This retrospective cohort study revealed that raised TyG index was independently correlated with greater risk of developing diabetes among apparently healthy adults
Table 4
Effect of magnitude of TyG index on diabetes risk stratified by subgroups.
| Characteristics | No. of participants | HR (95%CI) | P -value | P for interaction |
| Age (year) | | | | < 0.0001 |
| < 40 | 106,447 | 4.53 (3.76–5.45) | < 0.0001 | |
| >=40, < 60 | 71,176 | 3.54 (3.19–3.93) | < 0.0001 | |
| >=60 | 23,675 | 2.67 (2.37-3.00) | < 0.0001 | |
| Sex | | | | 0.0150 |
| Male | 109,236 | 3.16 (2.90–3.45) | < 0.0001 | |
| Female | 92,062 | 3.84 (3.37–4.37) | < 0.0001 | |
| BMI(kg/m2) | | | | < 0.0001 |
| < 18.5 | 11,593 | 3.64 (1.53–8.64) | 0.0034 | |
| >=18.5, < 24 | 112,241 | 4.13 (3.62–4.71) | < 0.0001 | |
| >=24, < 28 | 60,886 | 3.22 (2.90–3.58) | < 0.0001 | |
| >=28 | 16,578 | 3.07 (2.64–3.56) | < 0.0001 | |
| SBP(mmHg) | | | | < 0.0001 |
| < 140 | 181,383 | 3.48 (3.19–3.79) | < 0.0001 | |
| >=140 | 19,915 | 2.89 (2.53–3.29) | < 0.0001 | |
| DBP(mmHg) | | | | 0.9984 |
| < 90 | 185,636 | 3.29 (3.04–3.56) | < 0.0001 | |
| >=90 | 15,661 | 3.43 (2.90–4.05) | < 0.0001 | |
| Smoker | | | | 0.6979 |
| Now | 11,071 | 3.03 (2.36–3.90) | < 0.0001 | |
| Once | 2419 | 4.27 (2.30–7.91) | < 0.0001 | |
| Never | 43,497 | 3.40 (2.84–4.08) | < 0.0001 | |
| Not recorded | 144,311 | 3.39 (3.13–3.69) | < 0.0001 | |
| Drinker | | | | 0.2174 |
| Now | 1243 | 5.31 (2.34–12.05) | < 0.0001 | |
| Once | 8496 | 3.65 (2.52–5.28) | < 0.0001 | |
| Never | 47,248 | 3.33 (2.84–3.90) | < 0.0001 | |
| Not recorded | 144,311 | 3.39 (3.13–3.69) | < 0.0001 | |
| Family history of diabetes | | | | 0.1175 |
| No | 197,196 | 3.39 (3.15–3.65) | <0.0001 | |
| Yes | 4102 | 3.07 (2.10–4.50) | <0.0001 | |
Note 1
the model was adjusted for sex, age, BMI, LDL-C, TC, Scr, ALT, AST, SBP, DBP, drinking, smoking and family history of diabetes.
Note 2
the model was adjusted for all above variables except the corresponding stratification variable.
in China (HR, 3.34; 95% CI, 3.11–3.60). Besides, a significant nonlinear relationship was observed and showed the risk of diabetes tend to ascend with increase of TyG index. Compared with the lowest quartile, individuals with the top quartile of TyG index demonstrated a sixfold greater risk of developing diabetes (Q4 vs. Q1; adjusted HR 6.26, 95% CI 5.15–7.60). Additionally, the results of subgroup analysis revealed this correlation existed regardless of participants being male or female, younger or older, or obese or nonobese, suggesting our results were robust and the TyG index was suitable for a wide range of subjects. Moreover, stronger associations were observed in participants with age < 40 years, BMI ≥ 18.5 kg/m2 and < 24 kg/m2, or SBP < 140 mmHg, or in females.
The TyG index, derived from FPG and TG, was proven as a marker of IR in many epidemiological studies [9–13, 21]. Compared with HIEC, the TyG index had high sensitivity (96.5%) as well as good specificity (85.0%) for diagnosing IR in a Mexican population [10], and was a more accurate predictor than HOMA-IR in a Brazilian study [11]. Moreover, consistent with our results, several studies suggested that high TyG index was relevant to future risk of T2DM in different races, as shown in reports from Korea, Singapore, and Europe [15–17, 22]. Similar results were observed in another Chinese cohort study [14] and the trend of nonlinear relationship of TyG index with diabetes risk was generally consistent with our study. However, the study only included 5,706 subjects with normal BMI and was conducted in rural areas. Therefore, its generalizability is relatively limited. This study was based on a large cohort of 201,298 apparently healthy adults across 32 sites in 11 cities, and is clearly applicable to a relatively wide range of individuals, and provides a stronger basis for clinical promotion and application. Similarly, the risk of diabetes in a Singaporean population elevated progressively across TyG index quartiles (Q) from Q1 to Q4 (Q4 vs. Q1; adjusted HR 5.30, 95% CI 2.21–12.71) [17]. However, potential confounders, such as serum lipid index (LDL-C and TC), drinking, smoking and family history of diabetes, were not sufficiently adjusted, and were notablely relevant to high risk of diabetes [23–28]. Fortunately, these confounding factors were taken into consideration in our study to avoid potential effects on the results.
Subgroup analysis and exploration of interactions is critical for clinical research, to better understand the actual relationships between independent variables and dependent variables [29]. Unfortunately, the related studies described above only used sex, and/or age as stratification factors for subgroup analyses [14–16], and no interactions were observed, which may hinder our understanding of the real association of TyG index with future diabetes risk. In this study, these factors, including BMI, sex, age, DBP, SBP, drinking, smoking and family history of diabetes, were taken as stratified variables, and stronger associations were observed in participants with age < 40 years, BMI ≥ 18.5 kg/m2 and < 24 kg/m2, or SBP < 140 mmHg, or in females. This association was particularly obvious in females, and was consistent with the cohort study by Zhang et al [14]. This may be because serum lipids in female hepatocytes were higher than that in male hepatocytes under fasting and glucose lipid loading [30, 31]. In clinical practice, obese and older individuals are generally considered the primary targets for diabetes screening. However, based on subgroup analysis, the TyG index appeared to be more sensitive for predicting risk of diabetes in younger individuals and those with normal BMI or SBP, suggesting it may be promising for screening risk of future diabetes, especially in individuals without high-risk factors such as hypertension, obesity and older age.
Islet β-cell dysfunction and IR remain the core pathological trait of T2DM [32]. Interestingly, the TyG index, besides being a substitute of IR, is associated with susceptibility of β-cells to glucotoxicity and lipotoxicity. Evidence suggested that elevated glucose levels can induced reactive oxygen species generation on islet β-cells, which in turn cause oxidative stress and β-cells dysfunction, and then lead to IR and T2DM [33–35]. Other studies revealed that long-term high free fatty acid content was related to prolonged exposure of TG in pancreatic islets, which may impair pancreatic β-cell function [36–38]. Furthermore, glycotoxicity and lipotoxicity were interactive rather than independent adverse effects on pancreatic β-cell [39–41]. Long term exposure of pancreatic beta cells to high fatty acids concentrations could result in impaired glucose-induced insulin secretion [42, 43] and increased β-cell death [44]. An intervention study confirmed that patients with impaired glucose metabolism had improved insulin secretion ability after being treated with n-3 fatty acids [45]. Besides, IR is largely attributable to the impairment of insulin-stimulated glucose absorption into skeletal muscle. When TG levels in peripheral blood and skeletal muscle were significantly increased, glucose metabolism in skeletal muscle would be impaired [46]. Therefore, to a certain extent, the TyG index reflects muscle IR [47].
This study had several advantages. First, it was based on a large sample cohort study with broad age spectrum. Therefore, there were sufficient subjects for analysis to guarantee dependability and robustness of results. Furthermore, the results are applicable to a relatively wide range of individuals. Other similar cohort studies had relatively small sample sizes and populations that tended to be older. Second, taking TyG index as continuous variable and categorical variable respectively, sensitivity analysis and trend test were carried out to improve the reliability of results and avoid the contingency in data analysis. Finally, subgroup analyse and interaction test were conducted to further prove the dependability of the results and identify potential interactions with other variables. Interestingly, we observed that the TyG index appears to be more sensitive for predicting risk of diabetes in females, younger individuals, and those with normal BMI or SBP. Therefore, the TyG index may represent a reliable predictor for screening diabetes in individuals without high-risk factors such as older age, obesity, and hypertension. Obviously, more evidence is needed to further validate our results.
This study also had limitations. Firstly, diabetes was diagnosed depending on FPG ≥ 7.0 mmol / L or self-reported diabetes, rather than by glycosylated hemoglobin or 2-hour oral glucose tolerance test, which was probably underestimated. Secondly, this study did not distinguish between types of diabetes. However, these findings may be more applicable to T2DM, which accounts for approximately 90–95% of all diabetes cases. Thirdly, as this large cohort study was conducted in China, our findings can not be generalized to other races and certain populations, such as children and pregnant women. Finally, the present report was a secondary analysis on the basis of existing database, and although numerous confounding factors had been adjusted, some variables not included in the database, such as physical activity, dietary factors, and lipid-lowering agents, failed to be adjusted. Therefore, potential effects of these residual confounding factors on the results cannot be ignored.