Preterm Birth and Childhood Wheezing Disorders: A Systematic Review and Meta-Analysis

1The impact of early life exposures on subsequent health and disease is increasingly recognised. Preterm birth is a common early life event, the adverse consequences of which can affect the entire life course. Worldwide, over 11% of babies are born preterm, a number that continues to rise in most regions. Prevention of preterm birth is currently limited by a knowledge gap regarding the mechanisms of normal and pathological labour onset. Given the increasing burden of preterm birth and the restricted scope for prevention, a focus on appreciating and containing its consequences is warranted.

2Respiratory distress syndrome is the most obvious direct manifestation of immaturity in preterm newborns. The combination of structural lung immaturity and pulmonary surfactant deficiency results in regional atelectasis and a varying degree of respiratory compromise. Ventilatory support and additional oxygen supplementation are common necessities, and many very preterm babies go on to develop chronic lung disease (bronchopulmonary dysplasia [BPD]). The majority of preterm babies are, however, born close to term, when respiratory compromise is a less common event. Whereas late preterm babies have long been regarded a normal variation of term babies, it is increasingly recognised that they are at risk of a range of adverse outcomes, including respiratory disease.

3Accumulating evidence now implicates preterm birth in the development of respiratory disease in later life. Small airway obstruction is evident through decreased forced expiratory volume in 1 s (FEV1) in children and adults born preterm, and a hypothesized link with chronic obstructive pulmonary disease in adulthood was recently confirmed. Along this line of evidence, a meta-analysis published in 2006 identified preterm birth as a risk factor for asthma in a mixed paediatric and adult population. Asthma is the most common chronic disease affecting children, and its link with preterm birth is of significant public health relevance given the increasing incidence of both entities,. Of note, the majority of cohorts aggregated in this meta-analysis were born before the 1990s. Important changes in neonatal clinical management have been introduced since, including increased use of antenatal steroids and postnatal surfactant, and a shift towards less aggressive respiratory management. These changes have selectively impacted the survival of extremely preterm infants, and shifted the occurrence of acute respiratory problems towards lower gestational age groups. Associated changes in long-term pulmonary outcomes may have occurred, and there is therefore a need for a comprehensive, rigorous assessment of the contemporaneous evidence base.

4We aimed to investigate the association between preterm birth and childhood wheezing disorders through a systematic review and meta-analysis of studies with populations born from the 1990s onwards. We focused on wheezing disorders in general rather than just asthma, given the difficulty in differentiating between the two, particularly among young children. Unlike previous work, we also considered the potential impact of confounding factors and explored the association by degree of prematurity. Based on aggregated association measures, we have furthermore estimated population-attributable risks (PARs) to quantify associated disease burden.

5This review was performed following the methods detailed in a systematic review protocol registered with PROSPERO (CRD42013004965; Text S1). Online databases were searched independently by two authors using the following search terms: PubMed and Embase: (preterm OR prematur*) AND (asthma OR wheez*); Google Scholar: asthma, wheeze, wheezing, preterm, premature, child, children, childhood; World Health Organization Global Health Library, World Health Organization Library Information System, SciELO, and Trip: ((preterm OR premature OR preterms OR prematurity) AND (asthma OR wheeze OR wheezing)). The searches covered the period from 1 January 1995 to 23 September 2013. No language restrictions were applied. Additional studies were identified by screening reference lists of articles of interest and tracing citations of articles through ISI Web of Knowledge. We asked an international panel of experts in the field (Text S2) to report any additional published, unpublished, or in progress studies that might have been missed.

6Epidemiological studies were eligible for inclusion if they reported an association between preterm birth (<37 wk versus ≥37 wk gestation [term]) and asthma or wheezing in children (aged 0.5 to 18 y). Studies reporting on populations that included children born before 1995 were eligible for inclusion only when at least 50% of the cohort was born from 1995 onwards and none of the children was born before 1990, so as to primarily include studies conducted in the period following the important changes in neonatal practice outlined earlier. When studies with overlapping data were identified, the most informative study was selected for inclusion in this report; the determination of which study was considered most informative was based on consensus, study size being an important determinant. Final study selection was based on a consensus decision between reviewers, with arbitration by a third reviewer in case of disagreement.

7Data were independently extracted from eligible studies by two reviewers (J. V. B. and M. J. L.); disagreements were resolved through discussion, with arbitration by a third reviewer if necessary (A. S.). The following study characteristics were extracted: authors, full reference, study design, location, sample size, inclusion and exclusion criteria, age range and birth year of study participants, method of ascertainment of exposure and outcomes, and outcome measure.

8Post-term (>42 wk gestation) births were excluded from the analyses, if possible. If studies reported follow-up at different time points, the most recent was selected. Similarly to previous work, we used only one outcome measure from each study in the analyses; for studies with multiple outcome measures, the following hierarchy was used to select the outcome measure used in the analyses, from most preferred to least preferred: asthma, persistent wheezing, recurrent wheezing, severe wheezing, and wheezing. Furthermore, we used the following hierarchy to select the highest ascertainment level if the outcome was measured multiple ways: clinician diagnosis, documented medication use as a wheezing disorder proxy, routinely collected health-care data, parent- or patient-reported clinician diagnosis, parent- or patient-reported medication use, and parent- or patient-reported symptoms. “Ever” wheezing or asthma was favoured over recent and current wheezing or asthma.

9Absolute patient counts were extracted to compose 2×2 tables according to preterm birth and wheezing disorder status, and these were used to calculate univariate odds ratios (ORs); in instances where relevant crude data were missing, we approached authors to request these. If needed, count data were calculated from provided percentages, and these were then rounded off to the nearest integer. Adjusted association measures were extracted from the most adjusted model presented.

10Study quality was assessed independently by two investigators using the Effective Public Health Practice Project quality assessment tool for quantitative studies. Sex, maternal smoking during pregnancy, and maternal atopy or asthma, or a family history of atopy or asthma, were deemed the most important confounders. Any disagreement was resolved by consensus and arbitration by a third reviewer, where necessary.

11Univariate association measures were pooled using Mantel-Haenszel analysis. Random-effects models were applied because of anticipated heterogeneity given between-country variation in clinical practice and differences in inclusion and exclusion criteria, outcome specification, and ascertainment level of exposure and outcome. Adjusted association measures were pooled via random-effects generic inverse variance analysis. Standard errors for study point estimates were calculated from the respective 95% confidence intervals as described in the Cochrane Handbook.

12For studies reporting adjusted association measures according to multiple gestational age strata, the least preterm stratum was selected to obtain the most conservative estimate. In a separate analysis, a linear association between gestational age at birth and wheezing disorder risk was investigated, pooling individual studies by random-effects generic inverse variance analysis. For this purpose, adjusted association measures for multiple gestational age strata were aggregated within studies using a fixed-effects log-linear dose–response regression model. The Q-statistic and I 2-test were used to assess heterogeneity among studies. Small-study effects were assessed using funnel plots and Harbord's modified regression test for unadjusted data and Egger's regression test for adjusted association measures.

13Subgroup analyses were performed according to age group (<5 y versus ≥5 y) and degree of prematurity (<32 wk versus 32–36 wk). When studies reported the association between preterm birth and wheezing disorders for both subgroups (i.e., follow-up at <5 y and at ≥5 y, and/or reported for <32 wk and 32–36 wk gestation), the corresponding association measures were included in both sides of the comparison.

14Sensitivity analyses were performed according to risk of bias (low, moderate, or high), study size (n<10,000 versus n≥10,000), and outcome definition (asthma versus wheezing) and ascertainment (clinician diagnosis and/or medication use versus parent-reported outcomes).

15Meta-regression analysis was performed to assess the independent effects of study size, mean age, wheezing type (wheezing versus asthma), diagnosis ascertainment, and publication year on the association between preterm birth and wheezing disorders.

16PAR was calculated using the following formula:where P e is the exposure prevalence and RRe is the relative risk due to exposure. RRe was calculated as follows:where PEER is the patient expected event rate. As estimations of PAR based on unadjusted association measures can be biased by confounding of the exposure–outcome relationship, we additionally calculated PAR using aggregated adjusted association measures via the following formula:where pd is the proportion of cases exposed.

17Analyses were performed using Stata 12 (StataCorp).

18The funders had no role in study design, collection, analysis, interpretation of the data, writing of the report, or decision to submit the work for publication.

19Forty-two eligible studies were identified –. Study selection is summarised in Figure 1. Five studies from Sweden,,,, and two from northern California, US, had significant population overlap,. Furthermore, several eligible studies described the same cohort: the Epicure cohort, UK and Ireland (n = 4),,,; the Boston Birth Cohort, Boston, US (n = 3),,; a birth cohort from southern New England, US (n = 2),; and cross-sectional surveys from Liverpool, UK (n = 2),. The most relevant study was selected from each set of overlapping studies. Characteristics of the 12 studies that were excluded, as well as reasons for exclusion, are given in Table S1. Meta-analysis was performed on the remaining 30 unique studies (Tables 1 and 2), which reported on a total of 1,543,639 individuals,–,–,,,–,,–,–.

20Four studies were deemed to have low risk of bias, 12 had moderate risk of bias, and 14 had high risk of bias (Table S2). Studies with high risk of bias were generally smaller (median size 756; range 90–7,925) than those with moderate risk (median size 10,246; range 560–397,852) and low risk (median size 6,992; range 1,448–708,907).

21Crude data were provided by authors of five studies,,,,, and were calculated for seven additional studies,–,,,. Crude data could not be retrieved for four studies, and these were included in the adjusted meta-analysis only,,,. Pooling of the 26 studies (1,500,916 individuals) for which unadjusted data were available showed a significant association between preterm birth and wheezing disorders (OR 1.71, 95% CI 1.57–1.87, p<0.001; Figure 2). Out of 93,616 children born preterm, 12,858 (13.7%) were diagnosed with a wheezing disorder, compared to 116,732 out of 1,407,300 children born at term (8.3%).

22Seventeen studies provided adjusted association measures that could be pooled in a meta-analysis (874,710 individuals). Although the variables for which individual studies adjusted varied (Table S3), the majority included the important confounders: sex, maternal smoking, and parental atopy or asthma. The final summary OR for the association between preterm birth and wheezing disorders was slightly attenuated as compared to the unadjusted analysis (OR 1.46, 95% CI 1.29–1.65, p<0.001; Figure 3).

23In order to investigate a possible “dose–response” relationship between gestational age at birth and wheezing disorder risk, we aggregated corresponding adjusted association measures from 17 studies (1,105,828 individuals; Figure 4). The pooled estimate of the linear association between gestational age and wheezing disorder risk thus obtained (OR 0.94, 95% CI 0.92–0.96, p<0.001) indicated a 6% (95% CI 4%–8%) decrease in wheezing disorder risk for every week increase in gestation length up to 40 wk.

24The strength of the association between preterm birth and wheezing disorders was similar between children aged <5 y and older children in both unadjusted and adjusted analyses (Figures S1, Figures S2, S3). The risk was considerably higher among children born very preterm (unadjusted: OR 3.00, 95% CI 2.61–3.44, p<0.001; adjusted: OR 2.81, 95% CI 2.52–3.12, p<0.001; Figure 5) when compared to moderately preterm children (unadjusted: OR 1.49, 95% CI 1.34–1.66, p<0.001; adjusted: OR 1.37, 95% CI 1.17–1.62, p<0.001; Figure 6).

25In adjusted analyses, the association between preterm birth and wheezing disorders was most pronounced in the higher quality studies, whereas studies with a high risk of bias had the highest point estimate in the unadjusted analysis (Figures S4, S5, S6). Aggregate ORs did not differ much between subgroups according to study size (Figures S7, S8, S9), diagnosis ascertainment (Figures S10, S11, S12), or wheezing type (Figures S13, S14, S15).

26To further investigate the independent effects of study size, population age, diagnosis ascertainment, wheezing type, and publication year, meta-regression analysis was performed. None of these factors had an independent effect on the strength of association between preterm birth and wheezing disorders (Table 3). The models' residual I 2 values suggested that additional unmeasured factors contributed to between-study heterogeneity.

27Funnel plot asymmetry was minimal for both unadjusted and adjusted association measures (Figure 7). Accordingly, Harbord's test (p = 0.55; unadjusted association measures) and Egger's test (p = 0.43; adjusted association measures) revealed no important small-study effects potentially indicative of publication bias.

28Based on unadjusted association measures, the PAR for childhood wheezing disorders associated with preterm birth was estimated at 4.0% overall, comprising 1.6% (95% CI 1.3%–1.9%) for very preterm birth and 2.4% (95% CI 1.7%–3.1%) for moderately preterm birth. PAR based on adjusted association measures was 3.1%: 1.2% (95% CI 1.1%–1.3%) for very preterm birth and an additional 1.9% (95% CI 1.0%–2.7%) for moderately preterm birth.

29In this meta-analysis of observational studies, using data from over 1.5 million children from across six continents, preterm birth was found to be associated with a 1.71 (95% CI 1.57–1.87) times increased risk of childhood wheezing disorders. The association was slightly attenuated by adjustment for potential confounding within individual studies (OR 1.46, 95% CI 1.29–1.65). Additional analyses strongly support a dose–response relationship between length of gestation and wheezing disorders (OR 0.94, 95% CI 0.92–0.96, per week increase), with children born very preterm having three times the risk of that of children born at term (OR 3.00, 95% CI 2.61–3.44). Consistency across sensitivity analyses further supports the robustness of the findings. Moderate preterm birth accounted for the majority of the explained variation in childhood wheezing disorders by preterm birth. Together, these findings are in line with the increasing recognition of the impact that early life influences have on subsequent health and disease, including pulmonary outcomes,. Furthermore, they highlight the pressing need for prioritisation of research into prevention of preterm birth and the aetiology of its adverse consequences for subsequent respiratory health.

30As with any systematic review we cannot exclude the possibility that potentially relevant studies were missed by our search strategy, although it included the primary online databases for medical research and the most rational search terms relevant to the research question. Identification of additional studies was performed through screening of reference lists and citations, and contacting an international panel of experts. Absence of important small-study effects suggests that bias resulting from selective study inclusion is likely to be minor.

31I 2 values suggested that considerable heterogeneity was present among the studies included in the meta-analysis, although these values may be inflated by the inclusion of some very large studies. Heterogeneity was handled by applying random-effects models. Low I 2 values in subgroup analyses according to age group suggest that between-study population age differences may contribute. Variation furthermore exists among doctors and parents, and between doctors and parents regarding the definition and perception of wheeze, as well as asthma. We therefore performed sensitivity analyses separating wheezing from asthma, and physician diagnoses from parental reports. The results suggest that, although perceptions may differ, this did not affect the estimation of the association between preterm birth and childhood wheezing disorders. Indeed, residual heterogeneity remained after adjustment for these and other factors via meta-regression analysis. Although study quality may be another responsible factor, the observation that aggregated association measures were most pronounced in studies with low risk of bias supports the validity of the conclusions.

32Several factors potentially confound the association between preterm birth and wheezing disorders. We therefore performed separate analyses aggregating adjusted association measures. This attenuated the association between preterm birth and wheezing disorders to some degree, as expected. The vast majority of studies adjusted for important potential confounders, although their number and nature varied between studies. A varying degree of residual confounding thus remains, which should be taken into account when interpreting the findings. From studies reporting association measures according to multiple gestational age strata we included the least preterm stratum. The aggregated point estimate is therefore likely to underestimate the actual association.

33Different methods for gestational age estimation exist. Although most studies failed to report their approach, first trimester ultrasound is increasingly used in higher income countries. Alternative estimation is usually based on the last menstrual period, which is known to overestimate the numbers of preterm and post-term births. Again, this is expected to result in underestimation of the association between preterm birth and wheezing disorders.

34In a previous meta-analysis, preterm birth was associated with a 1.37 (95% CI 1.30–1.43) times increased risk of asthma. To better reflect current neonatal practice, we included only cohorts born from the 1990s onwards, resulting in a much larger sample size and negligible overlap with the report by Jaakkola and colleagues. The substantially greater numbers of events in our review enabled us—for the first time—to investigate aggregated adjusted association estimates to account for potential confounding, explore the association with degree of prematurity (i.e., a dose-dependent relationship), and estimate the associated disease burden (i.e., PAR). These additional steps are crucial to interpreting these data, as the evidence is of necessity derived from observational studies, which are inherently at risk of bias. We can only speculate as to the underlying reasons for the association being more pronounced in our report, even after adjustment for confounding. The former review mainly included children born in the 1960s–1980s. These older cohorts are likely to be less preterm on average, which may explain their somewhat lower risk of developing asthma. Analysis by study characteristics in both meta-analyses suggests that the study characteristics are unlikely to explain the difference. The apparent increase over time in wheezing disorder risk associated with preterm birth indicates that changes in neonatal practice have generally failed to improve obstructive pulmonary outcomes in children born preterm.

35The current findings do not support prior suggestions that the association between preterm birth and wheezing disorders becomes less prominent with increasing age. Instead, the strength of the association was similar across age groups, suggesting that the pulmonary consequences of preterm birth tend to persist throughout the life course. This observation is supported by evidence from longitudinal studies showing temporal tracking of small airway disease among individuals born preterm,,, and by recent confirmation of a link between preterm birth and chronic obstructive pulmonary disease. The contribution of atopy to this association is unclear and requires further study.

36The processes underlying preterm birth are poorly understood. The incidence as well as the underlying causes of preterm birth are known to vary both geographically and temporally, at least partially driven by meteorological, socioeconomic, ethnic, and (epi)genetic variation,–. Environmental exposures, such as pollution and tobacco smoke, and individual behavioural differences (i.e., hygiene, smoking during pregnancy) furthermore contribute,. Many such factors have also been linked to the development of wheezing disorders,,. Through interference with the foetus's natural environment during critical windows of development, the processes underlying preterm birth challenge the developmental plasticity of the lungs and airways,,. Common antecedents of preterm birth such as inflammation, maternal smoking, metabolic derangement, hypoxia, and growth restriction have well-recognised adverse effects on lung maturation and structure, T cell polarisation and development, and airway reactivity,,–. Such alterations can differentially affect susceptibility of the lungs to injurious exposures that commonly follow preterm birth, including sepsis, respiratory infections, mechanical ventilation, and hyperoxia,–. For any of these factors, the likelihood of being exposed increases with decreasing gestational age at birth. Immediate adverse effects are seen in very preterm infants developing BPD, which is an independent risk factor for reactive airway disease,,,,. Preterm birth furthermore augments the association of several of these risk factors, such as antenatal inflammation and smoke exposure, with childhood wheezing,. Such mechanisms may all contribute to the observed dose–response relationship between gestational age and adverse respiratory outcomes, including wheezing disorders,. Genetic influences and gene–environment interactions are furthermore likely to play a role, and the link between preterm birth and asthma has been suggested to at least partially reflect a common genetic background,.

37The magnitude of the problem having been established, there is now a pressing need to address several knowledge gaps. Epidemiological studies should concentrate on temporal tracking of respiratory status into adulthood following preterm birth, with particular focus on different wheezing phenotypes, and estimating the relative impact of distinct early life factors on these patterns. Recent accomplishments in combining European birth cohorts can help provide the required numbers. There is, furthermore, a need for additional studies from low- and middle-income countries. More high-quality research is required to identify the underlying mechanisms and accordingly develop appropriate preventive and therapeutic measures. Innovative approaches may combine these aspects, as recently highlighted by a randomised controlled trial of repeated administration of a monoclonal respiratory syncytial virus antibody in moderately preterm infants. The intervention induced an important decrease in wheezing in the first year, establishing the causal link between respiratory syncytial virus and wheezing, as well as offering a possible solution. Immune modulation is beginning to fulfil its promise in neonatal medicine, and preterm babies may well prove particularly favourable targets for the immunomodulatory therapies that are likely to shape the future of asthma management,,. The potential for stratified medicine to benefit children born preterm needs to be more generally explored through appropriately designed trials and pre-specified subgroup analyses in future trials.

38This work provides compelling evidence that preterm birth is an important early life risk factor for wheezing disorders in childhood. Given the increasing incidence of both entities and their potentially lifelong consequences, there is an urgent need to identify the underlying mechanisms and explore the potential for preventive and therapeutic approaches.