Elevated triglyceride-glucose index predicts poor third-trimester glycemic control among gestational diabetes mellitus women: A prospective study in Shanghai

Abstract

BACKGROUND

The triglyceride-glucose (TyG) index is widely recognized as an effective marker for assessing insulin resistance (IR). Previous studies have predominantly utilized TyG index to predict the risk of developing gestational diabetes mellitus (GDM) in pregnant women.

AIM

To explore the association between TyG index and glycemic control in late pregnancy based on the group-based trajectory model (GBTM) among GDM women. To investigate the value of TyG index for predicting the effect of glycemic control among GDM women during late gestational period.

METHODS

GDM women were classified into good and bad plasma glucose control (PGC) groups by applying GBTM based on a series of abnormal plasma glucose (PG) prevalence obtained at each prenatal checkup. TyG index was calculated as ln[fasting triglyceride (mg/dL) × fasting PG (mg/dL)/2]. Logistic regression analysis was used to explore the association between TyG index and the prevalence of good PGC among GDM women. A P-value less than 0.05 was viewed as statistical significance.

RESULTS

Total 1118 GDM women were classified as good PGC (609, 54.5%) and bad PGC group (509, 45.5%). Logistic regression analysis indicated that the good PGC prevalence was negatively associated with the elevated TyG index [odds ratio (OR) = 0.73; 95%CI: 0.67-0.78], even after adjusting for potential confounders (OR = 0.76, 95%CI: 0.70-0.82). GDM women in higher TyG index categories had a significantly greater risk of poor PGC compared to those in the lowest, regardless of tertile or quartile classification of TyG index. Receiver operating characteristic (ROC) curve demonstrated that TyG index could be an independent predictor for glycemic control in late pregnancy [area under curve (AUC) = 0.74, 95%CI: 0.71-0.77], with a TyG index ≥ 9.3 indicating subsequently higher rates of PG abnormalities.

CONCLUSION

The elevated TyG index was negatively associated with good PGC in late pregnancy among GDM women. TyG index less than 9.3 could predict the good glycemic control in third trimester.

Key Words: Gestational diabetes mellitus; Triglyceride-glucose index; Group-based trajectory model; Abnormal plasma glucose prevalence; Predict

Core Tip: The elevated triglyceride-glucose (TyG) index was negatively associated with good plasma glucose control in late pregnancy among women with gestational diabetes mellitus (GDM). TyG index less than 9.3 could predict good plasma glucose control in third trimester, particularly among GDM women with older age (≥ 35 years), and overweight/obesity [body mass index (BMI) ≥ 24 kg/m²]. In the future, measuring fasting plasma glucose and triglyceride levels in pregnant women with GDM at 24-28 gestational weeks and calculating TyG index, then this index could serve as an effective tool for predicting glycemic control status in late pregnancy among GDM women.

INTRODUCTION

As a special form of diabetes, gestational diabetes mellitus (GDM) traditionally refers to abnormal glucose tolerance with onset or first recognition during pregnancy, contributing to diverse kinds of adverse pregnancy outcomes that impairs both maternal and fetal health worldwide[1-4]. The number of GDM cases is increasing globally, even having a 2-3 fold increase across multiple countries over the recent two decades. Therefore, a predictive indicator for gestational glycemic control is crucial for GDM women management.

Currently, when GDM is typically diagnosed at 24-28 gestational weeks, predictive indicators for the evaluation of glycemic control in late pregnancy are still limited. The triglyceride-glucose (TyG) index combines fasting plasma glucose (FPG) and triglyceride (TG) levels, reflecting the interaction between lipid and glucose metabolism[5]. TyG index was first applied to explore its relationship with cardiovascular and cerebrovascular diseases. Several large-scale studies in recent 5 years have indicated that TyG index exhibits better predictive accuracy in GDM or adverse pregnancy outcomes[6-8]. Furthermore, insulin resistance (IR) is crucial for the development of GDM, and TyG index is proposed to be an useful biomarker for IR identification[9,10]. Therefore, TyG index could be an effective indicator for monitoring glycemic control in late pregnancy. However, evidence regarding the association between TyG index and gestational glycemic control is still lacking.

In previous studies, the percentage of abnormal plasma glucose (PG) was calculated as the number of abnormal PG tests divided by the total number of PG tests during late pregnancy, and a percentage of abnormal PG < 20% was defined as good glycemic control among GDM women[11]. However, women with GDM are required to complete serial antenatal checkups before delivery, a single point could not completely depict the dynamic PG changes or glycemic control status in their subsequent antenatal checkups and might induce some bias. Therefore, a novel group-based trajectory model (GBTM) aiming to depict data with repeated measurements and analyze trend variation across different time points should be considered[12]. GBTM employs clustering analysis to identify potential subgroups that exhibit similar temporal trajectory patterns in observed indicators. This approach is especially useful for examining populations with substantial heterogeneity[13,14]. Our previous work indicated that GBTM could be used to classify GDM women into good plasma glucose control (PGC) and bad PGC groups based on a series of abnormal PG prevalence values calculated at each prenatal checkup[15]. These rates can reflect the dynamic profile of glucose metabolism during pregnancy, rather than capturing only instantaneous glycemic status at a single time point. A higher rate indicated bad PGC across gestation. In this study, we aim to identify these two PGC groups and further investigate the association between the increased TyG index level and bad PGC risk in late pregnancy.

MATERIALS AND METHODS

Study design and participants

In this study, we established a prospective cohort of women with GDM at Songjiang Maternal and Children’s Health-Care Hospital located in Shanghai since August 2019. Specific information regarding the cohort and sample size calculation are available in our previously published studies[11]. In this study, GDM was verified according to a diagnostic 75 g oral glucose tolerance test recommended by the International Association of Diabetes and Pregnancy Study Group at gestational weeks 24-28[16], which was in line with the recommended diagnostic criteria formulated by China in 2011[17]. Pregnant women without a previous history of diabetes before pregnancy were eligible for a clinical diagnosis of GDM when any one of the values reached or exceeded the above threshold: FPG level, 1-hPG level, and 2-hPG level should be lower than 5.1 mmol/L, 10.0 mmol/L, and 8.5 mmol/L[18]. We initially recruited 1162 singleton GDM women aged 18-45 years who completed their first prenatal examination in the hospital. The detailed inclusion criteria were as follows: (1) Without diabetes before pregnancy; (2) Singleton pregnancy; (3) Residence in Shanghai without migration plan in two years; (4) Being able to read and sign the informed consent form; and (5) Personal prenatal examination records are complete, especially for the information on the calculated index, such as FPG and TG. GDM women were excluded if they met any of the following criteria: (1) Receipt of oral hypoglycemic agents or insulin for glycemic regulation during pregnancy; (2) Long-term use of medications interfering with glucose and lipid metabolism; and (3) Physical exercise (PE) restrictions and psychological disorders. A total of 1118 GDM women were finally enrolled. The Institution Review Board of Songjiang Maternal and Children’s Health-Care Hospital reviewed and approved this study (No. 2019-12-003), and each GDM women provided the written informed consent before the questionnaire interview. This study was performed in line with the Declaration of Helsinki and registered in the Chinese Clinical Trial Registry (ChiCTR2000028832).

Data collection

A questionnaire was designed and conducted to collect information from all GDM women, which contained four parts. Part A covered the baseline demographic features (maternal age, education, employment, residency status, and monthly income). Part B concerned distinct information related to obstetric and childbirth history, height and weight before pregnancy, the blood tests covered lipid profiles and glucose metabolism indices, including TG index measured at 24-28 weeks of pregnancy and PG test records registered from the initial prenatal examination until the completion of delivery. Part C covered the perinatal outcomes (gestational weeks, delivery pattern, neonatal characteristics (sex and birth weight), adverse pregnancy complications (preterm birth, low birth weight, macrosomia, postpartum hemorrhage, polyhydramnios, etc.). Part D documented 20 types of PE (walking, house cleaning, stationary riding, jogging, yoga, oxygen dance, etc.) in late pregnancy, recording the monthly frequency and duration of each activity on average at gestational weeks 27-40[19]. Data in Parts A, B, and C were directly extracted from the hospital information system, and data in Part D was obtained through face-to-face interview conducted by trained investigators after the delivery of GDM women.

In this study, a standardized procedure was performed in each subsequent survey since the initial antenatal care visit. All records containing missing values were systematically flagged and subsequently cross-verified by two independent researchers based on the original paper documents, these incomplete values were then handled through the multiple imputation method.

Definition and index calculation

After the confirmation of GDM, subsequent antenatal checkups were conducted in gestational weeks 28-30, 30-32, 32-34, 34-36, 36-37, 37-38, and 38-39. The abnormal PG was defined as FPG ≥ 5.1 mmol/L after an overnight (at least 10 hours) fast or 2-hPG ≥ 8.5 mmol/L following consumption of a breakfast meal. Glycemic control status in third trimester was indicated by a series of longitudinal abnormal PG prevalence calculated at each prenatal checkup after GDM confirmation. The calculation was defined as the cumulative number of abnormal PG during the first and current prenatal checkups divided by the actual number of prenatal checkups (model A) or the total number of prenatal checkups completed up to the current time point (model B). For example, a GDM woman completed a total of 8 prenatal visits after her GDM diagnosis. If abnormal PG measurements were documented in 3 out of her first 5 prenatal checkups, the abnormal PG prevalence at the 5^th prenatal checkup was 37.5% in model A (3/8) and 60% in model B (3/5). A lower percentage or a decreasing trend of abnormal PG levels reflected better glycemic control in late pregnancy.

In this study, we applied GBTM to identify individuals who exhibited varying developmental trajectories of abnormal PG prevalence over the same temporal dimension, then classified them into good and bad PGC groups, and further examined the associated influencing factors. Detailed information on the overall trajectory curves was depicted in Figure 1. Adherence trajectories were computed using the hlme package in R for GBTM. The plot function within the hlme package was used to generate primitive trajectory curves. Subsequently, the ggplot function from the ggplot2 package was adopted to produce high-quality trajectory figures. The optimal number of trajectory groups and polynomial orders were determined through an iterative procedure. Firstly, a maximum polynomial order of 2 was specified for each trajectory group, and the model assumed that there was only one trajectory group to be fitted, followed by gradually increasing the number of groups (considering 1-3 classes). The appropriate number was ultimately certified according to the following objective indices: (1) Entropy > 0.8; and (2) Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC) reach their minimum values (Supplementary Table 1). In addition, subjective professional judgment was also incorporated for final determination. In GBTM, AIC and BIC tend to decline gradually as the number of groups increases, since additional groups improve model fit and better capture population heterogeneity. We retained the 2-group classification to enable direct comparison between good PGC and bad PGC groups and to facilitate clear interpretation when exploring factors associated with glycemic control. Secondly, the polynomial orders for each trajectory group were adjusted (sequentially fitting linear, quadratic, and cubic terms), and the optimal order was determined when all parameters reached a significant level. For the 2-group models, model 1-2, model 2-2 and model 2-3 all achieved excellent discrimination (P < 0.05), while cubic polynomial model could identify complex nonlinear changes in abnormal PG prevalence (Supplementary Table 2). Furthermore, we applied average posterior probability (AvePP) > 0.7 and odds of correct classification (OCC) > 5 to evaluate the model fit quality, and the proportion of GDM women in each trajectory group should not be less than 5% of the total number of GDM women (Supplementary Table 3). The implications of each index are available in our previously published research[15]. Finally, a trajectory model with two groups and cubic polynomial orders was recognized as the best-fitted solution.

Figure 1 Trajectories of abnormal plasma glucose rate at each preset time-point from prenatal examination after the diagnosis of gestational diabetes mellitus. PGC: Plasma glucose control; pred: Prediction; obs: Observation; CI: Confidence interval; GDM: Gestational diabetes mellitus.

Biochemical parameters, such as FPG, 1-hPG, 2-hPG, TG, were determined by using standard laboratory techniques. The TyG index was calculated according to the following formula: TyG = ln[fasting TG (mg/dL) × FPG (mg/dL)/2], and was further divided into three or four equal groups: The 1^st, 2^nd and 3^rd tertiles were defined as T1, T2 and T3, while the 1^st, 2^nd, 3^rd and 4^th quartiles were regarded as Q1, Q2, Q3 and Q4. In our study, the original clinical FPG data extracted from the hospital information system were documented in mmol/L. All FPG measurements were converted to mg/dL before TyG calculation with the conversion factor of 1 mmol/L = 18 mg/dL. Pre-pregnancy body mass index (BMI) was calculated as weight (kg)/height (m)². According to the standard of China Obesity Working Group[20], BMI < 24 kg/m² refers to low or normal weight and BMI ≥ 24 kg/m² is considered as overweight or obesity. The age of women with GDM was categorized into < 35 years and > 35 years. Education level was quantified according to the total completed years of formal schooling and was classified into 6-9 years (primary/junior high), 10-12 years (senior high), and > 12 years (college and above). The amount of total daily PE time was obtained through face-to-face interview and was classified into “< 30 minute/day“, “30-59 minute/day“, “60-89 minute/day“, “90-119 minute/day“, and “≥ 120 minute/day“.

Statistical analysis

In this study, R 4.3.2 and SPSS 23.0 software were applied for data analysis. Normally distributed quantitative variables were reported as mean and SD, and Student’s t test was applied to compare the difference between groups. Quantitative variables that exhibited a skewed distribution were expressed as median and interquartile range (IQR), and group comparisons were assessed by using non-parametric rank sum test. Qualitative variables were presented as n (%), and compared using the χ² test or Fisher’s exact test.

Univariate and multivariate logistic regression models were applied to calculate the odds ratio (OR) and 95%CI to explore factors associated with good PGC, as well as the correlation between TyG index and good PGC among women with GDM classified based on GBTM. The specific variables definition and coding schemes for factors in logistic regression models were depicted in Supplementary Table 4. In this study, subgroup analyses were conducted with the adjustment of age, pre-pregnancy BMI and daily PE time. Additionally, receiver operating characteristic (ROC) curves were applied to assess the predictive performance of TyG index combining with three covariates for gestational PGC in their late pregnancy among GDM women. In this study, a P- value less than 0.05 (two-tailed) was considered as statistically significant.

RESULTS

In this study, 1118 GDM women included 358 local residents (32.0%), with an average age of 30.5 years (SD = 4.3), and the percentage of GDM women with advanced maternal age (≥ 35 years) was 17.4%. 60.5% of them had an educational level of college and above, and the majority of GDM women were employed (1041, 93.1%), and more than three-quarters had a monthly income over 5000 RMB. Moreover, the overall median BMI was 22.7 kg/m² (IQR: 20.6-25.0), and 35.1% of women with GDM were overweight or obese (BMI ≥ 24 kg/m²) before pregnancy, while only less than half of them adhered to dietary guidelines recommended by doctors. Most GDM women (586, 52.4%) were multiparous, and the average gestational age at delivery was 38.5 weeks (SD = 1.2), with a low preterm birth rate of 5.1%. The median daily PE time was 65 minutes (IQR: 45-90), with 437 (39.1%) GDM women reporting less than 60 minute/day and 306 (27.4%) reporting more than 90 minute/day (Table 1).

Table 1 Baseline characteristics of women with gestational diabetes mellitus classified by the trajectory analysis (n = 1118), n (%).

Variables	Total GDM women (n = 1118)	Good PGC group (n = 609)	Bad PGC group (n = 509)	Statistic	P value
Age (years), mean (SD)	30.5 (4.3)	30.3 (4.1)	30.7 (4.6)	1.23	0.22
Age group (years)
< 35	924 (82.6)	514 (55.6)	410 (44.4)	2.87	0.09
≥ 35	194 (17.4)	95 (49.0)	99 (51.0)
Education
Primary/Junior high	231 (20.7)	106 (45.9)	125 (54.1)	19.84	< 0.01
Senior high	210 (18.8)	98 (46.7)	112 (53.3)
College and above	677 (60.5)	405 (59.8)	272 (40.2)
Monthly income (RMB)
≤ 5000	261 (23.3)	120 (46.0)	141 (54.0)	11.38	< 0.01
5001-10000	584 (52.2)	325 (55.7)	259 (44.3)
> 10000	273 (24.4)	164 (60.1)	109 (39.9)
Residency status
Local resident	358 (32.0)	200 (55.9)	158 (44.1)	0.41	0.52
Non-local resident	760 (68.0)	409 (53.8)	351 (46.2)
Balanced diet with suggestion
Yes	455 (40.7)	254 (55.8)	201 (44.2)	0.57	0.45
No	663 (59.3)	355 (53.5)	308 (46.5)
Pre-pregnancy BMI (kg/m2)
< 24 (low/normal weight)	726 (64.9)	440 (60.6)	286 (39.4)	31.41	< 0.01
≥ 24 (overweight/obesity)	392 (35.1)	169 (43.1)	223 (56.9)
Preterm delivery
Yes	57 (5.1)	30 (52.6)	27 (47.4)	0.08	0.78
No	1061 (94.9)	579 (54.6)	482 (45.4)
Pregnant status
Primipara	532 (47.6)	296 (55.6)	236 (44.4)	0.56	0.46
Multipara	586 (52.4)	313 (53.4)	273 (46.6)
Daily PE time (minute/day), median (IQR)	65 (45-90)	70 (48-90)	65 (45-85)	-2.68	< 0.01
Daily PE time group (minute/day)
< 30	52 (4.7)	25 (48.1)	27 (51.9)	13.68	< 0.01
30-59	385 (34.4)	196 (50.9)	189 (49.1)
60-89	375 (33.5)	204 (54.4)	171 (45.6)
90-119	220 (19.7)	122 (55.5)	98 (44.5)
≥ 120	86 (7.7)	62 (72.1)	24 (27.9)
Lab test at 24-28 gestational weeks
FPG (mmol/L), mean (SD)	5.1 (0.4)	4.9 (0.4)	5.3 (0.4)	11.86	< 0.01
1-hPG (mmol/L), mean (SD)	9.1 (1.9)	9.2 (1.9)	9.1 (2.0)	-1.64	0.10
2-hPG (mmol/L), mean (SD)	7.5 (1.5)	7.5 (1.5)	7.5 (1.5)	-0.76	0.45
TG (mmol/L), mean (SD)	2.8 (1.1)	2.6 (1.0)	2.9 (1.2)	5.44	< 0.01
CHOL (mmol/L), mean (SD)	5.9 (1.1)	6.1 (1.1)	5.8 (1.1)	-4.55	< 0.01

GDM: Gestational diabetes mellitus; PGC: Plasma glucose control; BMI: Body mass index; PE: Physical exercise; IQR: Interquartile range; FPG: Fasting plasma glucose; PG: Plasma glucose; TG: Triglyceride; CHOL: Cholesterol.

Classification of GDM women into distinct trajectory groups based on GBTM

Based on GBTM, 1118 GDM women were classified into good PGC group (n = 609, 54.5%) and bad PGC group (n = 509, 45.5%). GDM women with good PGC had higher proportions of college or above education (59.8%), monthly income > 5000 RMB (80.3%) and pre-pregnancy BMI < 24 kg/m² (60.6%), all showing statistically significant differences (P < 0.05). The median daily PE time among GDM women in good PGC group (70 minutes) was significantly higher than that in bad PGC group (65 minutes) (P < 0.05) and the good PGC prevalence was higher in GDM women with daily PE time of (30-59) minutes (50.9%), (60-89) minutes (54.4%), (90-119) minutes (55.5%), and ≥ 120 minutes (72.1%) (P < 0.05). Moreover, GDM women with good PGC had lower level of FPG and TG, but had slightly higher cholesterol level (P < 0.05; Table 1).

Association between TyG index and PGC

Figure 2 indicated that an augmented TyG index had a positive curvilinear association with the elevated abnormal PG prevalence (y = -3.92x² + 90.4x - 451, P < 0.05), which demonstrated that GDM women with lower TyG index level would have a declined risk of bad PGC in their late pregnancy.

Figure 2 The association between triglyceride-glucose index and the abnormal plasma glucose rate among gestational diabetes mellitus women, with the analysis of binomial regression. TyG: Triglyceride-glucose.

In this study, the mean value of TyG index was 9.3 for all GDM women, with 9.2 in good PGC group and 9.4 in bad PGC group. GDM women were then stratified into TyG index tertiles (T1, T2, T3) and quartiles (Q1, Q2, Q3, Q4), respectively. The proportion of GDM women classified into good PGC group decreased dramatically with an ascending TyG index categories, regardless of the tertile (69.2% in T1, 51.9% in T2, and 42.4% in T3) classification or quartile (71.0% in Q1, 58.6% in Q2, 47.0% in Q3 and 41.4% in Q4) classification (P < 0.05; Table 2).

Table 2 Triglyceride-glucose index in women with gestational diabetes mellitus classified by the trajectory analysis (n = 1118), n (%).

Variables	Total GDM women (n = 1118)	Good PGC group (n = 609)	Bad PGC group (n = 509)	Statistic	P value
TyG index, mean (SD)	9.3 (0.4)	9.2 (0.4)	9.4 (0.4)	8.15	< 0.01
TyG index, tertile
T1 (8.19-9.09]	373 (33.4)	258 (69.2)	115 (30.8)	55.56	< 0.01
T2 (9.09-9.39]	372 (33.3)	193 (51.9)	179 (48.1)
T3 (9.39-10.68]	373 (33.4)	158 (42.4)	215 (57.6)
TyG index, quartile
Q1 (8.19-8.99]	279 (25.0)	198 (71.0)	81 (29.0)	58.08	< 0.01
Q2 (8.99-9.23]	280 (25.0)	164 (58.6)	116 (41.4)
Q3 (9.23-9.47]	279 (25.0)	131 (47.0)	148 (53.0)
Q4 (9.47-10.68]	280 (25.0)	116 (41.4)	164 (58.6)

GDM: Gestational diabetes mellitus; PGC: Plasma glucose control; TyG: Triglyceride-glucose.

Logistic regression indicated that daily PE time over 120 minutes (OR = 1.25, 95%CI: 1.06-1.47) and college and above educational level (OR = 1.11, 95%CI: 1.03-1.19) were positively correlated with the prevalence of good PGC among GDM women, but pre-pregnancy BMI ≥ 24 kg/m² (OR = 0.88, 95%CI: 0.83-0.94) and TyG index were negatively associated with the good PGC prevalence (OR = 0.76, 95%CI: 0.70-0.82; Table 3).

Table 3 Factors associated with the rate of good plasma glucose control among women with gestational diabetes mellitus (n = 1118).

Variables	LRa1		LRb2		LRc3
	OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value
Daily PE time group (minute/day)
< 30	1.00		1.00		1.00
30-59	1.03 (0.89-1.19)	0.70	1.02 (0.88-1.17)	0.83	1.02 (0.89-1.18)	0.74
60-89	1.07 (0.92-1.23)	0.39	1.04 (0.90-1.19)	0.59	1.05 (0.91-1.20)	0.51
90-119	1.08 (0.93-1.25)	0.33	1.05 (0.91-1.22)	0.48	1.06 (0.92-1.23)	0.42
≥ 120	1.27 (1.07-1.51)	< 0.01	1.25 (1.06-1.48)	0.01	1.25 (1.06-1.47)	0.01
Age group (years)
< 35	1.00		1.00
≥ 35	0.94 (0.87-1.01)	0.09	0.94 (0.86-1.01)	0.10
Education
Primary/Junior high	1.00		1.00		1.00
Senior high	1.01 (0.92-1.11)	0.87	0.98 (0.90-1.08)	0.73	0.99 (0.90-1.08)	0.79
College and above	1.15 (1.07-1.24)	< 0.01	1.10 (1.00-1.21)	0.04	1.11 (1.03-1.19)	< 0.01
Monthly income (RMB)
≤ 5000	1.00		1.00
5001-10000	1.10 (1.02-1.18)	0.01	1.04 (0.96-1.13)	0.36
> 10000	1.15 (1.06-1.25)	< 0.01	1.05 (0.94-1.16)	0.39
Residency status
Local resident	1.00		1.00
Non-local resident	0.98 (0.92-1.04)	0.52	1.01 (0.95-1.08)	0.78
Pre-pregnancy BMI (kg/m2)
< 24 (low/normal weight)	1.00		1.00		1.00
≥ 24 (overweight/obesity)	0.84 (0.79-0.89)	< 0.01	0.89 (0.83-0.94)	< 0.01	0.88 (0.83-0.94)	< 0.01
Balanced diet with suggestion
Yes	1.00		1.00
No	0.98 (0.92-1.04)	0.45	0.99 (0.94-1.06)	0.87
Pregnant status
Primipara	1.00		1.00
Multipara	0.98 (0.92-1.04)	0.46	1.03 (0.97-1.09)	0.37
TyG index	0.73 (0.67-0.78)	< 0.01	0.76 (0.70-0.82)	< 0.01	0.76 (0.70-0.82)	< 0.01

¹LRa indicates univariate logistic regression model to explore the influence of variables on the rate of good plasma glucose control among women with gestational diabetes mellitus (GDM).

²LRb indicates multivariate logistic regression model to explore the influence of variables on the rate of good plasma glucose control among GDM women with the inclusion of all variables.

³LRc indicates multivariate logistic regression model adjusted for the influence of daily physical exercise time, education, pre-pregnancy body mass index, and triglyceride-glucose index on the rate of good plasma glucose control among GDM women.

OR: Odds ratio; PE: Physical exercise; BMI: Body mass index; TyG: Triglyceride-glucose.

In this study, TyG index was then classified into (T1, T2, T3) by tertile value and (Q1, Q2, Q3, Q4) by quartile value, respectively. In comparison with GDM women with the lowestr tertile of TyG index (T1), those with higher tertiles had lower prevalence of good PGC, the OR was 0.85 (95%CI: 0.80-0.91) for T2 and 0.79 (95%CI: 0.74-0.85) for T3, after adjusting for potential confounding factors. Similarly, GDM women with higher quartiles of TyG index also had lower prevalence of good PGC, the OR was 0.90 (95%CI: 0.83-0.97) for Q2, 0.81 (95%CI: 0.74-0.87) for Q3 and 0.78 (95%CI: 0.72-0.84) for Q4, after the adjustment for potential confounding factors (Figure 3).

Figure 3 The association between the tertiles and quartiles of triglyceride-glucose index and good plasma glucose control among women with gestational diabetes mellitus based on the univariate and multivariate logistic regression models. Multivariate logistic regression model adjusted for daily physical exercise time, education, and pre-pregnancy body mass index to access the rate of good plasma glucose control among women with gestational diabetes mellitus. OR: Odds ratio; TyG: Triglyceride-glucose.

Subgroup analyses

Subgroup analyses were implemented to explore the association between TyG index and the good PGC prevalence in women with GDM, by age (< 35 years, ≥ 35 years), pre-pregnancy BMI (< 24 kg/m², ≥ 24 kg/m²), and daily PE time (< 120 minute/day, ≥ 120 minute/day). Logistic regression indicated that GDM women with higher tertiles of TyG index (T2, T3 in tertile classification, and Q2, Q3, Q4 in quartile classification) had lower prevalence of good PGC among those with age < 35 years, pre-pregnancy BMI < 24 kg/m² and daily PE time < 120 minute/day. However, a statistically significant lower prevalence of good PGC was only identified among GDM women with higher tertiles of T3, Q3 and Q4 for age ≥ 35 years, T2, T3, Q3 and Q4 for pre-pregnancy BMI ≥ 24 kg/m² and T3, Q4 for daily PE time ≥ 120 minute/day. In summary, an elevated TyG index maintained significant association with lower prevalence of good glycemic control across almost all prespecified subgroups (Figures 4, 5, and 6).

Figure 4 The influence of age for the association between the tertiles and quartiles of triglyceride-glucose index and good plasma glucose control among women with gestational diabetes mellitus based on the univariate and multivariate logistic regression models. A: Subgroup with age < 35 years; B: Subgroup with age ≥ 35 years. Multivariate logistic regression model adjusted for daily physical exercise time, education, and pre-pregnancy body mass index to access the rate of good plasma glucose control among women with gestational diabetes mellitus. OR: Odds ratio; TyG: Triglyceride-glucose.

Figure 5 The influence of pre-pregnancy body mass index for the association between the tertiles and quartiles of triglyceride-glucose index and good plasma glucose control among women with gestational diabetes mellitus based on the univariate and multivariate logistic regression models. A: Subgroup with pre-pregnancy body mass index < 24 kg/m²; B: Subgroup with pre-pregnancy body mass index ≥ 24 kg/m². Multivariate logistic regression model adjusted for daily physical exercise time and education to access the rate of good plasma glucose control among women with gestational diabetes mellitus. OR: Odds ratio; TyG: Triglyceride-glucose.

Figure 6 The influence of daily physical exercise time for the association between the tertiles and quartiles of triglyceride-glucose index and good plasma glucose control among women with gestational diabetes mellitus based on the univariate and multivariate logistic regression models. A: Subgroup with daily physical exercise time < 120 minute/day; B: Subgroup with daily physical exercise time ≥ 120 minute/day. Multivariate logistic regression model adjusted for pre-pregnancy body mass index and education to access the rate of good plasma glucose control among women with gestational diabetes mellitus. OR: Odds ratio; TyG: Triglyceride-glucose.

ROC curve analysis

In this study, pre-pregnancy BMI, daily PE time, education and TyG index showed significant associations with gestational glycemic control among GDM women identified in logistic regression models. ROC curve analysis was further employed to evaluate both the independent and combined predictive efficacy of TyG index and other three covariates for glycemic control in late pregnancy among GDM women. Values of area under curve (AUC) were compared in each seperated subgroup, which was 0.63 for pre-pregnancy BMI, 0.53 for daily PE time, and 0.55 for education. Table 4 indicated that TyG index demonstrated superior predictive performance compared to other indicators, with the AUC of 0.74 (95%CI: 0.71-0.77), sensitivity of 0.70 and specificity of 0.69. Additionally, the optimal cutoff point for TyG index was 9.3, which was identified as the threshold for determining high abnormal PG risk. Furthermore, the AUC values for TyG index combined with each of the three covariates individually all approach 0.75. ROC curves incorporated with all four indicators had exhibited better predictive strength. In short, TyG index demonstrated effectiveness in predicting glycemic control status among women with GDM during the third trimester of pregnancy, both individually and in combination with other indicators (Figure 7).

Figure 7 Receiver operating characteristic curves of four indicators for predicting plasma glucose control after the diagnosis of gestational diabetes mellitus. AUC: Area under curve; TyG: Triglyceride-glucose; BMI: Body mass index; PE: Physical exercise.

Table 4 Parameters related to the receiver operating characteristic curves of four indicators for predicting plasma glucose control after the diagnosis of gestational diabetes mellitus.

Variables	AUC (95%CI)	Sensitivity	Specificity	PPV	NPV	YI
TyG	0.74 (0.71-0.77)	0.70	0.69	0.73	0.66	0.39
Pre-pregnancy BMI	0.63 (0.59-0.66)	0.51	0.73	0.69	0.55	0.24
Daily PE time	0.53 (0.50-0.57)	0.19	0.86	0.63	0.47	0.06
Education	0.55 (0.52-0.58)	0.65	0.45	0.58	0.52	0.10
TyG + pre-pregnancy BMI	0.75 (0.72-0.78)	0.68	0.72	0.75	0.66	0.41
TyG + daily PE time	0.74 (0.71-0.77)	0.61	0.78	0.77	0.63	0.40
TyG + education	0.74 (0.71-0.77)	0.70	0.70	0.73	0.66	0.40
TyG + pre-pregnancy BMI + daily PE time	0.75 (0.72-0.78)	0.62	0.78	0.77	0.63	0.40
TyG + pre-pregnancy BMI + education	0.75 (0.72-0.78)	0.64	0.76	0.77	0.64	0.41
TyG + daily PE time + education	0.74 (0.71-0.77)	0.75	0.65	0.72	0.68	0.40
TyG + pre-pregnancy BMI + daily PE time + education	0.75 (0.72-0.78)	0.66	0.74	0.75	0.64	0.40

AUC: Area under curve; PPV: Positive predictive value; NPV: Negative predictivevalue; YI: Youden index; TyG: Triglyceride-glucose; BMI: Body mass index; PE: Physical exercise.

DISCUSSION

This study enrolled over 1000 women with GDM to evaluate the predictive value of TyG index measured at gestational weeks 24-28 for subsequent glycemic control in late pregnancy. To our knowledge, this study represents the first investigation about the correlation between TyG index and abnormal PG prevalence among women with GDM. Our findings revealed that an elevated TyG index was significantly associated with an increased risk of poor glycemic control. Moreover, this association remained consistent across subgroups, further confirming the robustness of the results. In terms of predictive performance, TyG index yielded a favorable AUC value. Based on these analyses, our study identified 9.3 as an optimal cutoff point, suggesting that pregnant GDM women with a TyG index exceeding this threshold may require prioritized clinical intervention to reduce the risk of adverse glycemic outcomes during late gestational period.

An interplay between defective pancreatic β cell function and IR during pregnancy is a critical underlying mechanism contributing to the onset and development of GDM[21,22], characterized by a progressive loss in insulin secretion from pancreatic β cells within the setting of IR, resulting in reduced efficiency of insulin in mediating glucose uptake and utilization. Several methods are available for IR evaluation, including homeostatic model assessment of IR (HOMA-IR), quantitative insulin sensitivity check index (QUICKI), sex hormone binding globulin (SHBG) and TyG index. The clinical application of HOMA-IR and QUICKI is limited by performance variability depending on maternal characteristics and testing time[23], coupled with lacking consensus on standardized cutoff values for diagnosing pathological IR currently.

TyG index, which is calculated based on two routinely measured biochemical parameters (TG and FPG), both of which can be easily obtained from standard antenatal examinations. Previous studies have demonstrated that TyG index serves as a practical and reliable screening tool for the early identification of women at elevated risk for GDM. For instance, a prospective cohort study conducted in Korea involving over 38 thousands primiparous women reported that a 1-unit increase in TyG index significantly increased the risk of GDM by 1.81 times, even after fully adjusting for risk factors[6]. Similarly, data from National Health and Nutrition Examination Survey revealed that TyG index correlated positively with GDM (OR = 3.43, 95%CI: 1.20-9.85)[7]. A prospective cohort study from the China Maternal Nutrition and Health Scientific Survey found that women in the group with the highest TyG index had a significantly higher risk of developing GDM compared to those in the lowest quintile group (OR = 3.25; 95%CI: 2.23-4.75)[24]. Additionally, a retrospective cohort analysis comprising 631 singleton pregnant women in Wenzhou found that TyG index exhibited an advanced diagnostic performance for predicting GDM, yielding an AUC of 0.67[25]. Since TyG index offers superior predictive capacity for metabolic dysfunction and reflects the glucolipid metabolic status at the time of GDM diagnosis, these findings suggest that it may also exhibit stronger predictive ability for glycemic control, which is consistent with the findings in this study.

Over recent decades, the prevalence of GDM has increased significantly among women with advanced maternal age and overweight/obesity[26,27]. Previous studies have identified various factors contributing to the rising trend of childbearing among older women, including “three-child” policy implementation, advances in assisted reproductive technology, and delayed marriage age[28]. Additionally, the observed prevalence appears to be propelled by a complex interaction of multiple determinants, including excessive consumption of processed food, sedentary behaviors and inadequate PE[29-31]. In this study, our analysis revealed that GDM women with pre-pregnancy BMI < 24 kg/m² and exercise time ≥ 120 minute/day would maintain optimal PGC. However, no significant association was observed between poor PGC and advanced maternal age, which might be attributed to the substantial disparity in sample size between age groups. Therefore, we stratified the analysis by age, pre-pregnancy BMI and daily PE time groups in order to accurately access potential effect modification. Subgroup analyses consistently reinforced the robustness of our primary conclusions. Therefore, early detection of women at high GDM risk by TyG index is crucial for them to sustain appropriate glycemic level during pregnancy.

The primary strength of this study lies in its use of real-world clinical data from a prospective cohort in Shanghai, which includes a large sample size of 1118 participants. The prospective design and robust sampling enhance the generalizability of our findings. Furthermore, recall bias is substantially reduced through the direct extraction of biochemical parameters – such as FPG, 1-hPG, 2-hPG, and TG – as well as maternal baseline characteristics from electronic hospital records. This process ensures high data integrity, with a missing rate below 5%. Additional strengths include the calculation of abnormal PG percentage, which offers an objective measure of glycemic control throughout gestation, and the application of GBTM to identify temporal trajectories of this metric across eight antenatal visits. Most importantly, this is the first study to report a positive association between TyG index – measured at 24-28 gestational weeks – and the risk of poor glycemic control in late pregnancy. These results support the potential use of TyG index as an effective predictor for identifying individuals at increased risk of suboptimal glycemic management.

This study has some limitations. First, the addition of all three covariates yielding merely an absolute AUC increment below 0.05 indicates limited incremental predictive value. This phenomenon may be attributed to the weak individual discriminative ability of these covariates, all exhibiting low Youden index values. Second, GDM women were enrolled only from one hospital, which ensured rigorous adherence and high internal validity, but limited the generalizability of our findings to different populations across diverse geographic and demographic settings. Therefore, our findings require further validation in multi-center prospective cohorts. Third, the detailed dietary patterns and some genetic factors (GDM family history, TyG index-related genetic polymorphisms, etc.) associated with PGC among GDM women were not collected and analyzed, indicating residual confounding may remain unaccounted for. Therefore, it is necessary to incorporate a broader range of factors in future studies. Fourth, the lack of fasting insulin measurements precluded a direct comparison between TyG index and other established indicators of IR. It would be valuable for future work to incorporate additional biomarkers to evaluate their synergistic predictive utility alongside the TyG index. So, further investigations are needed to address these limitations and validate our findings.

CONCLUSION

An elevated TyG index is associated with an increased risk of poor glycemic control during late pregnancy in women with GDM. As an easily accessible and cost-effective screening tool, TyG index may serve as a practical clinical biomarker to identify suboptimal glycemic regulation in this population. Specifically, women with a TyG index exceeding 9.3 should be considered for early and targeted interventions, particularly those with additional risk factors such as advanced maternal age, overweight or obesity, and insufficient physical activity.

ACKNOWLEDGEMENTS

We thank Professor Michael Engelau (Center for Disease Control and Prevention, United States) and Professor Klossi Chen (University of California, Los Angeles, United States) for providing suggestions and comments on this manuscript.