Factors associated with serious outcomes of pneumonia among children in a birth cohort in South Africa

Background Child hospitalization for pneumonia remains common, and pneumonia is a major cause of child mortality. Early identification of clinical factors associated with serious outcomes may help target risk-mitigation strategies. Methods Pneumonia cases occurring in the Drakenstein Child Health Study, a prospective birth cohort outside Cape Town, South Africa were analysed, and factors associated with serious outcomes of pneumonia were identified. Pregnant women were enrolled antenatally, followed through pregnancy, and mother-child pairs from birth to 2 years. Active surveillance for pneumonia was done. Children hospitalized with pneumonia had chest radiography and blood drawn for inflammatory markers; course, outcome and duration of hospitalization were investigated. Serious outcomes were defined as in-hospital mortality or admission to intensive care unit (ICU). Prolonged hospitalization was also explored as a proxy for severity. Features associated with serious outcomes or prolonged hospitalization were analysed using modified Poisson regression. Results Among 1143 live born infants, there were 174 hospitalized pneumonia events in 133 children under 2 years. Three children (1.7%) died, 14 (8%) required ICU admission for respiratory support. In modified Poisson regression, age < 2 months, preterm birth, or hypoxia (oxygen saturation <92%) were significantly associated with serious outcomes. Preterm birth, low birth weight, HIV exposure, stunting, or underweight-for-age (UWFA) were associated with prolonged hospitalization. Chest radiography, elevated C reactive protein, white blood cell and neutrophil counts were not useful to predict death or ICU admission in children hospitalized with pneumonia. Conclusions In this cohort, death from pneumonia was rare, but clinical features associated with serious outcomes and prolonged hospitalization were identified. These may help with risk stratification, to identify children who may benefit from enhanced monitoring or earlier escalation to respiratory support.

Introduction ICU admission in this cohort. In addition, we investigate factors associated with prolonged hospitalization for pneumonia in this birth cohort.

Study design
The DCHS, a population-based birth cohort established in 2012 in a peri-urban community outside Cape Town, South Africa, has been described in detail [11,12]. The population is of low socio-economic status (SES); unemployment and maternal HIV infection are common. Pregnant women (20-28 weeks' gestation) were recruited and followed through pregnancy and delivery; mother-child pairs were followed through early childhood. For this study, follow-up through the first 2 years of life is included. University of Cape Town Faculty of Health Sciences Human Research Ethics Committee (HREC numbers 401/2009; 651/2013) provided ethical approval. Mothers provided written informed consent at enrolment, with annual renewal.

Study procedures
Trained study staff administered structured interviews and performed home visits antenatally and postnatally. Indoor air particulate matter of diameter 10μm or less (PM 10 ) was measured using an air sampling pump (AirChek 52; SKC, Eighty Four, PA, USA), placed in homes antenatally for 24 hours; the ambient standard was defined as 40 μg/m 3 [13]. Diffusion tubes, placed in homes for 2 weeks, measured toluene concentration (Markes thermal desorption tubes; Llantrisant, UK), as described previously [14]. Maternal smoking was measured by urine cotinine measures collected antenatally and at delivery; the higher measurement was used to assign smoking status, as described previously (non smoker/passive smoker <500ng/ml; active smoker > 500 ng/ml) [14].
Pneumonia surveillance was performed using active case finding: mothers were interviewed frequently (every 2 weeks in the first year, every 3 months in the second year) about respiratory symptoms (cough, fast breathing, difficulty breathing). Thus it was possible to retrospectively identify pneumonia events occurring at other facilities or outside the area; where possible, details of these pneumonia events were obtained from clinical records of other facilities [15]. Birth cohort participants presenting to clinics or primary care providers with cough, tachypnoea or difficulty breathing were referred to the research study for assessment in real time. Trained research staff confirmed tachypnoea and clinical signs. Pneumonia was diagnosed according to revised WHO guidelines when a child had cough or difficulty breathing, with either lower chest wall indrawing (LCWI) or age-appropriate tachypnoea (� 50 breaths per minute if 2-12 months, �40 breaths per minute if �12 months). Severe pneumonia was diagnosed if children were under 2 months old; or at any age, with a general danger sign (cyanosis, decreased level of consciousness, inability to feed, vomiting everything, seizures) [4]. In children with wheezing, pneumonia was not diagnosed if tachypnoea and LCWI resolved following a trial of inhaled bronchodilator (5 puffs of salbutamol via a metered dose inhaler with mask/spacer). Only community-acquired cases were included. Nosocomial pneumonia (new onset respiratory symptoms occurring after hospital admission) and congenital pneumonias (tachypnoea present on the first day of life) were excluded. Pneumonia treatment and decision to admit were at the discretion of treating doctors, not study staff. Indications for hospitalization included hypoxia (oxygen saturation <92% in room air) or inability to take oral medication. Children with severe pneumonia were admitted at Paarl hospital, the single regional hospital; children requiring intensive care unit (ICU) were referred to the affiliated unit 47km away. Children were transferred to ICU for ventilatory support if they remained hypoxic (oxygen saturation <92%) despite oxygen supplementation via nasal prongs or face mask; or for inotropic support if shock was refractory to intravenous fluid therapy.
Age at pneumonia was categorized as under 2 months, 2 to 12 months or over 12 months [4]. Within each age group, duration of hospitalization was calculated; duration >90 th centile for age group was considered "prolonged hospitalization". Events occurring 14 days following pneumonia were considered new-onset pneumonia. All children had anthropometry performed; stunting was defined as length-for-age <-2 Z score; and wasting as weight-for-length <-2 Z score on WHO growth charts. Severe tachypnoea or severe tachycardia were defined as respiratory rate or heart rate >99 th centile of normal for age [16].
Blood was drawn for full blood count, differential and C-reactive protein (CRP). CRP was modelled both as continuous and dichotomized at �40mg/L, as this has previously been explored as a proxy to distinguish bacterial from viral pneumonia [17]. Chest radiograph taken at the time of hospitalization was read by 2 independent doctors trained in the WHO standardized chest radiograph reporting methodology [18]. Radiographs were classified as no consolidation, "other" consolidation, or "primary end point consolidation" (PEPC). A third reader, blinded to readings of the others, resolved discordant results. Sensitivity of CRP� 40mg/L to predict PEPC was calculated as "true positives" (children with PEPC and CRP�40mg/L) divided by all those with "true disease" (all those with PEPC).
Usual care was provided at one of 2 local clinics including the national vaccination schedule: Bacille Calmette Guerin (BCG) at birth; 13 valent pneumococcal conjugate vaccine (PCV) at 6, 14 weeks and 9 months; hexavalent vaccine (diphtheria, tetanus, acellular pertussis, hepatitis B, polio and Hib at 6, 10, 14 weeks and 18 months); and measles vaccine at 9 and 18 months. HIV-exposed infants were tested with HIV-DNA PCR (Cobas Ampliprep/Cobas Taqman HIV-1, Roche Diagnostics Systems, Inc., Branchburg, NJ) between 6 and 10 weeks with repeated testing during hospitalization. Testing was repeated with age-appropriate tests at 9 and 18 months and after breastfeeding cessation according to local guidelines [19]. Co-trimoxazole prophylaxis was provided to all HIV-exposed infants from 6 weeks until breastfeeding cessation.

Statistical analysis
Continuous data were compared with median (inter-quartile range, IQR) and Wilcoxon ranksum test. Factors associated with death/ICU admission or with prolonged hospitalization were analysed with prevalence ratios from modified Poisson regression with robust variance estimation; results of regression are presented as risk ratios and 95% confidence intervals (CI), as these are equivalent to prevalence ratios [20,21]. Robust standard errors were calculated to account for clusters within individuals. Data were analysed using Stata version 11 (College Station, Texas, USA).
Of the 174 hospitalized pneumonia cases, there were 3 deaths (1.7%); 14 (8%) were admitted to ICU for respiratory support: 8 required intubation and mechanical ventilation and 6 received continuous positive airway pressure (CPAP), Table 2. One child died in ICU and 2 died prior to ICU transfer.

Fatal pneumonia or ICU admission
Fatal pneumonia or ICU admission were strongly associated with younger age: 2 of the 3 deaths and 8 of the 14 ICU admissions occurred in children under 2 months. In unadjusted modified Poisson regression, age <2 months, preterm birth, and hypoxia were significantly associated with serious outcomes, Table 3. WHO danger signs of decreased level of consciousness and inability to feed, although rarely observed, were strongly associated with serious outcomes. Low birth weight, maternal smoking, UWFA, stunting, grunting, radiological consolidation (PEPC), and elevated CRP appeared to be associated with increased risk of serious outcomes, but precision for these estimates was limited. Female sex, wasting, duration of illness prior to presentation, and cough not observed were not associated with serious outcomes. Several factors were identified which, although previously associated with pneumonia incidence, were not associated with serious outcomes: male sex, household crowding, HIV exposure, incomplete vaccinations, and inadequate breastfeeding. Children exposed to significant household air pollution, measured by elevated antenatal PM 10 concentration, had a 3-fold increased risk of a serious outcome, but precision for this estimate was limited by small sample size (OR 3.17, 95% CI 0.38-26.46).  Table 3. Prolonged hospitalization was associated preterm birth, low birth weight, HIV-exposure, never breastfeeding, decreased level of consciousness, UWFA and stunting. Hypoxia, lower chest indrawing and other WHO danger signs were not associated with increased risk of prolonged hospitalization, Table 3. PEPC on chest radiograph, elevated CRP, total white cell and neutrophil counts were also associated with prolonged hospitalization.

Discussion
In this well-vaccinated birth cohort with negligible HIV infection and reasonable nutritional status, 1.7% of hospitalized pneumonia cases died and 8% required ICU admission. Good access to care and referral may have also contributed to low pneumonia mortality. Young age < 2 months, hypoxia, or preterm birth were strongly associated with death or ICU admission; while low birth weight, stunting, UWFA, HIV exposure, and WHO-defined radiographic consolidation were strongly associated with prolonged hospitalization. Hypoxia has been considered the most important feature of pneumonia severity [5,[22][23][24]. However, in our cohort, peripheral oxygen saturation <92% was a reason to consider hospitalization, and was very common, occurring in 33% of hospitalized cases. Hypoxia was associated with serious outcomes, but was so common in this cohort that it did not have good discriminatory value among hospitalized children to identify those who require intensive monitoring or additional respiratory support. There are no reliable clinical signs to predict hypoxia in children [25], which makes pulse oximetry screening and oxygen supply essential for all facilities treating children [26].
Preterm delivery confers multiple risks for childhood morbidity, including less transplacental antibody transfer, anaemia, suboptimal breastfeeding, delayed/missing vaccinations, and slower growth [27,28]. Preterm birth is useful as a risk stratification marker where gestational age is accurately known; front-line health care workers can quickly identify at-risk children from patient-held birth records, and prioritise them for urgent assessment and pulse oximetry testing.
The original definition of WHO standardized radiographic consolidation was intended as a proxy marker of bacterial pneumonia for use in vaccine studies [18], and was strongly associated with mortality [29]. However, there is increasing evidence that much severe pneumonia post PCV and Hib is viral: 33% of pneumonia hospitalizations in this cohort were associated with RSV [10], and in the PERCH study, viruses (especially RSV) were predominant causes of hospitalized pneumonia among children with positive chest radiology [30]. In this well-vaccinated cohort, neither radiological consolidation nor elevated CRP were significantly associated with death/ICU admission. Furthermore, although PEPC was associated with higher median CRP, there was no CRP threshold that strongly predicted PEPC. It is increasingly accepted that CRP is a poor marker of bacterial infection [31,32]. This is consistent with the epidemiologic change that has occurred since WHO radiographic scoring was developed [33]; so PEPC or elevated CRP should no longer be considered a proxy for bacterial infection. Since neither radiological consolidation nor elevated CRP strongly predicted death or ICU admission, and are not widely available in LMIC, they could be excluded from severity scores in favour of more easily-obtainable clinical features.
A recent systematic review of pneumonia risk stratification scores for children in lowresource settings also described hypoxia, LCWI and decreased level of consciousness as strongly predictive for mortality, but that wheeze was protective [7]. A pneumonia risk stratification prediction model developed and validated in a high-income country reported retractions were associated with severe outcomes, but did not report prevalence of wheeze [6]. In our analysis, both wheezing and LCWI were common, but not associated with serious outcomes. This is consistent with evidence that led to WHO reclassification of LCWI as a sign of pneumonia for ambulatory therapy, not a marker of severe pneumonia requiring parenteral therapy [4]; in these young children, where oxygen saturation monitoring was available, LCWI indrawing did not add value in identifying those at risk of severe outcomes [34].
The PERCH severity score identified infants < 1 year, female sex, decreased level of consciousness, fast breathing without observed cough, grunting, hypoxaemia (peripheral oxygen saturation less than 90%), duration of illness more than 3 days and severe wasting as risk factors for mortality [8]. The score had a moderate overall predictive accuracy (C statistic 0.76% in a validation data set), but lower predictive accuracy in some age groups (eg age 6-11 months, where mortality was highest) and some sites, including South Africa, where mortality was lowest. The authors recommended that the PERCH score should be validated with other datasets. In our cohort, some of the PERCH predictors were confirmed: young age and hypoxia were strongly associated with serious outcomes. Decreased level of consciousness was infrequently identified, limiting its practical usefulness as a predictor of serious outcomes. Other PERCH predictors performed poorly in our study: female sex, grunting, cough not observed, and longer duration of illness were not associated with serious outcomes; no wasted child died or required ICU admission. Furthermore, the PERCH tool may be impractical for frontline clinicians in LMIC, as it requires careful history from the mother or caregiver, and accurate plotting of weight-for-length, which is often poorly recorded in critically ill children.
A strength of this study is the availability of detailed, longitudinal exposure history prior to pneumonia. Certain environmental factors, measured before the pneumonia event (antenatal maternal smoking, household crowding) were associated with hospitalization for pneumonia [9], but were not associated with serious outcomes. This suggests that while maternal smoking and household crowding were significant risk factors in the causal pathway for developing pneumonia, they may not be associated with severity; or may have been too common in this cohort to predict serious outcomes with adequate discrimination. Although elevated household PM 10 level was not significantly associated with risk of serious outcomes, precision for this estimate may have been limited by sample size. Indoor air pollution plays a critical role in child lung health [14]; this could be investigated as modifiable factor in adequately-powered prospective interventions.
Prolonged hospitalization may be used as a valid measure of pneumonia severity [35]; however, in children with complex co-morbidities, duration of hospitalization may reflect the underlying medical or social problems, and not the severity of the respiratory infection. Preterm birth, HIV exposure, stunting, underweight-for-age, and elevated CRP were significantly associated with prolonged hospitalization. We cannot attribute causality to these associated factors, as reasons for prolonged hospitalization may have been due to the severity of the respiratory infection, but also may have been due to other associated illnesses, nutritional rehabilitation or addressing the social concerns or feeding practices. We previously showed that HEU children have increased incidence of pneumonia in the first 6 months of life [9]; in this analysis we show that they are not at greater risk of serious outcomes, although they are at increased risk of prolonged hospitalization.
Limitations of this study include inadequate power to detect an association, as both death and ICU admission were such rare events; even when they are combined as a composite endpoint of "death or ICU admission", power is limited. Furthermore, use of a composite endpoint requires caution, as some clinical features may reflect the bias of the treating clinician, eg clinicians could be more likely to refer a low birth weight child to ICU, even if the severity of the disease did not require respiratory support. However, this is unlikely as the ICU in the regional referral hospital was 47km away, is a very limited resource, and all children admitted to ICU received invasive or non-invasive ventilation. Some authors consider non-ventilated ICU admissions "moderate" severity, vs death/ventilation ("severe") [5]. As we had few events with the composite outcome, we did not analyse "moderate" outcomes separately, but considered all ICU admissions as serious. Prediction models should be derived from one dataset and then applied to a different dataset, or validated with internal bootstrap [8,36,37]. With this small dataset, we were unable to derive a true prediction model; but results compare favourably with predictors from other local and international models. Pneumonia surveillance was strong and consistent throughout the study period; follow up of individuals and overall cohort retention was good, so missed pneumonia hospitalizations or deaths are unlikely. Other missing data, particularly CRP and white cell counts, would not have made substantial impact on outcomes, as most cases with serious outcomes had all blood results available.
In conclusion, we observed low mortality from pneumonia in this cohort. Most of the PERCH severity predictors performed poorly in this cohort. Clinical factors associated with death or ICU admission included age under 2 months, preterm birth or hypoxia. These clinical measures are easily obtained, do not require radiography, blood tests or interpretation of growth charts, and further underscore the urgent need for more widespread availability of pulse oximeters and oxygen in LMICs. Presence of these risk factors should alert clinicians to identify children who may require additional monitoring or early escalation of care; clinicians should be encouraged to engage in appropriate risk-stratification to identify at-risk children.