Article Text

Original research
Association of maternal lipid levels with birth weight and cord blood insulin: a Bayesian network analysis
  1. Jingya Wang1,2,
  2. Yashu Kuang1,3,4,
  3. Songying Shen1,3,4,
  4. Malcolm James Price2,
  5. Jinhua Lu1,3,4,
  6. Naveed Sattar5,
  7. Jianrong He1,3,4,
  8. Marta Pittavino6,
  9. Huimin Xia1,4,
  10. G.Neil Thomas2,
  11. Xiu Qiu1,4,7,
  12. Kar Keung Cheng2,
  13. Krishnarajah Nirantharakumar2
  1. 1Division of Birth Cohort Study, Guangzhou Women and Children's Medical Center, Guangzhou Medical University, Guangzhou, China
  2. 2Institute of Applied Health Research, University of Birmingham, Birmingham, UK
  3. 3Guangdong Provincial Key Laboratory of Research in Structure Birth Defect Disease, Guangzhou Women and Children's Medical Center, Guangzhou Medical University, Guangzhou, China
  4. 4Guangdong Provincial Clinical Research Center for Child Health, Guangzhou Women and Children's Medical Center, Guangzhou Medical University, Guangzhou, China
  5. 5Institute of Cardiovascular and Medical Sciences, University of Glasgow, Glasgow, UK
  6. 6University of Geneva, Geneva, Switzerland
  7. 7Department of Women's Health, Guangdong Provincial Key Clinical Specialty of Woman and Child Health, Guangzhou Women and Children's Medical Center, Guangzhou Medical University, Guangzhou, China
  1. Correspondence to Dr Xiu Qiu; qxiu0161{at}; Dr Krishnarajah Nirantharakumar; k.nirantharan{at}; Dr Jingya Wang; J.Wang.6{at}


Objective To assess the independent association of maternal lipid levels with birth weight and cord blood insulin (CBI) level.

Setting The Born in Guangzhou Cohort Study, Guangzhou, China.

Participants Women who delivered between January 2015 and June 2016 and with umbilical cord blood retained were eligible for this study. Those with prepregnancy health conditions, without an available fasting blood sample in the second trimester, or without demographic and glycaemic information were excluded. After random selection, data from 1522 mother–child pairs were used in this study.

Exposures and outcome measures Additive Bayesian network analysis was used to investigate the interdependency of lipid profiles with other metabolic risk factors (prepregnancy body mass index (BMI), fasting glucose and early gestational weight gain) in association with birth weight and CBI, along with multivariable linear regression models.

Results In multivariable linear regressions, maternal triglyceride was associated with increased birth weight (adjusted β=67.46, 95% CI 41.85 to 93.06 g per mmol/L) and CBI (adjusted β=0.89, 95% CI 0.06 to 1.72 μU/mL per mmol/L increase), while high-density lipoprotein cholesterol was associated with decreased birth weight (adjusted β=−45.29, 95% CI −85.49 to −5.09 g per mmol/L). After considering the interdependency of maternal metabolic risk factors in the Network analysis, none of the maternal lipid profiles was independently associated with birth weight and CBI. Instead, prepregnancy BMI was the global strongest factor for birth weight and CBI directly and indirectly.

Conclusions Gestational dyslipidaemia appears to be secondary to metabolic dysfunction with no clear association with metabolic adverse outcomes in neonates. Maternal prepregnancy overweight/obesity appears the most influential upstream metabolic risk factor for both maternal and neonatal metabolic health; these data imply weight management may need to be addressed from the preconception period and during early pregnancy.

  • General endocrinology
  • Lipid disorders
  • Maternal medicine

Data availability statement

All data relevant to the study are included in the article or uploaded as online supplemental information. No additional data available.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • To the best of our knowledge, this is the first study to date focusing on maternal lipids during pregnancy that weighs key maternal metabolic risk factors systematically using an innovative additive Bayesian network analytical tool.

  • Given the practical constraints, maternal insulin resistance status during pregnancy, as a potential significant explanatory factor, was not included in the current analysis, although maternal fasting glucose level was included.

  • Due to the limited ability of detecting feedback loops, partial results of Bayesian network analysis should be interpreted with a degree of caution.

  • Risk of bias due to self-reported prepregnancy weight can not be excluded.


Unfavourable fetal growth is often considered to attribute to an adverse intrauterine nutritional exposure that is largely dependent on maternal metabolic status.1 2 Other than those established maternal metabolic risk factors (prepregnancy body mass index (BMI), gestational weight gain (GWG) and gestational hyperglycaemia) for adverse birthweight outcomes,3–5 maternal lipid levels during pregnancy are increasingly recognised as ignored risk factors for adverse pregnancy outcomes in the last few years.6–8 With the evidence from 42 observational studies using classic statistical methods, our recent systematic review found that increased maternal triglycerides and decreased high-density lipoprotein cholesterol (HDL-C) were positively associated with high birth weight.9

However, the majority of previous studies only focused on the association between maternal lipid levels and birth weight, but ignored the underlying interdependency between lipids and other metabolic factors.10 11 Using data from 400 The Hyperglycemia and Adverse Pregnancy Outcome Study Caucasian mother–child pairs, two network analyses investigated the associations of maternal metabolites with maternal BMI, glucose level, birth weight or cord blood c-peptide.2 12 These studies only considered the interdependency of maternal metabolites and their association with a single phenotype rather than putting a broad range of metabolic traits and phenotypes together to draw a panoramic picture. Understanding how maternal lipid profiles influence neonatal metabolic health taking into consideration the clinical metabolic network into account is crucial to understand any potential underlying mechanism and future interventional studies.

In a birth cohort from Chinese pregnant women and their newborn babies, this study aimed to investigate the association of maternal lipid levels with birth weight and cord blood insulin (CBI) (indicators of neonatal developmental and metabolic conditions), with the consideration of the interdependency of maternal lipids with other maternal metabolic risk factors, including prepregnancy BMI, fasting glucose and GWG.

Materials and methods


This study is part of the Born in Guangzhou Cohort Study (BIGCS), a prospective cohort study conducted in the Guangzhou Women and Children’s Medical Centre (GWCMC). The design and methods of BIGCS have been described previously.13 In brief, eligible women with Chinese nationality, living in Guangzhou who are <20 weeks gestation and who intend to deliver at one of the two GWCMC campuses were recruited into BIGCS. This study was conducted in a planned subgroup of BIGCS in whom maternal and cord blood were analysed for metabolic parameters separately. Pregnant women attending BIGCS with a singleton pregnancy who delivered at GWCMC between January 2015 and June 2016 and had umbilical cord blood retained are eligible for this study. Women were excluded if (1) maternal blood samples were unavailable at 14–27 gestation week; (2) no records of maternal fasting glucose at 20–28 gestation week; (3) lacking maternal demographic information and (4) diagnosed with one or more health conditions prior to pregnancy, including type 1 or type 2 diabetes, thyroid dysfunction, hypertension, virus hepatitis and renal diseases. Sample sizes were calculated according to the association between maternal triglycerides (the potential weakest risk factors among maternal metabolic traits) with birth weight according to literature (online supplemental file S1). The eligible mother–child pairs were then selected for this study by computer-generated randomisation.

Maternal data collection

Maternal information, including age, height, prepregnancy weight, parity, date of last menstrual period, monthly income, education levels and ethnicity, was collected through semistructured questionnaire. Maternal BMI was calculated by dividing weight in kilograms by height in metres squared. Based on the recommendations of the China Obesity Task Force of the Chinese Ministry of Health, maternal prepregnancy BMI is classified into two groups: lean group (<24 kg/m2) and overweight group(≥24 kg/m2).14 At 22–28 weeks gestation, women attending their second prenatal visit underwent a standard 2 hours 75 g oral glucose tolerance test (OGTT). Women with OGTT results which met or exceeded at least one threshold of the International Association of Diabetes and Pregnancy Study Groups criteria (fasting plasma glucose ≥5.1 mmol/L, 1-hour glucose ≥10 mmol/L and 2 hours glucose ≥8.5 mmol/L) were diagnosed as having gestational diabetes mellitus (GDM).15 Maternal fasting glucose concentration was obtained from OGTT test zero-time value in hospital records. Maternal second-trimester weight was measured to the nearest 0.1 kg using an electronic scale. Maternal early GWG was calculated by subtracting prepregnancy weight from maternal second-trimester weight, with documentation of the gestational age at measurement.


The outcomes of this study are neonatal birth weight and CBI. We considered birth weight as it is an important indicator of prenatal developmental conditions for newborns. Birth weight has been associated with both short-term and long-term health outcomes, including stillbirth, infant mortality, obesity, type 2 diabetes and cardiovascular diseases.16–18 Similarly, CBI, which is synthesised only by fetal pancreas,19 is also considered as an important biomarker that could reflect insulins resistance status in neonates, since insulin plays a central role in fetal growth and development.20

Biochemical test

Mothers were asked to keep overnight fasting status before taking blood samples in the second trimester. Maternal real fasting status was asked again and recorded before collecting samples. Maternal blood samples during the second trimester were used for testing the level of lipid profiles. Venous umbilical cord blood plasma samples were collected for insulin tests. Sample collection, delivery, pretreatment and measurements were blinded. All blood samples were stored and delivered to the pretreatment laboratory centre. Blood samples were then separated into serum and plasma by immediate centrifugation and were stored in EDTA tubes in the Biobank at −80°C until analysis. Maternal plasma lipids, including total cholesterol (TC), HDL-C, low-density lipoprotein cholesterol (LDL-C), and triglycerides, and CBI levels were measured using commercial kits in fully automated clinical analyser (Roche Diagnostics, Mannheim, Germany). Intraday and interday coefficients of variation (CVs) were consistently less than 2% for all assays.

Neonatal anthropometry data collection

For participating children, birth information, including birth characteristics, delivery mode and perinatal outcomes, was obtained from routine medical records. Gestational age was estimated from ultrasound examination during the first or second trimester. Birth weight and other information, including gestational age at delivery, mode of delivery, neonatal sex and pregnancy complications, were obtained from hospital records. Birth weight was measured to the nearest 50 g using an electronic scale by midwives immediately after delivery. Birth weight Z-Score and percentile adjusted for gestational age at delivery and neonatal sex were calculated using Intergrowth 21st Newborn Size Standard and Tools.21 Large-for-gestational age (LGA) was defined as a birth weight larger than the 90th percentile for gestational age by sex, while small-for-gestational age (SGA) was defined as a birth weight smaller than the 10th percentile based on the same birth weight reference.

Patient and public involvement

No patient involved.

Statistical analysis

Classic statistical methods

Continuous variables were summarised as mean±SD or median (IQR), and categorical variables were summarised as counts with percentages. Pearson’s correlation was used to assess the impact of the long-term −80°C storage on insulin concentrations in EDTA tube. Multiple imputation was used to handle missing data. Adjustments were then made to account for any degradation by correcting the initial value using linear regression methods (online supplemental file S2). Similarly, maternal lipid levels were adjusted for gestational age using regression model to account for timing of blood sampling (online supplemental file S3).22

Initially, multivariable linear regression models were used to estimate the association of maternal lipid levels with neonatal birth weight and CBI level. Covariates in the regression model included maternal age, ethnic group, parity, gestational age, maternal fasting status, neonatal sex and early pregnancy cigarette exposures. Delivery mode and sample storage duration were added in the model for CBI level. Thereafter, same model was used to investigate the association of maternal lipid Z-Scores with birth weight Z-Score and CBI Z-Score. To investigate the independent associations of maternal lipid levels with neonatal birth weight and CBI, model was further adjusted for maternal prepregnancy BMI Z-Score, maternal fasting glucose Z-score, GWG Z-score and gestational age of maternal weight measurement during pregnancy. Sensitivity analyses were conducted to compare the estimate differences between GDM and non-GDM participants, primiparous women and non-primiparous women, lean and overweight group, as well as before and after multiple imputations. All statistical tests were two tailed and a p<0.05 was considered statistically significant. Statistical analyses were performed in Stata V.14.0.

Additive Bayesian networks analysis

To further assess the interdependency between maternal metabolic risk factors and their association with birth weight and CBI, additive Bayesian Network (ABN) model—an unsupervised machine learning method—was conducted. Bayesian network analysis is a form of structure discovery statistical modelling that derives, from empirical data, a graphical network describing the dependency structure between variables, shown as directed acyclic graphs (DAGs).23 ABNs comprise DAGs where each node in the graph comprises a generalised linear model or a generalised linear mixed model. ABN model is suitable for analysing highly complex epidemiological data comprising many interdependent variables.24

Ten variables were chosen for ABN based on prior knowledge gained from literature and findings of the classical statistical analyses. These 10 variables were maternal age, maternal prepregnancy BMI, maternal fasting glycaemia in OGTT, early GWG, maternal fasting plasma HDL-C and triglycerides in the second trimester, birthweight Z-Score, CBI, gestational age at delivery and neonatal sex. GWG was adjusted for gestational age at weight measurement in mid-pregnancy. CBI was adjusted for sample storage duration. All continuous variables were standardised to Z-Scores to eliminate the influence of different measurement units. Mother–child pairs with missing data were excluded (n=93/1522, 6%).

First, an optimal DAG with the best goodness of fit (highest log marginal likelihood) was identified. Next, parametric bootstrapping (12 800 samples) was performed to address the potential overfitting. Full technical details are provided in online supplemental file S4, supplemental figure S1–S5. ABN analysis was conducted in R V.3.4.4 (The R Foundation for Statistical Computing) using ‘abn’ package.24

Figure 2

Additive Bayesian network graph. β, standardised regression coefficient; BMI, body mass index; GWG, gestational weight gain; HDL-C, high-destiny-lipoprotein cholesterol; TG; triglycerides.


Figure 1 shows the flow diagram of study. A total of 5497 pregnant women attending BIGCS between January 2015 and June 2016 are eligible for this study. Women whose blood samples were unavailable at 14–27 gestation week (n=902), who had no records of maternal fasting glucose at 20–28 gestation week (n=343), lacked maternal demographic information (n=39) and diagnosed with health condition prior to pregnancy (n=118) were excluded. The remaining 4039 women were subjected to randomisation, resulting in 1522 women were selected for maternal lipid profiles and CBI testing. The baseline characteristics of participants are shown in table 1. The majority (91.20%) of maternal blood samples were collected after overnight fasting. Maternal mid-pregnancy weight, fasting glucose and lipids profile were measured at a mean of 20.0 (SD=4.0), 24.6 (SD=1.4) and 20.5 (SD=3.5) gestation weeks, respectively. Cord blood samples were stored for a median of 488 (IQR 394–707) days before analysis.

Table 1

Baseline characteristics table

Table 2 presents the associations between the association of maternal mid-pregnancy lipid levels with neonatal birth weight and CBI level. The elevated maternal triglycerides level was associated with increased birth weight (adjusted β=67.46, 95% CI 41.85 to 93.06 g per mmol/L) and CBI level (adjusted β=0.89, 95% CI 0.06 to 1.72 μU/mL per mmol/L). Conversely, the increased maternal HDL-C level was negatively associated with decreased birth weight (adjusted β=−45.29, 95% CI −85.49 to −5.09 g per mmol/L), but not associated with CBI level (adjusted β=−0.82, 95% CI −2.12 to 0.48 μU/mL per mmol/L). No statistically significant association was observed between maternal TC and LDL-C levels and birth weight/CBI.

Table 2

Association of maternal lipid levels with birth weight and cord blood insulin level

Following further adjustment of prepregnancy BMI, GWG and maternal glucose level, only the association between maternal triglyceride Z-Score and birth weight Z-Score remains statistically significant (standardised β=0.07, 95% CI 0.03 to 0.11) (online supplemental table S1 and S2). No statistically significant result was observed in the sensitivity analysis (online supplemental table S3). It is worth noticing that the association estimate of maternal glucose Z-Score with birth weight Z-Score differed between non-GDM and GDM participants (standardised β: 0.06 vs 0.16, p=0.06), while the association estimate of maternal prepregnancy BMI Z-Score with birth weight Z-Score differed between primiparous and multiparous women (standardised β: 0.18 vs 0.28, p=0.06). Estimates for the association of maternal HLD-C Z-Score (standardised β: −0.03 vs −0.13, p=0.17), prepregnancy BMI Z-Score (standardised β: 0.03 vs 0.14, p=0.38) and glucose Z-Score (standardised β: 0.10 vs 0.19, p=0.13) with CBI Z-Score differed between non-GDM and GDM women. Between lean and overweight women, a difference in the association between maternal HDL-C Z-Score and CBI Z-Score was observed (standardised β: −0.03 vs −0.12, p=0.24). All estimates remain the same before and after multiple imputations.

Figure 2 and online supplemental table S4 show the optimal summary DAGs inferred by ABN analysis. Neither triglycerides nor HDL-C were linked to birth weight or CBI in the ABN results after accounting for the interdependency of maternal metabolic factors (including lipids). Maternal prepregnancy BMI was associated with all other maternal metabolic parameters in pregnancy (glycaemia: β=0.14, 95% CI 0.09 to 0.19; early GWG: β=−0.12, 95% CI −0.17 to −0.06; triglycerides: β=0.23, 95% CI 0.18 to 0.28; HDL-C: β=−0.12, 95% CI −0.17 to −0.07) and birth weight (β=0.27, 95% CI 0.22 to 0.32). An indirect association between prepregnancy BMI and CBI was also observed. Early GWG was associated with birth weight Z-Score (β=0.17, 95% CI 0.12 to 0.22). Maternal glycaemia was associated with CBI (β=0.12, 95% CI 0.07 to 0.17). Birth weight was also associated with CBI (β=0.24, 95% CI 0.19 to 0.29).


To the best of our knowledge, this is the first large prospective birth cohort study that considered the interdependency of maternal lipids profiles with other metabolic risk factors (maternal prepregnancy weight, GWG and gestational glycaemia level) when assessing the association between maternal circulating lipids levels and birth outcomes. We found that maternal mid-pregnancy HDL-C and triglycerides levels may not independently associated with birth weight and CBI level using Bayesian network analysis, although statistically significant associations—consistent with previous studies—were observed using conventional multivariable linear regression models.

Our results demonstrated that lipid pathways may not be meaningfully involved in the metabolic network pathway between mothers and neonates, and instead be a proxy measure for maternal metabolic health. Similarly, a previous study using Mendelian randomisation, another established causal inference method, analysing data from 30 487 women in 18 studies concluded that genetically higher maternal fasting HDL-C/triglycerides was not potentially causally associated with higher birth weight.25 Thus, both detailed pathways analyses in this paper and genetic finding suggest that lipid pathways are not causally related to birth weight.

Although the underlying mechanism of the absence of association between maternal lipid profiles and birth weight and CBI remains unclear, it is biologically plausible. Pregnant women with a higher prepregnancy BMI, increased GWG or/and glucose intolerance often have associated dyslipidaemia.1 26–29 The transportation efficiency of free fatty acids and glycerol broken down from triglycerides across placenta is significantly influenced by maternal insulin level, although it cannot cross the placenta.19 30 Elevated insulin concentration in fetus triggered by the increased maternal hyperglycaemia/hyperinsulinaemia could promote the circulating free fatty acids, triacylglycerol and glucose uptake in adipose and muscle tissue, block glycogenolysis and gluconeogenesis in the liver, and stimulate glycogen synthesis, resulting in enlarged adipose tissue in fetus.31–35 Conversely, the enlarged adipocytes could further induce insulin resistance, leading to an elevated insulin level in fetus.31 36 Therefore, although the glycerol and free fatty acids delivered from the maternal side are the essential nutrient substrate for neonatal de novo lipogenesis, it is plausible that the maternal circulating lipid levels, the placental transport of lipids and the uptake of lipids in the neonatal side are mainly determined by the maternal insulin resistance status driven by the body weight, insulin resistance status and glycaemic level in mothers.

Since 1950s, Pedersen hypothesis suggested that the increase in glucose transport from GDM mothers to fetus induces fetal hyperinsulinaemia, and results in subsequently adipose accumulation in neonates.37–39 In turn, the enlarged adipocyte will gradually become resistant to insulin to avoid further expansion, therefore, contributing to the increased insulin secretion in neonates.40 Hence, there may be a bidirectional relationship between birth weight and CBI. Due to the nature of DAG and lacking dynamic data, the ABN analysis might have a limited ability on exploring feedback loops. That means that the current DAG generated by the ABN analysis missed an arc from CBI level to birth weight, and that may be the reason why no arc from maternal glucose level to neonatal birth weight was observed.

Our ABN analysis suggested that the maternal prepregnancy BMI is the overweight/obesity is the most influential upstream metabolic risk factor for both maternal and neonatal metabolic health. Previous evidence shows maternal overweight/obesity has a persistent impact on the quality and content of oocytes as well as the epigenetic and translation profile of the developing embryo, thereby leading to altered phenotypes in offspring.41–44 Compared with lean women, women with a higher prepregnancy BMI tend to have increased risks of excessive GWG as well as GDM and often present with more severe dyslipidaemia during pregnancy.45–47 The hyperinsulinaemia and systemic low-grade inflammation driven by maternal obesity, gestational hyperglycaemia and gestational dyslipidaemia could potentially induce functional changes in the placenta, leading to elevated nutrients uptake and transport.48 49 Stress induced by the overwhelming nutrient exposure in utero can systematically shape fetal development, leading to permanent changes of metabolic function in offspring.50


If replicated, the findings in this study may have implications for antenatal/prenatal counselling and managing to avoid adverse weight-related birth outcomes. Most current clinical guidelines on preconception and antenatal care only focus on weight management during pregnancy. Recommendations on prepregnancy weight management are limited and ambiguous.51–53 Our results provide further important evidence on the clinical importance of maternal prepregnancy high BMI for both maternal and neonatal health outcomes. Dietary and/or physical activity interventions initiated before or in early pregnancy to reduce prepregnancy BMI, GWG and insulin resistance, rather than maternal lipid levels, would likely be most effective, although the effects need further investigation in future randomised trials.

Strengths and limitations

The major strengths of this study are the prospective design based on relatively large sample size, standardisation of strength of association for the comparison among maternal metabolic risk factors, and the use of powerful analytical tools for interpretation of multidimensional data. The study also has some limitation. First, given the practical constraints, maternal fasting glucose and triglycerides levels were measured only once during pregnancy. Therefore, we could not investigate the dynamic long-term influences of maternal metabolic risk factors in detail, although such levels generally track well over gestation. Second, the average prepregnancy BMI of included women and incidence of LGA/SGA babies in this study were significantly lower than for people living in the northern part of China. The relative healthiness of our cohort suggests that our results might underestimate the true impact of maternal metabolic disorders on neonatal health outcomes if extrapolated to this wider population. Third, the prepregnancy weight was self-reported, which might potentially underestimate the true value. However, evidence suggests that utilisation of self-reported or measured prepregnancy weight for prepregnancy BMI classification results in identical categorisation for most women.52 Fourth, as we mentioned earlier, maternal insulin resistance might be a significant explanatory factor in the whole pathway but was not estimated in the current analysis due to practical constraints. Although we include maternal fasting glucose level in the second trimester as a proxy of maternal insulin resistance, it may not sufficiently reflect the reality. Further analyses including maternal insulin resistance status are warranted.


In this cohort study, we found that mid-pregnancy maternal HDL-C and triglycerides concentrations was not independently associated with birth weight or CBI level after taking the interdependency of maternal lipids profiles with other metabolic risk factors into account. Pregnancy BMI, fasting glucose and GWG are three metabolic risk factors independently associated with increased birth weight and/or insulin secretion in neonates, with the prepregnancy BMI being the most influential upstream risk factor. Interventions initiated before or in early pregnancy to reduce prepregnancy BMI, GWG and insulin resistance would likely be most effective and need further investigation in future randomised trials.

Data availability statement

All data relevant to the study are included in the article or uploaded as online supplemental information. No additional data available.

Ethics statements

Patient consent for publication

Ethics approval

Ethical permission for the study was granted by the Guangzhou Women and Children’s Medical Center Ethics Committee (Reference number: 2013053107). Participants gave informed consent to participate in the study before taking part.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • XQ and KN are joint senior authors.

  • Twitter @MetaMedTeam

  • Contributors KN, XQ, KKC and JW conceived the research question. JW defined the question, designed the study and conducted data analysis. KN and MJP advised on study design and contributed to data analysis. KKC, H-MX, GNT and XQ provided input for study design. YK, JH, SS, J-HL and JW contributed to the sample collection, storage and biochemical tests, and data preprocessing. MP checked the ABN analysis externally. All authors contributed to the interpretation of the results. JW led the writing of the manuscript with critical input from all other authors (YK, SS, MJP, J-HL, NS, JH, MP, H-MX, GNT, XQ, KKC and KN). JW, XQ and KN had full access to all data (including statistical reports and tables) in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. JW is the guarantor.

  • Funding This work was supported by National Natural Science Foundation of China [grant numbers 82173525 and 82273642]. JW was supported by the LiSiguang scholarship provided by the University of Birmingham and the China Scholarship Council jointly (award number: 201508060007).

  • Disclaimer The funder had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

  • Competing interests None declared.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.