Design and population
The nested case–control study was conducted in the Tongji-Shuangliu Birth Cohort (TSBC) [13], which was started from March 2017 in the Shuangliu Maternal and Child Health Hospital in Chengdu. Until June 2019, 6143 pregnant women were enrolled during their first prenatal examination (6–17 weeks of pregnancy). Women were included if they met the following criteria: (1) women aged 18–40 years with singleton pregnancy; and (2) gestational age less than 15 weeks. Participants were excluded if they (1) conceived the fetus using assisted reproductive technology, such as in-vitro fertilization and intrauterine insemination; (2) reported severe chronic disease or infectious disease like cancer, tuberculosis, and HIV infection; or (3) refused to sign the written informed consent or had no ability to complete the questionnaire independently. Structured questionnaires were administered at enrollment, and blood samples were obtained for future analyses. The original cohort study was approved by the Ethics Committee of Tongji Medical College, Huazhong University of Science and Technology, and informed consent was obtained from all participants.
GDM diagnosis and matching to controls
GDM was diagnosed at 24–28 weeks of pregnancy according to the International Association of Diabetes in Pregnancy Study Groups criteria using the standard 75 g 2-h oral glucose tolerance test (OGTT): (1) fasting plasma glucose ≥ 5.1 mmol/L; and/or (2) 1-h plasma glucose ≥ 10.0 mmol/L; and/or (3) 2-h plasma glucose ≥ 8.5 mmol/L [14]. A total of 347 GDM women were diagnosed, of whom 14 did not provide sufficient blood samples for C-peptide measurements at enrollment and 1 had data missing for key covariates. We included 332 eligible GDM cases, and matched them individually to 664 pregnant women with normal glucose tolerance at 1:2 on maternal age (± 3 years) and gestational age (± 4 weeks) (Additional file 1: Figure S1).
Measurement of serum C-peptide and other biomarkers
Measurement of metabolic biomarkers were conducted using fasting blood samples collected at enrollment. Serum C-peptide, insulin, and leptin were measured using the Metabolic Group 1 (hu) Singleplex Assays on the Meso Scale Discovery (MSD) U-PLEX Metabolic Platform (MSD, Rockville, Maryland, US). The intra- and inter-assay coefficients of variation for C-peptide were 3.7% and 10.3%, separately. Fasting blood glucose (FBG) was measured using the Glucose Assay Kit (Sichuan Maccura Biotechnology, Chengdu, China) by the method of GOD-PAP (glucose oxidase-phenol and 4 aminophenazone). Glycosylated hemoglobin (HbA1c) was measured using a DCA Vantage Analyzer (Siemens Healthcare Diagnostics, Marburg, Hessen, Germany). Serum high-sensitivity C-reactive protein (hs-CRP) and adiponectin were tested using the R&D enzyme-linked immunosorbent assays (R&D Systems, Minneapolis, Minnesota, US). Total cholesterol, triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C) were measured via the Mindray BS-200 chemistry Analyzer (Mindray Medical International, Shenzhen, China). The homeostatic model of insulin resistance (HOMA-IR) was used to estimate insulin resistance and calculated based on the following formula: HOMA-IR = fasting blood glucose (mmol/L) × fasting insulin (mIU/L)/22.5 [15]. Missing values for FBG (n = 11), HbA1c (n = 17), and serum lipids (n = 5) were imputed using median values by GDM status in the study.
Measurement of covariates
Data of sociodemographic information, history of disease and reproduction, and lifestyle and behaviors were obtained via questionnaire interviews at enrollment. Anthropometric measurements were conducted at enrollment per standard protocols. Pre-pregnancy body mass index (BMI) was calculated according to the formula: BMI = weight (kilogram)/height2 (meter), in which pre-pregnancy weight was self-reported. Waist–hip ratio (WHR) was defined as waistline (cm) divided by hipline (cm). Education level was categorized according to years of education: ≤ 12 years and > 12 years. Smoking status and alcohol consumption were both categorized as never, former, and current. Blood pressure was measured twice using Omron electronic sphygmomanometer (Omron, Kyoto, Japan), and the average value was calculated. Physical activity in metabolic equivalent of task (MET)-hours per week was evaluated using the Chinese version of the Pregnancy Physical Activity Questionnaire [16], which has been validated among Chinese pregnant women [17]. Parity was classified into 0 and ≥ 1. Parental history of diabetes and history of GDM were both defined as yes and no.
Statistical analysis
For descriptive analyses, continuous variables were reported as mean and standard deviation (SD) or as median and interquartile range (IQR), and categorical variables as frequency and percentage. Baseline characteristics among C-peptide quartile groups were compared using Kruskal–Wallis test or analysis of variance for continuous variables and chi-square test for categorical variables. In addition, baseline characteristics between GDM cases and controls were compared by univariable conditional logistic regression.
Partial Spearman regression was used to examine the relationship of C-peptide with multiple metabolic biomarkers including WHR, blood pressure, FBG, fasting insulin, HOMA-IR, HbA1c, total cholesterol, TG, LDL-C, HDL-C, hs-CRP, adiponectin, and leptin in early pregnancy among all included pregnant women, with adjustment for maternal age, gestational age, education level, smoking status, alcohol consumption, physical activity, pre-pregnancy BMI, parental history of diabetes, history of GDM, parity, and GDM status.
Multivariable conditional logistic regression models were used to estimate odds ratios (ORs) and their 95% confidence intervals (CIs) between early-pregnancy serum C-peptide and risk of GDM. C-peptide was assessed as a categorical variable (quartiles based on the concentration among the control group), and as continuous variables (on the natural log scale and for each 1-SD change). Covariates were sequentially adjusted for in two models: maternal age (continuous, years), gestational age (continuous, weeks), and education level (≤ 12 years and > 12 years) in Model 1; additionally, smoking status (never, former, and current), alcohol consumption (never, former, and current), physical activity (continuous, MET-hours per week), pre-pregnancy BMI (continuous, kg/m2), parental history of diabetes (yes and no), history of GDM (yes and no), and parity (0 and ≥ 1) in Model 2. In sensitivity analyses, we separately adjusted for insulin (continuous, uIU/mL), HOMA-IR (continuous), and leptin (continuous, ng/mL) in multivariable conditional logistic regression model due to their stronger correlations to C-peptide.
P values for trend were estimated by modeling the median value of each C-peptide quartile as a continuous variable. We used restricted cubic splines with five knots at the 5th (reference), 27.5th, 50th, 72.5th, and 95th centiles to model the non-linear association between C-peptide and GDM. To investigate whether the association was modified by the baseline characteristics, we conducted subgroup analyses by maternal age (< 30 and ≥ 30 years), pre-pregnancy BMI (< 24.0 and ≥ 24.0 kg/m2), and parental history of diabetes (yes and no). Interactions (effect modifications) were assessed via the likelihood ratio test by adding an interaction term of a stratifying variable and C-peptide.
We calculated C-statistics based on logistic regression models to assess the predictive ability of early-pregnancy C-peptide for GDM. Four models were established in our analyses: Model 1 included conventional predictive factors for GDM including maternal age, gestational age, pre-pregnancy BMI, physical activity, parental history of DM, and history of GDM; Model 2 included conventional predictive factors and C-peptide; Model 3 included conventional predictive factors and FBG; Model 4 included conventional predictive factors, FBG, and C-peptide. To compare the discriminative performance, the Delong test was used to compare the C-statistics. Moreover, we used net reclassification improvement (NRI) [18] and integrated discrimination improvement (IDI) [19] statistics to measure the utility of C-peptide in GDM prediction.
Data analyses were performed by STATA 15.0 (Stata Corporation, College Station, TX, US). Partial Spearman’s correlation coefficients were visualized by GraphPad Prism 8 (GraphPad Software Inc., San Diego, CA, USA). NRI and IDI were calculated by comparison of predictive models using SAS 9.4 (SAS Institute, Cary, NC, USA). Two-sided P < 0.05 was considered to indicate statistical significance.