Fasting Metabolite Panel for OGTT-Free Diagnosis of Gestational Diabetes Mellitus: A Machine Learning Approach Validated in Dual Cohorts

Abstract

Objective

This study aimed to develop a fasting serum metabolite-based method for screening and risk assessment of gestational diabetes mellitus (GDM), potentially reducing dependence on the oral glucose tolerance test (OGTT).

Methods

Using a retrospective discovery cohort (n = 435; April-May 2021) with prospective validation (n = 473; November 2018-May 2021) design, 1,053 pregnant women completing standard 75g OGTT were initially enrolled. Fasting serum samples underwent targeted metabolomic profiling. A diagnostic model was constructed using machine learning (random forest) in combination with univariate analysis and rigorous validation protocols. Model performance was evaluated using the area under the receiver operating characteristic curve (ROC).

Results

Eight metabolites demonstrated significant differential expression between GDM and non-GDM groups (FDR <0.05). Based on the feature importance rankings, we developed a multivariate logistic regression model incorporating seven metabolites: 2-hydroxybutyric acid, 1,5-anhydroglucitol, glycine, 3-methyl-2-oxobutyric acid, 3-methyl-2-oxovaleric acid, tyrosine, and oleic acid. The composite model (fasting glucose + risk factors + metabolites) demonstrated significantly higher discriminative performance in the discovery cohort (AUC = 0.78) compared to fasting glucose alone (AUC = 0.62), with sustained performance in external validation (AUC = 0.71).

Conclusion

This fasting metabolite detection protocol demonstrates promising potential for GDM screening and risk stratification, offering the prospect of reducing reliance on OGTT in specific clinical settings.

Introduction

Gestational diabetes mellitus (GDM) is an abnormality of glucose tolerance first detected during pregnancy and poses a significant clinical challenge.¹ The prevalence of GDM in China ranges from 14% to 21%, associated with both short- and long-term adverse effects on maternal and child health.^2,3

The oral glucose tolerance test (OGTT) is recommended for diagnosing GDM in China.⁴ This method requires women to consume 75 g of glucose (dissolved in 300 mL of water and consumed within 5 min) after an initial fasting blood draw, followed by additional blood draws one and two hours later. Pregnant women are prohibited from eating or exercising during the test. However, this induces anxiety and physical discomfort due to the multiple blood draws and long waiting time.⁵ In addition, rapid ingestion of glucose solution can cause nausea and vomiting, which is particularly challenging for some women.^5,6 Recent studies demonstrate that 7–10% of pregnant women develop emesis post-glucose ingestion, which may hinder the completion of the OGTT.^7,8 Rare but severe adverse events, including hyperglycemic hyperosmolar state, further limit its safety.⁹

Notably, women with GDM exhibit a higher risk of developing type 2 diabetes mellitus (T2DM),^10–12 suggesting shared pathophysiology such as insulin resistance and β-cell dysfunction. With advancements in high-throughput metabolomics platforms, numerous metabolite biomarkers have been identified for early prediction and diagnosis of T2DM.^13–15 The OGTT evaluates β-cell function and glucose regulation capacity, requiring short-term glucose consumption and repeated blood sampling. This raises a key question: Does the fasting metabolic profile capture sufficient information to diagnose GDM and eliminate the need for the OGTT? Unfortunately, current studies have not yielded satisfactory alternatives.^16–18 Consequently, we aimed to develop a fasting plasma metabolite panel, validated in dual cohorts, as a reliable OGTT-free diagnostic tool for GDM.

Methods

Study Design and Cohort Recruitment

This study utilized a dual-cohort design, incorporating a retrospective discovery cohort and a prospective validation cohort. The discovery cohort comprised pregnant women who underwent the OGTT between April and May 2021. The validation cohort was derived from a prospective cohort study from November 2018 to May 2021. Baseline characteristics for the discovery cohort were extracted from electronic hospital records, whereas corresponding data for the validation cohort were collected via investigator-administered questionnaires. Inclusion criteria were: 1) completion of a 75-g OGTT as per Chinese guidelines; 2) age ≥18 years at enrollment; and 3) provision of fasting samples collected under standardized conditions (fasting ≥8 h, no caloric intake or exercise before sampling). Exclusion criteria were designed to minimize potential confounders associated with GDM, including: 1) fasting glucose ≥5.1 mmol/L (the diagnostic cutoff for GDM per Chinese guidelines); 2) multiple pregnancies (to avoid placental metabolic heterogeneity); 3) pregnancies with assisted reproductive technology (potential hormonal interference); 4) BMI ≥28 kg/m² (chosen to exclude severe obesity-related metabolic disturbances while maintaining BMI as a continuous variable within the non-obese population for risk stratification); 5) a previous history of GDM. The strict exclusion criteria mentioned earlier aim to reduce metabolic variability and improve the internal validity of the biomarker discovery process. By excluding populations with known significant metabolic confounders, we intended to focus on fasting metabolic signatures specific to GDM rather than those linked to related complications. Although this method may limit immediate generalizability, it allows for a more accurate assessment of the diagnostic potential of fasting metabolites in a controlled setting.

Diagnosis of GDM

GDM was diagnosed according to Chinese guidelines,⁴ in which a positive diagnosis was confirmed if any of the following 75-g OGTT plasma glucose thresholds were met or exceeded: fasting: 5.1 mmol/L; 1-hour: 10.0 mmol/L; 2-hour: 8.5 mmol/L. A subgroup of women with isolated post-challenge hyperglycemia (here termed TGP-GDM) was also defined, characterized by fasting glucose <5.1 mmol/L but with both 1-hour glucose ≥10.0 mmol/L and 2-hour glucose ≥8.5 mmol/L.^19,20

Sample Collection and Analysis

All fasting blood samples used for metabolomics analysis were collected, processed, and stored according to a strictly standardized protocol to minimize potential pre-analytical variability. All samples were collected, centrifuged, and analyzed within our laboratory. Blood samples should be processed within 30 min of collection, centrifuged at 1500g for 10 min to obtain serum. After centrifuging each sample, visually inspect and exclude any hemolyzed samples. Subsequently, aliquot the serum and store it at −80°C for subsequent testing. Each sample underwent only one freeze-thaw cycle. The laboratory staff responsible for testing were blinded to the identity and clinical information of the pregnant woman corresponding to each sample. Glucose was measured using the hexokinase method (Beckman Coulter Au5800, USA), with an intra-assay CV of 1.5%. Our laboratory is certified by external quality assurance programs of both the Zhejiang Provincial Clinical Laboratory Center and the National Clinical Laboratory Center. Processed serum samples were subsequently used for targeted metabolomic analysis. To eliminate potential biases, metabolomics laboratory personnel use de-identified codes for all data transmission and analysis without knowing the clinical outcomes. Eventually, academic researchers associate and integrate the data with the clinical results.

Design of Metabolite Biomarker Panel

To identify potential biomarkers for diagnosing GDM, we reviewed publications on metabolite biomarkers associated with insulin resistance (IR) and T2DM. The following criteria guided the selection of metabolites: 1) metabolites have been studied in ≥2 independent studies; 2) metabolites represented diverse metabolic pathways; 3) to ensure the quantitative accuracy of LC/MS, the selected metabolites must have commercially available chemical standards and stable isotope markers. Supplementary Table 1 details the selected metabolite biomarkers, associated pathways, and key references.

Metabolite Quantification Methodology

We have developed an absolute quantitative method for detecting metabolite biomarkers in serum. Initially, serum samples were precipitated with methanol, and stable isotope-labeled compounds were added as internal standards. Subsequently, the supernatant was obtained through centrifugation, followed by detection analysis using the Sciex Exion LC/Sciex 6500+ triple quadrupole tandem mass spectrometry system. Reverse-phase chromatography was initially employed, using a gradient elution with 0.01% formic acid in water/acetonitrile over 4 min. An electrospray ionization (ESI) source was used in both positive and negative ion mode with multiple reaction monitoring (MRM) scan mode to detect nine organic acids and lipids, including 2-hydroxybutyric acid, 3-hydroxybutyric acid, 3-methyl-2-oxobutyric acid, 3-methyl-2-oxovaleric acid, 4-methyl-2-oxopentanoic acid, pantothenic acid, palmitoyl carnitine, oleic acid, and palmitoyl-lysophosphatidylcholine. Subsequently, ion-pair chromatography was performed using a gradient elution from 0.05% perfluoropentanoic acid in water to 0.05% perfluoropentanoic acid in acetonitrile over 3 min. An ESI source was used in positive-ion mode with MRM scanning to detect eight amino acids: glycine, isoleucine, leucine, 2-aminoadipic acid, phenylalanine, tyrosine, valine, and serine. Lastly, hydrophilic interaction chromatography (HILIC) was used with a gradient elution from 0.1% triethylamine in acetonitrile to 0.1% triethylamine in water over 2 min. ESI source was used in negative-ion mode with MRM scanning to detect 1,5-anhydroglucitol. Each analytical batch included one standard curve and six quality control (QC) samples at the beginning and end of the run to monitor analytical performance. The standard curve comprised eight concentrations of standard substances. In comparison, the QC samples consisted of spiked samples at three different concentrations, with two replicates per concentration. Quantification was performed using weighted linear least-squares regression, with the ratio of compound peak area to the corresponding internal standard peak area used as the dependent variable.

Statistical Analyses

Data were analyzed using IBM SPSS (version 23.0) and R (version 4.1.3). The normality of continuous data was assessed using the Shapiro–Wilk test. Continuous variables that conform to a normal distribution are presented as mean ± standard deviation (SD), and those that do not conform are expressed as median (interquartile range). Welch’s t-test compared differences between the GDM and non-GDM for normally distributed continuous variables. For non-normally distributed continuous variables, the Mann–Whitney U-test was applied. Categorical variables are represented as n (%), with group differences analyzed using the Pearson χ² test or Fisher’s exact test. Fold change and p-value for each metabolite were calculated to compare the GDM and non-GDM groups. Random forest, a tree-based machine learning method, was employed to rank biomarker candidates based on mean decrease accuracy, which indicates each biomarker’s contribution to the group differences. Biomarker candidates were then incrementally added into a logistic regression model to develop a diagnostic model for GDM identification, and ROC curves were generated. To ensure that the selected metabolite biomarker panel was optimal (exhaustion test), all possible combinations of seven metabolite biomarkers (> 30,000) were tested and ranked by area under the curve (AUC) and sensitivity in both the discovery set and the validation cohort. The diagnostic performance of the model was subsequently validated in the validation cohort. Statistical significance was defined as p <0.05.

Results

Baseline Characteristics of the Discovery and Validation Groups

Of the 1,053 pregnant women initially recruited, 145 were excluded because they did not meet the inclusion criteria (Supplementary Table 2). Ultimately, the discovery cohort comprised 435 participants, whereas the validation cohort comprised 473 participants (Table 1). The incidence of GDM differed significantly between cohorts (discovery: 15.63% [68/435] vs validation: 24.73% [117/473]; p <0.001). Compared with the discovery cohort, the validation cohort demonstrated significantly greater proportions of advanced age (54.33% [257/473] vs 15.63% [68/435]; p <0.001), along with elevated fasting glucose (4.43 ± 0.28 mmol/L vs 4.36 ± 0.29 mmol/L; p <0.001), 1-h (8.33 ± 1.69 mmol/L vs 7.80 ± 1.56 mmol/L; p <0.001) and 2-h (7.26 ± 1.43 mmol/L vs 6.88 ± 1.31 mmol/L; p <0.001) post-load glucose levels (Table 1). Subsequent subgroup analysis of GDM cases showed no significant inter-cohort differences in glucose homeostasis markers or BMI (p >0.05).

Table 1.

Baseline Characteristics of the Participants in This Study

Characteristics	N	Discovery Cohort			Validation Cohort			p1 (All)	p2 (GDM)	p3 (Non-GDM)
		All-D	GDM-D	Non-GDM-D	All-V	GDM-V	Non-GDM-V
All, n (%)	908	435	68	367	473	117	356
Age, mean (SD), years	908	30.36 ± 3.95	31.53 ± 4.25	30.15 ± 3.86	34.48 ± 4.43	35.81 ± 4.44	34.05 ± 4.35	< 0.001	< 0.001	< 0.001
Age category, n (%), years	908							< 0.001	< 0.001	< 0.001
<35		367 (84.37)	49 (72.06)	318 (86.65)	216 (45.67)	36 (30.77)	180 (50.56)
≥35		68 (15.63)	19 (27.94)	49 (13.35)	257 (54.33)	81 (69.23)	176 (49.44)
BMI, mean (SD), kg/m2	908	20.75 ± 2.42	21.57 ± 2.39	20.60 ± 2.40	21.11 ± 2.47	21.24 ± 2.65	21.07 ± 2.41	0.028	0.387	0.009
BMI category, kg/m2, n (%)	908							0.451	0.402	0.885
<24		386 (88.73)	59 (86.76)	327 (89.10)	412 (87.10)	96 (82.05)	316 (88.76)
≥24		49 (11.26)	9 (13.23)	40 (10.90)	61 (12.90)	21 (17.95)	40 (11.24)
BMI*, mean (SD), kg/m2	908	23.34 ± 2.49	24.07 ± 2.43	23.19 ± 2.49	23.75 ± 2.60	23.61 ± 2.60	23.80 ± 2.60	0.016	0.245	0.002
Primigravida, n (%)	908							< 0.001	< 0.001	< 0.001
Yes		227 (52.18)	34 (50.00)	193 (52.59)	131 (27.70)	25 (21.38)	106 (29.78)
No		208 (47.82)	34 (50.00)	174 (47.41)	342 (72.30)	92 (78.63)	250 (70.22)
OGTT time, week	908	24.83 ± 1.25	24.85 ± 1.11	24.83 ± 1.28	25.28 ± 1.06	25.28 ± 1.15	25.28 ± 1.03	< 0.001	0.015	< 0.001
Fasting glucose (mmol/L)	908	4.36 ± 0.29	4.48 ± 0.32	4.34 ± 0.28	4.43 ± 0.28	4.51 ± 0.30	4.40 ± 0.27	< 0.001	0.449	0.006
1h-glucose (mmol/L)	908	7.80 ± 1.56	10.09 ± 1.01	7.37 ± 1.25	8.33 ± 1.69	10.17 ± 1.19	7.72 ± 1.36	< 0.001	0.626	< 0.001
2h-glucose (mmol/L)	908	6.88 ± 1.31	8.86 ± 1.22	6.52 ± 0.95	7.26 ± 1.43	8.97 ± 1.29	6.70 ± 0.94	< 0.001	0.575	0.009
HBA1C (%)	886	5.11 ± 0.30	5.23 ± 0.30	5.09 ± 0.29	5.07 ± 0.33	5.13 ± 0.28	5.04 ± 0.34	0.038	0.025	0.069
GDM (%)	185 (20.37)	68 (15.63)			117 (24.73)			< 0.001

Notes: A continuous variable is presented as mean ± SD, while a categorical variable is defined as n (%). *: BMI at OGTT. p < 0.05 was considered statistically significant. p1: All-D VS All-V; p2: GDM-D VS GDM-V; p3: non-GDM-D VS non-GDM-V.

Abbreviations: D, discovery; V, validation; BMI, body mass index; OGTT, oral glucose tolerance test; GDM, gestational diabetes mellitus.

Association Between Metabolite Biomarkers and GDM

Serum concentrations of metabolite biomarkers in GDM and non-GDM groups are shown in Figure 1. In the discovery cohort, 1,5-anhydroglucitol levels were significantly reduced in the GDM group (p <0.050). In contrast, elevated concentrations were observed for 3-hydroxybutyric acid (p <0.0001), 2-hydroxybutyric acid (p <0.0001), pantothenic acid (p <0.050), palmitoyl carnitine (p <0.0001), oleic acid (p <0.0001), phenylalanine (p <0.050), and 2-aminoadipic acid (p <0.050). The remaining biomarkers did not exhibit statistically significant differences between groups. As detailed in Supplemental Table 3, there were distinct profiles between GDM and non-GDM groups, with oleic acid (p = 2.20E-07), 2-hydroxybutyric acid (p = 2.26E-05), and 3-hydroxybutyric acid (p = 0.0007) demonstrating the strongest discriminatory power.

Figure 1.

Serum concentrations of metabolite biomarkers in GDM and non-GDM groups in the discovery cohort Differences between GDM and non-GDM groups were analyzed using Welch’s t-test for normally distributed variables and the Mann–Whitney U-test for non-normal variables.

Note: *p <0.05, **p <0.01, ***p <0.001, compared with non-GDM women. NS: Not significant; Metabolite concentrations: μg/mL; Negative: non-GDM; Positive: GDM.

Development of the GDM Diagnostic Model

Random forest analysis ranked the importance of metabolite biomarkers, identifying oleic acid as the most crucial, followed by 2-hydroxybutyric acid, 3-hydroxybutyric acid, palmitoyl carnitine, and pantothenic acid (Figure 2). Based on the feature importance rankings, we developed a multivariate logistic regression model incorporating seven metabolites: 2-hydroxybutyric acid, 1,5-anhydroglucitol, glycine, 3-methyl-2-oxobutyric acid, 3-methyl-2-oxovaleric acid, tyrosine, and oleic acid (Table 2). In the discovery cohort, fasting glucose alone achieved an AUC of 0.62, while clinical risk factors (including age and BMI) showed an AUC of 0.59. The metabolite panel significantly outperformed these benchmarks with an AUC of 0.75. When fasting glucose, risk factors, and the metabolite biomarker panel were combined, the AUC increased to 0.78. Similar trends were observed in the validation cohort. Furthermore, we evaluated the ability to predict GDM in women diagnosed with TGP-GDM. The fasting glucose levels, risk factors, and metabolite biomarker panel showed a predictive AUC of 0.83 for TGP-GDM.

Figure 2.

Analysis of the discriminating abilities of the metabolite biomarkers between GDM and non-GDM in the discovery cohort. (A) A significance test of the metabolite biomarkers between GDM and non-GDM. (B) Metabolites ranked by the mean decrease in accuracy in the discovery cohort by random forest analysis.Random forest analysis was performed to rank metabolite importance based on mean decrease in accuracy. One thousand bootstrap iterations were used to determine statistical significance.

Table 2.

ROC Curve Evaluations of Biomarkers for GDM

Model	All GDM-AUC		TGP-GDM-AUC
	Discovery 95% CI	Validation 95% CI	Discovery 95% CI	Validation 95% CI
Model 1: Fasting glucose	0.62 (0.55–0.70)	0.61 (0.54–0.67)	0.71 (0.60–0.80)	0.68 (0.61–0.76)
Model 2: Risk factors	0.59 (0.56–0.69)	0.61 (0.52–0.65)	0.65 (0.54–0.75)	0.65 (0.56–0.72)
Model 3: Metabolites	0.75 (0.69–0.81)	0.69 (0.64–0.75)	0.78 (0.69–0.87)	0.79 (0.72–0.85)
Model 1 + Model 2	0.67 (0.60–0.73)	0.64 (0.56–0.68)	0.75 (0.66–0.83)	0.72 (0.63–0.79)
Model 1 + Model 3	0.76 (0.69–0.82)	0.71 (0.66–0.77)	0.82 (0.73–0.88)	0.82 (0.77–0.87)
Model 2 + Model 3	0.77 (0.71–0.83)	0.69 (0.65–0.75)	0.80 (0.71–0.89)	0.79 (0.72–0.85)
Model 1 + Model 2 + Model 3	0.78 (0.71–0.83)	0.71 (0.65–0.77)	0.83 (0.75–0.90)	0.83 (0.76–0.88)

Notes: Risk factors: Age and BMI.Metabolites: Tyrosine, Oleic acid, 2-hydroxybutyric acid, 1,5-anhydroglucitol, Glycine, 3-methyl-2-oxobutyric acid, and 3-methyl-2-oxovaleric acid TGP-GDM: fasting glucose <5.1 mmol/L; 1 h ≥10.0 mmol/L; 2h ≥8.5 mmol/L.

Abbreviations: ROC, receiver operating characteristic; AUC, area under the curve.

Discussion

Our study revealed the potential of a fasting serum metabolite biomarker panel for screening and risk assessment in GDM, suggesting the possibility of reducing reliance on OGTT in specific clinical settings. The combined model, incorporating fasting glucose, clinical risk factors, and seven metabolites, demonstrated consistent diagnostic performance, surpassing that of fasting glucose alone. Notably, the model showed enhanced accuracy in identifying TGP-GDM cases, a subgroup at higher risk of adverse outcomes. These results emphasize the promise of metabolomics to address the limitations of OGTT and detect key metabolic disturbances in GDM.

Metabolomics offers insights into metabolic changes associated with diabetes, providing potential for early diagnosis and treatment.^21,22 Compared to the non-GDM group, one metabolite biomarker decreased while seven increased in the GDM group. Additionally, we explored, for the first time, the association between five metabolite biomarkers and GDM. Specifically, 2-aminoadipic acid and palmitoylcarnitine were found to be elevated in the GDM, whereas 3-methyl-2-oxobutyric acid, palmitoyl-lysophosphatidylcholine, and 4-methyl-2-oxopentanoic acid showed no significant differences. This underscores both the physiological similarities and differences between GDM and T2DM. 1,5-anhydroglucitol is one of the essential polyols in the human body, functioning similarly to glucose because it is reabsorbed in the renal tubules.^23,24 However, when glucose levels are elevated, the renal tubule reabsorption of 1,5-anhydroglucitol is reduced, leading to increased urinary excretion and decreased serum levels. Our findings showed lower levels of 1,5-anhydroglucitol in GDM than in non-GDM, consistent with previous studies.^25,26 This suggests that 1,5-anhydroglucitol serves as a valuable marker for assessing glucose regulation. Beyond carbohydrate metabolism, other metabolic pathways, such as amino acid and lipid metabolism, are also critical in GDM. Amino acids can stimulate β-cell proliferation, thereby promoting insulin secretion. However, an increased concentration of amino acids may inhibit the degradation of insulin receptor substrate, leading to IR and adversely affecting insulin secretion.²⁷ A prospective cohort study of 646 obese women reported higher valine, leucine, and isoleucine levels in the GDM group than in the non-GDM group, with a similar trend observed for phenylalanine and tyrosine.²⁸ In contrast, our results showed elevated phenylalanine levels in the GDM group, whereas the other amino acids did not differ significantly. We attributed this discrepancy to variations in the study populations. Their study primarily included obese people with a BMI ≥30 kg/m², while our study excluded women with a BMI ≥28 kg/m².

2-hydroxybutyric acid indicates metabolic abnormalities in T2DM, with elevated levels suggesting a disruption in insulin-glucose homeostasis and predicting deterioration in glucose control.^29–31 Our findings revealed significantly higher levels of 2-hydroxybutyric acid in the GDM group compared with the non-GDM group. Similarly, oleic acid, a crucial monounsaturated fatty acid synthesized endogenously in the human body, showed a marked increase in the GDM group. Among the metabolite biomarkers, oleic acid showed the most significant difference between GDM and non-GDM (p = 2.20E-07), and random forest analysis further identified it as the most essential metabolite biomarker. Previous study has shown that oleic acid can prevent IR and exert protective effects against T2DM.³² Notably, elevated oleic acid levels have been observed in patients with impaired glucose tolerance, though no significant difference was observed in those with impaired fasting glucose.³³ Thus, oleic acid may be a selective biomarker for assessing isolated impaired glucose tolerance. We evaluate the diagnostic performance of our model. The AUC values for fasting glucose and individual risk factors were 0.62 and 0.59, respectively, indicating their limited utility in diagnosing GDM. However, the AUC of the seven metabolite biomarkers identified was 0.75. When combined with fasting glucose and risk factors, the AUC improved from 0.67 to 0.78. The validation cohort achieved an AUC of 0.71, indicating moderate discrimination. In clinical practice, this performance level suggests that the current metabolite detection protocol is not sufficient as a standalone diagnostic tool. Its main value lies in risk stratification, identifying pregnant women who need increased vigilance or prioritization for OGTT.

A similar trend was observed in the validation cohort. In addition, we evaluated the model’s diagnostic power in women with TGP-GDM, as the OGTT’s reliability in diagnosing GDM has been questioned, particularly for those with borderline glucose levels at 1 or 2 h.^34,35 Variations in glucose test results and sample processing time can affect diagnostic accuracy,^36–40 and a study suggests caution in interpreting a single positive glucose result on an OGTT.³⁵ Thus, we focused on women with positive glucose levels at both 1 and 2 h, as they may have more severe metabolic disorders and IR, as well as a higher risk of adverse pregnancy outcomes.^17,41 Therefore, we need to pay special attention to this part of pregnant women diagnosed with TGP-GDM. The high predictive accuracy of our model for TGP-GDM (AUC = 0.83) underscores its potential to identify high-risk pregnancies that require enhanced monitoring. However, research indicates that TGP-GDM is associated with a favorable metabolic phenotype, as reflected in maternal and neonatal outcomes.¹⁹

Our study represents one of the most significant targeted metabolomics investigations on GDM in China. To minimize the potential effects of gestational age on metabolite biomarkers, we used fasting samples from the OGTT, thereby ensuring consistent and reproducible results. We also employed separate validation cohorts, confirming that our model is stable and reliable. However, there are some limitations. First, this study excluded pregnant women with common GDM risk factors such as BMI ≥ 28 kg/m², a history of GDM, or multiple pregnancies, which may limit the clinical usefulness of this model. This approach aimed to reduce confounding factors to validate the feasibility of diagnosing fasting metabolites in a population with similar metabolic traits. Therefore, this study only provides an initial feasibility assessment. Future research should test this model in larger populations to determine its real-world effectiveness. Secondly, despite rigorous pre-analytical controls applied to the samples, unavoidable handling variations, such as minor delays and storage times, may still affect metabolite levels. Future multicenter studies should adopt more standardized protocols. Finally, although including a metabolite biomarker panel improved diagnostic precision, it could increase the financial burden on pregnant women. However, our study offers an alternative for pregnant women unable to undergo the OGTT.

This study confirmed the feasibility of a metabolomics screening approach, demonstrating that the identified fasting metabolite signature reflects underlying pathophysiological dysregulation associated with GDM. However, the current model is more appropriate as a screening or triage tool than as a definitive substitute for OGTT diagnosis. Future research should improve the model’s performance and validate it in diverse obstetric populations (including high-risk groups) to determine its actual clinical value.

Funding Statement

The National Key R&D Program of China (2018YFC1002702). The funders played no role in the research design, data collection and analysis, publication decisions, or paper preparation.

Data Sharing Statement

The data in the article were collected from pregnant women who visited the clinic and were not publicly available. Interested researchers may contact the corresponding author for relevant data.

Ethical Approval

This study was conducted in accordance with the ethical standards of the Declaration of Helsinki and was approved by the Ethics Committee of the Women’s Hospital Affiliated with Zhejiang University School of Medicine, which has approved this study (Approval No. 20180142; IRB-20210325-R). All participants provided written informed consent, and data anonymization protocols were strictly adhered to protect patient privacy.

Author Contributions

Binbin Yin: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing-original draft; Yiyun Shen: Conceptualization, Investigation, Methodology, Writing-original draft; Qianwen Zhang: Conceptualization, Data curation, Visualization, Writing - review & editing; Rongchang Chen: Conceptualization, Data curation, Formal analysis, Validation, Writing - review & editing, Software; Xue Zhang: Visualization, Writing - review & editing; Ziqing Kong: Conceptualization, Methodology, Supervision, Visualization, Writing - review & editing; Yuning Zhu: Conceptualization, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing - review & editing. All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors Qianwen Zhang, Rongchang Chen, Zhang Xue, and Kong Ziqing are affiliated with Calibra Scientific, Inc. All other authors declare no conflict of interest regarding the publication of this article.