fell into neither category (Table 1). This analysis is restricted to a comparison of the 1935 cases with radiographic pneumonia, 573 nonpneumonia cases, and 5325 controls.

The proportion cases with radiographic pneumonia varied by site, from 31% at Matlab, Bangladesh, to 60% at Dhaka, Bangladesh, and Soweto, South Africa (Table 2). The proportion of nonpneumonia cases among those enrolled varied by site from none in Dhaka, Bangladesh and 4.5% of cases in Soweto, South Africa, to 31% enrolled from Kilifi, Kenya (Table 2). Compared with nonpneumonia cases, cases with radiographic pneumonia were more likely to be female (44% vs 36%; P < .001), have an abnormally high or low white blood cell count (51% vs 37%; P < .001), or be HIV positive (9% vs 1%; P < .001); malaria was more common in the nonpneumonia group (8% vs 1%; P < .001) (Table 1).

Characteristics used to distinguish radiographic pneumonia from nonpneumonia cases, such as hypoxemia (45% vs 6%), tachypnea for age (84% vs 51%), and the presence of crackles (69% vs 18%) differed as expected. Cases in the nonpneumonia group were more likely to produce high-quality sputum specimens (75% vs 67%; P < .001). The low number of cases classified as nonpneumonia in South Africa reflects the standard clinical practice at that site to administer oxygen to all children with a diagnosis of pneumonia (often without obtaining a room air oxygen saturation measurement), which led to a high percentage (75%) of cases meeting the study definition of hypoxemia. No cases met the definition of nonpneumonia at the Dhaka, Bangladesh site; 40% had a normal chest radiograph but all had other respiratory findings that excluded them from nonpneumonia status (Text Box 1).

To determine the concordance of pathogen detection rates between NP/OP and IS specimens, and the added yield from testing IS specimens in addition to NP/OP specimens, we analyzed NP/OP swabs and IS specimens collected from the 1692 cases admitted to the hospital with radiographic pneumonia from whom paired specimens were available. Of these 1692, 99% had 1 or more pathogens identified on IS PCR; the median number of pathogens identified on PCR of IS or NP/OP was 4. Among those with paired specimens, high quality IS specimens were available in 1114 (65.8%). Using both NP/OP and high-quality IS specimens, bacteria were detected in 1080 (96.9%) specimens and viruses in 1072 (96.2%) specimens. Overall, the IS specimens increased the number of cases with a virus, bacteria, or any organism detected by 4.1%, 3.9%, and 1.3%, respectively (Table 3). Restricting this analysis to cases <6 months of age slightly reduced the added benefit (data not shown).

Among cases with radiographic pneumonia, pathogen detection by PCR was similar for high-quality IS specimens and NP/OP specimens for most pathogens; this is reflected by the moderate-to-high agreement between NP/OP and IS specimens (Table 3). A similar pattern was also observed comparing high-quality IS and NP/OP specimens from nonpneumonia cases (Supplementary Table 2). The added yield of an IS specimen for identifying a pathogen in cases with radiographic pneumonia varied by organism from 4.8% (95% CI, 3.4%–6.8%; Moraxella catarrhalis) to 81.8% (31.3%–212.0%; Salmonella species). Because the added yield calculation can be misleading with small sample sizes (eg, the 81.8% for Salmonella species reflects a difference of 8 vs 9 detections in NP/OP vs IS specimens), it is useful to visualize the pattern of detection of different pathogens by different specimen types.

As seen in Figure 1, the NP/OP specimens detected the majority of infections for most organisms. However, for certain organisms, IS specimens seem superior to NP/OP specimens (eg, in cases with radiographic pneumonia, parainfluenza 1 was detected from IS specimens alone in 2.8% vs 0.5% for NP/OP specimens alone). An analysis stratified by pneumonia severity (severe or very severe) and age <6 months showed pathogen detection rates from NP/OP and IS specimens very similar to those indicated in Table 3 (data not shown). To assess whether the added yield of the IS specimen was related to collection of a sputum specimen per se or to collection of an additional respiratory specimen of any type, we compared the added yield of low-quality IS specimens (ie, those presumably from, or contaminated by, secretions from the upper respiratory tract) with the yield of NP/OP specimens. The added yield of low-quality IS specimens was similar (Figure 1 vs Supplementary Figure 1).

There was no difference in the prevalence or the density of P. jirovecii by IS or NP/OP specimens among cases with radiographic pneumonia (Table 3; Supplementary Table 3). Among HIV-positive children with radiographic pneumonia, the overall P. jirovecii detection rate was higher (19.7%) than in all HIV-negative cases with radiographic pneumonia (10.6%). In the latter group of HIV-negative children, there was a higher prevalence of P. jirovecii (17% vs 9%; P = .01) among children who were severely malnourished (weight for age less than −3 z-scores below the median of the WHO child growth standards) than among those who were not. There was little added value in seeking the detection of P. jirovecii in IS versus NP/OP specimens. Among 81 HIV-positive children with radiographic pneumonia, P. jirovecii was detected in both IS and NP/OP specimens in 13 cases, IS specimens alone in 2, and NP/OP specimens alone in only 1. This represents an increase in yield of only 1% in the detection of P. jirovecii when assessing IS specimens in addition to NP/OP specimens (15 of 81 [18.5%] vs 14 of 81 [17.3%]).

Among cases with radiographic pneumonia, pathogen density in either IS or NP/OP specimens did not show a consistent pattern across pathogens. As indicated by the proportion of cases with a difference in PCR density >1 log₁₀ copies/mL, the density was greater in the IS specimen than in the NP/OP specimen only for parainfluenza 1, and greater in the NP/OP specimen for 3 organisms (Haemophilus influenzae, M. catarrhalis, and S. pneumoniae) (Supplementary Table 3).

To assess whether organism detection in IS specimens was associated with radiographic pneumonia, we compared its prevalence in the IS specimens from radiographic pneumonia and nonpneumonia case groups, restricting the analysis to high-quality IS specimens. For 4 organisms (H. influenzae, M. pneumoniae, parainfluenza 1, and respiratory syncytial virus [RSV]), detection in the IS specimen was associated with radiographic pneumonia case status; however, this association was no longer significant when accounting for the presence of the pathogen in the NP/OP specimen (Table 4). That is to say, no organism was found significantly more frequently in the IS specimen from cases with radiographic pneumonia compared with nonpneumonia cases. However, given the close correlation in presence of organisms between IS and NP/OP specimens (Supplementary Table 2), the presence of the above 4 organisms in either the IS or NP/OP specimen was significantly associated with radiographic pneumonia status compared with nonpneumonia status. An analysis stratified by pneumonia severity (severe and very severe) and age <6 months showed very similar results (Supplementary Tables 4a and 4b).

To test the hypothesis that organism density should be higher in IS specimens from cases with radiographic pneumonia than nonpneumonia cases if the IS specimen is more closely reflective of the cause of pneumonia, we compared median Ct values for each organism detected with IS specimen PCR. Among high-quality IS specimens, there were no statistically significant differences in organism density in the IS specimens from cases with radiographic pneumonia compared with nonpneumonia cases, with the exception of a marginally significant difference for cytomegalovirus (median Ct, 27.9 vs 28.5; P = .01) (Figure 2). Findings were similar among infants aged <6 months (data not shown), with the only difference being greater S. pneumoniae density in the nonpneumonia group (median Ct, 27.0 for radiographic pneumonia vs 25.6 for nonpneumonia cases; P = .002).

We compared organism detection in the NP/OP specimens between the nonpneumonia cases and the community controls to assess whether our mock control group (ie, nonpneumonia group) was similar to the community control group (Supplementary Table 5). Accounting for multiple comparisons, we found that some pathogens were identified less frequently among nonpneumonia cases (mock control group) than among the true controls (adjusted odds ratio for M. catarrhalis, 0.63 [95% CI, .51–.78]; P. jirovecii, 0.45 [.29–.70]); alternatively, some were more commonly found in nonpneumonia cases (adjusted odds ratio for influenza A, 3.32 [95% CI, 1.93–5.69]; influenza B, 4.07 [CI, 2.14–7.72]; parainfluenza 1, 4.66 [2.83–7.68]; RSV, 5.94 [4.34–8.12]).