Cumulative evidence for association of sepsis and retinopathy of prematurity.

BACKGROUND: Retinopathy of prematurity (ROP) is a retinal vasoproliferative disease affected by multiple factors such as infection and preterm birth. The role of sepsis in the development of ROP remains controversial. This systematic review and meta-analysis aimed to identify the impact of sepsis on ROP.
METHODS: The PubMed, Embase, and Cochrane Library databases were searched using terms related to sepsis and ROP. Cohort or case-control studies that reported the association of sepsis and ROP were eligible. The odds ratios (ORs) together with the 95% confidence interval (CI) were extracted from the studies or computed by authors if not provided.
RESULTS: Thirty-four studies were ultimately included in this meta-analysis. The pooled results showed that sepsis increased the risk for the development of any stage ROP (OR = 2.16; 95% CI: 1.65-2.82). Both early onset (OR = 2.50; 95% CI: 1.97-3.18) and late-onset (OR = 1.37; 95% CI: 1.22-1.55) sepsis were associated with severe ROP. Furthermore, both bacterial sepsis (OR = 1.74; 95% CI: 1.21-2.50) and fungal sepsis (OR = 2.96; 95% CI: 2.05-4.28) were also found to be associated with severe ROP.
CONCLUSION: Sepsis increased the risk of any stage ROP, especially for the severe ROP. Further high-quality clinical studies are needed to eliminate heterogeneity and publication bias to validate these findings.

Introduction

Retinopathy of prematurity (ROP) is a vasoproliferative disease that affects the developing retinal vascular system in premature infants.[ Blindness and visual disability are the main long-term visual outcomes of ROP.[ It has been reported that ROP blinds approximately 20,000 infants annually.[ With increased survival of premature infants due to advanced neonatal care, the number of children affected by ROP is rising in low- and middle-income countries.[ Increasing evidence indicates that infants with ROP are at increased risk of dysfunctions associated with nonvisual neural defects.[ Therefore, retinopathy and brain injury in premature infants may share an etiology with similar risk factors.

Infection is an important risk factor for neonatal brain damage in premature infants.[ Sepsis is a key cause of neonatal inflammation, which contributes to neonatal morbidity.[ Recently, neonatal inflammation has been reported to be associated with ROP.[ Although a meta-analysis has reported that systemic fungal infection is associated with the development of any stage ROP, including severe ROP, in very low birth weight infants,[ there have been conflicting findings regarding the role of fungal sepsis as an independent risk factor for severe ROP.[ In addition, bacterial sepsis has been identified as an independent risk factor in a study of extremely low gestational age newborns.[

Thus, the objective of this systematic review and meta-analysis was to investigate the impact of sepsis on the development of any stage of ROP, as well as of severe ROP, in particular.

Materials and methods

This study was conducted according to PRISMA and MOOSE Guidelines for meta-analysis.

Study identification and selection

The PubMed, Embase, and Cochrane Library databases were searched. The search keywords and Medical Subject Headings (MeSH) terms were the following: “Sepsis” OR “Bacteremia” OR “Fungemia” OR “Bacterial sepsis” OR “Fungal sepsis” AND “Retinopathy of Prematurity” OR “ROP” OR “Retrolental Fibroplasia.” The last search was updated on October 24, 2018. The search was limited to human studies and was restricted to English language reports. Potential studies were identified by initially screening the titles and abstracts of all studies. If the title and abstract suggested that the study discussed risk factors for ROP, the full text of the report was read independently by 2 of the authors (YL and HC) to determine its eligibility according the following inclusion criteria: studies investigated the association between sepsis and ROP; with the diagnosis of ROP based on the results of ophthalmoscopy; classification of the ROP stage from 1 to 5 according to the International Classification of ROP,[ with ROP defined as any stage ROP and severe ROP (stages 3 to 5, plus disease, surgical, and threshold ROP); sepsis was diagnosed as culture-proven sepsis (fungal sepsis was diagnosed by the positive fungal culture but not the indirect mycological tests) and clinical sepsis without culture evidence; case–control or cohort studies; and studies that reported risk ratios or odds ratios (ORs) and corresponding 95% confidence intervals (CIs) or with other data available for calculating the CIs. The following exclusion criteria were also applied: studies with overlapping populations; those that did not include a control group; and those for which the original data were not available. Therefore, to accurately evaluate the association between sepsis and ROP, we only included the studies of culture-positive sepsis in the meta-analysis. Any disagreements were reconciled by a third author (YT), who independently reviewed the studies, and then discussed disagreements with the initial 2 reviewers until a consensus was reached.

Data extraction

Two authors (YL and HC) independently collected data from each study and then compared results. Any disagreement was resolved by discussions with a 3rd author (TZ). The extracted data included the 1st author, publication year, country, study design, included population, diagnosis of ROP, sample size, birth weight, gestational age, impact of sepsis on ROP as primary outcome, risk of bias, and adjustment of confounding factors. The ORs together with the 95% CI were extracted from the studies or computed by authors if not provided.

Quality evaluation

To assess the methodological quality of each included study, 2 authors (YT and TZ) independently screened the study design, the representativeness of the study population, the quality of the statistical analysis, and the validity of outcomes by using the Newcastle–Ottawa scale (NOS).[ The NOS was widely applied to evaluate cohort or case–control studies, based on group selection (4 items); comparability between groups (1 item); and outcome and exposure assessment (3 items). The maximum score was 9 stars. Studies with at least 5 stars were considered to be high-quality studies.[ Disagreements between the 2 reviewing authors were examined by a 3rd author (JH) and resolved by discussion.

Statistical analysis

The ORs of each included study were pooled to assess the association between sepsis and any stage ROP or severe ROP in particular. The pooled OR was calculated by using a random-effects model if statistical heterogeneity was found among studies.[ Otherwise, a fixed-effects model was used. Q statistics and I2 tests were applied to estimate heterogeneity among studies. Data were considered statistically heterogeneous if P < .1 and I2 > 50%. Forest plots were used to show the ORs and 95% CIs of each individual study, as well as the pooled ORs and 95% CIs. To find the source of the heterogeneity, we performed subgroup analysis combined with meta-regression analysis according to the variance in the studies. Publication bias was evaluated by funnel plots and Begg test,[ where a value of P < .05 was considered to be statistically significant. The nonparametric “trim and fill” procedure was also performed to assess the possible effect of publication bias.[ All statistical tests were performed using Review Manager 5.3 or STATA 12.0 software.

Results

Study selection process and characteristics

The literature search identified 639 studies based on our search strategy. After careful screening, we excluded the studies with culture-negative sepsis (31 studies) because the diagnosis criteria of clinical sepsis in different studies were inconsistent and the confounders were commonly existed in those studies. Finally, 34 studies were ultimately selected for the meta-analysis. A flow diagram detailing the selection process is shown in Figure 1. Characteristics of the included studies are summarized in Table 1. These included studies were published between 1985 and 2018. The sample sizes (case/control) varied from a maximum of 2547/22,001 to a minimum of 25/49. All the studies involved preterm neonates (gestational age before 37 weeks, and birth weight <2500 g). For the assessment of ROP, the included studies measured or used data from medical records, based on ophthalmoscopy. According to the onset time and etiology of sepsis, 6 studies reported late-onset sepsis, 4 reported early onset sepsis, 10 reported bacterial sepsis, and 7 reported fungal sepsis. Outcomes were categorized into 2 broad categories: “any stage ROP” and “severe ROP.”

Figure 1

Flow chart of study selection process.

Table 1

Characteristics of included studies.

Quality of included studies

A total of 28 cohort studies and 6 case–control studies were selected in this meta-analysis. The methodologic quality scores ranged from 4 to 8 stars. Most studies were deemed to be high quality, except for 1 study.[ Three studies[ included all preterm neonates without restriction of birth weight or gestation age, but the selection of cases in the other studies may have lacked representativeness. Most studies adjusted for potential confounding factors, but 8 did not.[ Most ORs were evaluated by multiple logistic regression analysis, adjusted for gestational age (24 studies), birth weight (16 studies), oxygen use (14 studies), mechanical ventilation (11 studies), and sex (8 studies) (Supplemental Digital Content, Table S1). Thirteen cohort studies and 1 case–control study provided a no-response rate at the end of follow-up (Supplemental Digital Content, Tables S2 and S3). All included publications were assessed using the NOS (Supplemental Digital Content, Tables S2 and S3).

Sepsis and any stage ROP

Twenty-two ORs of any stage ROP were pooled in the meta-analysis. The pooled OR from the random-effects model was 2.16 (95% CI: 1.65–2.82) (Fig. 2). Substantial heterogeneity was observed (P < .001; I2 = 84%). According to the onset time and etiology of sepsis, we estimated the respective impacts of early onset sepsis, late-onset sepsis, bacterial sepsis, and fungal sepsis on any stage ROP. The pooled ORs are shown in Table 2. Fungal sepsis had a significant impact on any stage ROP (OR = 4.00; 95% CI: 1.71–9.33; P = .001).

Figure 2

Forest plot showing the association between sepsis and any stage of retinopathy of prematurity. The pooled odds ratios (ORs) of association between sepsis and any stage of retinopathy of prematurity were calculated. Diamond marker indicates pooled effect sizes. Some ORs could slightly differ (centesimal) from published values owing to the rounding of primary values. CI = confidence interval.

Table 2

The pooled ORs of association between retinopathy of prematurity and sepsis according to the onset time and etiology.

Sepsis and severe ROP

Twenty ORs of severe ROP were pooled in the meta-analysis. The pooled OR from the random-effects model was 1.87 (95% CI: 1.53–2.29) (Fig. 3). Substantial heterogeneity was observed (P < .001; I2 = 78%). Based on the onset time and etiology of sepsis, we estimated the respective impacts of early onset sepsis, late-onset sepsis, bacteria sepsis, and fungal sepsis on the development of severe ROP. The pooled ORs are shown in Table 2. We found that early onset sepsis (OR = 2.50; 95% CI: 1.97–3.18; P < .001), late-onset sepsis (OR = 1.37; 95% CI: 1.22–1.55; P < .001), bacterial sepsis (OR = 1.74; 95% CI: 1.21–2.50; P = .003), and fungal sepsis (OR = 2.96; 95% CI: 2.05–4.28; P < .001) each had a significant impact on the development of severe ROP.

Figure 3

Forest plot showing the association between sepsis and severe retinopathy of prematurity. The pooled odds ratios (ORs) of association between sepsis and severe retinopathy of prematurity were calculated. Diamond marker indicates pooled effect sizes. Some ORs could slightly differ (centesimal) from published values owing to the rounding of primary values. CI = confidence interval.

Subgroup analysis

With adjustment for confounding factors, study design, classification of birth weight or gestation age, and sample size, a stratified meta-analysis, using subgroups, was performed to explore study heterogeneity. Control for confounding factors (adjusted R2 = 26.32%, P = .059 for any stage ROP; adjusted R2 = 32.32%, P = .068 for severe ROP) and sample size (adjusted R2 = 19.23%, P = .031 for any stage ROP; adjusted R2 = 30.45%, P = .053 for severe ROP) were significant factors in study heterogeneity (Supplemental Digital Content, Table S4).

Publication bias

Asymmetries were shown in funnel plots (Fig. 4) and publication biases were found by Begg test in both any stage ROP (P < .001) and severe ROP (P = .001). Therefore, we performed a sensitivity analysis and used the “trim and fill” method to produce symmetrical funnel plots by imputing hypothetical negative unpublished studies (Fig. 4). When we incorporated these hypothetical studies, the pooled analysis continued to show a statistically significant association between sepsis and severe ROP (OR = 1.296; 95% CI: 1.047–1.605; P = .017). However, the pooled analysis showed no statistically significant association between sepsis and any stage ROP (OR = 1.276; 95% CI: 0.980–1.691; P = .07).

Figure 4

Funnel plots without and with trim and fill. The pseudo 95% confidence interval (CI) is computed as part of the analysis that produces the funnel plot, and corresponds to the expected 95% CI for a given standard error (SE). OR = odds ratio.

Discussion

To our knowledge, no previous meta-analysis had assessed the association between sepsis and the development of ROP based on currently available studies. The meta-analysis showed that sepsis was significantly associated with the development of any stage ROP as well as of severe ROP in particular. Based on the onset time and etiology of sepsis, fungal sepsis increased the risk for the development of any stage ROP. Moreover, early onset sepsis, late-onset sepsis, bacterial sepsis, and fungal sepsis each associated with the development of severe ROP.

The internal association underlying the sepsis and development of ROP may summarized by the relationship between inflammation and angiogenesis.[ Some proinflammatory proteins and angiopoietins have been reported to play multiple roles in vascular processes,[ which may promote the abnormal angiogenesis of retina. In this meta-analysis, a stronger association with severe ROP was revealed for early onset sepsis than for late-onset sepsis. Neonatal early onset sepsis is secondary to intrauterine infection in most cases, which may increase the risk for the development ROP.[ Moreover, infants with intrauterine infection after birth always have a worse respiratory condition, which is typically treated with a higher concentration of oxygen and more advanced mechanical ventilation (such as invasive mechanical ventilation), which also increases the likelihood of ROP occurrence.[ On the contrary, we found that the association of fungal sepsis with ROP was stronger than that of bacterial sepsis. In terms of the etiology of fungal sepsis, of the 7 studies included, 5 studies involved Candida sepsis.[ As far as the type of bacteria resulted in the sepsis, 4 studies identified the specific bacteria. Three of them[ investigated the relationship of coagulase-negative Staphylococcus (CoNS) sepsis and ROP. However, there was no study to identify the different impact factors on ROP between bacterial and fungal sepsis. Therefore, further studies should be carried out to explain this difference.

Subgroup analysis indicated that control for confounding factors is one of the significant factors in study heterogeneity. The various possible confounders and the control for different confounders in studies may affect the final pooled results. Gestational age and birth weight were the main adjusted confounders in the included studies, which are 2 well-recognized risk factors for ROP.[ Moreover, recent studies have identified the impact of the use of oxygen and mechanical ventilation on the development of ROP.[ Thus, it is important to adjust for the influence of oxygen and mechanical ventilation on the relationship of sepsis and ROP. However, of the 34 studies included, only 7 studies[ had concurrently adjusted for this 2 factors. In addition, although there are 14 studies which adjusted the confounder of oxygen use, the bias may still exist because it is difficult to adjust appropriately for oxygenation to normalizing the oxygenation among studies.

This study has several limitations. First, significant heterogeneity was found in the meta-analysis, caused by adjustment for confounding factors and sample size, according to a subgroup analysis and meta-regression analysis. The various possible confounders mentioned above might be the main source of the heterogeneity and skewing of results. Second, significant publication bias was also found in this meta-analysis. After the “trim and fill” procedure, the association between sepsis and severe ROP was significant and stable, while that of sepsis and any stage ROP was not. This instability of association of sepsis and any stage ROP may come from the strict inclusion of studies and some unpublished negative results. Our review only included English publications, some results in non-English studies may be missed.

Conclusion

Our systematic review and meta-analysis showed that neonatal sepsis increased the risk of any stage ROP as well as that of severe ROP, in particular. To eliminate the heterogeneity and publication bias of our results, more high-quality clinical studies are needed to provide convincing evidence for the impact of sepsis on ROP occurrence.

Acknowledgment

The authors appreciate the English editors from Editage Company for proof reading this manuscript.

Author contributions

Conceptualization: Jichong Huang, Dezhi Mu.

Funding acquisition: Dezhi Mu.

Methodology: Jichong Huang, Ying Tang, Tingting Zhu, Yafei Li, Hua Chun.

Software: Jichong Huang.

Validation: Dezhi Mu.

Writing – original draft: Jichong Huang.

Writing – review & editing: Ying Tang, Yi Qu, Dezhi Mu.

Dezhi Mu orcid: 0000-0002-2599-7041.