Prevalence and Risk Factors for Retinopathy of Prematurity in China: A Systematic Review and Meta-Analysis

Background The etiology of retinopathy of prematurity (ROP) is thought to be related to genetic susceptibility and environmental exposure factors. The purpose of this article was to estimate the prevalence of ROP in mainland China and to attempt to summarize the environmental risk factors for ROP in Chinese infants. Method We searched 9 databases for articles that were published before May 29, 2021, and studies describing the prevalence and risk factors for ROP in Chinese infants were included. The xed-effects model and the random-effects model were applied to the effect sizes (ES) and their 95% condence intervals (CIs) with I 2 ≤ 50% and I 2 >50% in the heterogeneity tests, respectively. Twenty-two separate populations were included in the meta-analysis of the prevalence of ROP. The prevalence of ROP in mainland China was 9.284% (95% CI: 6.546-12.022%). It was negatively correlated with birth weight (BW) and gestational age (GA). Fifty independent meta-analyses were observed to be related to environmental exposure factors of ROP. Thirty of the 50 meta-analyses had results that were signicant at p values less than 0.05. The rst three risk factors with the largest combined effect size were GA ≤ 34 w, bronchopulmonary dysplasia (BPD) and BW ≤ 2,000 g.


Introduction
Retinopathy of prematurity is a vascular proliferative retinal disease that occurs in immature infants. It almost always occurs in small for gestational age infants and a small number of low birth weights in the population of fullterm newborns. Due to the fact that ROP is a disease that can be detected early through screening, neonatal fundus screening should be seriously considered. Standardized screening criteria need to be formulated based on the investigations of the prevalence and risk factors for ROP in different countries and regions.
Before 2010, research on ROP screening has been mostly concentrated in economically developed provinces, municipalities and capital cities in eastern coastal areas, such as Guangdong, Beijing, Zhejiang, Shandong, Jiangsu and Shanghai 1 . When compared with developed countries, infants with larger birth weights and larger gestational ages were more likely to suffer from ROP in China, which were similar results observed in some developing countries (such as Vietnam) 2,3 . In the past 10 years (2008-2018), the number of ROP screenings in the eastern coastal provinces and cities has further increased, and the number of studies in western and southwestern China has also signi cantly increased. However, reports in Northeast China are still few 4 .
It seems that reports on the prevalence of ROP in various regions in China are relatively scattered, and there are few reviews describing the prevalence of ROP in premature infants in China. Moreover, China's vast territory and economic and medical levels vary. Although the Chinese guidelines for screening for retinopathy of prematurity were formulated in 2014 5 , the screening standards of various provinces and cities still varied, which made the screening results of various regions less comparable. Therefore, it is necessary to update a review.
Some hypotheses have been proposed to explain the development of ROP, but its exact mechanism is still unclear.
Lower birth weight and smaller gestational age are the most frequently mentioned risk factors for ROP. The longterm use of oxygen after birth is also an important factor in promoting the development of ROP; however, this relationship has not been veri ed. Polymorphisms in some genes (such as the gene encoding vascular endothelial growth factor) also partially explain the occurrence of ROP 6 . Other factors have been included by researchers in Chinese newborns, but the data are scattered, and some articles have obtained the opposite conclusion. Thus, a systematic review and quantitative meta-analysis need to be conducted.
This article introduces the development of retrieval strategies, the retrieval and screening of literature, literature quality evaluation, data extraction and merging and an analysis of the sources of bias. The innovations and shortcomings of this study will be discussed at DISCUSSION part of this article.

Methods
We conducted this review based on the preferred reporting items for systematic review and meta-analysis (PRISMA) protocols.

Search strategy
The retrieved literature originated from nine English and Chinese databases, including PubMed, the Cochrane Library, Embase, Medline, Web of Science, SinoMed, China National Knowledge Infrastructure (CNKI), VIP Database for Chinese Technical Periodicals (VIP) and Wanfang databases. The search strategy was a combination of subject terms, such as "premature infant", "retinopathy of prematurity", "ROP", "prevalence", "morbidity", "risk factors" and free words.
This section contains ve parts: search strategy, the inclusion and exclusion criteria of the studies, the quality assessment of studies, data extraction and statistical analysis. Each step was separately performed by two researchers without interference. Any inconsistent results between the two researchers were submitted to the third researcher to obtain an agreement.

Inclusion and exclusion criteria
Inclusion criteria 1) Prevalence of ROP: newborns with gestational ages less than 37 weeks or gestational ages greater than 37 weeks but with birth weights less than 2,500 g were included in this study; the population that was studied included infants from mainland China; and the included studies must have described the prevalence of ROP and its 95% con dence intervals (CIs) or other forms of data that can be transformed into CIs.
2) Risk factors for ROP: the restrictions on the study population were the same as previously described; and the included studies must have clearly described the exposure factors, outcome indicators, relative risk (RR) values or odds ratio values, as well as their 95% CIs. Other data that can be transformed into these values were also accepted.

Exclusion criteria
The following studies were excluded: studies that were not representative in the population, such as studies that were conducted only in infants with a certain disease and studies that were conducted only in the NICU, among other cases; reviews, comments, conference abstracts and case reports, among other types; articles that only reported the prevalence (i.e., the incidence) of newly diagnosed ROP; and duplicate reports.

Quality assessment of studies
We used the Agency for Health Care Research and Quality (AHRQ) scale to evaluate cross-sectional studies. Those studies with a score of ≥8 were classi ed as being high-quality studies, and those studies with a score of 4-7 and less than 3 were of medium-quality and low-quality, respectively. Newcastle-Ottawa Scale (NOS) was used to assess cohort studies and case-control studies. Studies with scores of no less than 6 were classi ed as being highquality studies.

Data extraction
The extracted information included the rst author, publication year, survey year, survey area, sample size, number of patients and other data containing demographic characteristics.

Statistical analysis
We used STATA software (Version 16.0; Stata Corporation) to analyze the extracted data. The Cochran's Q test and I 2 index were used to test the heterogeneity of the studies. Studies with lower heterogeneity (I 2 less than 50%, p>0.05) used a xed-effects model to combine the effect values. In contrast, other studies used a random-effects model. A subgroup analysis, sensitivity analysis, cumulative meta-analysis and meta-regression were used to analyze the sources of heterogeneity. The publication bias was analysis by funnel plot and Egger's test.

Search results and characteristics
This review included all of the eligible studies that were published as of May 29, 2021, and unpublished studies were not involved. A total of 3,868 articles were retrieved, and 22 and 30 articles describing ROP prevalence and ROP risk factors, respectively, that met the criteria were nally included ( Fig. 1). Among the 22 studies describing the prevalence of ROP, there were 6 high-quality studies, 15 medium-quality studies and only one low-quality study.
Among the 30 studies included in the analysis of ROP risk factors, data extracted from 28 studies could be metaanalyzed. 24 of these studies were of high-quality.

Sensitivity analysis for prevalence
A sensitivity analysis of the prevalence of ROP (Fig. 3) showed that the exclusion of each study one by one did not have a signi cant impact on the overall combined results, thus suggesting good stability of the results. The cumulative meta-analysis of ROP prevalence sorted by publication year is shown in Fig. 3.

Subgroup analysis of ROP prevalence based on geographic region
We conducted a subgroup analysis of the prevalence of ROP in eastern, western and central China based on the survey area of each study. The results showed that the prevalence of ROP was the highest in central China at 12.083% (95% CI: 9.838-14.328%), whereas the prevalence of ROP in western China was the lowest at 4.339% (95% CI: 0.822-7.855%). The prevalence of ROP in central China was observed to be located the two previous prevalences at 9.112% (95% CI: 6.526-11.699%). However, there was no signi cant difference in the prevalence between South and North China ( Table 1).
The prevalence of ROP based on gestational age When grouped by gestational age, the combined results of ROP prevalence are shown in Table 1. In newborns with a gestational age of less than 30 weeks, the prevalence of ROP was as high as 44.541% (95% CI: 17.808-71.275%), which was only 3.207% (95% CI: 0.962-5.451%) in newborns with a gestational age of more than 34 weeks. With the increase of gestational age, the prevalence of ROP showed a downward trend.

Meta regression
The meta-regression analysis of ROP prevalence is shown in Fig. 4. The regression model showed that the prevalence of ROP decreased with increasing publication year, and the relationship was signi cantly different (metaregression coe cient: -0.600, 95% CI: -1.152 to -0.0487, p=0.034). However, the variation between the studies was relatively large (Tau 2 =28.37), the 98.75% residual variation could be explained by the heterogeneity (I 2 =98.75%) and the covariate that was included in the model (the year of publication) could only be responsible for 14.5% of the total variation (Adj R-squared= 14.5%). Publication bias A funnel chart was used to test for publication bias (Fig. 5), and the Egger's test results (t=0.46, p=0.652) showed that the funnel chart was symmetrical, thus suggesting that the possibility of publication bias in this study was unlikely.

Meta-analysis of risk factors for ROP
The meta-analysis of the univariate analysis of 50 risk factors for ROP in the Chinese population is shown in Table   2.

Other risk factors for ROP
In addition, risk factors that were not included in the meta-analysis were mentioned in some articles (Table 3).
In addition to risk factors such as BW and GA, a study by Liu Q et al. 29 (with a sample size of 1,614 cases) also showed that acid-base imbalance, maternal cold, hypoproteinemia, erythropoietin, encephalopathy of preterm infants, myocardial injury, coagulation dysfunction and vasoactive substances were also risk factors for ROP (OR>1, p<0.05). Yau GS et al. 30 showed that hypotension, cardiotonic use, NSAID use, thrombocytopenia and light therapy were also risk factors for ROP (OR>1, p<0.05) in another study. A 5-year cohort study by Yang Q et al. 31 suggested that 1 min Apgar<4 points was a risk factor for ROP (RR>1, p<0.05). Furthermore, oxygen therapy during pregnancy and intravascular hemolysis were mentioned as being risk factors for ROP 32,33 . Vitamin A supplementation in infants was found to reduce the incidence of ROP in very premature infants (RR=0.601, 95% CI: 0.465-0.775,

Discussion
In this meta-analysis, 50 separate populations from more than 21 provinces and cities were included, with 22 of those populations describing the prevalence of ROP. More than half (14) of these studies originated from provinces and cities in eastern China, with Guangdong possessing the most studies (4), followed by Shandong (3) and Beijing (3). There were only 3 studies mentioning the prevalence of ROP in western China from Inner Mongolia (1), Ningxia (1) and Guizhou (1) provinces. Only one study from Northeast China with a sample size of 60 was included in this meta-analysis.
Although the included populations had the same GA and BW characteristics, considerable differences were still identi ed in the prevalence of ROP among the studies, especially in studies from different regions. In addition, the different proportions of very low birth weight infants and very premature infants in the studies may also be one of the reasons for this difference.
As early as the last century, lower birth weight and smaller birth gestational age were found to be two important risk factors affecting the development of ROP. In this review, we re-con rmed this view and found that GA≤34 w (OR: The most important point from this paper is that we used rigorous inclusion criteria in the included population; speci cally, babies who were born with a gestational age less than 37 weeks and full-term babies who were born with a gestational age greater than or equal to 37 weeks (but with a birth weight less than 2,500 g), were analyzed.
The vast majority of babies who may have ROP were included in this meta-analysis. The GA and BW of the population of each included study were the same, in order to obtain a more stable result.
The limitations of this review included the fact that the number of included studies describing the prevalence of ROP is low, some of which are low-and medium-quality studies. Data from northeastern and western China are sparse. When regarding ROP risk factors, the data that we extracted were unadjusted OR (or RR) values of the univariate analysis because the confounding factors in each study were not the same. Some risk factors are only mentioned in 3 or fewer articles. Furthermore, we didn't include any unpublished studies in this systematic review and metaanalysis.
As mentioned above, the prevalence of ROP in China is generally lower than was previously reported and will further decrease in the future and obtain a relatively stable level. There is not much difference in risk factors for ROP between China and other countries. More data, especially data from economically underdeveloped provinces, are urgently needed. Figure 1 PRISMA owchart for the selection of studies    Funnel chart of ROP prevalence