The association between triglyceride-glucose index and all-cause mortality in postmenopausal women: A cohort study from NHANES

Abstract

Elevated triglyceride–glucose (TyG) index has been linked to cardiovascular risk and metabolic disorders, but its relationship with all-cause mortality in postmenopausal women remains unclear. We aimed to investigate the association between baseline TyG index and all-cause mortality among postmenopausal women, and to determine whether a nonlinear (threshold) relationship exists. We conducted a prospective cohort study using data from the U.S. National Health and Nutrition Examination Survey 2001–2014. Participants were 3925 postmenopausal women (weighted mean age 63.1 years). Fasting subsample survey weights were applied, and the complex multistage sampling design (stratification and clustering) was fully accounted for in all analyses. TyG index was calculated from fasting triglyceride and glucose measurements. The main outcome was all-cause mortality. We used restricted cubic splines and design-based two-piecewise Cox proportional hazards regression to model nonlinear associations and identify a potential threshold. Over a median follow-up of approximately 9.9 years, 1088 deaths occurred (27.7% of participants). The association was nonlinear: the two-piecewise Cox analysis identified a significant threshold at TyG = 9.06 (95% confidence interval [CI] 8.92–9.26). Below the threshold, TyG index was not significantly associated with mortality (hazard ratio 0.88, 95% CI 0.69–1.11). Above this threshold, each 1-unit increase in TyG was associated with a 44% higher risk of all-cause mortality (hazard ratio 1.44, 95% CI 1.05–1.96, P = .020) after full adjustment for covariates. In this cohort of U.S. postmenopausal women, an elevated TyG index above 9.06 was associated with increased all-cause mortality. The relationship was nonlinear, with a risk threshold. This threshold should be considered exploratory and requires external validation before clinical use.

1. Introduction

Menopause represents a significant transition in women’s lives, marked by the cessation of ovarian function and a decline in estrogen levels. This hormonal shift is associated with an increased risk of metabolic disorders such as dyslipidemia, insulin resistance, and central obesity. Collectively, these factors contribute to an elevated susceptibility to cardiovascular disease (CVD) and all-cause mortality among postmenopausal women.^[1,2] Therefore, early identification and management of these metabolic risk factors are crucial for enhancing long-term health outcomes within this population.^[3]

The triglyceride–glucose (TyG) index, derived from fasting triglyceride and glucose levels, has emerged as a useful surrogate marker for insulin resistance.^[4] A considerable body of evidence has shown an independent relationship between a heightened TyG index and increased risks of adverse clinical outcomes, such as type 2 diabetes, atherosclerosis, and cardiovascular events.^[5–7] Its convenience and predictive capabilities have led to its growing adoption as a screening tool in both epidemiological research and clinical practice for identifying at-risk populations.^[8,9]

While the relationship between the TyG index and all-cause mortality remains inadequately characterized in postmenopausal women, emerging evidence suggests that the relationship may be nonlinear in broader populations. Studies involving large cohorts have indicated that both very low and very high TyG indices are associated with elevated mortality risks, hinting at a possible J-shaped or U-shaped correlation.^[10,11] For instance, a recent analysis involving over 3 million Chinese adults noted a U-shaped relationship between TyG index levels and mortality, pointing to an inflection point around a TyG index of approximately 9.0.^[12,13] Such findings raise the question of whether a similar nonlinear relationship exists among postmenopausal women, which could help identify high-risk segments of the population that might benefit from targeted metabolic assessment based on easily obtainable indices like the TyG.

Therefore, the present study aims to investigate the relationship between TyG index and all-cause mortality in postmenopausal women, with a particular focus on identifying potential nonlinear associations and threshold effects. Understanding this connection may provide valuable insights for individualized risk assessment and the development of effective prevention strategies in clinical and public health settings.

2. Methods

2.1. Study design and participants

We analyzed data from the National Health and Nutrition Examination Survey (NHANES) 2001–2014 (https://wwwn.cdc.gov/nchs/nhanes/default.aspx). NHANES is a continuous, nationally representative survey of the U.S. civilian, noninstitutionalized population, conducted by the Centers for Disease Control and Prevention. The NHANES study plan received approval from the Ethics Review Committee of the National Center for Health Statistics (https://www.cdc.gov/nchs/nhanes/irba98.htm), and all study participants provided written informed consent. The survey uses a complex multistage probability sampling design, with data collected through interviews, physical examinations, and laboratory tests. For the present analysis, we combined 7 consecutive NHANES two-year cycles (2001–2002 through 2013–2014). Participants eligible for inclusion were women aged 40 years or older who self-reported as postmenopausal (defined as having no menstrual periods for at least 12 months). Menopausal status was determined based on responses to reproductive health questionnaires.^[14] Those who answered “no” were then asked the reason for the absence of menstruation. In the present analysis, postmenopausal status was defined as having no menstrual periods for at least 12 months due to natural menopause (i.e., responding “menopause/change of life” to the reason for amenorrhea).^[15–17] Women who reported hysterectomy as the sole reason for amenorrhea were not included, to reduce the risk of misclassification. We restricted the sample to postmenopausal women who attended the morning examination session and provided fasting blood samples, as fasting values for triglycerides and glucose are required to calculate the TyG index. Among those eligible, we excluded individuals with missing data on fasting triglycerides or fasting glucose (needed for TyG calculation), or missing data on mortality status or key covariates. The final analytic sample consisted of 3925 postmenopausal women (Fig. 1).

Figure 1.

Flowchart of study population selection. NHANES = National Health and Nutrition Examination Survey.

2.2. Assessment of TyG index

The exposure of interest was the TyG index, calculated for each participant from fasting laboratory measurements. Serum triglycerides (mg/dL) and fasting glucose (mg/dL) were measured under standard NHANES protocols after an overnight fast. The TyG index was computed using the formula: TyG index = ln([Triglycerides [mg/dL] * Fasting glucose [mg/dL]]/2), which is a formula validated in prior literature.^[6,18] For some analyses, we examined TyG index both as a continuous variable (per 1-unit increment) and by categories. For descriptive analyses, the TyG index was categorized into tertiles based on the 33.3rd and 66.7th percentiles of the empirical distribution in the study sample (cutoffs: 8.50 and 9.00). In further analyses, a specific high-risk threshold was identified via spline and piecewise regression methods.

2.3. Assessment of all-cause mortality

The outcome was all-cause mortality during follow-up. Mortality status and timing were determined through linkage of NHANES participants to the National Death Index using a unique identifier. Follow-up time was calculated from the date of NHANES exam (baseline) until date of death or end of follow-up, whichever came first. For this analysis, we used the NHANES public-use linked mortality data available through December 31, 2015.^[19] Thus, participants were followed for mortality from their baseline survey year (2001–2014) up to the end of 2015. The maximum possible follow-up was about 14 years (for those entering in 2001–2002), and the minimum was about 1 year (for those entering in 2013–2014). We did not impose additional censoring criteria beyond the administrative end of follow-up in 2015. All-cause mortality was defined as death from any cause. The vital status ascertainment in NHANES has been validated and is based on probabilistic matching to national death records; we treated it as complete for our analysis. These mortality files are available for online access (https://www.cdc.gov/nchs/data-linkage/mortality-public.htm).

2.4. Assessment of covariates

We included a range of baseline covariates known or suspected to confound the association between metabolic factors and mortality. Covariates were chosen a priori based on literature and data availability in NHANES. Demographic factors included age (years), ethnicity (categorized as Non-Hispanic White, Non-Hispanic Black, Mexican American, or Other), education level (less than high school, high school, or above), marital status (married/living with partner, widowed/divorced/separated, or never married), and poverty-to-income ratio (PIR, classified as poor, near poor, middle income, or high income). Lifestyle factors included smoking status (current smoker, former smoker, or never smoker) and alcohol use (classified as nondrinker, former drinker, mild drinker, moderate drinker, or heavy drinker, based on self-reported drinking frequency and quantity). We also adjusted for baseline hypertension status (yes/no, defined as having measured blood pressure ≥ 140/90 mm Hg or using antihypertensive medication or self-reported physician diagnosis of hypertension)^[20] and body mass index (BMI, in kg/m²). BMI was treated as a continuous variable in primary models; we also categorized BMI as 13.4 to 25.99, 26 to 31.33, or 31.34 to 64.80 in stratified analyses. We did not adjust for diabetes status or lipid-lowering medication use in the main models, since fasting glucose and triglycerides (which constitute the TyG index) are intermediate variables on the causal pathway; including such variables could induce overadjustment. Similarly, we did not include other metabolic laboratory measures as covariates to avoid multicollinearity with the TyG index. Specifically, fasting triglycerides and fasting glucose, as the components of the TyG index, were not included as separate covariates in any model to avoid collinearity.

Covariates were selected a priori based on clinical knowledge and published evidence on established confounders of the relationship between insulin resistance and mortality. No stepwise or data‑driven selection procedure (e.g., P‑value‑based or change‑in‑estimate methods) was applied, to avoid overfitting and to maintain transparency. Multicollinearity among covariates was assessed using the variance inflation factor; all variance inflation factor values were below 2.0, indicating no significant collinearity.

All covariate information was obtained at the baseline NHANES examination through standardized questionnaires, physical measurements, and lab tests.

2.5. Statistical analysis

All analyses were conducted using R (version 4.3.0) and EmpowerStats (www.empowerstats.com, X&Y Solutions, Inc.) software. A two-sided P < .05 was considered statistically significant.

Missing values were present in several covariates but not in the exposure (TyG index) or the outcome. Because the proportion of missing data was modest across covariates (ranging from 0% to approximately 8.54%), we applied the indicator (dummy variable) method to retain all observations and avoid unnecessary loss of statistical power. For categorical covariates with missing data (e.g., smoking status, alcohol use, education level, and marital status), a separate “missing” category was created. For continuous covariates with missing values (e.g., PIR and BMI), median imputation was performed together with a missing‑indicator dummy variable to flag imputed observations. A summary of missing proportions per variable is provided in Table S1, Supplemental Digital Content 1.

All estimates were weighted using the mobile examination center fasting subsample sample weights. Because 7 NHANES 2-year cycles were combined (2001–2014), a new combined weight variable was created by dividing the 2-year fasting sample weight by 7 (WTSAF2YR/7) in accordance with NHANES analytic guidelines. The complex survey design was accounted for by specifying the stratification variable (SDMVSTRA) and primary sampling unit variable (SDMVPSU) in all analyses. Variance estimation was performed using Taylor series linearization. Descriptive baseline characteristics are presented as survey-weighted mean (95% confidence interval [CI]) for continuous variables and survey-weighted percentage (95% CI) for categorical variables. For survival analyses, design-based Cox proportional hazards regression was performed using survey-weighted Cox models (R survey package, svycoxph function) with the combined weight, strata, and primary sampling unit variables specified. The threshold effect analysis, two-piecewise Cox models, and restricted cubic spline (RCS) models were all fitted under the same survey design specification. Descriptive baseline characteristics are presented as survey-weighted mean (95% CI) for continuous variables and survey-weighted percentage (95% CI) for categorical variables. The revised weighted Table 1 is in the main text; the original unweighted version is provided as Table S2, Supplemental Digital Content 2.

Table 1.

Baseline characteristics of study participants according to TyG index tertiles, weighted.

Characteristic	Total	Triglyceride-glucose index			P-value
		T1 (6.97–8.50)	T2 (8.5–9.00)	T3 (9.00–12.26)
N	3925	1308	1308	1309
Age (yrs)	63.14 (62.68–63.60)	61.49 (60.73–62.24)	64.13 (63.35–64.92)	63.96 (63.24–64.68)	<.0001
BMI (kg/m2)	29.32 (29.01–29.62)	26.90 (26.40–27.40)	29.79 (29.28–30.29)	31.53 (31.07–32.00)	<.0001
Triglyceride (mg/dL)	139.47 (135.16–143.78)	75.18 (73.81–76.56)	124.52 (123.02–126.02)	226.82 (218.24–235.39)	<.0001
TyG index	8.76 (8.73–8.79)	8.15 (8.13–8.17)	8.75 (8.74–8.76)	9.45 (9.42–9.49)	<.0001
Glucose (mg/dL)	108.9 (107.5– 110.3)	95.9 (95.2– 96.7)	104.2 (103.1– 105.3)	128.2 (124.6– 131.9)	<.0001
Ethnicity (%)					<.0001
Non-Hispanic White	77.49 (74.85–79.93)	77.02 (73.91–79.85)	76.45 (72.96–79.62)	79.09 (75.92–81.95)
Non-Hispanic Black	9.53 (8.18–11.07)	13.31 (11.32–15.58)	8.65 (7.14–10.43)	6.21 (4.95–7.75)
Mexican American	4.52 (3.55–5.76)	2.86 (2.16–3.77)	4.91 (3.68–6.53)	5.99 (4.55–7.85)
Other Hispanic	3.74 (2.75–5.06)	2.83 (1.95–4.09)	4.48 (3.27–6.12)	3.97 (2.64–5.94)
Other Race	4.72 (3.86–5.77)	3.98 (2.97–5.32)	5.51 (4.13–7.30)	4.74 (3.41–6.56)
Marital status (%)					.3156
Married/Living with Partner	58.84 (57.13–60.53)	61.62 (58.15–64.97)	57.29 (53.85–60.65)	57.35 (54.06–60.57)
Widowed/Divorced/Separated	36.07 (34.37–37.82)	33.81 (30.40–37.40)	37.69 (34.61–40.88)	36.93 (33.97–40.00)
Never married	5.08 (4.20–6.14)	4.57 (3.47–6.00)	5.02 (3.64–6.89)	5.72 (4.12–7.89)
Poverty income ratio (%)					<.0001
Poor	9.93 (8.51–11.55)	9.16 (7.07–11.79)	9.58 (8.06–11.36)	11.14 (9.13–13.54)
Nearly poor	22.30 (20.66–24.04)	16.86 (14.73–19.22)	24.10 (21.33–27.10)	26.53 (23.53–29.77)
Middle income	27.64 (25.56–29.82)	26.19 (22.98–29.69)	27.61 (24.20–31.30)	29.29 (26.13–32.67)
High income	33.16 (30.37–36.08)	40.91 (36.74–45.22)	31.73 (27.79–35.95)	25.98 (22.21–30.14)
Missing	6.97 (6.01–8.07)	6.88 (5.31–8.87)	6.98 (5.55–8.74)	7.06 (5.52–8.98)
Education level (%)					<.0001
Below high school	7.66 (6.72–8.73)	5.39 (4.27–6.79)	7.43 (6.09–9.05)	10.44 (8.82–12.32)
High school	39.34 (37.12–41.60)	31.04 (27.91–34.34)	42.53 (39.53–45.58)	45.33 (41.59–49.12)
Above high school	52.97 (50.55–55.38)	63.57 (60.08–66.92)	50.01 (46.74–53.28)	44.18 (40.27–48.16)
Missing	0.03 (0.01–0.07)	0.00 (0.00–0.00)	0.03 (0.01–0.12)	0.05 (0.02–0.17)
Smoking status (%)					.0937
Never	55.41 (52.98–57.82)	57.28 (53.18–61.28)	56.92 (53.57–60.21)	51.77 (48.00–55.53)
Former	28.85 (26.77–31.02)	28.50 (25.38–31.83)	28.40 (25.05–32.01)	29.71 (26.44–33.20)
Current	15.67 (13.72–17.85)	14.22 (11.00–18.19)	14.63 (12.25–17.39)	18.37 (15.84–21.20)
Missing	0.06 (0.02–0.26)	0.00 (0.00–0.00)	0.05 (0.01–0.20)	0.15 (0.03–0.89)
Alcohol use (%)					.0013
Never	19.53 (17.78–21.42)	18.63 (16.16–21.39)	18.83 (16.46–21.46)	21.27 (18.52–24.30)
Former	23.37 (21.61–25.23)	20.16 (17.37–23.27)	23.09 (20.63–25.75)	27.25 (24.60–30.06)
Mild	34.38 (31.86–37.00)	35.50 (31.94–39.23)	36.32 (32.48–40.35)	31.14 (27.58–34.93)
Moderate	15.31 (13.76–17.01)	18.11 (15.71–20.79)	13.30 (10.76–16.32)	14.25 (11.73–17.21)
Heavy	7.22 (6.07–8.56)	7.46 (5.61–9.86)	8.35 (6.47–10.70)	5.78 (4.37–7.60)
Missing	0.19 (0.08–0.46)	0.14 (0.04–0.50)	0.11 (0.02–0.51)	0.32 (0.07–1.34)
Hypertension (%)					<.0001
No	39.56 (37.63–41.54)	50.75 (46.89–54.59)	36.77 (33.44–40.24)	29.94 (26.56–33.55)
Yes	60.44 (58.46–62.37)	49.25 (45.41–53.11)	63.23 (59.76–66.56)	70.06 (66.45–73.44)
Diabetes status (%)					<.0001
No	59.55 (57.13–61.92)	79.42 (76.47–82.08)	61.28 (57.65–64.78)	35.53 (31.89–39.35)
Yes	23.02 (21.25–24.88)	9.34 (7.65–11.37)	18.81 (16.34–21.57)	42.65 (38.67–46.74)
IFG	8.34 (7.35–9.44)	3.97 (2.76–5.70)	8.81 (7.23–10.69)	12.73 (10.53–15.32)
IGT	9.10 (7.84–10.54)	7.26 (5.52–9.50)	11.10 (8.78–13.95)	9.08 (7.12–11.52)
All-cause mortality, n (%)	1088 (27.72%)	293 (22.40%)	380 (29.05%)	415 (31.70%)	<.0001
Cardiovascular mortality, n (%)	355 (9.04%)	99 (7.57%)	112 (8.56%)	144 (11.00%)	.0070

Data are presented as mean (95% confidence interval) for continuous variables or number (%) for categorical variables. P-values for continuous variables were derived from Kruskal–Wallis tests; categorical variables with expected cell counts < 10 were analyzed using Fisher exact test. All‑cause and cardiovascular mortality are presented as n (%) of deaths within each tertile over the follow‑up period.

BMI = body mass index, IFG = impaired fasting glycemia, IGT = impaired glucose tolerance, TyG = triglyceride‑glucose.

Baseline characteristics were summarized by TyG index tertiles (low, middle, and high). Continuous variables are presented as mean ± SD; categorical variables as percentages. Differences across 3 groups were tested using one-way ANOVA (continuous) or chi-square tests (categorical). Crude mortality rates (per 1000 person-years) by tertile were calculated; Kaplan–Meier curves were plotted.

We used a generalized additive model and a RCS to model TyG index and risk of death in our preliminary analysis to visualize its dose–response curve. A two-piecewise Cox proportional hazards model formally tested for a threshold effect.^[21] An iterative algorithm identified the inflection point (TyG value) maximizing the model log-likelihood. The log-likelihood ratio test compared the two-piecewise model to a linear model; a significant result (P < .05) indicated a threshold effect.^[6] Hazard ratio (HRs; 95% CIs) below and above the threshold were reported. Differences in all-cause mortality between TyG index groups were evaluated using Kaplan–Meier survival analyses. To evaluate potential effect modification, analyses were stratified for significant covariates by age, BMI, ethnicity, marital status, education level, smoking status, PIR, alcohol use, and hypertension. In a sensitivity analysis, we excluded participants who died within the first 2 years of follow-up to mitigate potential reverse causation bias from preclinical disease. The proportional hazards assumption was evaluated by testing the Schoenfeld residuals. A two-sided P value > .05 was considered to indicate no violation.

3. Results

3.1. Participant characteristics

A total of 3925 postmenopausal women were included in this study. The baseline characteristics of the study population stratified by TyG index tertiles are presented in Table 1. Significant differences in demographic, metabolic, and clinical variables were observed across T1, T2, and T3 TyG tertiles. Compared with participants in T1, those in T3 had higher mean values for age (63.96, 95% CI: 63.24–64.68 vs 61.49, 95% CI: 60.73–62.24 years), BMI (31.53, 95% CI: 31.07–32.00 vs 26.90, 95% CI: 26.40–27.40 kg/m²), triglyceride levels (226.82, 95% CI: 218.24–235.39 vs 75.18, 95% CI: 73.81–76.56 mg/dL), and fasting glucose (128.2, 95% CI: 124.6–131.9 vs 95.9, 95% CI: 95.2–96.7 mg/dL; all P < .0001). Ethnicity distribution varied significantly (P < .0001); the proportion of Mexican Americans increased from T1 to T3 (2.86–5.99%), while Non-Hispanic Black participants were more prevalent in T1 (13.31%) than in T3 (6.21%), Socioeconomic factors also differed: the combined proportion of poor and nearly poor individuals was higher in T3 (37.67%) than in T1 (26.02%), and the percentage with education below high school was greater in T3 (10.44%) than in T1 (5.39%; P < .001). Regarding health behaviors, current smoking was more common in T3 (18.37%) than in T1 (14.22%), although this difference did not reach statistical significance (P = .0937). Alcohol use differed significantly across tertiles (P = .0013), with T3 having higher proportions of never drinkers (21.27%) and former drinkers (27.25%) compared to T1 (18.63% and 20.16%, respectively). Comorbidities were markedly elevated in T3, including hypertension (70.06% vs 49.25% in T1) and diabetes mellitus (42.65% vs 9.34% in T1; P < .0001). Marital status was the only variable without significant inter-tertile differences (P = .3156). The cumulative all-cause mortality rates in the low, middle, and high TyG tertiles were 22.40%, 29.05%, and 31.70%, respectively (P < .001, Table 1). For cardiovascular mortality, the corresponding rates were 7.57%, 8.56%, and 11.00% (P = .007, Table 1). The revised weighted Table 1 has been moved into the main manuscript; the original unweighted version is provided as Table S2, Supplemental Digital Content 2.

3.2. Association between TyG index and mortality

A total of 3925 participants were included in this study, with 1088 all-cause mortality events recorded during the follow-up period. A Cox proportional hazards model was employed, incorporating the TyG index as a RCS to explore its association with all-cause mortality. The results are presented as a smoothed curve (Fig. 2). As shown in Figure 2, after adjustment for covariates, there was a nonlinear relationship between the TyG index and risk for all-cause mortality. At TyG values below approximately 9.0, the logarithm of relative risk remained near zero, suggesting little to no effect of the TyG index on all-cause mortality within this range. However, as the TyG index increases above 9.0, the logarithm of relative risk begins to rise, and this upward trend becomes more pronounced for TyG index values exceeding 10.0, indicating an increased risk of all-cause mortality associated with higher TyG index levels.

Figure 2.

Association between TyG index and log relative risk of all-cause mortality. Adjusted for age, ethnicity, marital status, poverty income ratio, education level, smoking status, alcohol use, hypertension, and body mass index. TyG index = triglyceride-glucose index, Log RR = logarithm of relative risk.

As shown in Table 2, a threshold effect analysis was performed to further investigate the nonlinear association between the TyG index and all-cause mortality risk. All models were adjusted for age, ethnicity, marital status, PIR, education level, smoking status, alcohol consumption, hypertension, and BMI.

Table 2.

Threshold effect analysis of TyG index on all-cause mortality, weighted.

Model	HR (95% CI)	P-value
Model I (Linear association)	1.17 (1.01–1.35)	.026
Model II (Threshold effect)
Inflection point (K)	9.06	–
95% CI for inflection point	8.92–9.26	–
TyG index < 9.06	0.88 (0.69–1.11)	.304
TyG index > 9.06	1.44 (1.05–1.96)	.020
P for Log-likelihood ratio	–	.001

Cox proportional hazards models were used to estimate HR and 95% CI. Adjusted for age, ethnicity, marital status, poverty income ratio, education level, smoking status, alcohol use, hypertension, and body mass index.

CI = confidence interval, HR = hazard ratio, TyG = triglyceride-glucose.

In the fully adjusted linear model (Model I), each unit increase in the TyG index was associated with a significantly higher risk of all-cause mortality (HR: 1.17, 95% CI: 1.01–1.35, P = .026). Using a two-piecewise Cox proportional hazards model (Model II), a significant threshold effect was identified, with an inflection point at a TyG index of 9.06. When the TyG index was below 9.06, there was no significant association with all-cause mortality (HR: 0.88, 95% CI: 0.69–1.11, P = .304). In contrast, a TyG index above 9.06 was associated with a 44% higher risk of all-cause mortality (HR: 1.44, 95% CI: 1.05–1.96, P = .020). The 95% CI for the inflection point (TyG = 9.06) was 8.92 to 9.26 (Table 2). The log-likelihood ratio test indicated a statistically significant improvement of the threshold model over the linear model (P = .001). The proportional hazards assumption was assessed using Schoenfeld residuals. The detailed results are provided in Table S3, Supplemental Digital Content 3. Although the global test showed a P value of .0347, the primary exposure variable (TyG index) satisfied the PH assumption (P = .115), indicating that the effect of TyG on mortality remained stable over time.

Finally, Kaplan–Meier survival curves were generated to evaluate the association between TyG index tertiles and all-cause mortality over a follow-up period of up to 200 months (Fig. 3). As shown in Figure 3A, participants in the highest TyG tertile exhibited significantly lower survival probabilities for all-cause mortality compared to those in the middle and lowest tertiles. Similarly, for cardiovascular mortality (Fig. 3B), a higher TyG index was associated with poorer survival. The log-rank test revealed significant differences in all-cause mortality among the 3 groups (Chi-square = 17.15, P = .0002). Kaplan–Meier curves with 95% CIs for (A) all-cause and (B) cardiovascular mortality are presented in Figure S1, Supplemental Digital Content 4. The overall density distribution of the TyG index is shown in Figure S2, Supplemental Digital Content 6.

Figure 3.

Kaplan–Meier survival curves for (A) all-cause mortality and (B) cardiovascular mortality, stratified by tertiles of the triglyceride-glucose (TyG) index. TyG index tertiles: low, 6.97 to 8.50; middle, 8.50 to 9.00; and high, 9.00 to 12.26. The number of patients at risk at selected time points is shown below the x-axis for each tertile.

3.3. Stratified analyses

The results of stratified sensitivity analyses indicated that the direction of associations between the TyG index and mortality was generally consistent with the main findings across strata (Table 3), although the magnitude of risk varied in some subgroups. For all-cause mortality, the TyG index exhibited stronger effects in younger age groups (HR = 1.95, P < .001) and among Non-Hispanic Black individuals (HR = 1.61, P < .001). In the stratified analysis by marital status, married/partnered individuals showed significantly elevated hazards (HR = 1.48, P < .001). For cardiovascular mortality, similar patterns emerged, with notable risks in Non-Hispanic Black (HR = 1.83, P = .001) and moderate alcohol users (HR = 2.19, P = .003). BMI groups uniformly showed elevated all-cause mortality risks (HR range: 1.35–1.40, P < .001), but cardiovascular effects were less consistent. In the analysis stratified by diabetes status, the TyG index was not significantly associated with all-cause or cardiovascular mortality in any glycemic subgroup, and no significant interaction was observed (all P for interaction > .05). Additionally, a sensitivity analysis excluding participants who died within the first 2 years of follow-up showed results consistent with the primary findings (Fig. S3, Supplemental Digital Content 7).

Table 3.

Stratified analysis of TyG index on mortality.

Stratified variable	N	All-cause mortality HR (95% CI)	P-value	Cardiovascular mortality HR (95% CI)	P-value
Age (yrs)
40–59	1224	1.95 (1.55–2.46)	<.0001	2.12 (1.40–3.21)	.0004
60–69	1285	1.40 (1.15–1.69)	.0006	1.87 (1.31–2.68)	.0006
70–85	1416	0.94 (0.82–1.07)	.3480	0.91 (0.73–1.13)	.3908
Ethnicity
Non-Hispanic White	2093	1.26 (1.12–1.41)	<.0001	1.17 (0.95–1.44)	.1288
Non-Hispanic Black	734	1.61 (1.28–2.02)	<.0001	1.83 (1.28–2.62)	.0009
Mexican American	574	1.15 (0.89–1.50)	.2888	1.48 (0.92–2.39)	.1042
Other Hispanic	308	1.70 (1.01–2.85)	.0450	2.56 (1.06–6.19)	.0365
Other Race	216	1.30 (0.68–2.48)	.4321	2.55 (0.88–7.35)	.0830
Marital status
Married/Living with Partner	1965	1.48 (1.28–1.70)	<.0001	1.44 (1.10–1.88)	.0073
Widowed/Divorced/Separated	1716	1.11 (0.98–1.26)	.1033	1.17 (0.95–1.44)	.1497
Never married	243	1.44 (0.95–2.18)	.0878	2.16 (0.93–5.02)	.0721
Poverty income ratio
Poor	620	1.04 (0.84–1.28)	.7207	1.09 (0.77–1.55)	.6263
Nearly poor	1066	1.17 (0.99–1.37)	.0612	1.23 (0.94–1.62)	.1304
Middle income	999	1.22 (1.00–1.49)	.0531	1.29 (0.90–1.85)	.1604
High income	905	1.36 (1.07–1.73)	.0126	0.99 (0.61–1.59)	.9583
Missing	335	1.41 (1.05–1.89)	.0223	1.59 (0.97–2.62)	.0668
Education level
Below high school	587	0.88 (0.71–1.09)	.2428	0.92 (0.64–1.34)	.6760
High school	1597	1.17 (1.02–1.35)	.0235	1.21 (0.95–1.53)	.1253
Above high school	1736	1.46 (1.24–1.71)	<.0001	1.46 (1.10–1.93)	.0085
Smoking status
Never	2302	1.33 (1.17–1.50)	<.0001	1.20 (0.97–1.49)	.0866
Former	1059	1.26 (1.07–1.49)	.0054	1.46 (1.10–1.95)	.0098
Current	560	1.08 (0.84–1.38)	.5483	1.42 (0.92–2.20)	.1135
Alcohol use
Never	965	1.37 (1.15–1.63)	.0004	1.33 (0.99–1.80)	.0601
Former	1034	0.95 (0.82–1.11)	.5464	0.90 (0.69–1.17)	.4161
Mild	1156	1.47 (1.21–1.78)	<.0001	1.38 (0.97–1.95)	.0729
Moderate	481	1.29 (0.93–1.78)	.1243	2.19 (1.30–3.70)	.0032
Heavy	280	1.47 (0.94–2.32)	.0935	2.36 (0.96–5.84)	.0623
Hypertension
No	1334	1.28 (1.05–1.56)	.0157	1.11 (0.72–1.70)	.6355
Yes	2591	1.17 (1.05–1.30)	.0045	1.18 (0.99–1.41)	.0636
Body mass index (kg/m2)
13.40–25.99	1278	1.40 (1.19–1.66)	<.0001	1.50 (1.12–2.02)	.0066
26.00–31.33	1283	1.35 (1.15–1.58)	.0002	1.32 (0.99–1.75)	.0547
31.34–64.80	1283	1.39 (1.16–1.66)	.0003	1.32 (0.98–1.77)	.0687
Diabetes status
No	2099	1.11 (0.94–1.32)	.2209	1.12 (0.82–1.55)	.4695
Diabetes mellitus	1132	1.03 (0.89–1.18)	.7047	0.95 (0.74–1.20)	.6520
IFG	333	1.01 (0.66–1.53)	.9796	1.14 (0.55–2.33)	.7282
IGT	361	0.86 (0.54–1.38)	.5404	0.82 (0.38–1.79)	.6209

CI = confidence interval, HR = hazard ratio, IFG = impaired fasting glycemia, IGT = impaired glucose tolerance, TyG = triglyceride-glucose index.

4. Discussion

In this cohort study of U.S. postmenopausal women, we found that the TyG index was independently associated with all-cause mortality, and that the association was markedly nonlinear. Specifically, mortality risk remained stable at lower to moderate TyG levels but increased sharply above a threshold of approximately 9.0. After adjusting for comprehensive demographic, socioeconomic, lifestyle, and health-related covariates, women with TyG values ≥ 9.06 had a 44% higher mortality risk compared to those below this threshold. To our knowledge, this is one of the first studies to demonstrate a threshold effect of TyG index on mortality in a general population sample of women. These findings suggest that the TyG index, a simple surrogate of insulin resistance, is associated with elevated mortality risk in postmenopausal women and may serve as a potential adjunctive risk indicator alongside traditional risk factors, although prospective studies are needed to confirm its clinical utility. However, we acknowledge that this threshold was identified through a data-driven approach and should be interpreted as exploratory. External validation in independent cohorts is needed before the TyG threshold can be recommended for clinical use.

Our results align with and extend the growing body of literature on TyG index and health outcomes. Several recent large-scale studies have reported positive associations between TyG and mortality. For example, He et al examined 3.5 million Chinese adults and found that higher TyG index was associated with increased all-cause and cause-specific mortality.^[22] In that study, the relationship appeared to be approximately linear over most of the TyG range, although extremely low TyG levels (possibly indicating malnutrition or other illness) were also associated with higher mortality, yielding a U-shaped curve. In contrast, our cohort of postmenopausal women did not show evidence of increased risk at the low end of TyG (perhaps because very low TyG values were uncommon in our relatively well-nourished U.S. sample). Instead, we identified a threshold relationship driven by high TyG, aligning with Zhao et al, who reported escalating mortality above TyG ≈ 8.8–9.0 in diabetic NHANES participants.^[23] Our findings in a general cohort of postmenopausal women, without selection for diabetes, echo those observations and suggest that a TyG threshold near 9 may reflect a reproducible marker of elevated risk, pending external validation. Similarly, a recent analysis of NHANES adults by Yi et al also found a nonlinear association between TyG and mortality, with curves flattening at lower TyG and rising at higher values.^[20] Collectively, these studies indicate that the relationship between the TyG index and mortality is nonlinear across populations, with disproportionately elevated risk beyond thresholds reflecting advanced insulin resistance.

The biological mechanisms underlying the association between the TyG index and mortality stem from the implications of insulin resistance and related metabolic dysfunction. A high TyG index indicates elevated levels of circulating triglycerides and glucose, which are characteristic of insulin resistance syndrome.^[24] This metabolic state can initiate several adverse pathways, including endothelial dysfunction, atherosclerosis, pro-inflammatory and pro-thrombotic effects, as well as oxidative stress.^[3] Over time, these processes can increase the risk of developing CVDs, such as coronary artery disease and stroke, both of which are leading causes of mortality.^[7] Evidence suggests that a significant portion of the mortality risk associated with the TyG index is mediated by cardiovascular causes.^[25] Additionally, insulin resistance and hyperinsulinemia may enhance growth factor signaling and cellular proliferation, linking a high TyG index to cancer mortality as well.^[26] Elevated triglyceride levels may also indicate abnormalities in lipoprotein particles (e.g., small dense LDL and low HDL), which further amplify cardiovascular risks.^[27] Furthermore, the TyG index can serve as an indicator of nonalcoholic fatty liver disease or ectopic fat deposition, conditions that are associated with increased mortality through liver disease, cardiovascular issues, and complications from diabetes.^[28] Notably, the TyG index has shown better correlation with visceral adiposity and liver fat than some traditional measures, such as BMI, thus providing a more nuanced understanding of risk factors that may not be apparent when relying solely on BMI.^[29]

Another interesting aspect of our findings is the lack of increased mortality at lower TyG values. In some cohorts, particularly older or frail populations, very low lipid levels or low BMI can be associated with higher mortality, a phenomenon often interpreted as reverse causation stemming from chronic illness or malnutrition.^[7] In our sample of postmenopausal women, we did not observe a significant increase in risk in the lowest TyG group. This could be because our population had few individuals with truly low TyG – even the lowest tertile had a mean TyG of ~8.1, which corresponds to fairly normal triglyceride and glucose levels. Thus, our study does not contradict the possibility of a U-shaped curve in certain contexts, but it suggests that within the range of typical to high metabolic values, the primary concern is the upper end. Women who maintain a low TyG (through healthy diet, physical activity, etc) did not show any mortality penalty in our analysis; on the contrary, they had the best survival, though the difference between low and moderate TyG was minor after adjustments.

From a clinical and public health perspective, these results highlight the potential utility of TyG index as a simple screening tool. Both triglycerides and glucose are routinely measured in clinical practice and annual health evaluations. Calculating the TyG index requires no additional cost or specialized tests beyond these standard measures.^[30] In clinical settings, a TyG index in the high range (for example, >9) was associated with significantly elevated mortality risk in this observational study, though its clinical utility as a screening trigger requires prospective evaluation. If these associations are confirmed in prospective studies, lifestyle modifications targeting weight loss, improved diet, and increased physical activity may be considered for individuals with elevated TyG index. Importantly, TyG index integrates information about 2 metabolic parameters; thus it might identify risk even when each individual component (glucose or TG) is only moderately elevated. For instance, a woman with “borderline” high fasting glucose (e.g., 110 mg/dL) and “borderline” high triglycerides (e.g., 160 mg/dL) would have TyG around 9.3, which according to our findings places her at elevated risk, though neither value alone might alarm a clinician following traditional thresholds.^[31] In essence, TyG captures the interaction of dysglycemia and dyslipidemia, which together are more deleterious than either abnormality alone.^[32]

Our study focused on postmenopausal women, a group that merits attention because CVD risk accelerates in women after menopause. The loss of estrogen is associated with adverse changes in body fat distribution and lipid profiles; typically, triglycerides rise, and HDL cholesterol falls.^[33] It is plausible that the TyG index is especially prognostic in this demographic as it reflects those menopause-related metabolic shifts.^[1] Prior studies in women have found TyG index to be associated with the development of CVD and type 2 diabetes.^[8,34] We add evidence that it also associated with hard endpoints such as mortality. This underscores the importance of monitoring metabolic health in postmenopausal women. While conventional risk scores (like Framingham or pooled cohort equations) include factors such as blood pressure, cholesterol, and diabetes, they do not include triglycerides or insulin resistance explicitly. The TyG index may serve as a complementary risk indicator in women, and future research could evaluate whether its addition to existing risk prediction models improves risk stratification. For example, future studies could assess whether incorporating the TyG index into risk prediction models improves the identification of at-risk individuals compared with traditional measures alone.^[35] Further research could explore if adding TyG index to risk prediction models yields better discrimination of outcomes in women.

In our primary models, we did not adjust for diabetes status, as the TyG index reflects insulin resistance, which lies on the causal pathway to type 2 diabetes and its complications. Adjusting for an intermediate variable can introduce overadjustment bias and collider bias, potentially distorting the true exposure–outcome association.^[36] To assess whether confounding by diabetes severity could explain our findings, we conducted stratified analyses by diabetes status. The TyG–mortality association did not differ significantly across subgroups (P for interaction > .05), and the HRs within each stratum were consistent with the overall pattern, supporting our primary modeling strategy. These results suggest that the prognostic value of the TyG index is not merely a proxy for diagnosed diabetes. However, we acknowledge that the stratified analysis may have been underpowered to detect modest effect modification, and we cannot fully exclude residual confounding by diabetes duration, severity, or treatment.

This study has several limitations. As an observational analysis, it cannot establish causality and may be affected by unmeasured confounders. The single baseline measurement of the TyG index could lead to regression dilution bias, and the use of sampling weights was limited to maximize power. In addition, we did not adjust for the use of statins or glucose-lowering medications, which may affect both TyG index components and mortality. However, adjusting for these medications could introduce overadjustment bias, as they are prescribed in response to the metabolic abnormalities captured by the TyG index. The lack of detailed medication data (e.g., duration and adherence) remains a limitation. The findings on all-cause mortality may not extend to specific causes of death, and the results from this cohort of U.S. postmenopausal women may not be generalizable to men, premenopausal women, or populations in other settings with different metabolic risk profiles.

5. Conclusion

In conclusion, we found that among postmenopausal women in the United States, a higher TyG index is associated with increased risk of mortality, and notably this risk becomes substantial above a threshold TyG value of roughly 9.0. This nonlinear association suggests that moderate increases in TyG (reflecting mild insulin resistance) may not markedly affect survival, but that more pronounced elevations, reflecting greater insulin resistance, are associated with considerably higher mortality risk.

Acknowledgments

The authors thank the staff and the participants of the NHANES study for their valuable contributions.

Author contributions

Conceptualization: Juanjuan Yao, Dongxue Zhang.

Data curation: Juanjuan Yao.

Formal analysis: Juanjuan Yao, Dongxue Zhang.

Supervision: Dongxue Zhang.

Validation: Juanjuan Yao.

Writing – original draft: Juanjuan Yao, Dongxue Zhang.

Writing – review & editing: Juanjuan Yao.