Background Previous observational studies have claimed that birth weight and childhood wheezing disorders are associated although the results remained inconsistent. One systematic review and two systematic reviews that included meta-analyses reported inconsistent results. We aimed to conduct a systematic review and meta-analysis to investigate this.
Methods An online search of published papers linking childhood asthma and wheezing disorders with birth weight up to February 2014 was carried out using EMBASE and Medline medical research databases. Summary ORs were estimated using random-effects models. Subgroup meta-analyses were performed to assess the robustness of risk associations and between-study heterogeneity.
Results A total of 37 studies comprising 1 712 737 participants were included in our meta-analysis. The unadjusted summary ORs for risk of childhood wheezing disorders associated with low birth weight (<2.5 kg) were 1.60 (95% CI 1.39 to 1.85, p<0.001) and 1.37 (95% CI 1.05 to 1.79, p=0.02) when compared with ≥2.5 and 2.5–4.0 kg birthweight groups, respectively. The overall summary OR for high birth weight (>4 kg) as compared to the 2.5–4.0 kg birthweight group was 1.02 (95% CI 0.99 to 1.04, p=0.13). There was substantial heterogeneity in the unadjusted low birth weight risk estimates which was not accounted for by predefined study characteristics. There was no significant heterogeneity in the high birth weight risk estimates. There was some evidence of funnel plot asymmetry and small study effects in the low birth weight (2.5 vs ≥2.5 kg and <2.5 vs 2.5–4 kg) OR estimates.
Conclusions Our results suggest that low birth (<2.5 kg) is an independent risk factor for wheezing disorders during childhood and adolescence although there was substantial heterogeneity among the risk estimates. However, we found no significant association of high birth weight with wheezing disorders.
- BIRTH WEIGHT
- META ANALYSIS
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Asthma is described as the most common chronic disease in children.1 According to self-reported symptoms, the prevalence of childhood wheezing disorders have increased markedly in the past and are projected to rise in the current decade worldwide.2 There have also been increases in childhood atopic diseases (eczema and rhinoconjunctivitis) and obesity in recent decades.3 ,4 This may indicate that the increases were mainly due to environmental and life-style changes. Observational epidemiological studies also suggest that childhood wheezing disorders have strong links with viral respiratory infections,5 parental smoking6 and childhood overweight.7 ,8
Low birth weight (<2.5 kg) is the most important factor affecting neonatal and postnatal mortality.9 ,10 Low-birth weight infants are also more likely to develop health problems,11–13 including respiratory disorders, asthma in particular,14 in their childhood and adulthood life. Past epidemiological studies have also reported that there is a link between low birth weight and childhood wheezing disorders, although results remained inconsistent.15 Syntheses of studies have been carried out in the past,16 ,17 however, the results were inconsistent and the methodologies applied by the authors were less rigorous.
In a meta-analysis of nine observational epidemiological studies, it was reported that there was an increase of 20% in childhood asthma risk for high birth weight children.16 However, the studies included in this meta-analysis of high birth weight and childhood asthma used a variety of definitions for high birth weight and risk estimations. One of the studies used 3.8 kg,18 three used 4 kg19–22 and another used 4.5 kg23 as cut-off points, while three others used different birthweight measurements;24–26 four used RR18 ,22 ,23 ,27 and five used OR19–21 ,25 ,26 which could potentially affect the summary risk estimate.
From a meta-analysis of nine studies, Mu et al17 have recently reported that low birth weight increases the risk of asthma by 28% and 34% for studies that used two and three birthweight categories respectively. However, the population's age and birthweight categorisation were not consistent across the studies included. For example, one of the studies used data driven quartile birthweight categories,28 another had a mixture of child and adult populations,21 and three others were treated as adult studies27 ,29 ,30 although the participants were children. And also, one other included study31 used ‘asthma attack’ as an outcome measure for asthma while this may underestimate the true number of cases as many asthmatics may not experience any ‘attack’ at all.
Until February 2014, more than 40 studies that investigated birth weight and childhood wheezing disorders were published. After the recent published meta-analysis,17 five studies that comprised greater than one million children have been carried out;32–36 however, the results remain inconsistent. Therefore, we aimed to provide an up-to-date investigation of the association between birth weight and childhood wheezing disorders through a systematic review and meta-analysis of studies, using consistent exposure (birthweight) and risk estimation definitions, and the standard WHO definition of age range for children and adolescents.37
The review was carried out in accordance with the PRISMA guidelines for systematic reviews and meta-analyses38 while a protocol was registered with PROSPERO.39 An online search was carried out using the EMBASE and MEDLINE databases (see online supplementary table S1). Two authors (TFM and RCP) independently carried out title and abstract reading. Eligible papers were those published as an article, in English, until February 2014 and reported original research on birth weight and wheezing disorders in children 0–19 years of age. Papers were excluded if birth weight was modelled as a continuous variable, no comparison group or risk estimates of birth weight were presented, the study included adult population with no separate data available for children and adolescents (see online supplementary figure S1).
Studies were selected by two independent reviewers who also carried out data extraction. Differences were resolved by consensus. The following study characteristics were extracted: authors, year of publication, country, study design, sample size, study age group, outcome (diagnosis) terms used, exposure (birthweight) categories used, exposure categorisation method, outcome ascertainment and exposure ascertainment.
Exposure variable (birth weight)
Authors of the included studies used four types of exposure categorisation techniques. For comparability and not to lose data due to variation in categorisation methods, standardisation was undertaken: (1) Where authors assumed the CDC40 and ‘recent’ WHO method41 (<2.5 kg=Low, 2.5–4 kg=normal and >4.0 kg=high) or the ‘old’ WHO method (<2.5 kg=low and ≥2.5 kg=normal),10 the reported adjusted risk estimates and data on the number of cases and non-cases of each weight comparison group were combined for meta-analysis without any change (2) Where authors adopted two or three birthweight categories with CDC or WHO ‘normal’ category as a reference and where the number of participants in each categories were available, the stratum based number of cases and non-cases were aggregated before being combined with the other studies for meta-analysis of unadjusted risk estimates. (3) Where authors adopted two or three birthweight categories with the CDC or WHO normal category as a reference and provided adjusted risk estimates, the stratum based risk estimates were aggregated using recommendations from Hamling et al42 before being combined with the other studies for meta-analysis of adjusted risk estimates. (4) Where authors adopted data driven multiple categories that could not be converted to either of the standard formats, the risk estimates were compiled in a table for descriptive analysis.
Study authors used one or multiple outcome terms in their reporting. Again, for comparability among studies, where authors used a single outcome, for example, asthma or wheezing, the quoted outcome term by the author and its risk estimate was assumed for analysis. However, where the authors used multiple outcome terms, a term that was highest in the hierarchy and its risk estimate were assumed for analysis. For example, if asthma and wheezing were used together, asthma was preferred over wheezing.
Papers included in this review were assessed for risks of bias using Newcastle-Ottawa quality assessment scale.43 Two authors carried out assessment of the studies (see online supplementary table S4).
In the meta-analyses of all studies, random effects models were preferred as we made an assumption that the studies were not functionally identical and the aim of our meta-analysis was to generalise about other populations in different parts of the world.44 Estimates were pooled using the DerSimonian and Laird45 method.
If studies presented stratum-specific estimates (eg, by gender), then to provide correct measures of heterogeneity, the risk estimates were aggregated using fixed effect models before being combined with the other studies for meta-analyses of adjusted risk estimates in a random-effects model. Likewise, where authors reported the number of cases and non-cases in each stratum, the total number of cases and non-cases were aggregated before being combined with the other studies for meta-analyses of unadjusted risk estimates of all studies.
To quantify between-study heterogeneity, the Cochrane Q-test46 and the I2 measure of the proportion of the total heterogeneity explained by between study variation47 were used. Subgroup meta-analyses and sensitivity analysis of unadjusted risk estimates were performed on nine covariates (study characteristics) in order to assess the robustness of the risk associations and levels of between-study heterogeneities. To account for the sources of between-study heterogeneity, meta-regression48 of unadjusted risk estimates were performed using Restricted Maximum Likelihood.
In investigating evidence of publication bias and small study effects, symmetry funnel plots and bias test models49 ,50 were used. Five per cent significance levels and 95% CIs were adopted throughout. Meta-analyses were carried out in Stata software V.12.51
A total of 48 studies that reported either the risk estimates or number of cases and non-cases of wheezing disorders in each exposure group were included in the review. The studies were from Europe (48%), Americas (27%), Asia (17%) and Oceania (8%). Thirty-seven of the total 48 studies either used the standard birthweight categories or presented data that were convertible to the standard formats. These studies were included in the quantitative analysis (table 1).
Eleven of the 48 studies used data driven birthweight categories which were found to be inconvertible into the standard formats (see online supplementary table S2). The cut-off point ranges for birthweight categories were: 2.0–3.2, 2.1–3.2 and 3.5–4.5 kg, for the ‘Low’, ‘Normal’, and ‘High’ birthweight categories, respectively.23 ,24 ,52–59 One other study used 2.7 kg as a cut-off point.60 The variation in the cut-off points made it difficult to aggregate these studies for meta-analysis; hence they were only described (see online supplementary table S3).
Quality of studies
With a maximum score of 9 points available for each article, of the 37 included in the meta-analysis: 13 scored >75%, 18 scored 50–75%, and six scored <50% and their risks of biases can be interpreted as ‘low’, ‘moderate’ and ‘high’, respectively. Of the 11 articles included in the descriptive analysis (see online supplementary table S2), 4 scored >75%, 6 scored 50–75%, and 1 scored <50%.
Low birth weight and childhood asthma and wheezing disorders
A total of 30 studies contributed data on the number of cases and non-cases of childhood wheezing disorders that included a total of 1 453 042 children. An overall risk estimate of the studies that compared <2.5 and ≥2.5 kg of birth weight showed that there was a significant increased odds of wheezing disorders (OR: 1.60; 95% CI 1.39 to 1.85, p<0.001) for <2.5 kg of birth weight (figure 1). There was substantial heterogeneity among the studies (I2=82% (95% CI 74% to 88%). A meta-analysis of 11 studies that comprised 105 071 children and provided adjusted ORs for the same birthweight comparison groups also showed an increase of risk by 63% (OR=1.63, 95% CI 1.32 to 2.01, p<0.001) for the <2.5 kg birthweight children (see online supplementary figure S2).
The summary risk estimate of 10 studies that provided data on 2.5–4.0 and <2.5 kg birthweight comparison groups showed that there is 37% associated risk for the <2.5 kg birthweight children (OR=1.37, 95% CI 1.05 to 1.79, p=0.02), and the between-study variation was very high (I2=83%, 95% CI 68% to 89%; figure 2). There was not enough data to carry out meta-analysis of adjusted risk estimates for these birthweight comparison groups—only one study contributed (OR=1.28, 95% CI 0.81 to 2.03, p=0.3)81
High birth weight and childhood asthma and wheezing disorders
A total of 10 studies provided data on the number of cases and non-cases of wheezing disorders on 2.5–4 and >4 kg birthweight comparison groups that comprised a total of 781 928 children (figure 3). The overall OR for >4 kg birth weight on childhood wheezing disorders was 1.02 (95% CI 0.99 to 1.04, p=0.13), which was not significantly different from 1. There was no significant heterogeneity among the studies’ OR estimates (I2=0%; 95% CI 0% to 45%). When further investigated if the non-significant heterogeneity was due to the presence of Kallen et al34 study that has dominated the pooled risk estimate, the summary risk estimate and the level of heterogeneity remained stable (OR=1.03, 95% CI 0.92 to 1.15 ; Q=6 (df=8), p=0.63, I2=0%). There were not enough data to carry out meta-analysis of adjusted risk estimates for these birthweight comparison groups—only one study contributed (OR=0.72, 95% CI 0.42 to 1.23, p=0.23).81
A subgroup meta-analyses of 20 studies that contributed data on wheezing disorder cases and non-cases in the low (<2.5 kg) and normal (≥2.5 kg) birthweight categories showed that the summary risk estimates remained significant in all subgroups of the a priori defined covariates, except if wheezing was used as an outcome term or diagnosis was reported by a parent or the studies were low quality (table 2). When the same analysis was carried out on the studies that reported adjusted ORs for the same birthweight comparison groups, there was no statistically significant risk of association between low birth weight and wheezing disorders if birth weight was extracted from e-records or the study age group were ‘5 years and above’ or the studies were high quality (see online supplementary table S5).
When subgroup analyses of 10 studies that contributed data on the wheezing disorder cases and non-cases in the low (<2.5 kg) and normal (2.5–4 kg) birthweight groups were performed, the results showed inconsistent risk of association across all the predefined study characteristics. For example, there was no significant association between low birth weight and wheezing disorders if studies used asthma as an outcome term or sample size of less than 1000 was used or studies were published before 2000 (table 3).
Subgroup meta-analyses of 10 studies that contributed data on the cases and non-cases of wheezing disorders in the high (>4 kg) and normal (2.5–4 kg) birthweight categories showed that the risk of association was not significant across all categories of the predefined study characteristics and the study quality levels (table 4).
When investigating the sources of between-study heterogeneities of the unadjusted low birthweight ORs, results showed that 59% (p=0.06) of the variance was explained by the a priori selected covariates in the <2.5 and ≥2.5 kg birthweight comparisons (see online supplementary table S6). However, none of the variance was explained by the a priori selected covariates in the <2.5 and 2.5–4 kg birthweight comparisons (see online supplementary table S7).
Investigating biases (small study effects)
The funnel plots showed no evidence of asymmetry for the high (>4 vs 2.5–4 kg) birthweight unadjusted ORs (see online supplementary figure S3c). However, there was some evidence of funnel plot asymmetry for the low birth weight (2.5 vs ≥2.5 kg and <2.5 vs 2.5–4 kg) unadjusted OR estimates (see online supplementary figures S3a,b) and low birth weight (2.5 vs ≥2.5 kg) adjusted OR estimates (see online supplementary figure S3d). This was also reflected in Egger's tests, with no evidence of small-study effects for online supplementary figure S3c (p=0.99), but some evidence of asymmetry in effects for online supplementary figure S3a (p=0.02), online supplementary figure S3b (p=0.02) and online supplementary figure S3d (p=0.02).
In this meta-analysis, we have found that low birth weight was associated with increased risk of childhood wheezing disorders. The risk of association of high birth weight was not statistically significant in contrast to a previous meta-analysis that reported high birth weight was associated with asthma risk (relative risk; RR 1.2, 95% CI 1.1 to 1.3).16 However, it must be noted that the studies included in the previous meta-analysis had used different cut-off points and measurement types for high birth weight, and risk estimation methods (RR and OR).
Our pooled risk estimates for low birth weight are moderately higher than those of a recent meta-analysis by Mu et al17 that reported ORs of 1.28 (95% CI 1.09 to 1.50) and 1.34 (95% CI 1.13 to 1.60) for studies that used two and three birthweight categories, respectively. However, the birthweight categorisation and the age of study population used by the studies in their meta-analysis were not consistent, and the fact that we have included more studies than theirs may have possibly influenced the difference in robustness of the summary risk estimates.
The studies that were not included in our meta-analysis reported inconsistent risk of association for the low birthweight categories (see online supplementary table S3), although we noted that a recent ISAAC phase III study that used similar birthweight categories has reported an OR of 1.20 (95% 1.12 to 1.30).14 However, all the studies reported that there was no risk of association for the high birthweight group which agreed with our findings.
On the basis of our pooled OR results, we noted that the adjusted and unadjusted summary ORs for two birthweight categories were almost identical. This may strongly suggest that low birth weight is an independent risk factor for childhood wheezing disorders although readers have to bear in mind that our analyses also showed some evidence of bias in our funnel plots and Egger's test of bias 49 ,50 which may indicate that there was potential publication bias towards studies that showed no significant risk of association.86
On the basis of subgroup analyses of the adjusted and unadjusted subgroup analyses of two birthweight categories, we observed that the summary ORs were lower for the studies that were published after 2000 than for those published before. It is also known that the prevalence of low birth weight has been falling87 while wheezing disorders have been increasing for the last decades in the developed countries.3 This may imply that the risk of association of low birth weight with wheezing disorders is genuine although there could be an overestimation as ‘wheezing’, which is the key symptom for asthma and wheezing disorders, can also be caused by other illnesses such as pneumonia, bronchiolitis and other viral infections in children under five.88
On the basis of heterogeneity measures (Q-test and I2), we observed that there was a considerable level of between-study variation in the low birthweight unadjusted risk estimates although this could be due to high precision or high sample size studies in our analyses,89 as noted in the forest plot (figure 1). The studies were mostly precise and had narrow CIs. However, there was no significant heterogeneity among the unadjusted risk estimates of high birth weight and asthma and this could be due to having less precise risk estimates with wider CIs, demonstrate by the forest plot (figure 2).
Our work has limitations and results should be interpreted cautiously. First, in our low birth weight and wheezing disorders summary risk estimates, we have found that there was a significant and substantial level of between-study variation that was not explained by our a priori selected covariates. We also had a significant funnel plot asymmetry and small-study effect bias estimate in our results for the studies that compared normal (≥2.5 or 2.5–4 kg) and low (<2.5 kg) birthweight categories. Second, as in any systematic review and meta-analysis, we cannot rule out the possibility of potentially relevant studies being missed by our search strategy. Third, our results are based on epidemiological observational studies and are solely dependent on the quality of the primary studies included. More importantly, we did not identify any studies conducted in developing countries to include in our meta-analysis so our results may not be relevant in regions not represented in this review.
The strength of our work is that we were able to produce consistent risk estimates due to our use of harmonised data. Combining adjusted risk estimates was a primary choice among previous authors. This technique may, however, under or overestimate the association between exposure and outcome variables due to exclusion of studies that used non-standard birthweight categories or combining all irrespective of the type of exposure categorisation method used. In order to improve validity of the summary risk estimates, we implemented data harmonisation techniques and were able to include more studies than if we were to use previous authors’ techniques: most importantly, we were able to produce more consistent summary risk estimates of low (<2.5 kg) and high (>4 kg) birth weight on asthma and wheezing disorders than if we were to combine multiple cut-off points as used by previous authors. The other strength of this work is also that we extracted and analysed adjusted and unadjusted risk estimates, which can be used as an internal validation with each other.
In conclusion, our results suggest that low birth (<2.5 kg) is an independent risk factor for wheezing disorders during childhood and adolescence. However, we found no significant effect of high birth weight on asthma or wheezing disorders.
What is known on this subject
Low birth weight has moderate risk of association with childhood asthma.
Previous meta-analyses reported inconsistent risk of association of high birth weight with childhood asthma.
What this study adds
There is strong risk of association of low birth weight with wheezing disorders.
There is no significant risk of association between high birth weight and childhood asthma and wheezing disorders.
There is significance between-study heterogeneity and small study effect among studies that compared the risk of low birth weight (<2.5 kg) with the normal (≥2.5 or 2.5–4.0 kg) birth weight group.
Contributors TFM, RGF and RCP conceived the idea. TFM performed the literature search. TFM, RGF and RCP contributed to study design, literature search, data extraction and study selection. TFM performed all the statistical analyses, interpretation of results and drafted the manuscript. DCG provided advice on statistical analyses and interpretation of results. DCG, RGF, RCP revised and commented on the manuscript. All authors approved the final version of the manuscript.
Funding TFM is funded by the Hall Dorman Research PhD Studentship.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.