Examining the predictive accuracy of metabolomics for small-for-gestational-age babies: a systematic review

Introduction To date, there is no robust enough test to predict small-for-gestational-age (SGA) infants, who are at increased lifelong risk of morbidity and mortality. Objective To determine the accuracy of metabolomics in predicting SGA babies and elucidate which metabolites are predictive of this condition. Data sources Two independent researchers explored 11 electronic databases and grey literature in February 2018 and November 2018, covering publications from 1998 to 2018. Both researchers performed data extraction and quality assessment independently. A third researcher resolved discrepancies. Study eligibility criteria Cohort or nested case–control studies were included which investigated pregnant women and performed metabolomics analysis to evaluate SGA infants. The primary outcome was birth weight <10th centile—as a surrogate for fetal growth restriction—by population-based or customised charts. Study appraisal and synthesis methods Two independent researchers extracted data on study design, obstetric variables and sampling, metabolomics technique, chemical class of metabolites, and prediction accuracy measures. Authors were contacted to provide additional data when necessary. Results A total of 9181 references were retrieved. Of these, 273 were duplicate, 8760 were removed by title or abstract, and 133 were excluded by full-text content. Thus, 15 studies were included. Only two studies used the fifth centile as a cut-off, and most reports sampled second-trimester pregnant women. Liquid chromatography coupled to mass spectrometry was the most common metabolomics approach. Untargeted studies in the second trimester provided the largest number of predictive metabolites, using maternal blood or hair. Fatty acids, phosphosphingolipids and amino acids were the most prevalent predictive chemical subclasses. Conclusions and implications Significant heterogeneity of participant characteristics and methods employed among studies precluded a meta-analysis. Compounds related to lipid metabolism should be validated up to the second trimester in different settings. PROSPERO registration number CRD42018089985.

Fetal growth restriction (FGR) and small for gestational age (SGA) infants are major 3 concerns in modern obstetrics. [1-3] SGA is commonly used as a proxy for FGR, [4] 4 despite the subtle differences between these two pathological conditions. The 5 prevalence of both varies according to criteria applied and on the population and 6 setting, although it reaches as much as 25% in low and middle-income countries. [5] 7 SGA newborns may have adverse health effects, such as stillbirth, [4] perinatal 8 asphyxia, [6] impaired neurodevelopment, [7] and increased cardiovascular risk. [8,9] 9 To date, there are no robust prediction tools for SGA using clinical factors, [10,11] 10 ultrasound data, [12,13] or placental biomarkers. [14] 11 For hypothesis generating or validation purposes, metabolomics is a novel 12 area of biomarker, discovery, development and clinical diagnostics in translational 13 medicine. [15,16] Metabolomics is the study of all metabolites [15,16] in a given 14 sample, i.e. low molecular weight compounds (50-2000 Da) that are intermediates of 15 biochemical reactions and metabolic pathways, considered to directly reflect cellular 16 activity and phenotype. [15,16] Recent studies have evaluated the pathophysiology 17 [17-20] of SGA with metabolomics. However, little is known about the potential of 18 metabolomics to identify predictive compounds of SGA. In this context, the main objective of this systematic review was to assess the 2 accuracy of metabolomics techniques in predicting SGA. As a secondary aim, we 3 intended to determine which metabolites are predictive of this condition. 4 5 METHODS 6 The protocol for this systematic review was published previously.
[23] This study 7 follows international guidelines for transparency (PROSPERO, CRD 42018089985) 8 and respects the Preferred Reporting Items for Systematic Reviews and Meta-9 Analysis (PRISMA) statement.
[24] This systematic review was conducted without 10 any public involvement, and ethical approval was unnecessary.  OR 'metabolit* 'H NMR' OR 'proton NMR' OR 'proton nuclear magnetic resonance' 2 OR 'liquid chromatogra*' OR 'gas chromatogra*' OR 'UPLC' OR 'ultra-performance' 3 OR 'ultra performance liquid chromatograph*') AND ('pregnan*' OR 'antenat*' OR 4 'ante nat*' OR 'prenat*' OR 'pre nat*') AND ('screen*' OR 'predict*' OR 'metabolic 5 profil*'). 6 7 Outcomes and subgroup analysis 8 The primary outcome was SGA, as a surrogate for FGR and defined as birthweight 9 <10 th centile, by population-based or customized charts. Secondary outcomes were 10 birthweight ≤5th or ≤3rd centile. 11 The intended subgroup analysis comprised: type of metabolomics technique 12 applied (nuclear magnetic resonance, NMR; gas or liquid chromatography coupled 13 with mass spectrometry, GC-MS or LC-MS respectively); maternal health status 14 before pregnancy (women with versus without any chronic health condition); type of 15 SGA suspected during pregnancy (early versus late SGA); and type of pregnancy 16 (singleton versus multiple pregnancy). duplicate data, in which case the most complete publication was included for final 2 analysis. 3 Two researchers (DFBL and ACM) independently selected studies, extracted 4 data and discussed discrepancies. One additional reviewer (EFMJ or RTS) helped to 5 decide, by majority, when no consensus was reached. 6 Piloted standardized forms were applied for data extraction, including 7 pregnancy characteristics and experimental details. The Human Metabolome 8 Database (HMDB) [25] and the Kyoto Encyclopedia of Genes and Genomes [26] 9 were used for matching chemical class and metabolic pathways of each metabolite, 10 respectively.  interpretation were standardized or metabolite threshold was defined before the 2 experiments (for targeted analysis); (iii) if the adequacy and reasons for choosing the 3 reference birthweight chart had been explained; or, (iv) if large for gestational age 4 babies had been excluded from the final comparative analysis. 5 6 Data synthesis 7 A quantitative summary of data was performed when any predictive accuracy 8 measures could be extracted. Authors were contacted to provide additional 9 information, when necessary. However, only Delplancke et al [28] replied. The 10 estimation of likelihood ratios and hierarchical summary receiver operator 11 characteristic curve [29] were planned, as well as assessment of heterogeneity and 12 publication bias.

15
Main findings 16 In this first systematic review of metabolomics and adverse pregnancy endpoints, we 17 presented techniques and metabolites, which were studied for the prediction of SGA. 18 Any effect on birthweight has important implications for perinatal research, since it is 19 related to short and long-term outcomes, [43][44][45][46]  Interpretation of metabolomics findings in pregnancy can be challenging. 25 Firstly, maternal metabolites concentrations are influenced by placental transfer to and from the fetus. The 'mirror effect', seen for maternal plasma and venous cord 2 blood metabolites at birth, [51] cannot be ruled out when only maternal specimens 3 are studied. Secondly, maternal exposure to distinct compounds may affect 4 metabolite levels. Statistically significant differences between SGA infants and 5 controls may not express the totality of underlying pathological pathways and have 6 no clinical meaning. Finally, it is unclear when the processes leading to SGA are 7 initiated. The disruption in maternal metabolism can theoretically occur at any time. 8 In general the lower the gestational age at which the condition is suspected, the 9 more severe the phenotype will be at birth. [52,53] Thus, the description of clinical 10 data in translational studies must deal with all these confounding factors. 11 Gestational age at sampling is probably the most important parameter for 12 prediction purposes. With timely prediction, women could be referred to specialized 13 care, have increased surveillance, and this in turn may lead to a reduction in 14 perinatal mortality. There are temporal changes in the maternal metabolome during 15 pregnancy; [28,54-57] therefore, it is reasonable to expect distinctive metabolites at 16 different stages of pregnancy, as reported here. Unfortunately, a wide or unclear 17 definition of gestational age of sampling [34,36,38,40] render a more precise 18 interpretation impossible, and may limit the clinical application of these results. 19 In contrast, gestational age at birth and birthweight centile seem to be the 20 hallmarks of severity and prognosis of growth restriction. [6,58]  related chemicals or quantify exposure to tobacco smoke. Therefore, no relationship 10 between SGA and tobacco was found. Hence, we suggest that tobacco interferes 11 with ongoing metabolic pathological processes, or its disturbance is related to 12 additional metabolic pathways other than the one examined by the included studies. selection have hindered data interpretation and precluded these analyses. 18 The majority of included studies performed a targeted approach, i.e. a 19 hypothesis-testing evaluation, [16,50]  Most studies were ranked as 'low risk' of bias or applicability to the review question. 16 However, the lack of clear descriptions of laboratory experiments, including sample 17 preparation and storage, and blinding of the researchers to the case/control status, 18 are major pitfalls of the included studies.   16,25] However, findings of this 2 systematic review must be interpreted with caution. The type of samples used may 3 have influenced LC-MS (2 nd trimester maternal blood) and GC-MS (2 nd trimester 4 maternal hair) findings in individual studies. Furthermore, the prediction of SGA in 5 the context of maternal disorders, suspected FGR and twin pregnancies is an open 6 field for future metabolomics studies, and environmental exposure investigation as 7 well. 8 Surprisingly, none of the studies used ≤3 rd centile of birthweight as a cutoff 9 or analyzed preterm deliveries and hypertensive syndromes. Considering our 10 findings and the different phenotypic manifestations of SGA, we envision a better 11 performance when (i) cutoffs other than the 10 th centile are tested; (ii) data on 12 gestational age at sampling and at birth are standardized; and (iii) other pregnancy-13 related syndromes are considered, especially hypertension. Thus, future 14 metabolomics results should advance in these critical points. 15 Finally, all detected biomarkers were related to lipid pathways and energy 16 metabolism. We consider that research efforts to predict SGA should focus on 17 compounds involved in these pathways, up to the 2 nd trimester of pregnancy.   None to declare.     Eligibility criteria 6 Specify study characteristics (e.g., PICOS, length of follow-up) and report characteristics (e.g., years considered, language, publication status) used as criteria for eligibility, giving rationale.

7-8
Information sources 7 Describe all information sources (e.g., databases with dates of coverage, contact with study authors to identify additional studies) in the search and date last searched.

6-7/ 9
Search 8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.

6-7
Study selection 9 State the process for selecting studies (i.e., screening, eligibility, included in systematic review, and, if applicable, included in the meta-analysis).

7-8
Data collection process 10 Describe method of data extraction from reports (e.g., piloted forms, independently, in duplicate) and any processes for obtaining and confirming data from investigators.

8
Data items 11 List and define all variables for which data were sought (e.g., PICOS, funding sources) and any assumptions and simplifications made.

Risk of bias in individual studies
12 Describe methods used for assessing risk of bias of individual studies (including specification of whether this was done at the study or outcome level), and how this information is to be used in any data synthesis.

8-9
Page 53 of 55 Risk of bias across studies 15 Specify any assessment of risk of bias that may affect the cumulative evidence (e.g., publication bias, selective reporting within studies).

RESULTS
Study selection 17 Give numbers of studies screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally with a flow diagram.

9
Study characteristics 18 For each study, present characteristics for which data were extracted (e.g., study size, PICOS, follow-up period) and provide the citations.

9-13
Risk of bias within studies 19 Present data on risk of bias of each study and, if available, any outcome level assessment (see item 12).

23-24
Results of individual studies 20 For all outcomes considered (benefits or harms), present, for each study: (a) simple summary data for each intervention group (b) effect estimates and confidence intervals, ideally with a forest plot.

20-22
Synthesis of results 21 Present results of each meta-analysis done, including confidence intervals and measures of consistency. 24 Risk of bias across studies 22 Present results of any assessment of risk of bias across studies (see Item 15). 9; 23-24 Additional analysis 23 Give results of additional analyses, if done (e.g., sensitivity or subgroup analyses, meta-regression [see Item 16]). NA

DISCUSSION
Summary of evidence 24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers).

24-28
Limitations 25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias).

28-29
Conclusions Results: A total of 9,181 references were retrieved. Of these, 273 were duplicate, 19 8,760 were removed by title or abstract, and 133 were excluded by full text content. 20 Thus, 15 studies were included. Only two studies used the 5 th centile as a cutoff, and 21 most reports sampled 2 nd trimester pregnant women. Liquid-chromatography coupled 22 to mass spectrometry was the most common metabolomics approach. Untargeted 23 studies in the 2 nd trimester provided the largest number of predictive metabolites, 24 using maternal blood or hair. Fatty acids, phosphosphingolipids, and amino acids 25 were the most prevalent predictive chemical subclasses.

11
 To our knowledge, this is the first systematic review to assess the predictive 12 accuracy of metabolomics for an adverse pregnancy outcome.

13
 Using SGA as surrogate for fetal growth restriction -just as in epidemiological 14 investigations -improves the translational potential of metabolomics.

15
 Identification of techniques, types of maternal samples and chemical classes 16 paves the way for future metabolomics investigations on fetal growth patterns.

METHODS
Protocol and registration 5 Indicate if a review protocol exists, if and where it can be accessed (e.g., Web address), and, if available, provide registration information including registration number.
6 Eligibility criteria 6 Specify study characteristics (e.g., PICOS, length of follow-up) and report characteristics (e.g., years considered, language, publication status) used as criteria for eligibility, giving rationale.

7-8
Information sources 7 Describe all information sources (e.g., databases with dates of coverage, contact with study authors to identify additional studies) in the search and date last searched.

6-7/ 9
Search 8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.

6-7
Study selection 9 State the process for selecting studies (i.e., screening, eligibility, included in systematic review, and, if applicable, included in the meta-analysis).

7-8
Data collection process 10 Describe method of data extraction from reports (e.g., piloted forms, independently, in duplicate) and any processes for obtaining and confirming data from investigators.

8
Data items 11 List and define all variables for which data were sought (e.g., PICOS, funding sources) and any assumptions and simplifications made.

Risk of bias in individual studies
12 Describe methods used for assessing risk of bias of individual studies (including specification of whether this was done at the study or outcome level), and how this information is to be used in any data synthesis.

8-9
Page 54 of 56 Risk of bias across studies 15 Specify any assessment of risk of bias that may affect the cumulative evidence (e.g., publication bias, selective reporting within studies).

RESULTS
Study selection 17 Give numbers of studies screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally with a flow diagram.

9
Study characteristics 18 For each study, present characteristics for which data were extracted (e.g., study size, PICOS, follow-up period) and provide the citations.

9-13
Risk of bias within studies 19 Present data on risk of bias of each study and, if available, any outcome level assessment (see item 12).

23-24
Results of individual studies 20 For all outcomes considered (benefits or harms), present, for each study: (a) simple summary data for each intervention group (b) effect estimates and confidence intervals, ideally with a forest plot.

20-22
Synthesis of results 21 Present results of each meta-analysis done, including confidence intervals and measures of consistency. 24 Risk of bias across studies 22 Present results of any assessment of risk of bias across studies (see Item 15). 9; 23-24 Additional analysis 23 Give results of additional analyses, if done (e.g., sensitivity or subgroup analyses, meta-regression [see Item 16]). NA

DISCUSSION
Summary of evidence 24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers).

24-28
Limitations 25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias).