Global Incidence of Necrotizing Enterocolitis: a Systematic Review and Incidence Meta-Analysis

Background: Necrotizing Enterocolitis (NEC) is a major source of morbidity and mortality in the Neonatal Intensive Care Unit (NICU), yet the global incidence is not systematically evaluated. We conducted a systematic review and Meta-Analysis of cohort studies reporting the incidence of NEC in infants with Very Low Birth Weight (VLBW). Methods: The databases searched included PubMed, MEDLINE, the Cochrane Library, EMBASE and grey literature. Eligible studies were cohort or population-based studies of newborns including registry data reporting incidence of NEC. Data were extracted from the selected papers and variables extracted included author and publication details, time period covered by the study, location and setting of the study and whether it was VLBW infants or preterm infants, incidence or number on NEC cases and size of population at risk were extracted from the selected papers. Additionally, risk of bias assessment of the included studies were carried out using a validated tool. Bias adjusted Quality Effect Model (QEM) were used to pool the estimates. Due to substantial heterogeneity, Random Effect Model (REM) estimates were also evaluated. The heterogeneity between studies were evaluated using the Cochrane Q statistics and Higgin’s I 2 value. Subgroup analysis and meta-regression were used to explore the sources of heterogeneity. Funnel plots as appropriate for ratio measures were used to assess publication bias. Results: A systematic and comprehensive search of databases identified 27 cohort studies reporting the incidence of NEC. The pooled estimate of the global incidence of NEC was 6.0% (95% CI: 4.0%-9.0%). There were substantial heterogeneity (I 2 = 100%) between studies. Funnel plots showed no evidence of publication bias. Conclusion: The incidence of NEC vary considerably, although meta-regression failed provide evidence for any substantial regional variation. High quality studies along with detailed risk factors of NEC is warranted, to predict and NEC.

distress syndrome (RDS) that reduced the mortality among preterm newborns (1). With better survival of premature babies, Necrotizing Enterocolitis (NEC) became more common and its burden became more prominent (2).
Multiple population-based studies, some based on large cohort studies, have reported the incidence of NEC to vary from 2%-13% in preterm and Very Low Birth Weight (VLBW) infants (2)(3)(4)(5)(6). The variation in the incidence were attributed to differences in the risk factor profiles as well as differing population at risk, detection rate and inclusion and exclusion criteria. There is no pooled estimate of the incidence of NEC worldwide. Furthermore, there is no incidence data from some regions such as North Africa, the Middle East or the Arab Gulf region, apart from a single study from the UAE (7).
With the continuing improvement in survival of preterm newborns, the modifiable risk factors of NEC need to be studies and made use of in developing appropriate interventions to reduce the incidence and impact of NEC. In this context, clinicians and researchers have attempted to identify the factors associated with risk and prognosis of NEC. It was reported as early as the 1980's, that there exist an association between rapid advancement of feeding and the onset of NEC (8). Subsequent reports showed preterm birth (9,10), small birth weight (9)(10)(11) and race (11) were also to be important risk factors. Contemporary reports confirm these initial reports and expand the list to include a few more.
This study aims to systematically review the incidence reported from different parts of the world to synthesize a global incidence of NEC in VLBW infants. The study also aims to explore the regional variability as well as other potential factors that can explain variability in the incidence.

Methods
The recommendations from the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) served as the guide in collating and reporting this review (17).

Eligibility criteria
Eligible studies included cohort or population-based studies of newborns including registry data. Both prospective and retrospective studies were included. Studies reporting the number, frequency or incidence of confirmed NEC in preterm infants or VLBW infants along with appropriate denominator were included. Studies that reported data on subgroups of infants with specific exposures such as congenital heart disease, perinatal infections, preterm rupture of membrane, or sepsis were excluded when the incidence could not be extracted. Studies with unclear case definitions of NEC were also excluded. Experimental studies that were assessing the effect of intervention of selected group of neonates were also excluded, as they would not provide incidence data, so do the case-control studies. Case series where there were no denominator data to compute the incidence were also excluded.
Incidence is used as opposed to prevalence because of the natural history of NEC and its short duration of disease. It is envisaged that findings form this study would provide a starting point for studies that aim to reduce the incidence of NEC.

Searches
The database search was performed in between 1st and 30th of December 2019. The databases

Search Strategy
The search strategy was developed by the authors included a comprehensive database search using broader search terms: "Enterocolitis, Necrotizing", "Epidemiology", "Incidence", "Cohort Studies", and "population-Based studies", "cohort studies", "epidemiological data", "prematurity", "Very low birth weight", "clinical study", "cohort analysis", and "'human". Medical Subject Heading (MeSH) terms were used when appropriate using the above terms with a combination of 'and' and 'or' in accordance with search engine specifications. The search string used for PUBMED is given in Supplementary file S1 as an illustration.

Study Records
The titles and abstracts of the publications identified through the comprehensive search were reviewed and duplicate entries were eliminated. Records thus extracted were individually reviewed against the eligibility criteria. The duplicate records and those not eligible were eliminated and a PRISMA flow chart was created.

Data Extraction
Data form the eligible studies were extracted and collated on to a master data table (Table 1). Name of the authors, year of publication, data on the time period covered by the study, location of the study, inclusion and exclusion criteria of the study, the reported population at risk and whether it was VLBW infants or preterm infants, case definition, incidence or number on NEC cases and size of population at risk were collected.

Risk of Bias Assessment
All the included studies were assessed for internal and external validity using the criteria put forward by Hoy et al, which is a validated tool for bias assessment in prevalence and incidence studies (18).

Data Synthesis
Pooling the incidence estimates was done after arcsine transformations of the data as it has been shown to stabilize variance and reduce bias (19). MetaXL software was used for arcsine transformation and pooling the data using the bias adjusted Quality Effect Models (QEM) (20).
Heterogeneity was assessed using the Cochrane Q test and Higgin's I 2 value. Smaller p values and I 2 > 50% were indicative of significant heterogeneity (21,22).
Fixed Effect Model (FEM) were first fitted but in the presence of substantial heterogeneity it would not be considered valid. Cochrane guidelines suggest use of Random Effect Model (REM) when significant heterogeneity is encountered (23). In addition we utilized Quality Effect model (QEM) to obtain bias adjusted pooled estimate in the presence of heterogeneity (20). Quality scores obtained using Hoy's criteria were used in fitting the Quality Effect Models. Forest plots were used to display the incidence of NEC among VLBW infants with corresponding 95% confidence intervals.
Hunter et al have shown the classical funnel plot to be in-appropriate for proportion studies such as prevalence or incidence studies. They have shown Hunter plot be more appropriate to assess the publication bias (24).
Subgroup analysis by region based on income category of the countries provided by World Bank and population at risk (VLBW or extremely premature) was used (25). This sub-group analysis was not apriori decision but an attempt to explain the variability in NEC due to substantial heterogeneity.
Groups consisted of high income countries (HIC) and low middle-income countries (LMIC). However, a-priori planned meta-regression was performed to evaluate if the publication year has any impact on the variability of the incidence and as a possible cause of heterogeneity. This was also thought to be important to understand if the long term trend in incidence of NEC to see if they are on a rise or decline.
The meta analyses were carried out using MetaXL (19) and the subgroup analysis and meta regression were carried out using Comprehensive Meta-Analysis (CMA-V3) software (26).

Study characteristics
The total number of publications identified for screening was 1694. The process of selection of eligible studies are depicted in the Fig 2. A total of 27 studies were found to fulfill the eligibility criteria and included in the review. The number of neonate included in these studies was 574,692. Of these, 39,965 neonates developed confirmed NEC. The studies covered a broader geographical areas globally (table 3). Some regions had multiple studies other areas had none. A total of eight studies were reported from the United States covering a number of states including: California, Texas, Atlanta, Connecticut, and New York (3,6,9,(27)(28)(29)(30)(31). Multiple studies were also reported from the Europe including Poland, Romania, Finland, Belgium, Sweden and Switzerland (12,13,28,(32)(33)(34). Also, four studies were done in China, Korea, Singapore and Malaysia (14,16,35,36). Three studies from Australia (4,37,38), one from the Middle East (7) and one from India (42).
The publication year of the studies ranged from 1988 to 2019, but the majority were carried out after 2000. Some of the studies focused on evaluating a certain exposure (7,9,33,37), however, the data presented in these papers were not limited to the exposure groups and data from the general population was extracted to compute the incidence.

Andersone et al. reviewed a cohort data from the California Office Statewide Health Planning And
Development [OS HPD] (28). Upon retrograde calculation of the number of NEC cases and dividing them by a total number of NICU preterm babies the incidence of NEC was 9.1%. Whilst, Patole et al conducted a retrospective cohort study reviewing 1,755 neonates who were less than 34 weeks of gestation (37). The aim was to study the effect on the incidence of NEC. In the control group (prior to the initiation of probiotic), there were 835 babies. Among those 250 were preterm with gestational age less than 28 weeks. Stage II or above NEC was found in 16 cases (6% of preterm controls).  1988, 1993-1994, and 1999-2000 (40). The analysis aimed to compare the outcome across the time periods. They showed that the incidence of NEC did not change over time.  (12). The incidence of NEC in less than 750g, 750-999 g, 1000-1499 g and 1500-2499 g were 5.31%, 4.16%, 1.52%, and 0.007%, respectively.
Verstrate et al. based on a retrospective cohort of 5134 neonatal intensive care unit admissions from a single hospital Belgium found 973 cases were born with a very low birthweight of less than 1500 g (32). The incidence of NEC with stage II or above, in this subgroup was 16.23% . Härkin et al.   1986-1987, 1992-1993, and 1998-1999 (4). The population included 1,655 cases from the three groups divided to 360, 622, and 673 cases in time periods 1986-1987, 1992-1993, and 1998-1999 respectively. Over the entire population the incidence was 7.67%. Wong et al. conducted a retrospective cohort study reviewing 2,549 neonates from 10 neonatal intensive care units serving New South Wales in Australia (38). This study population accounted for all preterm infants in the region of Australia between 1998 and 2004. The conducted the analysis complaining those exposed to steroids and those who were not.

Quantitative Analysis of Incidence
The FEM using the double arcsine transformation resulted in a pooled incidence of NEC in VLBW infants to be 7.0 % (95% CI: 7.0%-7.0%) (Fig. 3). However, due to significant heterogeneity as indicated by I 2 value of 100% and the Cochrane Q-statistics (value =7473; P <0.0001), a QEM was used to pool the data using the arcsine transformation and resulted in a pooled estimate of 6.0% (95% CI: 4.0% -9.0%) (Fig. 4). Furthermore, a sensitivity analysis using REM was also as a form of sensitivity analysis and the pooled estimate of REM was 7.0% (95% CI: 6.0% -8.0 %) (Fig. 5).

Publication bias
Hunter's modified funnel Plot (24) as appropriate for the incidence data used to evaluate the publication bias appear to not to show a serious concern (Fig 6). Further, the Eggers regression confirmed that publication bias was not statistically significant (two tailed p-value= 0.80). The Kendall's Tau test statistics was also not statistically indicating less likely that these studies encountered publication bias (two tailed p-value = 0.936).

Subgroup Analysis
There was no significant regional variation between North America, Western Europe and Australia as well as Asia as shown in the table 4. They were both similar to the global estimate as well as indicated by the 95% confidence intervals. There appear to some variation between HIC and LMIC countries, although these differences were not statistically significant. No significant variation between VLBW infants and extremely premature was found.

Meta-Regression
There was a statistically significant increase in the log event rate over time, quantified by the publication year. (Fig 7).

Discussion
This study is perhaps the first attempt to provide a pooled estimate of the incidence of Necrotizing Enterocolitis in VLBW infants. Seven out of 100 of all VLBW infants in NICU are likely to develop NEC as per our synthesis. However, there were considerable heterogeneity in the estimates across studies.
Such important variability may be driven by myriad of factors including the variation in the quality of health care systems.
Subgroup analysis based on geographic regions did not reveal any differences (i.e. South East Asia versus Europe, North America and Australia). However, when countries reporting the data on NEC were re-classified based on income levels using Word Bank classifications the incidence in high income counties (HIC) differed from the low and mid income countries (LMIC), although, there were no statistically significant differences. Such variation may be attributed to the low power of the studies from LMIC given that there is sparse data from these countries. However, it is also possible that slightly lower incidence reported in LMIC may be due to higher gestational age cutoff point for resuscitation used in case of extremely prematurity. It is also possible that the sicker babies in LMIC may have had higher risk of mortality. As a result the population of neonates in LMIC may appear healthier and at lesser risk of developing NEC.
The increase in the incidence of NEC over time which evident in fig-7 and confirmed by metaregression maybe attributed to multiple factors. Improvement in neonatal care and better survival of premature infants is a possible cause. Improvement in diagnosis and reporting NEC in general maybe another reason. These improvements along with lack of wide scale improvement and prevention strategies can explain the increase in incidence of NEC. Ahle et al. demonstrated a j-shaped distribution of incidence over time in Sweden. While the incidence was 150 per 10,000 liver birth among VLBW infants in the late 80 s, it increased to approximately 800 per 10,000 live births in VLBW in Sweden, a multiple fold increase (12). This increase may be related to variations in local health services. This is inconsistent with the finding of the analysis of the NICHD data base from the United States reported by Fanaroff and colleagues (40) which reviewed VLBW infants from three epochs: 1987-1988, 1993-1994, and 1999-2000. Their analysis compared the incidence across these three periods and incidence of NEC did not change over time. The data presented in our analysis represents a wider time period and a set of more diverse healthcare facilities.Due to sparse data available from developing and lower income countries, the pooled estimate may have limited external validity and not fully generalizable across all populations. The incidence of NEC may depends on many factors including the quality of neonatal care and health care system prevailing in the countries.
Our findings should be understood in the light of some of the limitations that this study encountered.
Only 12 out 26 studies could be considered to be of higher quality and this may be linked to the substantially heterogeneity that we encountered. Although, we employed quality effect models to adjust for variation in study qualities, substantial heterogeneity noted in this study does pose a threat to evidence synthesis. The diagnosis of NEC using Bell's criteria or similar definitions schemes is a day to day clinical challenge. To a certain extent, two clinicians may justifiably disagree on labeling a baby as confirmed NEC versus suspected NEC. This dependence of a list of elaborate criteria, vagueness of definition and lack of solid diagnostic are limitations attached to our attempt to provide a pooled estimate of the incidence of NEC.

Conclusions
Given the lack of high quality data related to incidence and factors related to NEC in VLBW infants, larger and higher quality studies are warranted. There were substantial variability in incidence reported from different parts of the world. This could partly be due to variation in the case definitions that was not always clear. Future studies focusing on standardized NEC outcome measures may help understand the burden of NEC in ICU much better.     Forrest plot obtained using Quality Effect Model  Hunter's plot used to assess the publication bias