Triglyceride glucose-waist circumference as a useful predictor for diabetes mellitus: a secondary retrospective analysis utilizing a Japanese cohort study

The data for this study was sourced from a research conducted by Takuro Okamura [15]. Researchers are permitted to utilize the data for secondary analysis in compliance with the Dryad terms of service, ensuring no harm is caused to the original authors.

The study received approval from the First Affiliated Hospital Ethics Committee of Shenzhen University. The research adhered to the Declaration of Helsinki guidelines during the original investigation. Each aspect of the study, including the declarations.

Initially, the study included 20,944 Japanese individuals who had undergone physical examinations. Subsequently, 5,531 participants were excluded from the study, resulting in a final sample size of 15,413 participants (8387 men and 7026 women) for data analysis (refer to Fig. 1 for a detailed flowchart) [15]. Participants were excluded if they met any of the following criteria: 1) Diagnosed with Type 2 diabetes. 2) Known liver diseases or alcoholic fatty liver disease, defined as daily alcohol consumption > 40 g for men or > 20 g for women and a drinking history > 5 years. 3) Fasting plasma glucose (FPG) ≥ 6.1 mmol/L, as this indicates a prediabetic or diabetic state. 4) Missing data for covariates. 5) Taking medications that could influence metabolic parameters, including diabetes management drugs (e.g., insulin, oral hypoglycemic agents), lipid-lowering drugs (e.g., statins, fibrates), antihypertensive drugs (e.g., beta-blockers, ACE inhibitors), and other medications known to affect glucose or lipid metabolism. 6) Identified as having TyG-WC outliers, which were excluded to reduce the influence of extreme values on the analysis [16].

The relationship between triglyceride glucose and waist size was handled as a continuous variable. The TyG-WC index is calculated by combining fasting plasma glucose (FPG) and triglycerides (TG) levels along with waist circumference (WC). The specific calculation formula is: [ln(TG(mg/dL)×FPG(mg/dL)/2)]×WC(cm). Where TG and FPG are measured in milligrams per deciliter (mg/dL), and WC is measured in centimeters (cm).) [14].

DM, a dichotomous variable with the values 1 for DM and 0 for non-DM, was our intriguing outcome variable. Participants with DM were those whose FPG was 7 mmol/l [17], whose HbA1c was not less than 6.5%, or who self-reported throughout the follow-up period.

The selection of variables for this study was informed. As covariates [15]: (1) The continuous variables considered in the study encompass age, BMI, alcohol consumption, baseline levels of diastolic and systolic blood pressure, gamma-glutamyl transferase, AST, ALT, TC, HDL-C, and glycosylated hemoglobin. (2) Categorical variables: fatty liver, regular exerciser, smoking status, and gender. For the initial trial, data on each participant’s lifestyle traits were acquired. The research team followed a standardized procedure to gather data from controlled laboratory tests [18]. The initial study determined the average weekly ethanol intake after evaluating the subjects’ alcohol consumption over the course of the previous month. Following fasting for the previous night, venous blood was taken for the testing of hematological indicators. Based on the findings of an abdominal ultrasound, a skilled technician determined that the patient had fatty liver.

TyG-WC quartiles using R language and EmpowerStats software, and appropriate statistical methods were employed to present the data. Differences between different TyG-WC groups were analyzed using chi-square tests, ANOVA, and Kruskal-Wallis H tests [19‐20], and Kaplan-Meier methods were used to assess the diabetes-free survival probability [21, 22]. Cox regression models were used to analyze the association between TyG-WC and DM risks, taking into account nonlinear relationships. We employed a stratified approach to assess the relationship between TyG-WC and diabetes risk, accounting for various confounding factors. Here, we outline the established schemes for Model I, II, and III presented in Table 1: Model I: This minimally adjusted model includes age and gender as covariates to control for the basic demographic variations that could influence the relationship between TyG-WC and diabetes risk. Model II: Building on Model I, this model further adjusts for lifestyle factors such as smoking status, alcohol consumption, and physical activity, which are known to impact both metabolic health and diabetes risk. Model III: This fully adjusted model incorporates all the variables from Model II, as well as additional clinical parameters including BMI, systolic and diastolic blood pressure, and lipid profiles (HDL-C, and LDL-C), to provide a comprehensive assessment of the independent association between TyG-WC and incident diabetes. Furthermore, in other sensitivity analyses, we disregarded persons who had fatty livers. TyG-WC was also shown to be positively linked with DM risk [23, 24]. The models were evaluated using generalized additive models (GAM) [25]. Finally, the predictive value for DM risk were assessed through ROC curve analysis [20, 26]. All results adhered to the STROBE statement, with p-value < 0.05 considered statistically significant.

Table 2 displays: Final analysis included 15,413 participants, of which 54.42% were male (Fig. 1). The participants had an average age of 43.72 ± 8.90 years. The follow-up duration ranged up to 13.0 years, with a median follow-up period of 5.4 years, 358 people finally developed diabetes.

The BMI, WC, TG, FPG, and TyG-WC were 22.08 ± 3.06 kg/m², and 76.37 ± 8.93 cm, 80.35 ± 57.23 mg/dL, 92.94 ± 7.43 mg/dL and 616.10 ± 105.88. We assigned participants into subgroups using TyG-WC quartiles (≤ 533.92, 533.92-608.15, 608.15-689.03, > 689.03). I In the highest TyG-WC group, participants tended to have higher values for age, SBP, DBP, BMI, WC, ALT, AST, GGT, TG, TC, HbA1c, FPG, and TyG. Additionally, they exhibited higher proportions of men, smokers, and heavy drinkers, along with lower HDL-C levels and lower rates of regular physical exercise. The distribution of TyG-WC levels was displayed in Fig. 2. With an average of 616.10 and a range of 339.45 to 940.03, it displayed a normal distribution. The participants were divided into two groups based on their diabetes mellitus (DM) status. Figure 3 illustrates the TyG-WC values for each group, showing that the DM group had higher levels of TyG-WC compared to the non-DM group. In age stratification with 10-year intervals, male participants exhibited a higher prevalence of DM than female participants across all age groups (Fig. 4). Moreover, we discovered that both among male and female individuals, the prevalence of DM rose with advancing age.

During the follow-up period, a total of 358 individuals developed incident diabetes, as detailed in Table 3. The overall incidence rate among all participants was 2.32% (95% CI: 2.08-2.56%). Specifically, the incidence rates for the four TyG-WC groups were 0.42% (95% CI: 0.21-0.62%), 1.01% (95% CI: 0.70-1.33%), 1.92% (95% CI: 1.49-2.35%), and 5.94% (95% CI: 5.20-6.69%), respectively. Furthermore, the cumulative incidence rates for the entire population and the four TyG-WC groups were 383.87 per 100,000 person-years, 67.31 per 100,000 person-years, 172.44 per 100,000 person-years, 317.76 per 100,000 person-years, and 970.99 per 100,000 person-years, respectively. Participants with high TyG-WC levels had a significantly higher incidence of diabetes compared to those in the lowest TyG-WC group (P < 0.0001 for trend), as illustrated in Fig. 5.

Table 4 displays: The findings indicated positive relationships between age, heavy alcohol consumption, smoking habits, SBP, DBP, BMI, WC, TG, TC, HbA1c, FPG, and TyG-WC with diabetes risk. Conversely, HDL-C was negatively correlated with the risk of diabetes in the univariate analysis. Male participants exhibited a higher likelihood of developing diabetes compared to females. Figure 6 displays the Kaplan-Meier curves illustrating the probability of diabetes-free survival.

An increase of 1 unit in TyG-WC was linked in the unadjusted model to a 0.9% increase in the probability of developing diabetes (HR:1.009, 95% CI:1.008 to 1.010, P < 0.001). The outcomes were statistically noteworthy. Each additional unit of TyG-WC could result in an enhanced risk of DM by 0.5% in the minimum adjusted model (HR: 1.005, 95% CI: 1.004 to 1.007, P < 0.001). The fully adjusted model did not appear to change as a result of the findings. Each additional unit of TyG-WC was associated with a 0.3% increase in the risk of developing diabetes (HR: 1.003, 95% CI: 1.001–1.005, P = 0.00102). The revised confidence interval underscores the model’s robustness in estimating the association between TyG-WC and diabetes risk, and the results are consistent with the previous analysis (Refer to Table 1 for more details).

According to the findings after controlling for confounding variables (HR:1.001, 95% CI: 0.998 to 1.004) (Table 5). Participants who consumed alcohol were likewise not included in the sensitivity analysis. The outcomes of the sensitivity analysis further validated the dependability and uniformity of our results.

In Fig. 7, we observed a non-linear relationship between TyG-WC and diabetes risk, analyzed using a GAM. The smooth curve fitting indicates that the risk of diabetes remains relatively stable at lower TyG-WC levels but increases sharply as TyG-WC exceeds a certain threshold. This trend was further examined using Cox proportional hazards regression, adjusting for confounders such as gender, age, alcohol consumption, smoking habits, physical activity, and biochemical markers including GGT, ALT, HDL-C, AST, and TC. The results revealed a significant increase in diabetes risk with higher levels of TyG-WC (P < 0.001), consistent with the non-linear trend depicted in Fig. 7. These findings suggest that while lower TyG-WC levels have limited impact on diabetes risk, higher levels are associated with a disproportionate increase in risk, highlighting the importance of monitoring individuals with elevated TyG-WC.

Regarding the stratification of liver enzymes in Table 6, the high and low groups for Alanine Aminotransferase (ALT), Aspartate Aminotransferase (AST), and Gamma-Glutamyl Transferase (GGT) were defined based on their respective median values within the study population: ALT: Participants were categorized into low ALT group (below the median value of 30 U/L) and high ALT group (at or above the median value). AST: The low AST group consisted of individuals with values below the median of 25 U/L, while the high AST group had values at or above the median. GGT: For GGT, the boundary was set at the median value of 40 U/L, with participants below this value classified as the low GGT group and those at or above as the high GGT group. These stratifications allowed us to explore the modifying effects of liver function on the relationship between TyG-WC and diabetes risk, providing insights into potential mechanistic pathways. A stronger association was observed in individuals with high levels of ALT and AST (Hazard Ratio (HR): 1.004, 95% Confidence Interval (CI): 1.002–1.007; HR: 1.005, 95% CI: 1.003–1.007, respectively). Conversely, the association between TyG-WC and diabetes risk was not significant in those with low ALT and AST levels (HR: 0.999, 95% CI: 0.995–1.004; HR: 0.998, 95% CI: 0.994–1.003, respectively). These findings suggest that the relationship between TyG-WC and incident diabetes remained robust across most subgroups(see Table 6).

Furthermore, a ROC curve was constructed to evaluate the predictive performance for diabetes risk assessment (Fig. 8). The Area Under the Curve (AUC) was largest for TyG-WC in predicting diabetes (AUC: 0.772, 95% CI: 0.7479–0.7964, P < 0.0001), followed by TyG- Body Mass Index (BMI) (AUC: 0.7662, 95% CI: 0.7416–0.7909, P < 0.0001) and TyG index (AUC: 0.7441, 95% CI: 0.7185–0.7698, P < 0.0001). The superior AUC value of TyG-WC indicates its higher predictive capability for incident diabetes compared to the other variables (see Table 7).