Effect of low birth weight on childhood asthma: a meta-analysis

We performed a systematic literature search of the PubMed database from 1966 to November 2013 using the following search command: birth weight and “asthma or wheezing or wheeze”. The abstracts identified by this search were screened to eliminate the obviously irrelevant abstracts (such as reviews, systemic reviews, and case reports). The criteria for paper inclusion were as follows: (i) case–control or cohort studies that focused on the association of low birth weight with childhood asthma (less than 16 years old); (ii) studies that presented the odds ratio (OR) or risk ratio (RR) estimates with the corresponding 95% confidence intervals (CIs), or sufficient data for calculation; (iii) normal birth weight as the reference category; and (iv) studies published in English up to October 2013. All publications providing sufficient information regarding the relation between low birth weight and childhood asthma were included, irrespective of whether this issue was their primary or secondary objective. We excluded studies that used asthma-like symptoms or other manifestations of respiratory impairment as an outcome, such as wheezing.

The outcome of interest was childhood asthma. The definition of asthma was based on a physician’s diagnosis, or a history of asthma reported by their parents. Our primary predictor variable was low birth weight, which was defined as a birth weight of less than 2500 g. For any study to be included in the present study, its outcome and predictor definitions had to be consistent or adaptable with our definitions. Both outcome and predictor variables were dichotomous.

Four investigators independently carried out data extraction of the following items: authors, publication year, study design, country, population age, diagnosis of asthma, adjustment for confounding (family history, air pollution, sex, gestation age, smoking, and caesarean etc.), and effect size and corresponding estimates with 95% CIs. Two reviewers completed the quality assessment independently. A set of structured criteria modified from previous studies (Newcastle–Ottawa scale for cohort study) were used to complete the quality assessment of publications. A higher score indicates higher quality. In the case of disagreement of extracted data, discrepancies were resolved by discussion and achieved consensus.

We abstracted the multivariate-adjusted risk estimates (OR or RR). The unadjusted risk estimates were calculated using original data when the adjusted was unavailable. The CIs were transformed to a log scale and then the standard error was calculated. A pooled summary relative risk was calculated according to the fixed-effects model and a random-effects model. Statistical heterogeneity was assessed with the I-squared (I²) value, which represents the percentage of total variation across different studies due to heterogeneity rather than chance. I² values of 25%, 50%, and 75% have been related to low, moderate, and high heterogeneity, respectively [10]. A random-effects model was applied when there was notable heterogeneity (I² index ≥50%), and otherwise, the fixed-effects model was used. Subgroup analyses were stratified by geographic area and adjusted for the major confounder. Meta-regression analysis was used to investigate the potential sources of heterogeneity. Cumulative meta-analysis in the order of publication year was conducted to check homogeneity of the results. Publication bias was assessed by using the funnel plot, the rank correlation test, and the graphical test [11‐13]. All statistical analyses were carried out by STATA version 12.0 (Stata Corp., College Station, TX).

A description of the search process is shown in Figure 1. Of the 692 papers found by the search, 13 cohort studies met the inclusion criteria and were included [14‐26]. Among these, only five studies provided the adjusted effect estimates of low birth weight versus childhood asthma [14‐16, 22, 23]. Of the 13 cohort studies selected, 11 were published in 2000–2013, and two were published in 1997 and 1998. The majority of the cohorts (n = 6) were in Europe (Denmark, Finland, Germany, Sweden, and the United Kingdom) [14, 16, 20, 21, 24, 26]. Four cohorts were studied in North America (US and Canada) [19, 22, 23, 25], one in Oceania (New Zealand) [17], one in South America (Brazil) [15], and one in Asia (Taiwan) [18]. Detailed characteristics of the studies that were included are shown in Table 1.

Eleven studies used the definition of low birth weight compatible with our criteria (<2500 g). Birth weight in the other two studies was less than 3000 g [16, 18]. General variance-based methods using CIs were applied to combine the results according to birth weight. The definition of asthma was based on a physician’s diagnosis. Children with wheeze or wheezing were excluded.

Two studies reported data for the effect of low birth weight on the risk of future asthma as adjusted ORs [14, 16]. These studies used multivariate logistic regression to obtain results. The other three studies presented adjusted RRs. The rest of the eight studies presented prevalence based on birth weight. Prevalence was converted into crude RRs, and ORs and RRs were treated as comparable measures of effect (RR). The ranges of ORs and RRs varied from 0.9 to 2.6 (Table 1).

Five of the 13 estimates were directly from the original studies, and eight were transformed from the information available to comply with our categorization used in the meta-analysis. Figures 2 and 3 show forest plots, which provide study-specific and pooled RRs (95% CIs), of development of asthma for low birth weight. When compared with normal birth weight, the pooled RRs from the fixed-effects and random-effects models were 1.162 (95% CI, 1.128–1.197) and 1.152 (95%, 1.082–1.222) for low birth weight (I² = 40.2%, P = 0.066), respectively. Both models showed similar summary effect estimates, indicating a significantly increased risk of asthma in relation to low birth weight. Table 2 shows the pooled RRs (95% CIs) among different subgroups stratified by five main characteristics of the studies. The subgroup analyses showed an even greater effect estimate in the six studies conducted in Europe compared with other geographic areas. Additionally, subgroup analyses showed a greater effect estimate in the eight studies with a small sample size (<5000 subjects) compared with those with a larger sample size, and a greater effect estimate in the seven recently published studies compared with those in the older six studies. The effect estimates stratified by study population age and adjusted RRs were similar to the overall pooled RRs.

The I² value representing the overall between-study heterogeneity was 40.2% (P = 0.066), which was moderate and acceptable in the present meta-analysis. Clinical heterogeneity among subgroup studies is shown in Table 2. Additionally, statistical heterogeneity for a greater effect estimate from subgroups stratified by geographic area, population size, and publication year was low, varying from 0% to 21.7%. A meta-regression model was used to assess potential determinants of heterogeneity. Five factors (geographic area, population size, publication year, adjusted RR, and study population age) showed no statistical significance in either univariate analysis or the multivariate model (P >0.05, Table 3).Cumulative meta-analyses of all of the studies showed that the estimates gradually became consistent, and the corresponding CIs narrowed down in the order of publication year (Figure 4). We also performed sensitivity analysis, and the pooled results did not appear to change.

We assessed the potential publication bias by using a funnel plot (Figure 5). The funnel plots did not show any remarkable asymmetry. Additionally, Egger’s and Begg’s tests showed no evident publication bias. P values for the two tests were 0.653 and 0.951, respectively.