Objective To conduct a systematic review of the association between chorioamnionitis (CA) and bronchopulmonary dysplasia (BPD) in preterm infants.
Methods The authors searched Medline, Embase, CINAHL, Science Citation Index and PubMed, reviewed reference lists and contacted the primary authors of relevant studies. Studies were included if they had a comparison group, examined preterm or low birthweight infants, and provided primary data. Two reviewers independently screened the search results, applied inclusion criteria and assessed methodological quality. One reviewer extracted data and a second reviewer checked data extraction. Studies were combined with an OR using a random effects model. Meta-regression was used to explore potential confounders.
Results 3587 studies were identified; 59 studies (15 295 patients) were included. The pooled unadjusted OR showed that CA was significantly associated with BPD (OR 1.89, 95% CI 1.56 to 2.3). Heterogeneity was substantial (I2=66.2%) and may be partially explained by the type of CA. Infants exposed to CA were significantly younger and lighter at birth. The pooled adjusted OR was 1.58 (95% CI 1.11 to 2.24); heterogeneity was substantial (I2=65.1%) which may be due to different variables being controlled in each study. There was strong evidence of publication bias which suggests potential overestimation of the measure of association between CA and BPD.
Conclusions Unadjusted and adjusted analyses showed that CA was significantly associated with BPD; however, the adjusted results were more conservative in the magnitude of association. The authors found strong evidence of publication bias. Despite a large body of evidence, CA cannot be definitively considered a risk factor for BPD.
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.
Bronchopulmonary dysplasia (BPD) is one of the most frequent sequelae in very preterm infants and results in increased healthcare costs, prolonged hospital stays with frequent readmissions, and deleterious effects on subsequent growth and neurodevelopment.1 2 BPD was first described as severe lung injury resulting from mechanical ventilation and oxygen toxicity.3 4 With the widespread use of surfactant treatment for respiratory distress syndrome (RDS), together with antenatal glucocorticoids and new ventilator strategies, preterm infants developing BPD nowadays are generally more immature, and have antenatal and postnatal histories that differ from those of preterm infants in the pre-surfactant era.5,–,7 They usually initially present with mild RDS responding to surfactant therapy and are not exposed to high concentrations of oxygen and airway pressures. This new BPD is characterised by arrest of lung development and interference with alveolarisation.8 9 Despite advances in perinatal medicine, its incidence has not decreased and there is neither specific treatment nor preventive treatment for BPD.
What is already known on this topic
▶ Bronchopulmonary dysplasia (BPD) is one of the most frequent sequelae in very preterm infants.
▶ Numerous studies have explored the association between BPD and maternal chorioamnionitis with conflicting results.
What this study adds
▶ This is the most comprehensive and systematic review and analysis of the association between bronchopulmonary dysplasia (BPD) and chorioamnionitis (CA) to date.
▶ Data show some evidence of an association between BPD and CA; however, results adjusting for important confounders show more conservative measures of association.
▶ There is evidence of publication bias.
Chorioamnionitis (CA) is the leading cause of very preterm delivery and its incidence increases with decreasing gestational age (GA).10 11 Histological CA is defined by a maternal inflammatory response with neutrophilic infiltration of the membranes and/or chorionic plate, with or without a fetal inflammatory co-response (funisitis). Several early observational studies have suggested that the risk of developing BPD may be partially dependent on antenatal factors, with the highest risk in infants born to mothers with CA.12,–,16 Whereas prenatal inflammation promotes lung maturation and thereby decreases the severity of RDS, it also seems to increase the risk of BPD, especially when postnatal proinflammatory insults occur. From those data as well as experimental results, a complex view of the pathogenesis of BPD (‘multiple hits’ hypothesis) has emerged, which includes antenatal exposure to a proinflammatory environment together with various postnatal inflammation-triggering events such as mechanical ventilation, sepsis and patent ductus arteriosus.1 7 15 16 The repair processes following inflammatory lung injury would result in a remodelling of the lung with arrest of lung development and interference with alveolarisation.17 However, other cohort studies have found no association between CA and BPD.18,–,20 This inconsistency may be attributable to differences in populations, definitions and methods, and whether potential confounding factors such as GA were considered. In an effort to resolve the discrepancy observed across studies, we conducted a systematic review and meta-analysis to synthesise the literature that measures the association between CA and BPD in preterm infants.
A medical librarian developed comprehensive strategies (see online supplemental appendix A) for MEDLINE (1996 to July week 1 2009), Ovid MEDLINE In Process and Other Non-Indexed Citations (July 2009), Cochrane Central Register of Controlled Trials (2nd quarter 2009), EMBASE (1988–2009 week 27), CINAHL (1937 to July 2009), Science Citation Index Expanded (1900 to July 2009) and PubMed (2009). We also reviewed the reference lists and contacted topic specialists to identify additional potentially relevant studies. We did not limit the search by language or publication status.
Studies were included if they had a comparison group, examined preterm or low birthweight (BW) infants, and reported primary data that could be used to measure the association between exposure to CA and the development of BPD. We classified CA as either clinical, histological or microbiological based on previously reported definitions.21 We classified BPD according to the definition used by the authors in the primary report22: (1) ‘Northway’: x-ray abnormalities persisting beyond 1 month of age in patients who continued to require respiratory support (oxygen or ventilation); (2) ‘Bancalari’: need for oxygen on 28 of the first 28 days together with a compatible chest radiograph (or need for oxygen at 28 days of age, ie, the National Institutes of Health (NIH) consensus definition of mild BPD); (3) ‘modern’: need for additional oxygen at 36 weeks postmenstrual age (ie, the NIH consensus definition of moderate or severe BPD). To identify relevant studies, two reviewers independently screened the results of the searches and applied inclusion criteria using a structured form. Discrepancies were resolved through discussion or in consultation with a third reviewer.
We assessed methodological quality using the Newcastle-Ottawa Scale for case–control or cohort studies as appropriate.23 The scale assesses for potential selection bias, comparability of cohorts or cases and controls, and ascertainment of exposure (cohort studies) or outcome (case–control studies). Two reviewers independently assessed the methodological quality of each study. Discrepancies were resolved through discussion or in consultation with a third reviewer.
One reviewer (LH, HM or CS) extracted data from relevant studies using a predetermined data extraction form. A second reviewer (LH, CS or TL-M) checked data extraction for accuracy and completeness. All data included in meta-analyses were checked by the statistician (YL). Discrepancies were resolved by consulting the primary report. Data extracted from each study included citation information, language of publication, country where research was conducted, objectives, study design, definitions of CA and BPD, inclusion/exclusion criteria, patient characteristics (eg, GA, BW) and results (including raw numbers where available for unadjusted and adjusted analyses of CA and BPD).
The association between CA and BPD was measured using ORs. A comparison of the mean GA and BW for the CA exposed and unexposed groups was conducted using weighted mean differences (WMDs). We combined data using a random effects model. Heterogeneity was quantified using the I2 statistic; greater than 50% was considered substantial.24 25 We conducted an a priori subgroup analysis to investigate heterogeneity according to the type of CA. Because we observed substantial heterogeneity that was not adequately explained by the type of CA, we conducted post hoc analyses to assess whether heterogeneity was explained by definition of BPD,22 whether or not the primary objective of the study was to measure the association between CA and BPD, and whether studies only included very low BW (VLBW) or <32 weeks GA infants. Analyses were conducted based on raw data for CA and BPD (unadjusted OR) and using data from multiple logistic regression analyses within studies that adjusted for potential confounders (adjusted OR). The confounders included in the multiple regression analyses varied by study. Meta-regression was used to examine whether the association between CA and BPD was related to GA (ie, difference in mean GA between the exposed and unexposed groups). All results are reported with 95% CIs. Publication bias was assessed graphically using the funnel plot26 and statistically using Egger's and Begg's tests.27 28 The trim and fill method was applied to examine the association between CA and BPD while adjusting for publication bias.29 30 Analyses were conducted in STATA (v 11; StataCorp LP, College Station, Texas, USA).
Description of studies
We identified 3587 potentially relevant studies from the electronic databases. Fifty-nine studies (15 295 patients) met the inclusion criteria (figure 1).12,–,14 18 31,–,85
Studies were published between 1994 and 2009 (median year 2004). All except three studies were published in English.69 76 81 Studies were conducted in a variety of countries (see online supplemental appendix B), most commonly the USA. While all studies provided data to measure the association between BPD and CA, the objectives of the studies varied. In 34 studies, the primary objective was to measure the association between risk factors including CA and BPD; in 20 studies, the primary objective was to evaluate risk factors excluding CA and BPD. In another five studies, the primary objective was other than measuring risk factors for BPD (eg, randomised controlled trials of hydrocortisone therapy for prevention of BPD).
Definitions of BPD and CA varied across studies. BPD was defined according to Northway in 10 studies, Bancalari in seven studies, the NIH consensus (‘modern’) definition in 40 studies and other in two studies.22 The type of CA varied from histological (n=34), clinical (n=12), microbiological (n=3), clinical and histological (n=7), histological and microbiological (n=1), clinical and microbiological (n=1) to all three types (n=1). For those studies that evaluated clinical CA, the definition was not specified in five studies, while one study required more than one clinical sign or symptom, five studies required the presence of two clinical signs or symptoms and 10 studies required three or more signs or symptoms. The inclusion criteria varied across studies (see online appendix B). The majority of studies (n=45) examined VLBW infants (<1500 g), including one study of extremely low BW infants (<1000 g).
Studies included randomised trials (n=2), and prospective cohort (n=33), retrospective cohort (n=21) and case–control (n=3) studies. For the cohort studies and randomised trials, in all cases the non-exposed cohort was drawn from the same community as the exposed cohort. Common weaknesses were lack of representativeness of the exposed cohort and lack of control for potential confounders. In all studies, the non-response rate was the same for both groups or the non-respondents were described. Detailed results of the quality assessment are available from the authors.
Association between CA and BPD: unadjusted analyses
The pooled unadjusted OR from 57 studies showed that CA was a significant risk factor for BPD (WMD 1.89, 95% CI 1.56 to 2.3), although heterogeneity was substantial (I2=66.2%) (figure 2). The association remained significant for histological but not for other types of CA; infants exposed to histological CA were 2.19 times more likely to develop BPD (95% CI 1.76 to 2.72; n=37 studies). The effect of CA on BPD appeared to be stronger when mean GA at the study level increased; however, it was not statistically significant (p=0.124; n=52 studies). We conducted post hoc analyses to explore sources of heterogeneity. We observed no clear patterns for subgroups by CA type and BPD definition or CA type and population characteristic (ie, VLBW vs all preterm infants) (data not shown).
Exploring potential confounders
Since GA and BW are known to be important confounding variables,86 we wanted to see whether it was reasonable to assume that the distributions of GA and BW were the same in the exposed and non-exposed groups. Twenty-four studies reported the mean GAs for both exposed and non-exposed groups. A pooled WMD estimate of −1.42 indicated the infants exposed to CA were born 1.4 weeks earlier (figure 3). This estimate was statistically significant (95% CI −1.69 to −1.15). We examined BW in a similar fashion. The pooled estimate for 25 studies that reported mean BW by groups of exposed and non-exposed infants gave a WMD of 92.88, indicating that infants exposed to CA were born on average 92.88 g lighter (figure 4); this estimate was also statistically significant (95% CI −129.54 to −56.22). In summary, these analyses showed that the infants exposed to CA were born significantly earlier and lighter. Without controlling for GA and/or BW, the results (OR of having BPD for exposure vs non-exposure) would likely overestimate the association. Meta-regression demonstrated an increased risk when the absolute difference in GA between exposed and unexposed groups increased (p=0.022; n=24 studies; figure 5).
We performed an exploratory analysis based on adjusted results which were available in 17 studies (table 1). Overall CA was a significant risk factor for BPD (OR 1.58, 95% CI 1.11 to 2.24) (figure 6), although the heterogeneity was substantial (I2=65.1%). The association remained significant for histological and ‘combination’ (one study only) but not for clinical CA. The results were also quite heterogeneous within CA type. This could be due to the fact that various confounders were controlled in each study (table 1).
We tested for publication bias graphically and statistically. For the unadjusted estimate, there was strong evidence of publication bias (figure 7; p=0.042 based on Egger's test; p=0.03 based on Begg's test; n=57 studies). After applying the trim and fill method adjusting for publication bias, the pooled OR was 1.61 (95% CI 1.31 to 1.98; n=66 studies). There was also strong evidence of publication bias for the adjusted estimate (figure 8; p=0.01 based on Egger's test; p=0.3 (non-significant) based on Begg's test; n=17 studies). The pooled OR after applying the trim and fill method was no longer significant (OR 1.07, 95% CI 0.74 to 1.54; n=23 studies). We conducted subgroup analyses to explore whether studies with a primary objective of assessing the association between CA and BPD showed different results compared to other studies. In three of the four subgroupings (clinical CA and nature of primary objective, microbiological CA and nature of primary outcome, combination CA and nature of primary outcome), the results were more conservative for studies where the primary objective was not to measure the association between CA and BPD (data not shown). In one of these subgroupings (combination CA and primary objective did not include assessing the association between CA and BPD) the association was protective (OR 0.61, 95% CI 0.39 to 0.96).
Summary of findings
The objective of this systematic review was to synthesise the available evidence for the association between CA and BPD in an effort to resolve discrepant findings in the literature. The meta-analysis of unadjusted data showed a significant association between CA and BPD with the odds among infants exposed to CA 1.89 times the odds of those not exposed. The adjusted estimates were more conservative (OR 1.58) but remained statistically significant. These estimates for the association between CA and BPD may be conservative since the comparison groups within the studies were other preterm infants. Moreover, the comparison groups likely included preterm births resulting from vascular and placental pathologies, which are also associated with perinatal inflammation and neonatal morbidity such as BPD.87 88 Our analyses confirmed that GA and BW are confounding factors that need to be taken into account when assessing the association between CA and BPD. Moreover, our analyses provide data on the magnitude of the differences in GA and BW between infants exposed and non-exposed to CA.
The results concerning an association between CA and BPD are tempered by two factors. One is that there was substantial heterogeneity across studies in the magnitude of association, the direction of the association, and whether or not the association was statistically significant. We explored a number of variables in an effort to explain the heterogeneity. We grouped studies by type of CA and heterogeneity remained substantial within most groups. Within the CA groups we explored other potential sources of heterogeneity including definition of BPD and whether the study included only VLBW or all preterm infants. The subgroup analyses did not identify any single, consistent source of heterogeneity, thus the variation across studies is likely due to multiple factors and the relationship between these factors is likely complex. This underscores the ‘multiple hits’ hypothesis for the pathogenesis of BPD.15,–,17 20 Events such as exposure to antenatal steroids, postnatal sepsis, administration of surfactant, mode of ventilation and patent ductus arteriosus were not known to be potential confounding factors at the time many of these studies were undertaken and therefore were not recorded. Likewise, the definition of histological CA was not consistent across all studies and this inconsistency may account for part of the heterogeneity. For instance, recent studies suggest that markers of fetal inflammatory response (leukocytosis or funisitis) are associated with the development of BPD, whereas CA alone is not.89 90 Finally, the impact of ethnicity and the genetic background on the risk of developing BPD has become more apparent over the last few years and cannot be underestimated.91 This underscores the need for including all potential confounding factors in future prospective observational or interventional studies and performing analyses that adjust for potential confounders and for clarity on the role of confounders (eg, magnitude of impact) and potential interactions between confounders.
The second factor that tempered the results showing a significant association between CA and BPD was the finding of significant publication bias. The visual plots of asymmetry suggest that small studies showing no effect between CA and BPD are less likely to be published. After adjusting for publication bias, the measure of association based on unadjusted data was more conservative but remained statistically significant (OR 1.61). However, after controlling for publication bias for the adjusted data the result was no longer statistically significant and the point estimate was close to the null. We also explored whether there were differences in measures of association for studies where the primary objective was to assess the association between CA and BPD. For the unadjusted data, we found in three of the four subgroups that studies reporting CA versus BPD as the primary objective were more likely to report a significant association, whereas studies where the primary objective was not about the association between CA versus BPD showed no significant association and in one case showed a protective effect. These analyses need to be interpreted cautiously because they were post hoc and based on small numbers within the subgroupings; however, they do raise concerns around publication bias and selective outcome reporting.
Strengths and limitations
This is one of the first reviews to systematically and comprehensively identify and synthesise the evidence for an association between CA and BPD. We employed rigorous methods in conducting the systematic review including: an extensive and comprehensive search; duplicate screening, inclusion and data extraction to reduce bias; and meta-regression to control for potential confounders. One important limitation is that we were restricted to data presented in the published reports and we did not always have detailed information to control for important confounders. For instance, in some cases we had to rely on the average GA for the entire sample (including those who did not survive to the endpoint assessment, eg, 28 days or 36 weeks) when controlling for GA in our meta-regression. Further, we had to rely on the adjusted analyses as presented in the published reports and the variables which they included, which were not consistent across studies.
This is the most comprehensive review and analysis of data on the association of CA and BPD to date. This review clearly identified substantial heterogeneity across this body of evidence that substantiates the discrepant findings reported from individual studies. The presence of significant publication bias suggests that positive findings of an association between CA and BPD are more likely to be published and have likely led to an exaggeration of the magnitude of the association. When controlling for publication bias the results were more conservative and adjusted results were no longer significant. This review provides empirical evidence demonstrating the importance of accounting for potential confounders when interpreting data in this area.
The authors thank the following individuals for their contributions to this project: Carol Friesen (searching), Tamara Durec (searching), Carol Spooner (data checking, quality assessment), Heather McPhee (data extraction, quality assessment) and Annabritt Chisholm (article retrieval).
Funding Financial support for this project was received from the Stollery Children's Hospital Foundation and the Alberta Research Centre for Health Evidence.
Competing interests TL-M was an author on studies included in the systematic review.
Provenance and peer review Not commissioned; externally peer reviewed.