Association between maternal adiposity measures and infant health outcomes: A systematic review and meta‐analysis

Summary Maternal obesity increases risks of adverse fetal and infant outcomes. Guidelines use body mass index to diagnose maternal obesity. Evidence suggests body fat distribution might better predict individual risk, but there is a lack of robust evidence during pregnancy. We explored associations between maternal adiposity and infant health. Searches included six databases, references, citations, and contacting authors. Screening and quality assessment were carried out by two authors independently. Random effects meta‐analysis and narrative synthesis were conducted. We included 34 studies (n = 40,143 pregnancies). Meta‐analysis showed a significant association between maternal fat‐free mass and birthweight (average effect [AE] 18.07 g, 95%CI 12.75, 23.38) but not fat mass (AE 8.76 g, 95%CI −4.84, 22.36). Women with macrosomic infants had higher waist circumference than controls (mean difference 4.93 cm, 95% confidence interval [CI] 1.05, 8.82). There was no significant association between subcutaneous fat and large for gestational age (odds ratio 1.06 95% CI 0.91, 1.25). Waist‐to‐hip ratio, neck circumference, skinfolds, and visceral fat were significantly associated with several infant outcomes including small for gestational age, preterm delivery, neonatal morbidity, and mortality, although meta‐analysis was not possible for these variables. Our findings suggest that some measures of maternal adiposity may be useful for risk prediction of infant outcomes. Individual participant data meta‐analysis could overcome some limitations in our ability to pool published data.


| INTRODUCTION
Maternal obesity is arguably the leading challenge for pregnancyrelated clinical practice. 1 It is a risk factor for several adverse maternal, fetal, and infant outcomes. 2 There is a wealth of evidence demonstrating the association between maternal obesity, measured using body mass index (BMI), and increased risk of immediate adverse outcomes for the fetus and infant (hereon called "infant"), as well as lifelong health and well-being. [2][3][4] Immediate infant health outcomes include abnormal fetal growth, 5 congenital anomalies, 6 preterm birth, 7 smalland large-for gestational age (SGA and LGA) infants, 8,9 infant morbidity and mortality, and consequently, increased risk of childhood obesity 10 with associated long-term complications such as type 2 diabetes. 11 It is well accepted that the in-utero environment critically influences both short-and long-term health outcomes of infants, 12 making maternal obesity a priority research area relating to optimizing child health. Newborns exposed to insufficient or excess maternal nutrition are more likely to have had abnormal in-utero growth, including both fetal growth restriction and fetal overgrowth. This abnormal fetal growth contributes to both adverse birth outcomes and neonatal morbidity, including hypoglycemia, hypothermia, and neonatal intensive care unit (NICU) admission. The Barker hypothesis maintains that adverse nutrition during pregnancy increases the life-long risk of metabolic syndrome in infants, including obesity, diabetes, hypertension, hyperlipidemia, coronary artery disease, and stroke. 12 A suboptimal in-utero environment prompts epigenetic modification of genes critical for metabolic programming of the fetus. 13 Subsequent transgenerational transmission of these modifications then increases risks for generations to come. 14 Further, evidence is accumulating that a healthy microbiome confers positive metabolic programming of infants and children. 15 It is suggested that maternal obesity influences the microbial colonization in intrauterine environment. 16 Since the microbiome of the infant is largely inherited from the mother, modification could result in a more optimal intra-uterine environment and improve both immediate postnatal and longer-term health of infants.
Diet and physical activity interventions, which aim to reduce the infant health risks associated with maternal obesity, have been inconsistent in their findings to date. For example, while meta-analysis shows a general pattern of interventions reducing the risk of high birthweight outcomes, there is generally a lack of significant difference between intervention and control arms of trials. 17 As there is clear evidence of the potential benefits of changing the in-utero environment to improve infant outcomes, perhaps the lack of effect to date could be due to the current use of maternal BMI in identifying which pregnancies might be at a higher risk. BMI is used in international guidelines to diagnose an individual's weight status (i.e., obesity) for risk stratification of pregnant women with the aim of improving maternal and infant health outcomes. 18-20 BMI is a useful tool to identify population trends in obesity-related disease. 18 However, it is well established in non-pregnancy literature that BMI has high specificity but low sensitivity to detect excess adiposity in individuals and fails to identify half of people with excess body fat. 21,22 Alternative measures of body fat distribution have been more successful in predicting individual risk. For example, waist circumference (WC) has been used for a number of years to assess abdominal obesity as an alternative to, or alongside, BMI, as it is highly correlated with visceral fat. 21 In pregnancy, it has been suggested that early-pregnancy abdominal adiposity is associated with maternal metabolic consequences and could be a better marker of metabolic risks and fetal size than BMI alone. [23][24][25] The evidence base from risk prediction models (in non-pregnant populations) suggests that the use of WC as a continuous variable, adjusted for BMI, works better than BMI alone to identify individuals with a high-risk obesity phenotype. 26 Other measures, such as early pregnancy waist-to-hip ratio (WHR) and subcutaneous fat thickness, have also been suggested to better predict pregnancy outcomes than BMI, as these may be more reflective of abdominal obesity. 25,27,28 This evidence highlights the importance of better understanding of adiposityrelated risk in efforts to improve infant health in short and long terms.
However, there is a lack of robust evidence relating to maternal adiposity measures and infant health-related outcomes and whether these measures work better than BMI to predict risk. This systematic review and meta-analysis aimed to identify measures of early pregnancy adiposity that are associated with infant health outcomes which may be candidate alternative measures to the current use of BMI.

| METHODS
This systematic review was conducted alongside a systematic review and meta-analysis of maternal outcomes, and the details of the searches have been reported in full elsewhere. 29 The methods are also summarized here, with details on the search and amendments to the inclusion criteria specific to the aim of this paper. The systematic review was registered on PROSPERO (CRD42017064464). 30 Both observational cohort and cross-sectional studies were included; therefore, the meta-analysis of observational studies (MOOSE) guidelines were followed. 31 The search strategy was derived by an experienced information specialist using the concepts "Pregnancy," "Adiposity," "Prediction/ Risk," and "Outcomes" and was peer reviewed by another experienced information specialist using the PRESS checklist. 32  and citation searches were carried out using the Google Scholar "cited by" feature. Any new studies which met the inclusion criteria were also reference and citation searched. Authors were contacted when additional information was required for analyses (Table S2). Database searches were completed in April 2021, citation and reference list searches in June 2021, and contacting authors in January 2022.
Inclusion criteria were based on PECOS. 33 The population (P) were singleton pregnancies, with any exposures (E) of pre-or early-pregnancy measures of high adiposity (measured ≤20 weeks' gestation). We included prospective or retrospective observational peer-reviewed studies, including cohort, case control, and crosssectional studies (S) with a comparison group of low adiposity (C). We included any pregnancy outcomes relating to infant health (O). We excluded RCTs and studies which were restricted to sub-populations (e.g., adolescents and pre-existing type 2 diabetes), with the exception of BMI to explore associations across a range of BMIs. No country, language, or date restrictions were applied at the screening stage.
Screening results are reported using the PRISMA statement. 34 The data were extracted by one author and validated by a second (NH, LN, AO, AF, LH, AS, LC, and VS). A standardized protocol was used, which included the study context, design and conduct, the adiposity measures, infant outcomes, and results reported. Two authors independently carried out Newcastle-Ottawa quality assessments for cohort and case control studies to assess information bias, selection bias, and confounding 35 (Table S1). Conflicts in data extraction or quality assessment were resolved by discussion or a third author. We assessed all included studies for duplicate publication of the same population. Two studies 36,37 reported data from the same cohort.
There was overlapping data relating to fat-free mass and birthweight, and newborn anthropometry (head circumference, crown-heel length) reported by both papers; therefore, we excluded these data reported by one of the papers 36 from the analysis.
Methods of analysis have been reported elsewhere. 29 Metaanalysis was carried out when three or more studies reported data suitable for pooling on the associations between early pregnancy adiposity measure and infant outcomes. All studies were examined for quality prior to meta-analysis. To be eligible for meta-analysis, the combinations of adiposity exposures and infant outcomes, and type of data (e.g., odds ratio [OR] and mean difference [MD]) needed to be similar enough to justify pooling. Summary of ORs, MDs, and average treatment effects (AE) were calculated using the random effects model by restricted maximum likelihood. 38,39 The I 2 statistic was used to assess the heterogeneity among studies, 40 with a threshold of >75% representing significant heterogeneity. 41 Due to the small number of studies in each meta-analysis, we were not able to perform metaregression, sensitivity analysis, sub-group analysis, or tests for publication bias 42,43 as planned in our PROSPERO protocol. 30 The statistical analyses were conducted using metafor 44 packages for R version 4.0.4.
A narrative synthesis was carried using recommendations by Popay et al. 45 when meta-analysis was not possible. Data from each study were tabulated and grouped according to the outcome being reported, sub-grouped by the adiposity exposure, and patterns were described.
The quality score of studies ranged from four (medium quality) to eight (high quality) (Table S5A, B). Twenty-five cohort studies were rated as high quality, eight were rated as medium quality, and no studies were rated low quality (Table S5A). Cohort studies consistently scored highly across all assessment criteria (>70%), where adequate length of follow-up (Q6) was met by all included cohort studies (100%), and the lowest scoring item was adequacy of follow-up (Q7) (73%).
The case control study had a score of eight (high quality) (Table S5B).

| Ratios and birthweight
Three studies 49,57,64 reported data for maternal WHR. Two 49,64 reported significant associations including a 0.1 unit increase in F I G U R E 1 PRISMA Flow-chart of the study selection process. Adapted from: Page et al 34 maternal WHR predicting 120 g greater newborn weight, 49 and a positive correlation (r 0.6, p < 0.01). 64 The third study 57  The third study 57 reported VAT thickness as being independently associated with birthweight centile (adjusted r 2 15.8%, p = 0.002) ( Table S6). The fourth study 56 reported a 1-cm increase in VAT depth associated with a 1.5 higher birthweight percentile, but the association was not statistically significant (adjusted odds ratio [AOR] 1.5, 95%CI À0.03, 3.00). One study 57 reported a significant but weak positive correlation with the combination of VAT + SAT (r 0.39, p = 0.004) (Table S6).

| Maternal SFT and birthweight
Two studies 47,73 reported no significant associations between measures of maternal SFT (triceps, biceps, subscapular, and suprailiac) and birthweight. One 73 reported negative but no significant association per SD increase in triceps (standardized estimate of À18g, 95% CI À100, 65) or subscapular (À39 g, 95% CI À118, 39) SFT. One 47 reported positive but non-significant correlation coefficients for maternal triceps (r 0.06), subscapular (r 0.09), and suprailiac SFT (r 0.15) (p value not reported), whereas no correlation was found for maternal biceps SFT (r 0, p value not reported) (Table S6)  F I G U R E 2 Meta-analysis of the association between maternal fat mass (in kilograms) and birthweight (kg). Total sample size (n = 3071), CI confidence interval, RE-random effect F I G U R E 3 Meta-analysis of the association between maternal fat free mass (in kilograms) and birthweight (kg). Total sample size (n = 3071), CI confidence interval, RE-random effect An overview of the meta-analysis and narrative synthesis for high birthweight are presented for each adiposity exposure.

| Subcutaneous fat and high birthweight
There were three studies 23

| Maternal neck circumferences and high birthweight
One study 54 reported an AUROC of 0.65 (95% CI 0.53, 0.75) for maternal neck circumference >36.5 cm to predict macrosomia (Table S7A) and significantly higher median maternal neck circumference for macrosomic infants compared to controls (Table S7B).

| Maternal WHR and high birthweight
Four studies 25 F I G U R E 4 Meta-analysis of the association between maternal WC (mean differences, cm) and macrosomia. Total sample size (n = 1826), MDmean difference (cm), CIconfidence interval, RE-random effect F I G U R E 5 Meta-analysis of the association between maternal subcutaneous fat (mm) and LGA. Total sample size (n = 5159), ORodds ratio, CIconfidence interval, RE-random effect reported the AUROC data for maternal WHR and LGA; one 25 reported the AUROC as being 0.514 (p = 0.57), and the other 66 reported 0.713 (p not reported but author classified >0.7 as a predictive value).

| Types of fat and high birthweight
Three studies 55,56,62 reported data for maternal visceral fat. One reported significantly increased odds of LGA per 5-mm increase in maternal visceral fat depth (AOR 1.06, 95%CI 1.02, 1.11
Maternal adiposity measures were early pregnancy WC, WHR, subcutaneous fat, visceral fat, FM, FFM, and SFT measures. There was a significantly increased odds of birthweight <2500 g with every 5-mm increase in maternal subcutaneous fat 24 (Table S9A). Two 46,71 reported significant positive correlations (p < 0.05) between maternal FM (including total, leg and arm FM in kilogram and percent) and fetal mid-thigh soft-tissue measurements at 36 week's gestation, 46 and change in estimated fetal weight between second and third trimester (standardized β 0.36, SE 9.75, p < 0.01) 71 but not with femur length between second and third trimester (standardized β À0.05, SE 0.01, p = 0.88) 71 (Table S9A) (Table S9B). This study also found a significant positive correlation for maternal FFM and newborn head circumference (r 0.057, SE 0.020, p < 0.01) and crown-heel length (r 0.067, SE 0.029, p < 0.05), but not for maternal FM and newborn head, chest, abdominal or midupper arm circumference, or crown-heel length. 37 Maternal FM was significantly and positively correlated with infant percent FM measured at 2 weeks old (r 0.14, 95% CI 0.07, 0.20) in one study 50 (Table S9B).

| DISCUSSION
This systematic review has identified a large body of existing evidence from 34 studies, including data from 40,143 pregnancies, which report associations between early pregnancy adiposity measured ≤20 weeks' gestation and infant health-related outcomes. Maternal early pregnancy WC was the most frequently reported adiposity measure, followed by WHR and measures of FM and FFM. Due to both heterogeneity in reporting, and the limited number of studies reporting data for the same combinations of adiposity exposures and infant outcomes, only four meta-analyses were performed. These showed a significant association between maternal FFM and birthweight, and for WC and macrosomia, but not for FM and birthweight or subcutaneous fat and LGA. The narrative synthesis of data that could not be included in the meta-analysis suggests that higher maternal WC was associated with birthweightrelated outcomes, maternal WHR was associated with high birthweight, maternal visceral fat with birthweight and preterm birth, and maternal subcutaneous fat with low birthweight outcomes. While other significant results were observed for different combinations of adiposity exposure and infant outcomes, there was an over-reliance on data from one or two studies contributing to the narrative synthesis (e.g., fetal growth outcomes), and therefore, results should be interpreted with caution.
This systematic review has strengths and limitations. There are several systematic reviews reporting maternal obesity, measured by BMI, and increased risk of adverse infant health outcomes. 2,3 To our knowledge, this is the first systematic review and meta-analysis that explore the use of different early pregnancy maternal adiposity measures that could be used to predict a range of infant health outcomes. Strengths of this review include the rigorous search strategy, supplementary searches, and additional information obtained from authors to maximize the number of studies in the meta-analyses. The screening, data extraction, and quality assessments were carried out in duplicate to minimize human error and subjectivity. We also aimed to maximize the number of studies we were able to pool in meta-analysis by transforming the data where appropriate. This was possible only in a limited number of cases, due to the heterogeneity in reporting. For example, there was a variability in the exposure definitions (e.g., adiposity was reported as both continuous measures and by categories). This review also has several limitations. There was a lack of consistency between studies in how outcomes were reported. A range of measures of association were reported, including ORs, correlations, means, medians, and AUROC, with some results from adjusted models and others were univariate analysis. When adjusted models were reported, there was a lack of consistency in the variables included, although maternal age, BMI, parity, behavioral factors (e.g., smoking), and socio-demographic factors (e.g., ethnic group) were the most consistently included variables.
Despite a wealth of existing data, there is a lack of standardized reporting of the adiposity measures across studies limiting the ability to pool results in meta-analysis. For example, the gestational age when adiposity was measured in the included studies varied and we were not able to explore this in the meta-regression or sub-group analysis. Indeed, the small number of studies we were able to include in each meta-analysis meant we were not able to explore sources of heterogeneity using meta-regression, sub-group analysis, publication bias tests, or sensitivity analysis as planned in our protocol. 30 One way to overcome some of the challenges with heterogeneous reporting is to use individual participant data (IPD) meta-analysis methods. This would enable a standardized approach to applying definitions to, and analyzing, the data across studies to facilitate direct comparison of adiposity measures to determine which might be best at predicting individual risk. 75,76 IPD metaanalysis would also facilitate the direct comparison with BMI, or combining adiposity measures with BMI, in the same population of women.
This research supports the need for early intervention in the prevention of adverse infant-related risks, starting preconception. 77 The current evidence base suggests that large-scale behavioral interventions have been successful at improving maternal behavior 78 and weight-related outcomes 78 but shows limited impact on infant health outcomes. 17 It has been suggested that this evidence demonstrates that interventions during pregnancy are "too little, too late" to fundamentally improve child health outcomes, 77 and the preconception period presents a greater opportunity for intervention, based on the life course approach and the embryo development around the time of conception. 77,79 However, interventions to date have either been universal (i.e., no targeting of high risk groups) or have targeted women based on maternal BMI. As previously discussed, evidence from non-pregnant populations shows that BMI only identifies half of individuals with adiposity-related risk. 22 Similarly, current evidence demonstrates that approximately half of women with an obese BMI have uncomplicated pregnancies and do not require high-risk care, whereas almost half of women with an overweight BMI develop complications despite not being considered as high risk. 80 Thus, the (lack of) usefulness of BMI to predict individual risk (and therefore allocation of care/intervention) that is seen in non-pregnant populations appears to be replicated in pregnancy. Basing decisions to offer additional care on individuals BMI will result in some women receiving additional care without needing it, while others who would benefit from additional care are excluded; therefore, alternatives to BMI need to be explored in the pregnancy context. This systematic review indicates that there are potential alternative measures of maternal adiposity to BMI that are associated with adverse infant health outcomes, which warrant further investigation to explore whether they better predict risk and could be used to inform targeted intervention approaches. Examples of some potential strong predictor variables include WC, visceral fat, WHR, and FFM. However, the lack of ability to conduct thorough meta-analysis of all potential adiposity measures and infant health outcomes limits drawing firm conclusions on which specific adiposity measures might be most useful. A number of measures such as WC or WHR have been previously suggested as alternatives to BMI to predict pregnancy related risk as they are more reflective of visceral fat and abdominal adiposity. There is evidence that WC and WHR are largely unaffected until 20 weeks' gestation; 28,81 thus, the measurement collected in early pregnancy could be a reliable indicator of pre-or early-pregnancy adiposity status. In addition to WC and WHR, findings from this review also suggest the potential use of visceral fat, which could be implemented by incorporating into routine antenatal ultrasound scan appointments.
Further research is required to explore whether a different approach to identifying adiposity-related risk than current use of BMI, and targeting interventions to women with greatest risk, would result in improved infant outcomes. The strong evidence relating to the fetal environment being important for life-long health 12-15 along with evidence of antenatal interventions in other behavioral fields significantly improving infant health, such as smoking cessation, 17 suggests that the pregnancy period is an opportunity to improve infant health, along with preconception interventions. Rather than future intervention research repeating diet and physical activity interventions in populations of women above specific BMI thresholds, we should be exploring ways to build on the knowledge we have to optimize effectiveness.
Future studies should explore whether adiposity measures could be used in routine care to improve our ability to identify women and infants with the greatest risk, and whether additional clinical care or behavior change interventions are effective at improving infant health outcomes if they are targeted towards women with high adiposity.

TB is funded by the NIHR Applied Research Collaboration North East
and North Cumbria (Grant/Award Number: NIHR200173). We would like to thank Dr Finona Beyer for her help with translating the search strategy into CINAHL database. We would also like to thank authors: Dr Dilys Freeman and Dr Mehrabi Esmat who responded to our request for additional information.

CONFLICTS OF INTEREST
None to declare.