Bronchiolitis

We systematically reviewed all literature published from January 1, 1990 through March 31, 2012 to identify studies with data on risk factors for pediatric pneumonia. We searched a variety of databases-Medline (Ovid), Embase, CINAHL and Global Health Library using combinations of key search terms: pneumonia, low birth weight, undernutrition, breast feeding, crowding, smoking, indoor air pollution, immunization, HIV etc. (full search terms are available in Supplementary material). Hand searching of online journals was also performed by examining the reference lists for relevant articles. We did not apply any language or publication restrictions. Relevant full-text articles in foreign language were translated to English using Google translator.
We defined an episode of severe pneumonia in hospital setting as any child hospitalized overnight with an admission diagnosis of pneumonia or bronchiolitis. In community-based studies, the presence of lower chest wall indrawing in a child with cough and difficulty breathing with increased respiratory rate for age was used to define a case, using the same cut off values as in the WHO's case definition (4 ,5 ). We recognized that the eligible studies used varying case definitions for the putative risk factors. We therefore grouped the risk factor definitions into categories and analyzed the association between risk factor and outcome for each of these categories (Table 1). We classified the risk factors into three groups based on the consistency and strength of association with severe ALRI:
(i) those that consistently (ie, across all identified studies) demonstrated an association with severe ALRI, with a significant meta-estimate of the odds ratio, would be classified as “definite”;
(ii) those demonstrating an association in the majority (ie, in more than 50%) of studies, with a meta-estimate of the odds ratio that was not significant, would be classified as “likely;” and
(iii) those that were sporadically (ie, occasionally) reported as being associated with severe ALRI in some contexts were classified as “possible.” This classification is consistent with the one originally used by Rudan et al (2 (link)).
We included studies that reported severe pneumonia in children under five years of age (Table 2). Eligible study designs included randomized control trials (RCTs), observational studies (cohort, case-control, or cross-sectional) that assessed the relationship between severe pneumonia in children and any one of the putative risk factors. Studies were excluded if their sample size was less than 100 detected cases, if their case definitions did not meet our broad range of case definitions, or if the case definitions were not stated clearly and/or not consistently applied (Figure 1). Studies where health care workers went house to house to identify cases of pneumonia were considered as having active community-based case ascertainment. By contrast, studies where children with pneumonia presented to a health facility were considered as having passive hospital-based case ascertainment.
The included studies used either multivariate or univariate analyses to report the association between the putative risk factors and the outcome, ie severe pneumonia. Since the multivariate design takes into account the interaction with other risk factors and potential confounders, we decided to report the results of the meta-analysis of these data separately. We decided that if there was significant heterogeneity in the data, ie, I²>80%, (corresponding to P < 0.005) (6 (link)), then we would report the meta-estimates from the random effects model (7 (link)). Importantly, we hypothesized that the effects of the risk factors were likely to be different in developing countries and industrialized countries. Because of this, we decided to report the results separately for developing (Table 3) and industrialized countries (Table 4). We extracted all relevant information from each retained study (Supplementary Table S2(supplementary Table 2)) and assessed the quality of included studies using a modified GRADE scoring system (Supplementary Tables S1(supplementary Table 1)) (8 (link)). Briefly, we assessed each article against the GRADE criteria and calculated the overall score for each article. We then calculated the cumulative score for each risk factor after accounting for the included studies (Supplementary Table S3(supplementary Table 3)). We used Stata 11.2 (StataCorp, College Station, TX, USA) for the meta-analysis (Figure 2; Supplementary Figure S1(supplementary Figure 1)).

To estimate the proportion of influenza-attributable SARI, a periodic regression method was applied to the weekly number of hospitalizations for pneumonia and bronchitis [7 (link)]. We modelled the baseline level with a multivariable linear regression model using the following equation:
where y_w is, for week w, the number of pneumonia and bronchitis hospitalizations.
The baseline level in a given week w was obtained by fitting the model to the observations from 1/07/2012 to 30/06/2018 from which we removed the presence of influenza and bronchiolitis epidemics (Additional file 1) in the training dataset from the series. This trimming was an attempt to ensure that the dataset used to model the baseline level was free of influenza or respiratory syncytial virus (RSV) [8 (link)].
The number of excess hospitalizations for pneumonia or bronchitis was defined as the sum of the differences between the expected and observed values. This excess was considered attributable to influenza during periods defined as influenza epidemics.