The Etiology of Pneumonia in Zambian Children

Supplemental Digital Content is available in the text.

D espite globally improved child and infant mortality rates, many low-and middle-income countries have not yet achieved the Millennium Development Goal 4 target for under-5 mortality. 1 Much of this is due to pneumonia which remains the leading infectious cause of death among children under 5 years in developing countries. 2,3 Efforts to reduce under-5 pneumonia deaths include improved case management and availability of highly effective respiratory vaccines. Despite these efforts, approximately 6000 under-5 children die of pneumonia every year in Zambia, comprising 15% of all deaths in this age group. 4 To improve pneumonia outcomes further in Zambia, we need a more complete understanding of risk factors, and the microbiologic causes of pneumonia, specific to the local setting. A Zambian site was included in the Pneumonia Etiology Research for Child Health (PERCH) study, described in detail elsewhere. 5,6 Briefly, PERCH is a highly standardized 7-country case-control study of under-5 children hospitalized with World Health Organization (WHO) defined severe and very severe pneumonia (pre-2013), with a primary aim to determine the microbiologic etiology of pneumonia and risk factors for severe and very severe pneumonia.
This report is limited to the findings from the Zambia site, which is typical of the majority of countries in Eastern and Southern Africa with regard to community or environmental pneumonia risk factors. Because risk factors and etiology are likely to differ by HIV status, we restrict this analysis to the etiology of pneumonia among HIV-negative children enrolled at the Zambia site. Findings among the HIV-infected children are reported separately in this issue. 7 Zambia is a landlocked middle-income country in Southern Africa with relatively high infant mortality (56 per 1000) and HIV prevalence (19.4% among Lusaka women of child-bearing age). 8 Childhood vaccines are widely available and distributed, although pneumococcal conjugate vaccine (PCV) was not introduced until the end of the study. Detailed country characteristics are available in Supplemental Digital Content 1, http://links.lww.com/INF/ D850. This study was conducted at the University Teaching Hospital (UTH) in Lusaka, the densely populated capital of Zambia. 9 UTH is a 1500-bed academic and tertiary healthcare facility in Lusaka serving the greater Lusaka district, including the most impoverished segments of the population, and is a referral center for the entire country. Access to mechanical ventilation was limited and rarely used. Obtaining radiographs required taking children to the Radiology Department at some distance from the pediatric wards and at the caregivers' expense, therefore, were not routinely performed (for nonstudy patients). Oxygen, however, was routinely available. Nearly all children presenting to UTH are initially seen and evaluated at surrounding Lusaka clinics, transported (mostly by private means) to UTH and have often received initial antibiotics (typically benzylpenicillin G) from the clinic, if clinically indicated.

Participants
Study participants were children 1-59 months of age living within the Lusaka catchment area. Details of study methods have been published elsewhere, and are described in Supplemental Digital Content 1, http://links.lww.com/INF/D850. 5,10 Briefly, cases were children hospitalized between November 2011 and October 2013 with WHO-defined 11 severe or very severe pneumonia (pre-2013), including cough and/or difficulty in breathing plus danger signs (central cyanosis, difficulty breast-feeding/drinking, vomiting everything, multiple or prolonged convulsions, lethargy/unconsciousness, or head nodding) defined as "very severe pneumonia," or lower chest wall in-drawing in the absence of danger signs defined as "severe pneumonia." Due to constraints in processing samples on weekends and study staffing, enrolment occurred on weekdays from 07:30 to 18:00 hours. Nighttime admissions were eligible for participation the following morning.
Controls were children without case-defining pneumonia who were randomly selected from the community and frequency matched on age and season (within 3 weeks) to cases. They were not excluded if they had evidence of upper respiratory tract infections.
Cases and controls were excluded if they had been hospitalized within the previous 14 days or if they were a PERCH study participant within 30 days; cases were also excluded if lower chest wall in-drawing resolved with bronchodilator challenge (if applicable).

Clinical Procedures
Standardized study clinical examinations for cases occurred on admission and at 24 and 48 hours. One-month postdischarge, examinations occurred to determine clinical and vital status. Chest radiographs (CXR) were obtained at admission from cases and classified as normal, consolidation, other infiltrate, consolidation and other infiltrate or uninterpretable as defined by the WHO standardized interpretation of pediatric CXRs. 12,13 For controls, clinical assessments only occurred at time of enrolment in PERCH.
HIV exposure status of the child was obtained from the mother's health antenatal card or by maternal recall if card not available. All exposed children had HIV testing performed by DNA PCR or serology, as per Zambian National HIV Guidelines.

Statistical Analysis
Odds ratios and 95% confidence intervals (CIs) of pathogens detected on NP/OP PCR in cases compared with controls were calculated using logistic regression adjusted for age in months and presence of all other pathogens detected on NP/OP PCR to account for associations between pathogens. Logistic regression adjusted for age in months was used to compare clinical characteristics by case-control status and, among cases, by vital status. Results were stratified by age, severity, CXR status and mortality. P values <0.05 were considered significant.
The percent of pneumonia due to each pathogen was estimated using the PERCH integrated analysis (PIA) method, which is described in detail elsewhere. [22][23][24] In brief, the PIA is a Bayesian nested partially latent class analysis that integrates the results for each case from blood culture, NP/OP PCR, whole blood PCR for pneumococcus and induced sputum culture for Mycobacterium tuberculosis. The PIA also integrates test results from controls to account for imperfect test specificity of NP/OP PCR and whole blood PCR. Blood culture results (excluding contaminants) and induced sputum or gastric aspirate results for M. tuberculosis were assumed to be 100% specific (ie, the etiology for a case was attributed 100% to the pathogen that was detected in their blood by culture).
The PIA accounts for imperfect sensitivity of each test/pathogen measurement by using a priori estimates of their sensitivity (ie, estimates regarding the plausibility range of sensitivity which varied by laboratory test method and pathogen). Sensitivity of blood culture was reduced if blood volume was low (<1.5 mL) or if antibiotics were administered before specimen collection. Sensitivity of NP/OP PCR for S. pneumoniae and H. influenzae was reduced if antibiotics were administered before specimen collection (Supplemental Digital Content 2, http://links.lww.com/INF/D851).
As a Bayesian analysis, both the list of pathogens and their initial "prior" etiologic fraction (EF) values were specified a priori, which favored no pathogen over another (ie, "uniform"). The pathogens selected for inclusion in the analysis included any noncontaminant bacteria detected by culture in blood at any of the 9 PERCH sites, regardless of whether it was observed at the Zambia site specifically, Mycobacterium tuberculosis, and all of the multiplex quantitative PCR pathogens except those considered invalid because of poor assay specificity (Klebsiella pneumoniae 25 and Moraxella catarrhalis). A category called "Pathogens Not Otherwise Specified" was also included to estimate the fraction of pneumonia caused by pathogens not tested for or not observed. A child negative for all pathogens would still be assigned an etiology, which would be either one of the explicitly estimated pathogens (implying a "false negative," accounting for imperfect sensitivity of certain measurements) or not otherwise specified. The model assumes that each child's pneumonia was caused by a single pathogen. Analyses were adjusted for age (<1 vs. ≥1 year) to account for differences in pathogen prevalences by this factor; analyses stratified by pneumonia severity could not adjust for age due to small sample size. For results stratified by case clinical data (eg, to CXR+, very severe, vital status, etc.), the test results from all controls were used. However, for analyses stratified by age, only data from controls representative of that age group were used. A separate integrated etiology model was run with in-hospital vital status as a covariate, adjusting for the child's age.

S42
The PIA estimated both the individual-and population-level etiology probability distributions, each summing to 100% across pathogens where each pathogen has a probability ranging from 0% to 100%. The population-level EF estimate for each pathogen was approximately the average of the individual case probabilities and was provided with a 95% credible interval (95% CrI), the Bayesian analogue of the CI.
Statistical analyses were conducted using SAS 9.3 (SAS Institute, Cary, NC), R Statistical Software 3.3.1 (The R Development Core Team, Vienna, Austria) and Bayesian inference software JAGS 4.2.0 (http://mcmc-jags.sourceforge.net/). The R package used to perform the PIA, named the Bayesian Analysis Kit for Etiology Research, is publicly available at https://github.com/zhenkewu/baker.

Ethical Considerations
The study protocol was approved by the Institutional Review Boards at Boston University, the Johns Hopkins Bloomberg School of Public Health in the United States and by the ERES Converge Ethical Review Committee in Zambia. Parents or guardians of participants provided written informed consent.

Demographics and Clinical Characteristics
Among 17,592 children 1-59 months of age admitted to the Pediatric Wards at UTH during the study period, 737 (4.1%) met the case definition of WHO severe or very severe pneumonia ( Fig. 1). Six hundred seventeen (83.7%) of these consented and were enrolled. This report is limited to the 514 (83.3%) HIV-uninfected cases (severe: 354, 68.9%; very severe: 160, 31.1%); clinical and etiology details of the remaining 103 (16.7%) HIV-infected cases will be reported elsewhere. Six hundred one HIV-uninfected controls were consented and enrolled from Lusaka communities within UTH catchment and are included here (Fig. 1). A small proportion (69/601, 11.5%) of the controls had evidence of upper respiratory tract infections at the time of enrolment (Table 1). There was some seasonal variation in disease-particularly with respiratory syncytial virus (RSV) (Supplemental Digital Content 3, http://links.lww.com/ INF/D852), with a greater number of cases (313, 60.9% of total) being enrolled during the Zambia rainy period (November to May).
The median case age was 5 months (vs. 6 months for controls), and 77.6% were infants <1 year of age ( Table 1). The percentage of children who were HIV-exposed (ie, mother HIV infected at delivery) but uninfected was similar between cases (26.1%) and community controls (28.0%).
Both cases and controls were up-to-date for Haemophilus influenzae type b vaccination (79.1% and 86.8%, respectively). Among cases, 31.2% were moderately to severely malnourished, measuring ≤−2 Z scores in WHO weight-for-age calculations. Owing to referral practices for children seen at UTH, almost all cases (91.5%) received a dose of antibiotics at the referring clinic before enrolment and specimen collection.

Clinical Description of Cases
Interpretable CXR findings were available for 367 of 514 (71.4%) cases; of these, 208 (56.7%) showed evidence of either consolidation or "other infiltrate" (Table 1). Sixty-one cases did not have a CXR obtained due to early death (n = 30) or technical reasons (n = 31). An additional 86 were uninterpretable. CXR+ cases had longer duration of illness at enrollment and were more likely to present with hypoxia, fever, leukocytosis and elevated C-reactive protein (Supplemental Digital Content 4, http://links.lww.com/ INF/D853) than CXR normal cases (P < 0.05). Compared with severe pneumonia cases, very severe children were more likely to be hypoxic, severely anemic and be wheezing on auscultation (P < 0.05) (Supplemental Digital Content 4, http://links.lww.com/INF/ D853).
Malaria was infrequently detected at our Zambia site, with only 1 case (0.2%) and 5 controls (0.8%) testing positive by rapid diagnostic testing and thus not included as a potential cause in the integrated etiology analysis.

Mortality
While in-hospital follow-up at the Zambia site was available for all children, 1-month follow-up was less complete. Only approximately half of the cases-distributed equally by CXR positivity status-returned for the 1-month visit. We therefore focus here on in-hospital mortality.
There were 82 (16%) deaths among all HIV-uninfected cases, 76 of whom died in hospital. Of the 34 in hospital deaths with an interpretable CXR taken, 23 (67.6%) were CXR positive. Among the 61 children who did not get a CXR, 30 (49%) died in hospital (vs. in-hospital mortality among cases with an interpretable  ‡Restricted to those children >10 months of age with available measles vaccine data. §Defined as serum bioassay positive (cases and controls), antibiotics administered at the referral facility, or antibiotic administration before whole blood specimen collection at the study facility (cases only).

DISCUSSION
The clinical and etiology findings described here reflect the presentation of community-acquired childhood pneumonia among HIV-uninfected children in the setting of a typical sub-Saharan African city, Lusaka, Zambia-an impoverished, densely populated, urban population with high HIV and M. tuberculosis prevalences and limited access to high-quality health care. Our results should be contrasted with the findings from the 4 other sub-Saharan African PERCH sites with higher-quality health care (South Africa), little HIV (Kenya/Mali/Gambia) and a rural population (Kenya/Gambia).
The main findings at the Zambia PERCH site include (1) the commonest pathogens contributing to severe and very severe pneumonia are similar to the other PERCH sites from Africa and Asia; (2) RSV is the dominant cause of pneumonia, across age and in both WHO severity strata; (3) CXR+ pneumonia was caused by treatable organisms in a large portion (37.6%) of the cases; (4) pediatric M. tuberculosis is a significant cause of pneumonia for hospitalized children; and (5) children with severe and very severe pneumonia have a very poor prognosis in Zambia.
Although PERCH was designed to assess the presentation of childhood pneumonia in globally diverse geographic settings, it is noteworthy that we observed relatively little variation in the most prevalent pneumonia pathogens. In Zambia, we found that a viral etiology was a somewhat less common cause of pneumonia (EF: †Prematurity defined as <37 weeks gestational age or maternal report of premature. ‡Duration of illness defined as duration (in days) of cough, wheeze, fever or difficulty breathing, whichever is longest. §Restricted to those with an interpretable CXR (N = 34 for fatal cases and N = 333 for nonfatal cases). CXR obtained at admission. ¶Hypoxemia defined as oxygen saturation <90% or on supplemental oxygen if a room air oxygen saturation reading was not available. ║Severe anemia defined as hemoglobin ≤7.5 g/dL. **Leukocytosis count defined as >15 × 10 9 cells/L for children 1-11 months and >13 × 10 9 cells/L for children 12-59 months.
www.pidj.com | S47 50.9%) than at all the PERCH sites (EF: 61.4%) reported in the allsites analysis. 22 Similar to the other PERCH sites, the 10 most frequently identified pneumonia pathogens in Zambia accounted for over 80% of causes for pneumonia. The remaining 22 pathogens assessed in this study accounted for a small portion of the remaining cases of pneumonia.
RSV was the dominant cause of pneumonia in Zambia (EF: 26.1%) as it was in the other PERCH sites and exhibited a clearly seasonal pattern. This is consistent with findings from numerous sources. [25][26][27] However, it is notable that RSV contributes relatively little to the pneumonia mortality. Nevertheless, these findings underscore the importance of developing prevention and treatment strategies for RSV in Zambia and support the importance of developing an effective vaccine. Current treatment for severe RSV disease is solely focused on supportive care, including oxygen, which can be in short supply in Zambia.
Vaccine-preventable diseases (Haemophilus influenzae type b, vaccine-type S. pneumoniae, pertussis) were relatively rare among the PERCH cases (EF: 6.8, 95% CI: 0.02-14.4). Disease due to B. pertussis was infrequent (EF: 2.0%), presumably due in part to very high DTP vaccination rates (≈80%), but was more common in infants, an age group of higher pertussis mortality. Despite only Enterobacteriaceae includes E. coli, Enterobacter species, and Klebsiella species, excluding mixed gram-negative rods. Bacterial summary excludes Mtb. Analysis adjusted for age and vital status. Pathogens estimated at the subspecies level but grouped to the species level for display (Parainfluenza virus type 1, 2, 3 and 4; S. pneumoniae PCV10 and S. pneumoniae non-PCV10 types; H. influenzae type b and H. influenzae non-b; influenza A, B, and C). Description of symbols: Line represents the 95% credible interval. The size of the symbol is scaled based on the ratio of the estimated etiologic fraction to its standard error. Of 2 identical etiologic fraction estimates, the estimate associated with a larger symbol is more informed by the data than the priors. Potentially treatable causes for pneumonia comprised a large proportion of cases in Zambia. Six of the top 10 pathogens at our site (descending order-M. tuberculosis, E. coli, H. influenzae type b, Salmonella species, P. jirovecii and S. aureus) accounted for almost 40% of disease. Empiric antibiotic treatment, for commonly presenting bacterial pathogens, needs to be updated accordingly and guided by local antibiotic resistance patterns. Additional trimethoprim-sulfamethoxazole coverage for P. jirovecii, particularly for under-1 children exposed to HIV, should also be emphasized in Zambia. Further efforts for detecting M. tuberculosis in all children presenting with pneumonia in Zambia should be considered.
Overall in-hospital mortality among HIV-exposed-uninfected children was nearly double that in HIV-unexposed-uninfected cases (22% vs. 11%, P < 0.001). This may be due to the lower viral etiology (EF: 30.3% HIV exposed vs. 53.2% HIV unexposed), and higher bacterial (S. aureus, Salmonella species and E. coli) causes, along with P. jirovecii, among the HIV-exposed children. 7 In a nested substudy, we have previously shown that standardized empiric therapy for these HIV-exposed children improves outcome. 28 P. jirovecii was somewhat common (EF: 4.5% in CXR+ cases) among this HIV-uninfected cohort reported here and was associated with high in-hospital mortality. However, most of the P. jirovecii was found in the HIV-exposed children (10.9% vs. 0.7% HIV unexposed). 7 M. tuberculosis was one of the most common pathogens found at our site and was estimated to cause 12.8% of pneumonia CXR+/HIV-uninfected cases. M. tuberculosis as a common cause of pneumonia is consistent with others 29 who found a 7.5% prevalence of culture-confirmed M. tuberculosis among cases of severe pneumonia. This is also consistent with a recent autopsy study from UTH which found pulmonary tuberculosis in 8% of in-hospital deaths. 30 Tuberculosis is a potentially preventable and treatable disease in children, despite the difficulties in diagnosis, and ongoing efforts for improving diagnosis (use of induced sputum, multiple sputum samples; GeneXpert, Cepheid, Sunnyvale, CA), prevention (isoniazid for exposed infants, for example) and care (long-term appropriate antituberculosis regimens) should be made to improve child survival.
Our in-hospital case fatality ratio (CFR) (14.8%) is the highest among the 9 PERCH sites (11.1%) but close to the Mali site (10.4%). The increased CFR is partly explained by the high prevalence of well-established risk factors for poor outcomes for childhood pneumonia: malnutrition, very severe disease as defined by presence of danger signs, and hypoxia on presentation. Delayed access to care, evidenced by the longer duration of illness among the fatal cases, was also a factor. Other probable contributors to the higher mortality were the limited availability of mechanical ventilation, preponderance of bacterial etiologies and lack of standardized care among the pediatric hospital staff. 28 Nonetheless, with pulse oximetry more widely available, it is important to note that easily assessable information on hypoxia and WHO danger signs on admission are early indicators of children with pneumonia who are at high risk of death, and thus easy areas of focus when trying to improve outcomes.
There were limitations to the PERCH study, such as reliance on samples obtained from outside the lung and challenges distinguishing NP common carriage from infection, as discussed in the all-site paper. 22 Additionally, the PIA model assumes each case's pneumonia episode is caused by a single pathogen and it does not attempt to identify or quantify pathogen combinations. 22 The Zambia site had several other limitations. First, less than half of the cases (CXR+ and CXR normal) were seen at 1 month posthospitalization, which likely underestimated the mortality. Second, our rate of antibiotic pretreatment was high, owing to the near universal practice of giving 1 dose of antibiotic before referral to UTH, which likely diminished the isolation of bacterial pathogens from blood culture. Positive blood culture results are a key input into the PIA, and despite lowering the blood culture sensitivity priors in the PIA for those on antibiotics, the analysis may have underestimated bacterial pathogens. Third, 12% of cases did not have a CXR performed. Timely radiography was a challenge because radiographers were sometimes unavailable due to staffing issues or were located across campus, a particular issue for the sickest children who died shortly after enrolment. Fourth, we did not perform any autopsies on fatal cases at our site. Nonetheless, the distribution of pathogens found in pneumonia deaths in an in-hospital autopsy study conducted during the same period at UTH were similar, 30 where a large proportion of children had histopathologic evidence of bronchopneumonia (50%), mycobacteria (8%) and P. jirovecii (5%). Children enrolled in the autopsy study were slightly older (median age: 19 months) than PERCH cases (median age: 5 months) but had similar comorbidities, including high severe malnutrition and HIV prevalence. Fifth, to improve the specificity of WHO-defined pneumonia that comprised our inclusion criteria, the PERCH study chose to define cases based on a positive CXR for the primary analysis. We may have excluded some cases of true pneumonia among those subjects who died before obtaining a CXR. Finally, we may have omitted cases of true pneumonia that developed a positive CXR after admission or omitted some cases of pneumonia (P. jirovecii) which may not have had any CXR findings.
The results presented here provide a clinical and microbiologic view of HIV-uninfected children with community-acquired pneumonia in an urban setting in a typical sub-Saharan African city. We have purposely omitted the HIV-infected cases (described in detail elsewhere) to provide a sample of children more comparable with the other PERCH sites (except South Africa). In aggregate, our results show a similar etiologic spectrum of pneumonia to the other PERCH sites, albeit a higher mortality. It remains to be determined how much of this increased mortality is due to drug resistance, more virulent organism or health system effects.