U-Shaped Association Between the Triglycerides-Glucose index and the Risk of Incident Diabetes in Apparently Healthy Population: A Population-based Cohort Study

Although the triglycerides-glucose (TyG) index was thought to be a practical predictor of incident diabetes, the association between them has not been well characterized. The study aimed to further examine the association between the TyG index and incident diabetes in Japanese adults. The cases were extracted of the individual participating in the NAGALA (NAd in the Gifu Area, Longitudinal Analysis) study at Murakami Memorial Hospital from 2004 to 2015, and 14297individuals apparently healthy at baseline were included in the study. Cox proportional hazards models were used to evaluate the associations between baseline TyG levels and incident of T2DM, and a two-piecewise linear regression model was use to examine the threshold effect of the baseline TyG on incident diabetes using a smoothing function. The threshold level (i.e., turning point) was determined using trial and error. A log likelihood ratio test was also conducted to compare the one-line linear regression model with a two-piecewise linear model. with the TyG threshold of 7.97 in men and 7.27 in This information may be useful for reducing incident diabetes by maintaining the TyG index near these

The prevalence of diabetes has increased considerably worldwide over the past four decades [1], imposing a huge burden on mortality, morbidity, and health-care expenditure [1][2][3][4][5][6]. One of the effective interventions to prevent and overcome the burden of diabetes is identifying individuals at a high risk of developing diabetes [2,7]. To this end, novel markers or risk factors need to be investigated to inform the application of preventive models on populations [2,8,9].
Among the traditional risk factors, levels of fasting plasma glucose (FPG) and triglycerides (TG) are associated with an increased risk of type 2 diabetes mellitus (T2DM) [2,[9][10][11]. As a derivative of FPG and TG [12][13][14], the triglycerides-glucose (TyG) index seems to help better predict the risk of T2DM than either component factor and has recently been suggested as a predictor of this risk in individuals without diabetes [2,8,[15][16][17]. The TyG index was suggested to be considered as a screening tool for identifying people at a high risk of T2DM in clinical practice [18]. Although studies on the TyG index as a risk of T2DM have been conducted in Singapore, Korean, Thailand, and the white European population [17][18][19][20], no data on the Japanese population are currently available.
More importantly, previous cohort studies showed a positive relationship between high levels of the TyG index and the risk of incident diabetes [17][18][19][20], but the positive association disappeared at low levels of the TyG index [17,18,20]. This trend suggested a nonlinear association which had never been well characterized in previous studies. Additionally, previous studies did not exclude the individuals with impaired glucose tolerance or 5.7 ≤ HbA1c levels < 6.5%, thus, those result may not be well applied to the healthy population. This study, therefore, aimed to further investigate and characterize the relationship between the TyG index and incident diabetes in a large-scale Japanese cohort.

Design and participants
The data were derived from the population-based NAGALA (NA d in the Gifu Area, Longitudinal Analysis) cohort study conducted at Murakami Memorial Hospital (Gifu, Japan), which was described in detail previously [21].Brie y, The NAGALA was a cohort study of the individual who participated in the medical examination program at Murakami Memorial Hospital from May 1st, 1994 to Dec 31st, 2016 [21].A large proportion of the participants received one to two exams per year. In this study, the cases were extracted of the individuals participating in the examination program at Murakami Memorial Hospital from 2004 to 2015 [21]. The participants with alcoholic fatty liver, viral hepatitis, any medication usage, diagnosed with diabetes, fasting plasma glucose ≥ 6.1 mmol/L, orHbA1C ≥ 5.7% (considered to have pre-diabetes [22])at baseline were excluded. Additionally, individuals with missing data of covariates, were also excluded. Moreover, the statistical method boxplot was used to remove 180 outliers (86 men and 94 women) of TyG index. Eventually, 7857 men and 6440 women were included in this study. As sex differences in the risk, pathophysiology and complications of diabetes existed [23], the male and female participants were analyzed separately in this study. This study was approved by the Ethics Committee of Murakami Memorial Hospital.

Primary outcomes:
The outcome of incident Type 2 diabetes was de ned as HbA1c ≥ 6.5% or fasting plasma glucose ≥ 7 mmol/L according to the diagnostic criteria of ADA or self-reported [21,26].

Statistical analyses
Participants' characteristics were classi ed according to the level of baseline TyG index. Continuous variables are presented as mean ± S.D. or as median (Q1-Q3) for normally and abnormally distributed data separately. Categorical data are presented as frequencies (percentages). Statistical differences between the two groups were evaluated using the Mann-Whitney or Chi-square tests ( Table 1). Strati ed analyses were performed according to baseline age, BMI, waist circumference and fatty liver categories, and their interactions were tested (Fig. 1). Cox proportional hazards models were used to evaluate the associations between baseline TyG levels and incident of T2DM with or without adjustment for potential confounders, including baseline alcohol consumption, smoking status, systolic blood pressure, diastolic blood pressure, age, fatty liver, BMI, waist circumference, habit of exercise, family history of diabetes, HDL cholesterol and total cholesterol ( Table 2). A two-piecewise linear regression model was used to examine the threshold effect of the baseline TyG on incident diabetes using a smoothing function (Fig. 2, Table 3). The threshold level (i.e., turning point) was determined using trial and error: rstly, the turning points were selected according to a pre-de ned interval, and then the turning point was chosen which gave the maximum model likelihood. A log likelihood ratio test was also conducted to compare the one-line linear regression model with a two-piecewise linear model. Results were considered statistically signi cant when P < 0.05 (two-tailed). All statistical analyses were performed using the statistical packages R (The R Foundation; http://www.r-project.org; version 3.4.3) and EmpowerStats [27].  Continuous variables are presented as mean ± S.D. or as median (Q1-Q3) for normally and abnormally distributed data separately. Categorical data are presented as frequencies (percentages).

Participants' characteristics
The current study enrolled 7857 men and 6440 women. Table 1 presents the clinical and biochemical characteristics of the population studied. The average age, BMI, waist circumference, total cholesterol, triglycerides, fasting plasma glucose, the TyG index, blood pressure, alcohol consumption, the percentage of fatty liver, as well as the proportions of smokers and regular exercisers were signi cantly higher in men than those in women. In contrast, the average HDL cholesterol was lower in men than in women. Besides, 47 women and 182 men developed T2DM over a median follow-up period of 5.26 years and 5.88 years, respectively.
Univariate analysis of the association between incident diabetes and baseline characteristics The univariate analysis of the association between incident diabetes and baseline characteristics indicated that age, BMI, waist circumference, total cholesterol, fasting plasma glucose, triglycerides, the TyG index, fatty liver, blood pressure, and current smoking were all possible risk factors of incident diabetes. Meanwhile, the HDL cholesterol presented a protective effect in both women and men (Supplemental Table 1). Unlike women, men having a parent with diabetes may have a signi cantly higher risk for incident diabetes.

Strati ed analysis
Age, BMI, waist circumference, and fatty liver were identi ed risk factors for diabetes [28][29][30][31]. To evaluate and control for confounding factors, we measured the effect of the baseline TyG index on the risk of incident diabetes within each of the strata categorized according to the above-mentioned factors. In the analysis strati ed by age, BMI, waist circumference, and fatty liver, a positive association between the TyG index and the risk of incident diabetes was found in men, while no signi cant association was observed in females after adjusting for potential confounding factors (Fig. 1). These confounders included ethanol consumption, habit of exercise, HDL cholesterol, total cholesterol, smoking status, systolic blood pressure, diastolic blood pressure, and family history of diabetes, as well as age, BMI, waist circumference and fatty liver, except for the strati cation factor itself (Fig. 1). Moreover, the interaction analysis showed that age, BMI, waist circumference, and fatty liver did not play an interactive role in the association between incident diabetes and the TyG index in both genders (all P-values for interaction > 0.05) (Fig. 1).

Multivariable analysis
Multivariate regression analysis was conducted to assess the independent effects of the baseline TyG index on incident diabetes, ( Table 2). The risk of diabetes was strongly associated with the baseline TyG index (as a continuous variable) in the crude model in women (HR = 6.1, 95% CI 3.4 to 11.2; P < 0.001) and men (HR = 3.5, 95% CI 2.7 to 4.6; P < 0.001). The relationship remained statistically signi cant in women (HR = 2.7, 95% CI 1.4 to 5.1; P = 0.004) and men (HR = 2.0, 95% CI 1.5 to 2.6; P < 0.001) after adjusting for age, fatty liver, BMI, and waist circumference (multivariate-adjusted model 1, Table 2). Moreover, further adjustment for ethanol consumption, habit of exercise, HDL cholesterol, total cholesterol, smoking status, systolic blood pressure, diastolic blood pressure, and family history of diabetes did not alter the signi cant association in men (HR = 1.7, 95% CI 1.2 to 2.5; P = 0.002) (multivariate-adjusted model 2, Table 2). However, the risk between the TyG index and the risk of incident diabetes was not signi cant in women after further adjustment in multivariate-adjusted model 2 ( Table 2). We then divided the participants into tertiles according to the levels of TyG index. In both genders, the crude model, Model I or Model II did not show any typical dose-dependent positive relationship between the TyG index and the risk of incident T2DM, suggesting the existence of a nonlinear association.
Two-piecewise linear regression model analysis using a smoothing function Since the previous multivariable analysis suggested the nonlinear correlation between the TyG index and the risk of incident T2DM, a two-piecewise linear regression model was used to further elucidate their association using a smoothing function. Excitingly, adjusted smoothed plots showed a U-shaped association between the risk of incident diabetes and the TyG index in both genders (Fig. 2). According to the two-piecewise linear regression model, the TyG magnitude was signi cantly negatively correlated with the log Relative Risk (logRR) for incident diabetes with a TyG index of < 7.27 in women and < 7.97 in men after adjusting for confounding variables (Table 3 and Fig. 2). With the per-unit increase in the TyG index, the risk of T2DM decreased nearly 90% (HR = 0.09, 95% CI 0.01 to 0.93) in women and 80% (HR = 0.21, 95% CI 0.08 to 0.57) in men after adjusting for age, fatty liver, waist circumference, BMI, ethanol consumption, habit of exercise, HDL cholesterol, total cholesterol, smoking status, systolic blood pressure, diastolic blood pressure, and family history of diabetes. Furthermore, the risk of incident diabetes decreased to the lowest level as the TyG index increased up to the thresholds (TyG = 7.27 in women vs. TyG = 7.97 in men) ( Table 3 and Fig. 2). In contrast, the TyG magnitude was signi cantly positively associated with the risk of incident diabetes when higher than 7.27 (HR = 2.76, 95% CI 1.20 to 6.34) in women and 7.97 (HR = 2.42, 95% CI 1.66 to 3.53) in men (Table 3).

Discussion
To the best of our knowledge, this is the rst study to describe a U-shaped association between the TyG index and the risk of incident T2DM. Moreover, we revealed a turning point (TyG = 7.27 in women vs. TyG = 7.97 in men) using threshold effect analysis (Table 3 and Fig. 2). Besides, as far as we know, in Japan area, the is the rst study conducted to investigate the association between TyG and incident diabetes.
The association between the TyG index and the development of diabetes has been examined in previous studies in other areas [17,18,20,[32][33][34]. Ming Zhang et al. found that the risk of incident T2DM was increased with elevated TyG quartiles among 8003 Chinese participants [33]. Similarly, the Chungju Metabolic Disease Cohort (CMC) study revealed an increased risk of incident diabetes in participants with TyG index in Quartiles 2, 3 and 4, compared to those in Quarter 1 [32]. Consistently, different studies on a white European population, middle-aged Koreans, residents of the northern region of Singapore reported a signi cantly higher risk of incident diabetes among participants in the highest TyG quartile relative to the lowest quartile [17,18,20]. However, all of these studies did not exclude the individuals with prediabetes (impaired glucose tolerance or 5.7 ≤ HbA1c levels < 6.5% [35]).
In the present study, all the participants with impaired fasting glucose tolerance or HbA1c ≥ 5.7 were excluded, and some new insights were demonstrated. Overall, there was a U-shaped association between the TyG index and the risk of developing T2DM. Besides, the thresholds (TyG = 7.27 in women vs. TyG = 7.97 in men) were identi ed on which the risk of incident diabetes was the lowest in this population (Table 3 and Fig. 2). Well in line with certain previous studies [17,18,20,[32][33][34], the risk of developing T2DM signi cantly increased with the elevated TyG index in men with TyG > 7.97 and women with TyG > 7.29 (Table 3 and Fig. 2). However, a lower level of TyG index (< 7.27 in women and < 7.97 in men) substantially changed the association between the TyG index and incident diabetes. In participants with TyG levels lower than the thresholds, the risk of incident diabetes among women and men decreased nearly by 90% and 80%, respectively, with each increment in the TyG index after fully adjusting for confounders (Table 3 and Fig. 2). This has not been reported in the existing literatures.
Actually, consistent with our study ndings, previous cohort studies showed that after adjusting for confounders, the risk of incident diabetes increased signi cantly only in the third and fourth TyG quartiles [17,18,20,34],but not in the second quartile, when compared with that in the rst quartile [17,18,20]. This indicates a nonlinear association between the TyG index and incident diabetes. Furthermore, in the Vascular-Metabolic CUN cohort, the nonlinear relation became more obvious in the subgroup with the normal fasting glucose level in the strati ed analyses [17]. Interestingly, some previous studies suggested that the risk of incident Type 2 diabetes was elevated across all TyG quartiles [32,33]. However, most of these studies performed the analyses in all participants with fasting glucose level < 7 mmol/L and did not exclude those with impaired fasting glucose or with 5.7 ≤ HbA1c levels < 6.5% (prediabetes) [12,21,24] . Besides, the previous studies had a relatively small population scale. Hence, the nonlinear association between the TyG index and the risk of incident diabetes might have been neglected.
Although the U shape association between TyG and incident diabetes in apparent healthy individuals was rstly revealed in this study, there were several limitations that must be noted. Firstly, the current study was carried out only in a Japanese population, making the generalizability of its ndings to non-Japanese populations uncertain. This implies the need to verify our results by conducting future studies on other ethnicities. Secondly, the oral glucose tolerance test was not performed in this study, and the prevalence of incident diabetes might have been underestimated. Lastly, the effects of the TyG index on the risk of incident diabetes require further investigation.
Despite the limitations mentioned above, this study had several strengths. Firstly, the data set extracted from the NAGALA database was relatively large and complete, covering wide range of TyG levels. Secondly, this study was performed in an apparently healthy population because we excluded all individuals with HbA1c levels ≥ 5.7% or impaired fasting glucose. These two strengths allowed us to evaluate the association across an extensive TyG range in healthy individuals and establish the U-shaped association between the TyG index and incident diabetes. Thirdly, our analysis of the relationship between the TyG index and incident diabetes was adjusted for more potential confounding factors, when compared to previous studies, probably making the result more robust and reliable.

Conclusions
For the rst time, the current study reported that there was a U-shaped association between the TyG index and the risk of incident diabetes among healthy individuals, with a TyG threshold of 7.97 in men and 7.27 in women, raising a possibility of the TyG index as an intervention target for preventing incident diabetes.
This study is expected to encourage future clinical and mechanistic studies to con rm our current ndings and to better understand what roles the TyG index plays in the development of T2DM.

Consent for publication
Not applicable Figure 1 Page 20/20 The association between the TyG index and incident diabetes strati ed by age, fatty liver, waist circumference and BMI at baseline. Each strati cation was adjusted for ethanol consumption, habit of exercise, HDL cholesterol, total cholesterol, smoking status, systolic blood pressure, diastolic blood pressure and family history of diabetes as well as all other presented subgroups except the strati cation factor itself.

Figure 2
U shape relationship of the TyG index incident diabetes: women (A) and men (B). The solid black line represented the smooth curve t between the TyG index and incident diabetes. The dotted curves were the 95% CI of the t. The association was adjusted for age, fatty liver, waist circumference, BMI, ethanol consumption, habit of exercise, HDL cholesterol, total cholesterol, smoking status, systolic blood pressure, diastolic blood pressure and family history of diabetes.