The pre-pregnancy fasting blood glucose, glycated hemoglobin and lipid profiles as blood biomarkers for gestational diabetes mellitus: evidence from a multigenerational cohort study

Abstract Background Early prevention of gestational diabetes mellitus (GDM) is important to reduce the risk of adverse pregnancy outcomes and post-pregnancy cardiometabolic risk in women and offspring over the life course. This study aimed to investigate some blood biomarkers before pregnancy as GDM predictors. Methods We investigated the prospective association of blood biomarkers before pregnancy and GDM risk among women from the Mater-University of Queensland Study of Pregnancy (MUSP) cohort. A multiple logistic regression model was applied to estimate the odds of experiencing GDM by blood biomarkers. Results Out of 525 women included in this study, the prevalence of GDM was 7.43%. There was an increased risk of experiencing GDM among women who experienced obesity (Odds ratio = OR 2.4; 95% confidence interval = CI 1.6–3.7), had high fasting blood glucose (OR = 2.2; 95% CI = 1.3–3.8), high insulin (OR = 1.1; 95% CI = 1.0–1.2), high insulin resistance (OR = 1.2; 95% CI = 1.0–1.3) and low high-density lipoprotein (OR = 0.2; 95% CI = 0.1–0.7) before pregnancy. Adjustment for potential confounders, such as age, marital status, and BMI did not attenuate these associations substantially. Conclusion The pre-pregnancy fasting blood glucose, insulin, and insulin resistance were independent predictors of GDM. They may be used as early markers for predicting the incidence of GDM.


Introduction
Gestational diabetes mellitus (GDM) is a state of glucose intolerance, first detected anytime during the pregnancy, which does not meet the criteria for diagnosis of diabetes in non-pregnant women [1]. GDM is a common complication of pregnancy that affects about 12% of Australian pregnancies [2], potentially causing several short-and long-term health consequences for the mother and her child [3]. Most importantly, GDM is associated with an elevated risk of hypertensive disorders of pregnancy, operative delivery, and macrosomia [4][5][6]. It increases antenatal and postnatal care expenses [7]. In the long term, it is associated with an increased risk of metabolic disorders and cardiovascular diseases (CVDs) in the mother and her baby [8]. Therefore, early detection of GDM allows for risk management, improves the health of the mothers and newborn babies, and decreases the healthcare burden [7].
Pancreatic beta-cell dysfunction (lower insulin secretion) and peripheral insulin resistance play primary roles in GDM pathophysiology [9]. However, for most women with GDM, the pathological process has already begun before pregnancy [3,10], so early pregnancy lifestyle interventions for GDM prevention have had limited success [11]. The systematic review by Song et al. [12] has suggested that first-trimester lifestyle intervention may reduce GDM by 20%. However, preconception interventions may potentially be more successful [10]. Promising, preliminary research from retrospective bariatric surgery studies suggests that pre-pregnancy reduction in body weight may help in GDM prevention and reduce GDM recurrence [13].
Currently, risk factors including age, body mass index (BMI), previous history of GDM, and family history of diabetes, are used to identify women who would benefit from early pregnancy screening for GDM [14]. Studies have considered some biomarkers during pregnancy as GDM predictors [14,15]. For example; pre-pregnancy and early pregnancy fasting blood glucose (FBG) may help in excluding women who do not require further investigation for GDM, but it cannot replace oral glucose tolerance test (OGTT) in GDM diagnosis [16]. Many studies have reported that elevated first-trimester FBG (within the range of normoglycemia) is an independent risk factor for the later development of GDM [17][18][19].
Furthermore, all 10 studies included in the systematic review of Kattini et al. reported that glycosylated hemoglobin (HbA1c) was associated with an increased risk of GDM and that women with levels between 5.7% and 6.4% in early pregnancy, are more likely to develop GDM [20]. A recent study by Nissim et al. found that HbA1c !5.45% in the first trimester of pregnancy predicted GDM with 83.3% sensitivity and 69% specificity. Therefore, HbA1c in early pregnancy may serve as a simple, early predictor for GDM [21]. Multiple large-scale studies suggested that fasting blood glucose and HbA1c measurements in the first trimester may help diagnose women who would benefit from early treatment [14]. Furthermore, some studies reported increased fasting insulin in the first trimester of pregnancy in women who later develop GDM [22,23]. However, other investigators have found that fasting insulin was not an independent predictor for GDM after adjusting for clinical characteristics [22,24].
Moreover, other studies have reported increased lipid profiles in women with GDM compared to normal women, and lipid abnormalities are inducing factors for insulin resistance [9,25]. Thus, overall, the results of the studies considering lipid patterns and GDM are inconsistent and most of these studies have focused on the association of lipid profiles in early or throughout pregnancy and GDM risk rather than before pregnancy.
Several studies have tested whether such biomarkers can be identified and employed to identify women at risk of GDM [9,26,27]. However, most of these studies are cross-sectional and have examined the association during pregnancy rather than before pregnancy. Moreover, it is as yet unproven whether these biomarkers, together or separately, are of practical value as GDM prediction tools [14]. Detecting some pre-pregnancy blood biomarkers using a longitudinal design would provide valuable information to reduce the need for screening testing in all pregnant women and allow early intervention to improve GDM outcomes.
This study aimed to determine whether pre-pregnancy blood biomarkers predict GDM, using the Mater University of Queensland Study of Pregnancy (MUSP) cohort [28]. These blood biomarkers could be used as a pre-pregnancy screening tool for GDM and would decrease the need for further screening tests for women without risk factors and initiate early prevention and treatment for those women with GDM risk factors.

Study data
We used the Mater-University of Queensland Study of Pregnancy (MUSP) [28], which is a prospective cohort study of 7223 women (G1) and their offspring (G2) who received antenatal care at a major public hospital in South Brisbane between 1981 and 1984. The maternal cohort and their index children were subsequently followed up by maternal questionnaires at 6 months, 5, 14, and 21 years after childbirth. Mothers' cohort and offspring were followed up separately at 27 years and 30 years, respectively, due to the difference in the timing of the research grants. Recently (2016-2018), a 34-year follow-up of G2 and children-of-the offspring (G3) was undertaken. For this study, the participants are the female offspring (G2) of the original (G1) mothers at the 34-year follow-up. The analytical sample comprises a sub-sample of 525 offspring (G2) for whom we have information on some of their blood biomarkers (these women provided blood samples at the 30-year follow-up), their BMI, and additional factors at 30-year follow-up ( Figure 1). The ethics committees at the Mater Hospital and the University of Queensland approved each phase of the study. At each data collection phase, informed written consent was obtained from the adults. Full details of information about the study participants and measurements have been previously reported [29].

Measures
Blood biomarkers at 30-year follow-up At the 30-year follow-up of G2, fating blood samples were drawn from 1625 young adults (after written informed consent) to evaluate different biomarkers [30].
For those participants residing in Brisbane, samples were collected by Mater Pathology, Brisbane. For participants outside the Brisbane area, their specimen collection was performed by their nearest Sonic Healthcare clinic. Those participants from Brisbane who were having difficulty attending a Sonic clinic underwent home specimen collection with a registered nurse from LifeScreen. Detailed instruction about fasting blood sample collection was provided to all the participants who were advised to eat their normal evening meal by 7 pm and fast for a minimum of 9 h with blood sampling before 9 am (an overnight fasting blood sample). The blood samples were used for testing some blood biomarkers, such as FBG, HbA1c, high-density lipoproteins (HDL), and low-density lipoproteins (LDL). Accordingly, FBG, HbA1c, insulin, and lipid markers were considered as potential pre-pregnancy predictors for GDM in our study, considering their associations with GDM in previous epidemiological studies [1,9,14,15,[17][18][19]21,25,31].
The Glucose Oxidase method was used for assessing FBG and Cation exchange-HPLC method using Bio-Rad D10 for HbA1c calculation. Prediabetes was defined by FBG, ranging from 5.6-6.9 mmol/L or HbA1c 5.7-6.4% based on American Diabetes Association recommendations [32]. Phosphotungstate/Mg 2þ was used to assess serum lipid using the Ortho Clinical Diagnostics Vitros analyzer. According to the guidelines for preventive activities in general practice [33], 2.0 mmol/L for LDL, 1.0 mmol/L for HDL, and 5.5 for total Figure 1. Flowchart demonstrates the study exposures, outcome and the number (%) of G1, G2 and G3 retained in MUSP study at each phase (age) of data collection [29].
cholesterol: HDL ratio (TC/HDL) were considered as cutoff values. Insulin resistance was estimated using the homeostatic model assessment of insulin resistance score HOMA-IR, according to the formula fasting glucose (mmol/l) x fasting insulin (mU/ml)/22.5 [1]. Women with HOMA-IR >2 [34] and fasting insulin >10 mIU/L [35] were considered to have insulin resistance.

Gestational diabetes mellitus
The GDM status of G2 females were based on selfreport at the 34-year follow-up. Our expert recruiter asked the women by telephone about their pregnancy history, pregnancy-related complications, delivery, lifestyle, and birth outcomes. GDM data was recorded using the question: "Did the doctor ever diagnose you with diabetes (high blood sugar) during pregnancy?" with response options "yes" or "no." However, the diagnosis did not identify which pregnancy was affected by GDM and there was no information regarding treatment and biochemical confirmation of the diagnosis. In addition, at 30 years follow-up, G2 women were asked whether they had experienced type 1 diabetes mellitus (T1DM) or type 2 diabetes mellitus (T2DM). This information was used to know whether they experienced diabetes mellitus before age 30 and was used as an exclusion criterion.

Statistical analysis
Overall distribution (mean and standard deviation) of FBG, HbA1c, insulin, HOMA-IR, HDL, LDL, and TC/HDL ratio were calculated. Statistical analysis was performed using F-tests for mean values comparison and the X 2 test was used for categorical variables. Multiple logistic regression with adjustment for potential confounders was conducted. Comparing different models, we found which factor(s) confounded/mediated the association between the selected biomarkers and GDM. All analyses were interpreted as odds ratio (OR) with 95% confidence intervals (CI).
The MUSP cohort has some loss to follow-up, as observed in other cohorts. The main study had revealed that the participants' mothers (G1), who were lost to follow-up, were more likely to have been teenagers when they delivered, of lower educational status, lower socioeconomic status, been smokers during pregnancy, and have poorer mental health [29]. Therefore, standard analytic methods were applied to correct the potential bias induced by attrition. The proportion of missing data ranged from 0% to 35%. Missing values for variables were imputed using multiple imputation chained equations/MICE with 30 imputed datasets [36]. We assessed all covariates' influence in the primary analysis on our complete data in a logistic regression model. All analyses were undertaken using Stata version 16 (StataCorp, College Station, TX, USA).

Results
A sub-sample of 525 females during the 34-year follow-up of G2 provided information about GDM and for whom we have data on blood biomarkers before pregnancy (at 30-year follow-up) were included in this study. We divided the sub-sample (regardless the number of pregnancies) into women experienced GDM and women without GDM. As a result, 39(7.43%) of the women self-reported a diagnosis of GDM during pregnancy.
Participant women's characteristics stratified according to GDM status are shown in Table 1. Women who reported GDM were more likely to be married, nonsmokers, had annual family income >$50k, had high IPAQ scores, and BMI !30 kg/m 2 at age 30. The association between BMI and GDM was statistically significant and about 50% of the women who reported GDM experienced obesity before pregnancy. However, no significant differences were observed between GDM and non-GDM women regarding age, marital status, family income, smoking, alcohol consumption, and physical activity at 30-year follow-up. Table 2 demonstrates the mean values of the prepregnancy blood biomarkers with the risk of GDM.
Women who reported a diagnosis of GDM had significantly increased mean FBG, insulin, HOMA-IR, TC/HDL ratio, and decreased mean HDL concentrations before pregnancy compared to the non-GDM group. The univariate and multivariate-adjusted estimates of the continuous associations between pre-pregnancy blood biomarkers and the risk of GDM are presented in Table 3. In the univariate analysis, FBG (OR ¼ 2.2; 95% CI ¼ 1.3-3.8), insulin (OR ¼ 1.1; 95% CI ¼ 1.0-1.2), HOMA-IR (OR ¼ 1.2; 95% CI ¼ 1.0-1.3) and TC/HDL ratio (OR ¼ 1.5; 95% CI ¼ 1.2-2.0) were positively associated with increased GDM risk and HDL (OR ¼ 0.2; 95% CI ¼ 0.1-0.7) was inversely associated with GDM risk. Adjustment for potential confounders did not explain these associations. Furthermore, no statistically significant associations were noted between HbA1c and the risk of developing GDM in unadjusted and adjusted models.   Table 3. Associations between pre-pregnancy blood biomarkers (continuous variables) and GDM.  Table 4 shows the proportion of GDM status by the pre-pregnancy blood biomarkers groups. The women who reported GDM were more likely to have increased insulin resistance (about 46% had insulin level >10 mIU/L and almost 60% had HOMA-IR >2, p < .001), and about 10% were in pre-diabetes status before pregnancy based on FBG (p ¼ .006). Table 5 shows the multiple logistic regression models of the categorical blood biomarker groups. Significantly lower odds were observed for FBG, insulin, and insulin resistance in the adjusted models (OR ¼ 6.4; 95% CI ¼ 2.0-36.6), (OR ¼ 1.4; 95% CI ¼ 1.1-3.4) and (OR ¼ 3.0; 95% CI ¼ 1.2-7.3), respectively. The association between pre-pregnancy HbA1c and GDM was not statistically significant.

Discussion
The present study showed that the associations between pre-pregnancy FBG, insulin, and HOMA-IR and GDM were statistically significant and remained so after adjustment for various potential confounders indicating that metabolic impairments preceded GDM pregnancy. Therefore, they would be considered independent markers for the prediction of GDM. Our results suggest that pathophysiological changes related to glucose regulation, which may lead to GDM, are present before pregnancy and are not simply induced by pregnancy in women who develop GDM [37]. These findings are of clinical relevance, as they may be used in identifying women at risk of GDM before pregnancy and may allow for preconception interventions to reduce the risk of GDM in a future pregnancy. These may be more effective than interventions starting in early pregnancy for GDM prevention.
Study by Gunderson et al. considered multiple biomarkers together for GDM prediction using the clinical data of Coronary Artery Risk Development in Young Adults (CARDIA) [27]. It found that impaired fasting glucose, elevated fasting insulin, and low HDL levels were present in 41% of women (58 out of 141) before GDM pregnancy. Furthermore, the retrospective study of Noussitou et al. examined the relationships between GDM and metabolic syndrome found that the metabolic abnormalities were already present in about 26% of the included women before their GDM pregnancy and the associated increased risk of future T2DM [38].
In addition, studies of early pregnancy found that elevated first-trimester FBG (within the range of normoglycemia) was an independent risk factor for the later development of GDM [17,19] but did not measure other blood biomarkers. Moreover, higher first-trimester fasting glucose levels, even within the non-diabetic range, increased the risk of adverse Table 4. Generation 2 participants' pre-pregnancy blood biomarkers level (categorical variables) in GDM-group and non-GDM group. 34  pregnancy outcomes [17]. The FBG cannot replace OGTT in diagnosing GDM, but it could help exclude women who do not require further investigation for GDM [16]. The study of Immanuel [1] showed that insulin resistance contributed relatively to early GDM. In another study of early pregnancy, it has reported that there is a positive correlation between high serum insulin levels and GDM. In addition, women with elevated first-trimester insulin should be managed as same as those with GDM despite a negative OGTT. Furthermore, it concluded that serum insulin measurement before 16 weeks of gestation was reliable for predicting GDM [23].
This study has some strengths, as it used data from a large longitudinal cohort. The longitudinal design reduces the likelihood of recall bias. The availability of a wide range of blood biomarkers and confounders were also strengthened our study. However, there are several potential limitations to this study. First, we relied on self-reported GDM during the 34-year followup without confirmation by objective laboratory tests. Although the trained interviewer asked the women whether the doctor ever diagnosed them with diabetes during pregnancy, women in this cohort were highly likely to have underdiagnosed gestational diabetes mellitus. However, our prevalence of GDM was consistent with the last report from the Australian Institute of Health and Welfare [39]. Several studies have reported that a self-report of diabetes yielded high agreement when compared with medical records data and physical examination and HbA1c [40,41]. Therefore, this limitation does not affect the clinical utility of these biomarkers significantly. Second, although we considered age, marital status, family income, smoking, alcohol drinking, physical activity, and BMI as potential confounders, there are unmeasured confounders that may have impacted the associations but cannot be considered. Finally, the loss to follow up in the cohort. Nonparticipants at different follow-ups were likely to be from poorer backgrounds; their mothers were more likely to have lower educational attainment and younger at childbirth. Missing data because of attrition would be biased our results only if the associations we assessed were either non-existent or in the opposite direction in non-participants, which is unlikely. A variety of modeling strategies have been used in the MUSP study for attrition adjustment. However, these methods have minimally affected overall findings [42,43]. Therefore, multiple imputation chained equations were carried out to adjust for missing data in our analyses [36]. We suggested that missing data was unlikely to have biased our results.

Conclusion
FBG, insulin, and insulin resistance were independent pre-pregnancy predictors of GDM. Therefore, women with high FBG and insulin-resistance before pregnancy should be monitored and paid more attention to prevent GDM and its adverse outcomes. Pre-pregnancy biomarker measurements might be more beneficial in GDM prediction than early pregnancy values. Therefore, such analyses could be included in the routine health checks before pregnancy to identify women with a high risk of GDM in subsequent pregnancies. Future longitudinal studies are needed to examine the association between these biomarkers before and during pregnancy and GDM. Furthermore, further research is needed to determine the most appropriate risk-benefit use of pre-pregnancy blood biomarkers.