SGA as a Risk Factor for Cerebral Palsy in Moderate to Late Preterm Infants: a System Review and Meta-analysis

Small for gestational age (SGA) is an established risk factor for cerebral palsy (CP) in term infants. However, there is conflicting data on the association between SGA and CP in moderate to late preterm infants. The aim of the article was to explore the relationship between SGA and CP in the moderate to late preterm infants and its strength by meta-analysis. We performed a system search in OVID (EMBASE and MEDLINE) and WANFANG from inception to May 2016. The study-specific risk estimates were pooled using the random-effect model. A total of seven studies were included in the meta-analysis, consisting of three cohort and four case-control studies. A statistically significant association was found between SGA and CP in moderate to late premature infants (OR: 2.34; 95% CI: 1.43–3.82). The association were higher in the several subgroups: 34–36 week gestational age (OR: 3.47; 95% CI: 1.29–9.31), SGA < 2SDs (OR: 3.48; 95% CI: 1.86–6.49), and malformation included in CP (OR: 3.00; 95% CI: 1.71–5.26). In moderate to late premature infants, SGA is a convenient and reliable predictor for CP. More studies are needed to explore the underlying mechanisms between SGA and CP association.

Scientific RepoRts | 6:38853 | DOI: 10.1038/srep38853 intervention can be provided, we conducted a meta-analysis to assess SGA as a risk factor for CP in moderate to late premature infants.

Methods
Literature search. The electronic databases were searched from inception to May, 2016. A search was performed in OVID (EMBASE and MEDLINE) and WANFANG. For example, in MEDLINE, we used a combination of keywords, text words as well as word variants related to SGA, IUGR, CP, and searched medical subject headings: CP limited in epidemiology, etiology, prevention and control; fetal growth retardation and SGA limited in complications. A filter to identify case-control and cohort studies provided by BMJ Evidence Centre information specialists was used 43 . The research was also conducted in EMBASE using the similar terms. Furthermore, bibliography lists of relevant studies and previously published review articles were searched for potential studies. Only articles that were reviewed by peers were considered. The search strategy is described in the Appendix. Study selection. Inclusion criteria were as follows: (1) Case-control or cohort studies; (2) The study subjects included moderate or late premature infants; (3) The study aimed to explore the relationship between SGA and CP; (4) Relative risk (RR) or odds ratio (OR) and their corresponding 95% CIs of CP related to SGA were reported, or could be calculated from the data presented in the articles. We excluded animal studies, duplicates, case report studies, case series studies, and studies that evaluated the relationship between SGA and other neurodevelopmental conditions.
All titles and abstracts discussing the relationship between SGA and CP in premature infants were independently retrieved and screened by two authors (M.Z. and H.D.). If the titles or abstracts were considered as potentially relevant by both reviewers, the full-text of those identified studies were then assessed, with all references, author names, journal title, and funding sources concealed. Any disagreement was settled by discussion with the help of a third author (L.Z.).

Data extraction.
For articles meeting the criteria described above, the following data were independently extracted into standardized forms by two researchers (M.Z. and H.D.): publication year, birth year, malformation included in CP, the matched or adjusted variables, case-control or cohort study design, gestational age category, definition of SGA, study location, the first author and odds ratio with 95% confidence interval or crude data. When relative odds for the same study subjects were reported in multiple articles, the most recently published article was selected. When both crude and adjusted data were available in a study, only crude data were selected for the summary calculation since many included studies provided crude data and were not adjusted for the same confounders. The ORs were calculated with primary data presented in the articles and then pooled in the final meta-analysis, so as to maximize consistency of the individual risk estimates because most identified studies reported ORs rather than RRs. The OR can be a substitute for the RR when the prevalence of a disease is low (the prevalence of CP is 1.8-3.5 per 1000 live births). When inadequate information was provided for calculating OR, the original author was contacted for further information.
Quality assessment. Two authors (M.Z. and H.D.) independently evaluated each included article using the Newcastle-Ottawa Scale on the basis of the qualities of the study group selection, comparability of study groups, and ascertainment of outcome assessment. The total quality scores ranged from 1 to 9, with higher scores indicating higher quality. Any disagreement was settled by discussion with the help of a third author (L.Z.).
95% CIs were calculated with crude data presented in the articles using the Woolf method. When zero cell was encountered, 0.5 was added into cell with corresponding 2*2 table. Since some studies provided multiple effect estimates on the basis of separate gestational age, an overall effect estimate was calculated from the available crude data using inverse variance method. The homogeneity was evaluated by Q test, preserved p = 0.1 used as heterogeneous criteria. A summary OR was calculated using randomized effect model, which takes weighted mean of each effect estimate. This model was chosen because of anticipated significant heterogeneity between studies on the basis of study methods and population. Funnel plot was used to detect potential publication bias. The trim and fill analysis was performed to assess the possible effect of publication bias on the meta-analysis. Subgroup analysis was performed using the predefined stratification according to study design, definition of SGA, gestational age, malformation included in CP, with p = 0.05 indicating statistical significance.

Results
Literature selection. A total of 3519 articles were retrieved in accordance with the above search strategies.
Characteristics of included studies. Details of included articles such as first author, location, birth year, study design, gestational age, SGA definition, sample size, outcome assessment, assessment of study quality, and matched or adjusted variables are presented in Table 1. Most studies were from developed countries (in Europe), with only one study from a developing country (China). The year of birth of the participants ranged from 1967 to 2008. The total number of participants included in this meta-analysis was 135650. Four studies defined SGA infants with birth weight less than 2SDs below the mean weight for gestational age, while the others defined SGA infants with birth weight less than 10th percentile for the gestational age. The potential confounders controlled in the included studies were different in different studies, with two studies presenting crude estimates. There was no significant association between SGA and CP in moderate to late premature infants in two studies 14 seven studies, ORs ranged from 1.01 to 6.0. One article adjusted for the most confounders provided an adjusted OR (OR: 1.85; 95% CI: 1.25-2.75) in the 34-36 week gestational age and OR (OR: 1.10; 95% CI: 0.57-2.13) in the 32-33 week gestational age 5 . In most studies, the information on CP was extracted from registers, in which the diagnosis of CP was based on medical history, imaging data, and clinical multidisciplinary evaluations. Only two articles excluded participants with congenital malformation, which was an established risk factor for CP. The study quality scores ranged from 5 to 7. The median score was 6.
Overall pooled effect. The overall pooled effect calculated from the seven articles demonstrated a statistically significant association between SGA and CP in moderate to late premature infants (OR: 2.34; 95% CI: 1.43-3.82), as was shown in Fig. 2. Significant heterogeneity was determined in the meta-analysis (p < 0.001), and a random effects-model was used.
Subgroup analysis. The associations between SGA and CP in moderate to late premature infants in subgroup meta-analyses are shown in Table 2. The increased risk was more evident in several strata of study characteristics: using birth weight less than 2 SDs below the mean weight for gestational age to define SGA (OR: 3.48; 95% CI: 1.86-6.49), with gestational age limited in 34-36 weeks (OR: 3.47; 95% CI: 1.29-9.31), and malformation included in CP (OR: 3.00; 95% CI: 1.71-5.26). In most subgroup analyses, the SGA was a risk factor for CP and the associations were statistically significant. The risk of CP was higher in case-control studies than in cohort studies. However, case-control studies did not produce significant summary OR (OR: 2.48; 95% CI: 0.93-6.67), while cohort studies did (OR: 2.19; 95% CI: 1.15-4.18).

Sensitivity and publication bias analysis.
The association between SGA and CP in moderate to late premature infants was determined by studies with different definitions of SGA and study design. However, removal of any study did not appreciably change the association. We performed a publication bias analysis by the funnel plot. The graph appeared to be asymmetrical, suggesting the existence of a publication bias (Fig. 3). The trim and fill analysis was performed and compared with the original results, which produced higher OR (OR: 2.65; 95% CI: 1.65-4.28; p < 0.001).

Discussion
To our knowledge, no meta-analyses have previously explored the association between SGA and CP in moderate to late premature infants. Findings from the current research indicated that moderate to late premature infants with SGA had higher risk for CP. The finding was supported by five of seven studies included in this meta-analysis 5,33,[39][40][41] .
From our meta-analysis evidence, SGA was a risk factor for CP in moderate to late premature infants. The higher risk for CP among children born with SGA is poorly understood. The association might be explained by several hypotheses: (1) Some factors result in both abnormal growth and brain damage; (2) Brain damage leads  to abnormal growth or vice-versa; (3) Abnormal growth makes the newborns susceptible to negative perinatal events. According to the first two theories, SGA infants with CP are of prenatal origin. With the third, the perinatal events are the main sources, but with a latent prenatal factor. Current studies suggest that genetic and prenatal factors may be the primary causes for children born with SGA developing CP 24,25 . Besides, newborns with combined marked SGA and a major birth defect are at higher risk for CP 25 , as was confirmed by our subgroup analyses. Alternatively, some studies suggest that the risk of CP in children born with SGA is limited to those born with low Apgar scores [57][58][59] . Birth asphyxia may interact with poor fetal growth in the pathogenic pathway leading to CP 23 . Different pathophysiological mechanisms are probably associated with different intervention methods and long neurodevelopment outcomes. The potential causes should be actively explored in every infant with SGA.
In subgroup analyses, we showed that the pooled OR was not statistically significant in the subgroup, which defined SGA as < 10th. Defining SGA as < 10th is somewhat arbitrary because it may include not only pathological growth retardation but also constitutional smallness. This may be a reason why the relationship between CP and SGA has been described as paradoxical. Standard categorization of SGA is valuable for assessment of morbidity and mortality, better counseling, and potential early intervention. Our evidence supports that defining SGA as < 2SDs may be a better cutoff for distinguishing pathological growth retardation from the population. When stratified by study design, the risk of CP was higher but not statistically significant in case-control study subgroup than in cohort study subgroup. This may be attributed to small sample sizes in case-control studies.
The significant association between SGA and CP appeared to be higher in 34-36 week gestational age in the subgroup analyses. It was apt to think that the extended effect of prenatal factors on fetus would put the fetus at an increased risk of developing CP. Some studies indicated that spontaneous preterm labor with SGA might be a fetal adaptive response 14 . However, we could not extract data on 32-33 week gestational age from the original articles and statistically significant association existed between SGA and CP in 32-36 week gestational age. More detailed gestational age classification will be needed in future studies.
In most studies, CP was defined as a group of non-progressive motor impair syndromes attributable to lesions or abnormalities of the developing fetal or infant brain. But the definition of CP used in the included studies differed in some respects. Some studies excluded children with clearly identified postnatal causes or malformation, while others did not. The age of CP diagnosis varied among the studies. Of all included studies, three made the diagnosis in children at least three years old, four without explicit age limitation. In our subgroup meta-analysis, the studies were classified by whether they excluded CP infants with malformation. The association between SGA and CP was weakened in subgroup with malformation excluded. Genetic mutations were suggested to be the cause of CP, and that a standard definition of CP is needed.
This meta-analysis has some limitations. First, we found significant heterogeneity across studies, which may result from differences in study design, sample size, definition of SGA, analysis strategies, and participants' characteristics. But sensitivity analysis showed that the result was robust. The statistically significant associations were consistent in most subgroup meta-analyses. Second, matched or adjusted variables in most studies were different and two studies provided crude OR, which indicated the potential to wrongly attribute SGA to CP when other factors may contribute. However, a cohort study with the most adjusted confounders provided similar results 33 . Third, the funnel plot suggested a possible publication bias. The result from the trim and fill analysis was higher than the original outcomes. Finally, our pooled data contained few data of developing countries, thus lacking generalizability.
Among the children born with 32-36 week gestational age, strength of the association between SGA and CP was confirmed. Infants born in the 32-36 week gestational age with SGA showed up to 1.34 fold increase in the risk for CP (OR: 2.34; 95% CI: 1.43-3.82) as compared to infants with appropriate gestational age. In addition, the risk of developing CP was 3.48 (95% CI: 1.86-6.49) when SGA was defined as < 2SDs. Accurate prediction of the risk of developing CP in moderate to late premature infants with SGA would be very valuable for early interventions to reduce the frequency of those adverse events. Additional research is needed to investigate the underlying causes between SGA and CP, which will likely improve clinical care, enable a more informed prognosis for those children and future pregnancies of their mothers, and contribute to a better understanding of the important relationship between SGA and CP.