The Determinants for Adolescent Glycolipid Metabolism Disorder from Prenatal Period through Adolescence: a Cohort Study

To explore the prevalence of and risk factors for glycolipid metabolism disorder (GLMD) from the prenatal period through childhood and adolescence. A bidirectional cohort study which was established in 2014 and followed between March 1 and July 20, 2019 were designed to evaluate the risk factors for GLMD. Two stage stratied cluster sampling was used to include participants from four communities in two counties in Chongqing. 2808 healthy children aged between 6 and 9 years in 2014 entered the cohort in 2014, and followed in 2019 with a follow-up rate of 70%. 2,136 samples (aged 11·68 ± 0·60 years) were included. The prevalence of insulin resistance (IR), prediabetes/diabetes and dyslipidemia was 21·02%, 7·19% and 21·61%, respectively. Subjects with an urban residence, no pubertal development, dyslipidemia in 2014, higher family income, higher parental education, low quality of life (QoL), and unstable neurotic and psychotic personalities had signicantly elevated fasting insulin (FI) or homeostasis model assessment insulin resistance (HOMA-IR) levels; subjects with female sex, no pubertal development, dyslipidemia in 2014, obesity, gestational hypertension, maternal weight gain above Institute of Medicine guidelines, single parents, low QoL, introverted and unstable personality had increased triglyceride or triglyceride/high-density lipoprotein (HDL). Adolescent with rural residence and living convenience had higher HbA1c level. We observed that low QoL status, unstable and psychotic personality traits were associated with increased GLMD risk independent of obesity, suggesting that the community intervention to improve the QoL and psychological health of children is essential. in childhood tracks into adulthood and predicts the severity of atherosclerosis in adults 1 . The triglyceride/high-density lipoprotein cholesterol (HDL-C) ratio was used as an IR marker for overweight and obese children 4 and was also an index of GLMD. The prevalence of hyperglycemia ranged from 5·7% to 11·13% among children with obesity 5 . Despite having a lower prevalence than IR and dyslipidemia, hyperglycemia during childhood is a predictor of type 2 diabetes in adulthood 6 . Because childhood metabolic disorders can predict CVDs in adulthood 4,6 , determinants of GLMD need to be evaluated during this period. Therefore, it is important to increase our understanding of the prevalence and risk factors for GLMD during the perinatal, younger childhood and adolescence periods. level was associated with increased HOMA-IR 21 , and our cohort study rst found dyslipidemia and elevated fasting glucose at 6 ~ 9 years old were independent risk factors of HOMA-IR and dyslipidemia in adolescents (10 ~ 14 years old). And adolescents with menarche or spermarche had decreased IR and lipid levels, which indicated the prepubertal stage will impact GLMD among adolescents.


Introduction
The increased prevalence of glycolipid metabolism disorders (GLMD) in childhood and adolescents has a well-established association with adult type 2 diabetes and cardiovascular diseases (CVDs) 1 . GLMD in adolescents includes insulin resistance (IR), dyslipidemia and hyperglycemia. The prevalence of IR and dyslipidemia in children and adolescents ranged from 25·3% to 44·3% among children and adolescents according different region and different diagnosed criteria 2,3 . Dyslipidemia in childhood tracks into adulthood and predicts the severity of atherosclerosis in adults 1 . The triglyceride/high-density lipoprotein cholesterol (HDL-C) ratio was used as an IR marker for overweight and obese children 4 and was also an index of GLMD. The prevalence of hyperglycemia ranged from 5·7% to 11·13% among children with obesity 5 . Despite having a lower prevalence than IR and dyslipidemia, hyperglycemia during childhood is a predictor of type 2 diabetes in adulthood 6 . Because childhood metabolic disorders can predict CVDs in adulthood 4,6 , determinants of GLMD need to be evaluated during this period. Therefore, it is important to increase our understanding of the prevalence and risk factors for GLMD during the perinatal, younger childhood and adolescence periods.
Obesity is the main cause of GLMD, and our previous study revealed that obesity is positively associated with low-density lipoprotein cholesterol (LDL-C) and TGs, but negatively correlated with HDL-C 7 . Moreover, previous studies have shown an increased prevalence of GLMD in individuals with a sedentary lifestyle, unhealthy dietary habits, genetic factors, exposure to higher maternal fasting blood glucose (FBG) levels in utero 8 and gestational diabetes 9 . A study found that extraverted personality is positively correlated with triglycerides, FBG and metabolic syndrome (MS) score in adults 10 . However, to our knowledge, there are no studies exploring the correlation between quality of life (QoL) and GLMD or between personality traits and GLMD in children aged 10~14 years in a rural-urban cohort study. This cohort study included measures of perinatal variables, social economic status (SES), anthropometric variables and biochemical indexes in 2014 and 2019 along with QoL scores and personality characteristics in adolescents, providing an excellent opportunity to fully examine the risk factors for GLMD. Moreover, it is not clear to what extent these variables account for the variance in GLMD in the transition period from childhood to adolescence.

Patient and Public Involvement
The children and their guardians or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Subjects
Two stage strati ed cluster sampling was used to include participants from two counties in Chongqing that represent urban and rural areas; then, two regions per county were randomly selected, and nally, all children living in the selected region were informed and included if they satis ed the inclusion criteria. In addition, a birth-cohort in which retrospective and prospective variables were analysed to evaluate the development of glycolipid metabolism risk factors from the perinatal period through adolescence, and the perinatal variables were retrospectively collected, with physical examination conducted prospectively in 2014 and in 2019. Participants who met all the following criteria were recruited: (1) aged between six and nine years in 2014, (2) resided in the target region for more than six months, (3) did not have serious diseases (e.g., nephropathy, cardiovascular disease or cancer), and (4) obtained consent from the parents and children for participation. At baseline, all participants completed SES and family health history questionnaires and were recruited mainly from grade one and grade two from primary public-schools screening of children whose families were interested in health research. The questionnaires were administered and collected by the teachers, and the physical measures results were disseminated also by the teachers. Two thousand one hundred and thirty-six subjects (with a follow-up rate of 70%) were ultimately included (Fig. 1) and the difference between children with follow-up and withdraw was compared in Supplementary   Table S1. All work in this study was conducted in accordance with the ethical guidelines of 1964 Declaration of Helsinki and its later amendments 11 . The Institutional Review Board at the Children's Hospital of Chongqing Medical University gave its approval for the study (File No.: 2019-86). Informed consent was provided by all subjects and parents/guardians.

Demographic variables
Demographic information, SES (parents' occupation and education level, family income and single-parent status) were collected. Parental education level was measured on a four-point scale (≤9 years (compulsory education in China), 9~12, 12~15 and >15 years), and we combined bachelor and master's degrees as there were few parents with master's degrees. Prenatal variables included maternal preconception obesity, increased body mass index of mother during pregnancy, birth with cesarean section, premature delivery (<37 weeks), birth-weight, breastfeeding, gestational hypertension (GH) and gestational diabetes.
Family history of obesity and CVD was investigated. The degree of pubertal development was surveyed both by the visit of paediatrician and children or parents lling the questionnaires, which included the date of the rst menstruation and rst nocturnal emission, then the age was calculated.
The validity and reliability of the questionnaire were checked, and were described in detail in a previous publication 12 . The questionnaire was completed by the parents or guardians of the children after standard training by the research group.

Physical examination
Anthropometric measurements were conducted by well-trained pediatric nurses, and the protocol for these measurements was described in a previous publication 12 . Waist circumference or waist-height ratio (WHtR=waist circumference/height) was used as an alternative measure of central adiposity. Hip circumference was measured twice horizontally at the level of the pubic symphysis in the front and the gluteus maximus in the back, with the participant standing upright and with their legs together and placing their arms naturally at their sides; the mean value was used.
Blood pressure (BP) and heart rate were measured on three separate occasions with an OMRON arm-type electronic sphygmomanometer (HEM7051) using an appropriately sized BP cuff placed on the subject's right arm, with the subject in a seated position, which was described in a previous publication 12 .

Biochemical indexes
Venous blood (three ml) was drawn from each subject in the morning after at least 12 hours of fasting and 24 hours of abstaining from high-fat and spicy foods. The biochemical markers and glycosylated hemoglobin were measured within two hours after venous blood was drawn and the protocol was introduced in detail in our previous publication 12 . Moreover, the ratio of TG/HDL-C was used as a parameter to assess lipid metabolism 4 . Siemens Centaur XP was used to measure fasting insulin (FI), and HbA1c was measured by an automatic hemoglobin analyzer (ARKRAY, Japan).

Measurement of QoL and Eysenck's Personality Questionnaire (EPQ)
The QoL questionnaire for adolescents consists of 49 items, including four factors (psychosocial function, physical and mental health, living environment and QoL satisfaction) and 13 dimensions, such as self-satisfaction, relationship of teacher and pupil, physical feeling, companionship, parenthood, physical activity ability, learning ability and attitude, self-esteem, negative emotion, attitude towards doing homework, opportunity for activity, living convenience and others (picky-eating and surroundings). The order of presentation of the 49 items was randomized. Children rated the statements on a four-point scale, and the direction of response (positive or negative) varied item by item to limit response bias. Individual item values were recoded prior to analysis so that the direction was consistent. Responses were summed and normalized according to the age-, sex-and region-speci c norms of Chinese into a score ranging from 0 to 100, normalized with a mean of 50 and SD of 10, with higher scores suggesting a better QoL 40 .
The Chinese version of Eysenck's personality questionnaire of China 41 consisted of 88 items scored on a two-point scale (for positive items NO=0 and YES=1), including four domains: extraversion (E) (25 items), neuroticism (N) (23 items), psychoticism (P) (18 items), and lie scales (L) (22 items). High scorers on the E scale indicate sociable, exciting, pleasurable, carefree, and aggressive characteristics. A higher score on the N scale is more likely to be a worried and moody person who tends to suffer from emotional and psychosomatic disorders. The P scale was designed to measure behaviour patterns that might be considered schizoid or psychopathic in extreme cases. The L scale assesses response bias. Items in the E, N, P and L domains are summed and normalized to the age-and sex-speci c norm of China into a score ranging from 0 to 100, normalized with a mean of 50 and SD of 10. People are de ned as middle type, tendency type and typical type if the T score (E and N) ranges from 43·3~56·7, 38·5~43·3 or 56·7~61·5, <38·5 or >61·5. People were considered psychotic personality if T score of the P domain >56·7, and response bias was indicated when the T score of the L domain was >70.
Overweight was indicated by a body mass index 15 at or above the 85th percentile and below the 95th percentile, and obesity was de ned as a BMI at or above the 95th percentile according to the sex-speci c Centers for Disease Control BMI-for-age growth charts 15 . The de nitions of size for gestational age used the global reference for fetal-weight and birthweight percentiles 16 : birth weight at or above the 90th percentile indicated large for gestational age (LGA), and birth weight less than the 10th percentile indicated small for gestational age (SGA), using the parameters of mean birthweight at 40·5 weeks of 3,332·93 g and a variation coe cient of 14·36%. Maternal overweight and obesity was indicated by a BMI of 24~27·9 kg/m 2 and a BMI ≥ 28 kg/m 2 , respectively; BMI < 18·5 kg/m 2 was de ned as a low BMI 17 . According to guidelines of the Institute of Medicine (IOM) for maternal pregnancy weight gain 18 , the recommendation is for underweight, normal weight, overweight, and obese women to gain 12·5~18·0 kg, 11·5~16·0Kg, 7·0~11·5 kg, and 5·0~9·0 kg, respectively; if weight gain exceeded that range, weight gain was de ned as "above IOM guidelines", and if weight gain was below that range, was de ned as "below the IOM guidelines".

Statistical analyses
Differences in glycolipid metabolism indexes between two groups were assessed using Student's t-test, ANOVA was used to compare more than two groups, and post-hoc comparison was perfoemed using Student-Newman-Keuls test. Continuous variables (insulin, HOMA-IR, TG and TG/HDL) that did not satisfy a normal distribution were subjected to natural logarithmic transformation before analyses. The χ 2 test was used to test the difference in prevalence rates of GLMD. A generalized linear model (GLM) was used to analyze the risk factors that may impact glycolipid metabolism. To reduce the collinearity of variables, model 1 mainly included the variables measured prenatally and in 2014, and model 2 mainly included the variables measured in 2019. Finally, model 3 included all the variables that may impact GLMD. Moreover, multivariable logistic regression was performed using diagnosed GLMD as the dependent variables with the impact factors from perinatal to adolescent, as independent variables. Adjusted R 2 was calculated to locate the variance of dependent variables can be explained by independent variables. Participants with the missing responding variables were not included in the analyses. And the participants nish the follow were compared with the dropouts.
The data analysis was conducted using SAS 9.4 software (Copyright © 2020 SAS Institute Inc. Cary, NC, USA). A signi cant difference was de ned by an α level of 0.05.

General characteristics
The general characteristics of the subjects are presented in Table 1. A total of 2,136 samples were included, with a follow up rate of 70·0%. And the difference of characteristics of childhood between participants with follow-up and withdraw were described in Supplementary Table S1. The mean age was 11·68 ±0·60 years, and 52·25% (1,116/2,136) were males. Biochemical indexes; anthropometric, perinatal, and SES variables; QoL; and personality characteristics are shown in Table 1. LGA, large for gestational age; QoL, quality of life. † natural logarithmic transformation.
* the total sample size is unequal to 2136 in 2019 as there are missing data. Table 2 displays the glycolipid metabolism results in adolescents. Adolescents with the characteristics of urban residence, female sex, older age, no pubertal development, dyslipidemia, obesity, low QoL score and instable personality traits had higher FI or HOMA-IR, TGs or TG/HDL than their counterparts. Whereas, HbA1c was higher in rural children and those with pubertal development, obesity, or maternal prepregnancy obesity than in their counterparts. In addition, TGs and/or TG/HDL were elevated in children with mother who experienced weight gain above IOM guidelines (P<0·05), single-parents (P<0·05), maternal hypertension (GH) and introversive personality compared with their counterparts (P<0·05 and P=0·06). The levels of FI and HOMA-IR were higher in children with parents with higher education levels and family incomes than in their counterparts (P<0·01). Moreover, adolescents with psychotic personality had elevated FI and IR compared with the levels in their counterparts (all P<0·01).   # Natural logarithmic transformation were used to calculate the P value. ¶ Nine hundred and thirty-two samples were included. * P<0·05; ** P<0·01. †, ‡, § Difference of post-hoc analyses among groups using " † ", " ‡ ", " § " labelled, different symbols mean the difference existed between two groups.
Prevalence of glycolipid metabolism disorder in adolescents Table 2 displays the prevalence of childhood GLMD. Overall, the prevalence rates of IR, prediabetes/diabetes and dyslipidemia were 21·02%, 7·19% and 21·61%, respectively. The prevalence rates of IR and dyslipidemia were higher in children with the characteristics of older age, dyslipidemia in young childhood (6~9 years), obesity and lower QoL score than in their counterparts. Moreover, Children with urban residence, LGA status, higher family income and parental education, and unstable personality (P=0·07) also had an increased prevalence of IR. The prevalence of prediabetes/diabetes was higher in children with abdominal obesity in 2014 and maternal prepregnancy obesity than in their counterparts.

Risk factors of glycolipid metabolism indexes using a GLM
In the GLM 1 (Table 4) (adjusted for sex, age and region), the results showed that female, living in urban areas and variables measured in 2014 (FBG, BMI, waist circumference [WC]) were risk factors for FI and HOMA-IR levels (all P<0·05), and older age was a risk factor of FI and IR(P<0·01); variables in 2014 (FBG, dyslipidemia, BMI) were the risk factors of TG and TG/HDL level (all P<0·01), and FBG and BMI in 2014 were risk factors for HbA1c level (Supplementary   Table S2). Model 1 explained 12·43%, 11·92%, 6·79%, 10·32% and 7·06% of the variance in FI, HOMA-IR, TGs, TG/HDL and HbA1c levels, respectively The GLM (

Discussion
This study is the rst prospective cohort study that involves QoL and personality traits, in addition to perinatal, SES and physical measurements over an average 12-year follow-up from birth to adolescence in urban-rural areas, to ascertain the prevalence of GLMD and its' potential in uencing factors.
The prevalence of GLMD varies by region and age, and some variance is also attributed to different diagnostic criteria and methods. The current literature describes at least one lipid adverse level prevalence as 19%-25% in US children and adolescents 8,19 , and the prevalence of prediabetes/diabetes in other study 5 , was comparable with our study. An elevated prevalence of GLMD has been observed in children with obesity in cross-section study 20 . In this study, we found that childhood obesity is the strongest predictor of adolescent GLMD, even adjusted other risk factors. Moreover, the prevalence of HOMA-IR exceeded 44% in children who had obesity comparable with the result from Yin J et al. 2 , and the prevalence of dyslipidaemia reached 28·57% in children with abdominal obesity, suggesting that health care programmes should be conducted for children who are obesity or abdominal obesity combined with other risk factors.
In addition, a cross-sectional study revealed that elevated TG level was associated with increased HOMA-IR 21 , and our cohort study rst found dyslipidemia and elevated fasting glucose at 6 ~ 9 years old were independent risk factors of HOMA-IR and dyslipidemia in adolescents (10 ~ 14 years old). And adolescents with menarche or spermarche had decreased IR and lipid levels, which indicated the prepubertal stage will impact GLMD among adolescents.
Whereas, the transient IR phenomenon emerging during pubertal maturation is accepted as a physiological condition 2 , which may be caused by an inadequate β-cell response to the decrease in insulin sensitivity 22 . In addition, glycolipid indexes (except HbA1c) were higher in females than in males, which coincided with the results of Interator H. et al 23 , and the mechanism may be dependent on the difference in the age of prepubertal stages between males and females.
Maternal adverse perinatal experiences will impact GLMD in the offspring 24,25 . We found maternal prepregnancy obesity was a risk factor for irregular HbA1c level. An animal study found maternal obesity permanently alters the hypothalamic response to leptin and subsequently regulates appetite and pancreatic beta-cell physiology 24 , which causes maternal and off-spring changes in glycolipid levels. Moreover, our study found both maternal pregnancy weight gain above IOM guidelines and GH were risk factors for elevated off spring TGs, which coincided with the results from young adulthood 26 . This phenomenon can be explained by shared genes or lifestyle. However, the conclusion was controversial, as a study with a small sample size found no association between GH and lipid levels in adolescent 27 , this founding needs to be veri ed by a large cohort study. In addition, SGA and LGA correlated with elevated HOMA-IR prevalence, which coincided with other ndings 28 . Birth-weight was correlated with nutritional status in utero, which may cause IR later in life; moreover, LGA is correlated with adolescent obesity, which is essential to IR.
SES is negatively correlated with cardiovascular disease. Our current cohort study provided further support for this concept in the adolescent population. A previous study 29 revealed that the marital status of parents was the strongest socioeconomic predictor of young adult arterial stiffness, and we found the TG level was higher in single-parent adolescents. In addition, the relationship between parental education and the cardiovascular risk of adolescent is controversial, and our results showed a positive relationship between parental education or family income and FI or IR. Studies have revealed a positive correlation between parental education and childhood obesity 30 , and obesity was the strongest predictor of insulin sensitivity. Moreover, we observed that rural residents have lower FI, IR and TG levels, but higher HbA1c levels, which could be induced by different dietary habits, as rural children consume less fat but more carbohydrates.
Our study revealed QoL scores were mainly negatively correlated with GLMD in adolescents. Research on the relationship between QoL and GLMD is limited.
According to a previous cross-sectional study that included 74 diabetic adolescents 31 , no signi cant relationship between QoL and HbA1c levels was observed. However, a cross-sectional study found QoL scores were correlated with an increase in the components of MS, and the physical health domain of QoL had the most signi cant association 32 . In our study, we found 6 domains, four factors and the total QoL score were signi cantly negatively correlated with glycolipids indexes, and the effect was independent of obesity. Moreover, only the living convenience domain of QoL was positively correlated with HbA1c, as the child has greater chance of consuming high-sugar snacks. To our knowledge, this is the rst cohort study with a large sample size to explore the relationship of QoL with GLMD in adolescents.
Personality traits may be associated with GLMD, but the conclusion remains controversial, and few studies have been conducted in adult populations; no related cohort study has explored this relationship in adolescents. The results from Japanese adult 10 showed the E score was positively correlated with TGs, FBG and MS components, and the P score was positively correlated with FBG. However, we found the E score in adolescents was negatively correlated with TGs and TG/HDL, suggesting the E score may be age dependent and have a different impact on lipid metabolism between adults and adolescents. And study from Evans, B. E. 33 reported that extraverted adolescents have less cortisol activity, which is associated with FBG and FI 34 . Therefore, E personality may regulate glycolipid indexes through the hypothalamic-pituitary-adrenal axis pathway 33 , which may impact GLMD by regulating the reactivity of the sympathetic nervous system (SNS) 35 , and the SNS will have potent effects on insulin secretion and sensitivity 36 and on lipid metabolism 37 . Higher N scores in adolescents would be more prone to responding more strongly to a stressor and anxiety, which could explain our results that the N score was positively correlated with FI and IR after adjusting for covariates, and N also correlated with lipid metabolism, which coincided with the results of study in adults 38 .
Moreover, we found the P score was positively correlated with FI and IR, as P scores may be linked to the SNS, which may regulate insulin sensitivity 39 .
Our study has some limitations that should be considered when interpreting the results. First, as this was an ambispective cohort study, recall bias may exist for the prenatal variables. We checked the birth certi cates to verify the birth weight, stature and gestational age. Second, data on GH and diabetes were collected using a questionnaire, and recall bias existed. However, the perinatal information both were collected both in 2014 and in 2019 independently. Finally, we were collected QoL and personality traits variables in a cross-sectional manner, which makes it di cult to draw conclusions regarding causality relationships of QoL and personality traits with GLMD.
In conclusion, the prevalence of GLMD and high glycolipid levels was elevated in adolescents with the features of obesity, maternal prepregnancy obesity, GH, SGA, LGA, single-parents status, low QoL score, and unstable or psychotic personality. SES was positively correlated with HOMA-IR. To our knowledge, this is the rst study to explore the relationship of QoL and personality characteristics with glycolipid indexes in a large-sample-size cohort study of adolescents, and the correlation was signi cant after adjusting for covariates. Our study emphasizes the importance of reducing or controlling adiposity of pre-pregnancy mother and children, increasing QoL, developing good personality traits, emphasizing the importance of providing support for single-parent children and reducing or preventing GH.

Declarations
Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Data are available from Xiao-