First-trimester lipid- and glucose-derived indices for predicting gestational diabetes mellitus: development of a combined model in a single-center retrospective cohort

Background

Gestational diabetes mellitus (GDM) is associated with substantial maternal and neonatal morbidity. Early identification of women at risk remains a clinical priority. This study aimed to develop and evaluate a simple first-trimester risk prediction model for subsequent GDM using fasting triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and glucose-derived indices (TyG, TG/HDL-C, and Lipid-IR).

Methods

This single-center retrospective cohort included 679 pregnant women with first-trimester fasting lipid and glucose measurements. Of these, 342 developed GDM (diagnosed at 24–28 weeks using a two-step approach and Carpenter–Coustan criteria), and 337 remained normoglycemic. Baseline demographic, anthropometric, laboratory, and obstetric/neonatal variables were compared. TyG was calculated as ln[(fasting triglycerides (mg/dL) × fasting glucose (mg/dL)) / 2], TG/HDL-C as TG (mg/dL) divided by HDL-C (mg/dL), and Lipid-IR as ln(2 × TG (mg/dL) + total cholesterol (mg/dL)). Independent predictors were identified through multivariable logistic regression. Receiver operating characteristic (ROC) analysis evaluated discriminative ability and optimal cut-off points using Youden’s J statistic.

Results

Compared with controls, women who developed GDM were older and had higher pre-pregnancy and current body weights, body mass index (BMI), and waist circumference (all p ≤ 0.003). They also exhibited higher fasting glucose, HbA1c, insulin, TG, total cholesterol, and liver enzyme levels, and lower HDL-C concentrations (all p ≤ 0.03). Cesarean delivery was more frequent among women with GDM (58% vs. 32%; p = 0.001), with higher birth weights and lower 5-minute Apgar scores. In multivariable models, Lipid-IR (OR 1.85), TG/HDL-C (OR 2.12), and TyG (OR 1.63) were independently associated with GDM (all p < 0.001). Discrimination was strong, with AUCs of 0.88 for TyG, 0.82 for Lipid-IR, and 0.79 for TG/HDL-C; combining all three indices increased the AUC to 0.92 (sensitivity 89%, specificity 85%).

Conclusion

In this single-center retrospective cohort, first-trimester lipid–glucose indices—particularly the TyG index were strongly associated with subsequent GDM, and a combined model showed good discrimination for risk stratification. External validation and assessment of clinical impact are required before routine clinical implementation.

Clinical trial number

Not applicable.

Clinical implication

Simple lipid–glucose–derived indices can be incorporated into early prenatal screening to identify women at high risk for GDM, enabling early risk stratification and intensified follow-up prior to OGTT, rather than treatment initiation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12902-026-02192-3.

Highlights

Early lipid–glucose indices predict gestational diabetes: First-trimester TyG index, TG/HDL-C ratio, and Lipid-IR were identified as independent predictors of gestational diabetes mellitus (GDM).

Strong predictive accuracy of the TyG index: The TyG index achieved the highest discriminative performance (AUC = 0.88; sensitivity 85%; specificity 82%).

Combined model provides superior performance: Integration of TyG, TG/HDL-C, and Lipid-IR indices enhanced diagnostic accuracy (AUC = 0.92; sensitivity 89%; specificity 85%).

Feasibility in routine prenatal care: These indices can be automatically calculated from standard first-trimester lipid and glucose tests, allowing pre-OGTT risk stratification.

Clinical implications: Early identification of at-risk women may enable timely lifestyle interventions, reducing macrosomia and adverse obstetric outcomes.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12902-026-02192-3.

Key findings

• TyG index, TG/HDL-C ratio, and Lipid-IR were all independently associated with GDM.

• TyG showed the strongest individual predictive ability (AUC 0.88).

• The combined model of all three indices achieved AUC 0.92, outperforming individual parameters.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12902-026-02192-3.

Concept

Study design

Gestational diabetes mellitus (GDM) remains a major cause of maternal and neonatal morbidity. Early detection before the oral glucose tolerance test (OGTT) may improve outcomes.

Retrospective cohort of 679 pregnant women (342 GDM, 337 controls) with first-trimester fasting lipid and glucose data.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12902-026-02192-3.

Introduction

Gestational diabetes mellitus (GDM) is defined as glucose intolerance that develops during pregnancy, typically diagnosed in the second half of gestation. It exerts significant short- and long-term effects on both maternal and fetal health, increasing the risk of preeclampsia, macrosomia, cesarean delivery, and congenital metabolic disorders [1]. Recent meta-analyses have reported that the prevalence of GDM varies between 5.2% and 13.7%, depending on diagnostic thresholds and study populations, with a prevalence of 13.7% in cohorts utilizing a one-step screening approach [2]. Data from Canada and the United States indicate that the prevalence of GDM increased from 6.1% to 10.4% between 2005 and 2019 [3].

Traditionally, GDM screening is performed using the oral glucose tolerance test (OGTT) at 24–28 weeks of gestation. However, the clinical utility of shifting this screening to earlier stages of pregnancy has become an emerging topic of interest. The 2024 American Diabetes Association (ADA) Standards of Care recommend assessing HbA1c and glucose parameters during the first trimester in high-risk women [4]. Pregnancy is characterized by progressive physiological insulin resistance driven by placental hormones, maternal adipose tissue expansion, and inflammatory signaling, resulting in adaptive changes in glucose and lipid metabolism [1]. In women who later develop GDM, these metabolic alterations may emerge earlier and more prominently, with higher fasting triglycerides, lower HDL-C, and subtle increases in fasting glucose even in the first trimester [2]. Because triglyceride-rich lipoprotein metabolism is closely linked to hepatic insulin resistance, combined lipid–glucose surrogates may capture early dysmetabolism before conventional screening tests are performed [3].

Accordingly, simple indices derived from routinely available fasting parameters—such as the triglyceride–glucose (TyG) index, the TG/HDL-C ratio, and lipid-based composite insulin resistance surrogates—have been proposed as practical tools for early risk stratification. If validated, these indices could help identify women who may benefit from earlier counseling and closer follow-up, rather than replacing the diagnostic OGTT.

Metabolic alterations occurring early in pregnancy have been investigated for their association with subsequent GDM development. Several studies have demonstrated that first-trimester serum triglyceride levels and the HDL-cholesterol ratio are significant predictors of GDM risk. Ma et al. reported that both the first-trimester TyG index and the TG/HDL-C ratio exhibit high diagnostic performance for predicting GDM, with areas under the curve (AUCs) of 0.88 and 0.79, respectively [5]. Similarly, a multicenter study found that the TyG index measured at the first prenatal visit was a useful predictor of GDM, with an AUC of 0.686 [6].

The objective of this study was to develop and evaluate an early-pregnancy (first-trimester) risk prediction model for subsequent GDM using three routinely available lipid–glucose-derived indices (TyG, TG/HDL-C, and Lipid-IR). We further aimed to report model discrimination (AUC), calibration (Hosmer–Lemeshow goodness-of-fit), and an operational predicted-probability threshold to facilitate practical risk stratification prior to standard OGTT screening.

Materials and methods

Study design and ethical approval

This study was designed as a single-center retrospective cohort investigation. The study protocol was approved by the Ethics Committee of Health Sciences University Izmir Tepecik Training and Research Hospital (21 Aug 2025; Ref. No. 05/28) and conducted in accordance with the principles of the Declaration of Helsinki. This study was conducted and reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines; the completed checklist is provided as a supplementary file.

Participants

Between June 2022 and January 2025, pregnant women whose fasting triglyceride (TG), glucose, and high-density lipoprotein cholesterol (HDL-C) levels were measured during the first trimester and who were subsequently diagnosed with GDM during the second trimester were included in the GDM group. During the same period, women with first-trimester measurements of these parameters who maintained normal glycemic levels throughout pregnancy were included as controls. All participants underwent routine GDM screening at 24–28 weeks using the institutional two-step approach; controls were defined as women who screened negative on the 50-g GCT and, when indicated, had a normal 100-g OGTT according to Carpenter–Coustan criteria.

Women with pregestational type 1 or type 2 diabetes, multiple gestations, or pre-existing metabolic disorders (including polycystic ovary syndrome (PCOS)), as well as chronic systemic diseases (e.g., hypertension, thyroid disorders) were excluded. Additionally, those without first-trimester follow-up data or whose delivery occurred at another institution after initial monitoring at our center were excluded from analysis.

Diagnosis of gestational diabetes mellitus

At our institution, GDM is diagnosed using a two-step approach. All pregnant women undergo a 50-g glucose challenge test (GCT) between 24 and 28 gestational weeks. Those with 1-hour plasma glucose ≥ 140 mg/dL undergo a diagnostic 100-g OGTT after an overnight fast (≥ 8 h). GDM is diagnosed when at least two values meet or exceed the Carpenter–Coustan thresholds. Following diagnosis, all patients initially receive dietary therapy; those who fail to achieve glycemic control within two weeks are transitioned to insulin therapy. Both diet-controlled and insulin-requiring cases were classified within the GDM group. Among women diagnosed with GDM, cases were further categorized according to treatment requirement as (i) diet-controlled GDM (medical nutrition therapy only) and (ii) insulin-requiring GDM. Subgroup analyses were performed to evaluate whether the predictive performance of the indices differed by GDM severity. For subgroup analyses, GDM cases were stratified by treatment requirement into (i) diet-controlled GDM (medical nutrition therapy only) and (ii) insulin-requiring GDM.

Data collection

For each participant, demographic and clinical variables including age, parity, smoking status, use of assisted reproductive techniques, fasting blood glucose (FBG), first-trimester systolic and diastolic blood pressure, HDL-C, and TG levels were recorded. Gestational age was calculated based on the last menstrual period and confirmed by crown–rump length measurement during first-trimester ultrasonography. Laboratory data were obtained from routine venous blood samples collected during standard prenatal visits at our institution. First-trimester blood samples were collected between 7 and 11 gestational weeks, in the morning, after an overnight fast (≥ 8–12 h).

Calculation of indices

A total of 679 participants were analyzed, comprising 342 women with GDM and 337 healthy controls, using first-trimester fasting TG, HDL-C, and glucose measurements.

• The TyG index was calculated according to the formula proposed by Simental-Mendía and Guerrero-Romero [7] as:

• The TG/HDL-C ratio was calculated by dividing fasting triglycerides (mg/dL) by HDL-C (mg/dL) [8]:

  

• The Lipid-IR index was defined as a composite indicator of lipid and glucose metabolism [8] and calculated as:

  

Statistical analysis

All analyses were performed using IBM SPSS Statistics version 22.0 (Armonk, NY, USA). Continuous variables were expressed as mean ± standard deviation (SD) or median [interquartile range, IQR], and categorical variables as number (percentage). The distribution of continuous variables was assessed using the Shapiro–Wilk test and graphical inspection. Comparisons between two groups were performed using the independent samples t-test for normally distributed variables and the Mann–Whitney U test for non-normally distributed variables. Categorical variables were compared using Pearson’s chi-square or Fisher’s exact test when expected frequencies were < 5.

Effect sizes were presented as mean/median differences with 95% confidence intervals (CI) for continuous variables and as odds ratios (OR) with 95% CI for categorical variables. Multivariable logistic regression analysis was conducted with GDM status (yes/no) as the dependent variable. Variables that were clinically relevant and/or showed a univariate association with p < 0.10 were included in the model. Multicollinearity was evaluated using the variance inflation factor (VIF); variables with VIF > 5 were excluded or adjusted. Model calibration was assessed using the Hosmer–Lemeshow goodness-of-fit test, and discrimination by the area under the receiver operating characteristic (ROC) curve (AUC).

ROC analyses were performed for the TyG index, Lipid-IR, and TG/HDL-C ratio. Optimal cut-off points were determined using Youden’s J statistic, and corresponding sensitivity and specificity were reported. Diagnostic performance of the combined model was evaluated based on AUC values derived from predicted probabilities of the multivariable logistic regression model. To develop the combined model, TyG, TG/HDL-C, and Lipid-IR were entered simultaneously into a multivariable logistic regression, and predicted probabilities were obtained from the fitted equation. The optimal probability threshold was determined using the Youden index from ROC analysis of predicted probabilities. A two-sided significance level of α = 0.05 was adopted for all tests. Cases with missing data were excluded using the listwise deletion method; no additional imputation was performed. ROC analyses were repeated for [1] diet-controlled GDM vs. controls and [2] insulin-requiring GDM vs. controls. Optimal cut-offs were determined using the Youden index for each comparison. Where applicable, AUCs were compared using appropriate methods.

Combined model and predicted probability

To operationalize the combined model, we fitted a multivariable logistic regression model including TyG, TG/HDL-C ratio, and Lipid-IR as predictors. The model-generated predicted probability (p) of GDM was calculated using the logistic function:

The optimal probability threshold was determined from the ROC curve of predicted probabilities using Youden’s J index; the selected cut-off was p ≥ 0.30.

Results

A total of 679 pregnant women were included, comprising 342 with gestational diabetes mellitus (GDM) and 337 with normal glucose tolerance. Among women diagnosed with GDM, 113 required insulin therapy, while the remaining cases were managed with diet/lifestyle measures. Women with GDM were significantly older than controls (31.8 ± 5.9 vs. 28.5 ± 5.2 years, p < 0.001). There was no difference in height (p = 0.12). Pre-pregnancy weight (70.5 ± 12.4 vs. 64.2 ± 8.7 kg, p = 0.003), current weight (81.2 ± 13.1 vs. 74.8 ± 9.5 kg, p = 0.001), BMI (29.8 ± 4.1 vs. 27.1 ± 3.5 kg/m², p < 0.001), and waist circumference (83.6 ± 8.9 vs. 78.4 ± 7.3 cm, p = 0.001) were all higher in the GDM group. While gravida, parity, and abortions tended to be higher in GDM, only the number of curettages reached statistical significance (0.4 ± 0.6 vs. 0.2 ± 0.4, p = 0.03). Gestational age at assessment was similar between groups (24.3 ± 1.2 vs. 24.6 ± 0.8 weeks, p = 0.22). Baseline demographic and clinical characteristics are summarized in Table 1.

Table 1.

Baseline demographic and clinical characteristics of the study population

Variable	Control (n = 337)	GDM (n = 342)	p-value
Age (years)	28.5 ± 5.2	31.8 ± 5.9	< 0.001
Height (cm)	162.3 ± 5.1	163.7 ± 6.3	0.12
Pre-pregnancy weight (kg)	64.2 ± 8.7	70.5 ± 12.4	0.003
Current weight (kg)	74.8 ± 9.5	81.2 ± 13.1	0.001
BMI (kg/m²)	27.1 ± 3.5	29.8 ± 4.1	< 0.001
Waist circumference (cm)	78.4 ± 7.3	83.6 ± 8.9	0.001
Gravida (number)	2.8 ± 1.5	3.1 ± 1.7	0.09
Gestational week at OGTT (24–28 weeks)	24.6 ± 0.8	24.3 ± 1.2	0.22
Parity	1.6 ± 1.1	1.9 ± 1.3	0.07
Abortions	0.4 ± 0.7	0.6 ± 0.9	0.06
Curettages	0.2 ± 0.4	0.4 ± 0.6	0.03

Data are expressed as mean ± SD. GDM: gestational diabetes mellitus; BMI: body mass index. Comparisons were made using independent-samples t-test or Mann–Whitney U test where appropriate. Two-tailed p < 0.05 was considered statistically significant

OGTT glucose levels were significantly higher at all time points among women with GDM (0 h: 112.6 ± 25.8 vs. 85.2 ± 10.4 mg/dL; 1 h: 158.4 ± 32.7 vs. 110.5 ± 18.3 mg/dL; 2 h: 135.9 ± 28.5 vs. 92.7 ± 12.6 mg/dL; all p < 0.001). Fasting glucose (108.2 ± 22.4 vs. 84.3 ± 9.8 mg/dL, p < 0.001) and HbA1c (6.9 ± 1.2 vs. 5.3 ± 0.6%, p < 0.001) were higher in the GDM group. Serum creatinine was modestly higher (0.61 ± 0.18 vs. 0.54 ± 0.12 mg/dL, p = 0.02), with a trend for BUN (p = 0.08). The lipid profile was more atherogenic in GDM: triglycerides (234.6 ± 98.5 vs. 182.4 ± 75.3 mg/dL, p = 0.004) and total cholesterol (238.7 ± 67.8 vs. 219.5 ± 45.2 mg/dL, p = 0.01) were higher, and HDL-C was lower (58.3 ± 12.7 vs. 68.5 ± 14.2 mg/dL, p < 0.001); LDL-C showed a non-significant increase (p = 0.09). Insulin levels were higher in GDM (15.6 ± 7.8 vs. 12.1 ± 4.3 µIU/mL, p = 0.006). Liver enzymes (GGT, ALT, AST) were also higher in the GDM group (p = 0.03, p = 0.01, and p = 0.02, respectively). Composite metabolic indices differed significantly: Lipid-IR (10.3 ± 1.8 vs. 8.9 ± 1.2, p < 0.001), TG/HDL-C (4.1 ± 2.3 vs. 2.8 ± 1.5, p < 0.001), and the TyG index (9.90 ± 1.38 vs. 8.75 ± 0.92, p < 0.001) were all higher in the GDM group. Laboratory and metabolic parameters are detailed in Table 2.

Table 2.

Laboratory and metabolic parameters at OGTT screening (24–28 weeks) according to GDM status

Variable	Control (n = 337)	GDM (n = 342)	p-value
OGTT 0 h (mg/dL)	85.2 ± 10.4	112.6 ± 25.8	< 0.001
OGTT 1 h (mg/dL)	110.5 ± 18.3	158.4 ± 32.7	< 0.001
OGTT 2 h (mg/dL)	92.7 ± 12.6	135.9 ± 28.5	< 0.001
Fasting glucose (mg/dL)	84.3 ± 9.8	108.2 ± 22.4	< 0.001
BUN (mg/dL)	14.2 ± 4.1	15.8 ± 5.3	0.08
Creatinine (mg/dL)	0.54 ± 0.12	0.61 ± 0.18	0.02
Triglycerides (mg/dL)	182.4 ± 75.3	234.6 ± 98.5	0.004
Total cholesterol (mg/dL)	219.5 ± 45.2	238.7 ± 67.8	0.01
HDL-C (mg/dL)	68.5 ± 14.2	58.3 ± 12.7	< 0.001
LDL-C (mg/dL)	124.6 ± 38.7	136.2 ± 52.4	0.09
HbA1c (%)	5.3 ± 0.6	6.9 ± 1.2	< 0.001
Insulin (µIU/mL)	12.1 ± 4.3	15.6 ± 7.8	0.006
GGT (U/L)	16.2 ± 6.5	19.4 ± 9.1	0.03
ALT (U/L)	14.8 ± 5.2	18.3 ± 7.6	0.01
AST (U/L)	18.3 ± 6.7	21.5 ± 9.4	0.02
Lipid-IR	8.9 ± 1.2	10.3 ± 1.8	< 0.001
TG/HDL-C ratio	2.8 ± 1.5	4.1 ± 2.3	< 0.001
TyG index	8.75 ± 0.92	9.90 ± 1.38	< 0.001

Data are presented as mean ± SD. OGTT: oral glucose tolerance test; BUN: blood urea nitrogen; HDL-C/LDL-C: high-/low-density lipoprotein cholesterol; HbA1c: glycated hemoglobin; GGT: γ-glutamyl transferase; ALT: alanine aminotransferase; AST: aspartate aminotransferase. Indices calculated as: TyG = ln[(TG × fasting glucose)/2], TG/HDL-C = TG ÷ HDL-C; Lipid-IR = ln(2×TG + total cholesterol). Two-tailed p < 0.05 indicates significance

Although gestational age at delivery was slightly lower in GDM, the difference was not significant (36.8 ± 1.9 vs. 37.2 ± 1.5 weeks, p = 0.06). Birth weight was higher in GDM (3580 ± 550 vs. 3250 ± 420 g, p = 0.008). The 1-minute Apgar score did not differ (p = 0.06), whereas the 5-minute Apgar score was lower in GDM (7.5 ± 1.4 vs. 8.1 ± 1.1, p = 0.022). Cesarean delivery was more frequent (58% vs. 32%, p = 0.001) and IUGR was more common (22% vs. 8%, p = 0.03) in the GDM group. NICU admission (12% vs. 5%, p = 0.15) and instrumental/complicated delivery (6% vs. 2%, p = 0.08) were higher without reaching significance; sex distribution was similar (p = 0.45). Perinatal outcomes are summarized in Table 3.

Table 3.

Delivery and neonatal outcomes

Variable	Control (n = 337)	GDM (n = 342)	p-value
Gestational age at delivery (weeks)	37.2 ± 1.5	36.8 ± 1.9	0.06
Birth weight (g)	3250 ± 420	3580 ± 550	0.008
Apgar 1 min	7.2 ± 1.3	6.8 ± 1.5	0.06
Apgar 5 min	8.1 ± 1.1	7.5 ± 1.4	0.022
Cesarean delivery, n (%)	108 (32.0)	198 (57.9)	0.001
IUGR, n (%)	27 (8.0)	75 (21.9)	0.03
NICU admission, n (%)	17 (5.0)	41 (12.0)	0.15
Instrumental/complicated delivery, n (%)	7 (2.1)	21 (6.1)	0.08
Neonatal sex — Male, n (%)	182 (54.0)	198 (57.9)	0.45
Neonatal sex — Female, n (%)	155 (46.0)	144 (42.1)

Continuous variables are expressed as mean ± SD; categorical variables as n (%). GA: gestational age; IUGR: intrauterine growth restriction; NICU: neonatal intensive care unit. Group comparisons: independent-samples t-test or Mann–Whitney U (continuous), Pearson chi-square or Fisher’s exact (categorical). Two-tailed p < 0.05 considered significant

In multivariable logistic regression (Table 4), Lipid-IR (OR 1.85, 95% CI 1.42–2.41), TG/HDL-C (OR 2.12, 95% CI 1.68–2.67), and TyG (OR 1.63, 95% CI 1.39–1.92) were independently associated with GDM (all p < 0.001). ROC analyses showed strong discrimination (Table 5; Fig. 1), with the highest AUC for TyG (0.88, 95% CI 0.84–0.91) and an optimal cut-off of ≥ 9.21 (sensitivity 85%, specificity 82%).

Table 4.

Multivariable logistic regression analysis for predictors of GDM

Variable	Odds Ratio (OR)	95% Confidence Interval	p-value
Lipid-IR	1.85	1.42–2.41	< 0.001
TG/HDL-C ratio	2.12	1.68–2.67	< 0.001
TyG index	1.63	1.39–1.92	< 0.001

ORs correspond to one-unit increases in each index. CI: confidence interval. The model included TyG, TG/HDL-C ratio, and Lipid-IR (index-only multivariable model). Two-tailed p < 0.05 denotes statistical significance

Table 5.

Receiver operating characteristic (ROC) analysis and diagnostic performance of lipid–glucose indices

Parameter	AUC (95% CI)	Optimal Cut-off	Sensitivity (%)	Specificity (%)
Lipid-IR	0.82 (0.78–0.86)	≥ 9.5	78	76
TG/HDL-C ratio	0.79 (0.74–0.83)	≥ 3.2	72	74
TyG index	0.88 (0.84–0.91)	≥ 9.21	85	82
Combined model (predicted probability)	0.92 (0.89–0.94)	Predicted probability ≥ 0.30	89	85

AUC: area under the ROC curve; CI: confidence interval. Optimal cut-off points determined by Youden’s J statistic. Sensitivity and specificity values correspond to respective thresholds. The combined model AUC derived from predicted probabilities of multivariable logistic regression incorporating TyG, TG/HDL-C, and Lipid-IR indices. For the combined model, the cut-off represents the predicted probability threshold from the multivariable logistic regression model (not a p-value). Predicted probabilities were computed using the logistic regression equation reported in the Statistical analysis section (β0 and β coefficients)

Fig. 1.

Receiver operating characteristic (ROC) analysis and diagnostic performance of lipid–glucose indices

ROC analyses demonstrated good to excellent discrimination for subsequent GDM (Table 5; Fig. 1). The TyG index showed the highest performance among single indices (AUC 0.88, 95% CI 0.84–0.91), with an optimal cut-off of ≥ 9.21, yielding 85% sensitivity and 82% specificity. Lipid-IR also performed well (AUC 0.82, 95% CI 0.78–0.86; cut-off ≥ 9.5; sensitivity 78%; specificity 76%), followed by the TG/HDL-C ratio (AUC 0.79, 95% CI 0.74–0.83; cut-off ≥ 3.2; sensitivity 72%; specificity 74%). A combined model incorporating TyG, TG/HDL-C, and Lipid-IR further improved discrimination (AUC 0.92, 95% CI 0.89–0.94), achieving 89% sensitivity and 85% specificity. For the combined model, the predicted probability of GDM was calculated from the logistic regression equation (see Statistical analysis), and ROC analysis of these probabilities yielded an optimal cut-off of p ≥ 0.30.

Subgroup analyses (Table 6), stratified by treatment requirement, suggested that predictive performance varied by GDM severity. In insulin-requiring GDM, the TyG index achieved an AUC of 0.92 (95% CI 0.88–0.96) with an optimal cut-off of ≥ 9.44 (sensitivity 90%, specificity 83%), whereas in diet-controlled GDM the AUC was 0.86 (95% CI 0.82–0.90) with a cut-off of ≥ 9.09 (sensitivity 82%, specificity 80%). Similar subgroup-specific patterns were observed for the TG/HDL-C ratio and Lipid-IR, with generally higher AUC values in the insulin-requiring subgroup. Overall, these findings suggest that first-trimester lipid–glucose indices—particularly TyG—may be especially informative for identifying women at risk of more severe GDM requiring pharmacologic therapy.

Table 6.

ROC performance of first-trimester indices stratified by GDM treatment requirement

Index	Diet-controlled GDM (n = 229) vs. Controls (n = 337) AUC (95% CI)	Cut-off	Sens (%)	Spec (%)	Insulin-requiring GDM (n = 113) vs. Controls (n = 337) AUC (95% CI)	Cut-off	Sens (%)	Spec (%)
TyG	0.86 (0.82–0.90)	≥ 9.09	82	80	0.92 (0.88–0.96)	≥ 9.44	90	83
TG/HDL-C	0.77 (0.72–0.82)	≥ 3.10	70	73	0.83 (0.78–0.89)	≥ 3.30	78	76
Lipid-IR	0.80 (0.75–0.85)	≥ 9.4	74	75	0.86 (0.81–0.92)	≥ 9.7	80	82
Combined model (predicted probability)	0.90 (0.87–0.93)	Predicted probability ≥ 0.30	86	83	0.94 (0.91–0.97)	Predicted probability ≥ 0.30	91	86

For the combined model, the cut-off represents the predicted probability (p) obtained from the logistic regression equation; the threshold was selected based on the optimal ROC point (e.g., Youden’s J). Predicted probabilities were computed using the logistic regression equation reported in the Statistical analysis section (β0 and β coefficients)

Discussion

This study evaluated the predictive value of first-trimester serum lipid parameters specifically, the triglyceride-glucose (TyG) index, TG/HDL-C ratio, and Lipid-IR index for GDM. Our findings demonstrate that these indices are independent risk factors for GDM and that their combined use substantially improves diagnostic performance.

In the present study, maternal age and pregestational BMI were significantly higher among women with GDM, reinforcing the central role of baseline metabolic risk in GDM pathogenesis. This finding is consistent with a landmark systematic review and meta-analysis demonstrating a strong, graded association between pre-pregnancy BMI and GDM: compared with normal-weight women, the odds of GDM were nearly doubled in overweight women and increased approximately threefold to fivefold in obese categories [9]. This reinforces the pivotal role of age and BMI in the pathogenesis of GDM. Together, these data support the clinical importance of early risk profiling based on maternal age and BMI.

We also observed higher neonatal birth weight, lower 5-minute Apgar scores, and a higher cesarean section rate in the GDM group, consistent with the adverse impact of intrauterine hyperglycemia and early dysmetabolic milieu on perinatal outcomes. The 2024 ADA Standards of Medical Care in Diabetes emphasize that maternal hyperglycemia is associated with increased risks of macrosomia and obstetric intervention [4]. The dyslipidemia observed in GDM characterized by elevated TG and reduced HDL-C has likewise been associated with macrosomia and perinatal morbidity in large-scale reviews and meta-analyses [10, 11]. Our results suggest that early metabolic indices not only predict GDM risk but may also have clinical relevance for neonatal outcomes.

Consistent with prior research, triglyceride levels were significantly higher and HDL-C levels lower in the GDM group (TG: 234.6 ± 98.5 mg/dL vs. 182.4 ± 75.3 mg/dL, p = 0.004; HDL-C: 58.3 ± 12.7 mg/dL vs. 68.5 ± 14.2 mg/dL, p < 0.001). A large meta-analysis by Hu et al. involving 97,880 women from 292 studies found that TG levels were approximately 20% higher across all trimesters in GDM pregnancies [10]. Similarly, Rahnemaei et al. analyzed 33 studies (n = 23,792) and reported significant increases in both total cholesterol and TG levels among women with GDM (SMD = 0.23 mg/dL and SMD = 1.14 mg/dL, respectively) [11]. Ryckman et al., in a meta-analysis of 60 studies, also demonstrated consistently higher TG and lower HDL-C levels across all trimesters in GDM cases [12]. Collectively, these findings underscore the critical role of early lipid regulation in glycemic control during pregnancy.

In our study, ALT, AST, and GGT levels were significantly elevated in the GDM group, suggesting hepatic insulin resistance as part of the metabolic response to pregnancy. This observation aligns with previous studies linking hepatic enzyme elevation with lipid-based insulin resistance markers, including TG/HDL-C and Lipid-IR [8, 10]. Mild hepatic enzyme elevation in early pregnancy may thus serve as an early indicator of later glycemic dysregulation.

The TG/HDL-C ratio was also independently associated with GDM in our analyses (AUC = 0.79), consistent with studies identifying TG/HDL-C as a practical lipid-based marker of insulin resistance during pregnancy [5]. Notably, in a large cohort evaluating lipid ratios from early to mid-pregnancy, higher TG/HDL-C levels (evaluated in quantiles rather than a single universal cut-off) were associated with increased odds of GDM, with the highest category showing roughly a twofold elevation in risk compared with the lowest category [13]. These findings align with our results and support TG/HDL-C as a useful early screening component when interpreted within population-appropriate thresholds [5, 13].

Higher insulin concentrations in the GDM group further support the presence of early insulin resistance reflected by TyG, TG/HDL-C, and Lipid-IR. The TyG index, in particular, serves as a practical and reliable surrogate marker of insulin resistance in early pregnancy [5, 8, 11]. From a pathophysiological perspective, our findings are plausible. Elevated triglycerides in early pregnancy may reflect increased hepatic VLDL production and impaired triglyceride clearance—metabolic features that frequently accompany insulin resistance [3, 4]. Simultaneously, lower HDL-C may indicate altered lipoprotein remodeling and a more atherogenic lipid profile, which has also been associated with insulin-resistant states [5, 6]. When fasting glucose is integrated with triglycerides in the TyG index, the resulting composite metric may more sensitively capture early hepatic insulin resistance than either parameter alone, explaining its superior discrimination in our cohort [7, 8].

When all three indices were combined, the model achieved an AUC of 0.92 with 89% sensitivity and 85% specificity, outperforming each parameter individually. Although no prior meta-analysis has assessed this exact combination, Hu et al. [10] suggested that simultaneous evaluation of multiple lipid-based markers may provide additive predictive benefit. The superior performance of the combined model likely reflects the complementary metabolic information captured by these indices, supporting the rationale for multivariable lipid-based screening models. Although the combined model uses three indices, it does not necessarily increase clinician workload. The inputs (fasting TG, HDL-C, total cholesterol, and fasting glucose) are routinely reported by laboratories, and the indices and predicted probability can be automatically calculated in a spreadsheet, calculator, or electronic medical record. Therefore, the combined model can be operationalized as a single probability output (p) with a predefined threshold, rather than requiring manual computation at the point of care.

From a clinical perspective, these indices can be automatically calculated from routine first-trimester biochemistry panels to provide pre-OGTT risk stratification. The 2024 ADA guidelines recommend early evaluation in high-risk pregnancies [4]; our findings suggest that incorporating lipid–glucose–derived indices into this framework could enhance early discrimination and enable timely lifestyle interventions, ultimately improving maternal and neonatal outcomes [4, 10].

Importantly, these indices should be interpreted as early risk-stratification tools rather than diagnostic tests. A higher first-trimester index may ethically justify only low-risk, guideline-concordant actions such as intensified lifestyle counseling, closer gestational weight-gain monitoring, and consideration of earlier confirmatory testing or glucose monitoring, rather than pharmacologic treatment based solely on an index.

Model performance may be further improved by integrating lipid–glucose indices with readily available clinical predictors (e.g., maternal age, pre-pregnancy BMI, family history of diabetes, prior GDM, or prior macrosomia). In addition, alternative surrogates that incorporate anthropometric measures—such as TyG-BMI or related composite scores—may capture both biochemical and phenotypic insulin resistance and could be compared head-to-head with TyG, TG/HDL-C, and Lipid-IR. However, insulin-based indices (e.g., HOMA-IR or QUICKI) would require fasting insulin measurements, limiting feasibility in routine care.

Conclusion

The present study demonstrates that simple indices derived from routinely measured first-trimester lipid and glucose parameters-particularly the TyG index, TG/HDL-C ratio, and Lipid-IR are effective tools for early prediction of GDM. Incorporating these indices into early screening protocols may enable risk stratification before OGTT and facilitate timely lifestyle modification, potentially reducing rates of neonatal macrosomia and obstetric complications. Future multicenter, prospective studies are warranted to validate these findings across diverse populations and to standardize optimal cut-off thresholds.

Study limitations

This study has several limitations. First, its retrospective single-center design may limit generalizability and introduces potential selection bias. Second, the study cohort is enriched for GDM (i.e., the case proportion is substantially higher than the population prevalence), which can affect the calibration of predicted probabilities and the transferability of a single operating threshold (e.g., p ≥ 0.30) to other settings; therefore, recalibration may be required before use in different populations. Third, although we excluded major confounders such as PCOS and clearly defined controls using the institutional two-step screening approach, residual confounding and outcome misclassification remain possible (e.g., unmeasured family history of diabetes, previous GDM, gestational weight gain, dietary patterns, socioeconomic factors, and other metabolic conditions). Fourth, we did not perform external validation, and the apparent model performance may be optimistic without resampling-based internal validation; thus, independent validation in prospective multicenter cohorts is warranted. Fifth, listwise deletion was used for missing data, which may introduce bias if data were not missing completely at random. Finally, because the indices share overlapping biochemical components, multicollinearity and reduced interpretability are possible despite VIF assessment. These limitations should be considered when interpreting the findings and planning implementation studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Mücahit Furkan Balcı: conceptualization, data acquisition, manuscript drafting. Celal Akdemir: data collection, critical review, approval of final draft. Fatih Yıldırım: data acquisition. İbrahim Karaca: statistical analysis, manuscript editing. Suna Yıldırım Karaca: writing and statistical analysis support. All authors reviewed and approved the final version of the manuscript.

Funding

This study received no external financial support.

Data availability

All data generated or analyzed in this study are included in this article. Further inquiries can be directed to the corresponding author, Dr. Mücahit Furkan Balcı.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Health Sciences University, Izmir Tepecik Training and Research Hospital (Approval date: 21 Aug 2025; Ref. No. 05/28). Given the retrospective design and the use of anonymized data, the requirement for individual informed consent was waived.

Use of artificial intelligence tools

ChatGPT (OpenAI, San Francisco, CA, USA) was used to assist with English language editing and phrasing of the manuscript. The authors carefully reviewed and revised all outputs and take full responsibility for the content of the final version. No generative AI tool was used for data analysis or interpretation.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.