Association of Triglyceride Glucose-Body Mass Index and Incident Diabetes Mellitus: A Secondary Retrospective Analysis Based On a Chinese Cohort Study

The triglyceride glucose-body mass index (TyG-BMI) has been proposed as a marker of insulin resistance (IR). However, evidence for the relationship between TyG-BMI and the incidence of diabetes mellitus remains limited. This study investigated the association between TyG-BMI and diabetes occurrence in Chinese individuals. This retrospective study included a cohort of 204978 non-diabetic individuals using data from healthy screening program data in China between 2010 and 2016. The independent and dependent variables are TyG-BMI and incident of diabetes, respectively. Cox proportional hazards regression analysis was used to evaluate hazard ratios (HRs) and 95% condence intervals (95% CIs) for the relationship between TyG-BMI and incident diabetes. Generalized additive models were used to identify non-linear relationships. Subgroup analysis helped better understand other factors that may affect the association between TyG-BMI and diabetes to identify potential special populations. And the data were downloaded from the DATADRYAD website. blood pressure; DBP:diastolic blood pressure. GAM:Generalized additive models; SD:standard deviation. IQR:interquartile range.

patients would reach 700.2 million [1]. China has the largest number of diabetic patients. Diabetes mellitus(DM) and its complications have resulted in a severe economic burden of mortality and disability.
Thus, Early identi cation and prediction of diabetes are crucial.
Type 2 diabetes(T2DM) is more common among obese individuals than nonobese individuals [1]. Insulin resistance (IR) and the consequences of compensatory hyperinsulinemia are vital pathological mechanisms of diabetes mellitus and obesity [2]. Thus, recognition of IR before the manifestation of clinical diabetes mellitus is of paramount importance.
Recently, triglyceride glucose body mass index(TyG-BMI), which combines triglyceride(TG), fasting plasma glucose(FPG), and obesity status, has been considered to identify IR more reliably than TyG. [3,4] Triglycerideglucose (TyG) index, which is estimated using the formula Ln(fasting triglycerides(mg/dl)× fasting blood glucose(mg/dl)/2), is an alternative for identifying insulin resistance in apparently healthy subjects [5][6][7][8][9]. Some studies revealed the TyG index was relevant with a high risk of diabetes. TyG-BMI = TyG × BMI [3]. Body mass index (BMI) is an easily detectable, inexpensive, and non-invasive measurement parameter closely related to IR. Ectopic obesity is the biggest risk factor for type 2 diabetes [10]. A Chinese study indicates that BMI is also independently related to the higher risk of diabetes among young people [11]. A Korean study suggests that triglyceride glucose-body mass index is a simple and clinically useful proxy for insulin resistance in non-diabetic individuals [11]. At present, there are few articles about the relationship between TyG-BMI and diabetes.
In this study, we performed a secondary data analysis based on previously published data. In that paper, the author investigated the association of body mass index and age with incident diabetes [11]. On this secondary analysis, TyG-BMI was used as an independent variable, and outcome variables and other covariates are consistent with those in the original research. The purpose of this study was to explore the association of TyG-BMI with incident diabetes.

Study population and design
In this population-based cohort analysis of a medical program established by the Rich Healthcare Group in China between 2010 and 2016, we investigated the effect of TyG-BMI on incident diabetes. Data were downloaded from the DATADRYAD website (www.datadryad.org), allowing others to use the database for free. In keeping with the terms of service, this research cites data packets shared by Chen Ying et al. [11]. The database materials included the following variables: sex, age, BMI, drinking, smoking, family history of diabetes, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), TG, FPG, Serum urea nitrogen(BUN), serum creatinine (Scr), aspartate aminotransferase (AST), alanine aminotransferase (ALT), systolic blood pressure (SBP), diastolic blood pressure (DBP), incident diabetes at follow up and follow-up time. The authors declared that they had relinquished copyright and relevant ownership of the database in the original paper. As for ethics approval, the study was a retrospective analysis approved by the rich healthcare group review committee.
The original study enrolled 685,277 participants ≥ 20 years old with at least two visits covering the period 2010-2016 across 32 sites and 11 cities in China (Shanghai, Beijing, Nanjing, Suzhou, Shenzhen, Changzhou, Chengdu, Guangzhou, Hefei, Wuhan, Nantong). The data we got has been ltered. Subjects with the following conditions were excluded: (1) missing information about weight, height, gender, fasting plasma glucose value at baseline; (2) extreme BMI values (< 15 kg/m 2 or > 55 kg/m 2 ); (3) visit intervals less than 2 years; (4)participants were diagnosed with diabetes at baseline and with unde ned diabetes status at follow-up. Finally, in the original study, Ying Chen et al. [11] kept 211,833 participants. We excluded missing values of baseline TG(n = 4,887) and zero values of baseline TG(n = 860) from the analysis cohort for further research. And then, TyG-BMI was calculated as the formula BMI×Ln(fasting triglycerides(mg/dl)×fasting blood glucose(mg/dl)/2). We excluded outliers of TyG-BMI(< means minus three standard deviation (SD) or > means plus three SD) (n = 1,108). Finally, a sum of 204978 participants was selected in our study. Figure 1 depicted the participants' selection process.

Data Collection And Measurements
The original database contained participants' clinical history and lifestyle factors based on a standardized questionnaire regarding demographic characteristics, lifestyle factors, personal medical history, and family history of chronic disease. Trained staff measured height, weight, and blood pressure.
Body weight was measured in light clothing with no shoes to the nearest 0.1 kg. Height was measured to the nearest 0.1 cm. BMI was derived from weight in kilograms divided by height in meters squared. Standard mercury sphygmomanometers measured blood pressure. Fasting venous blood samples were collected after at least a 10 hours fast at each visit. TG, TC, LDL-C, HDL-C and Plasma glucose levels were measured on an autoanalyzer (Beckman 5800). Plasma glucose levels were measured by the glucose oxidase method. The formula of the TyG index was Ln(fasting triglycerides(mg/dl)×fasting blood glucose(mg/dl)/2)).The target-independent variable is TyG-BMI, which equals the BMI×TyG index. The dependent variable is incident diabetes, which was de ned as fasting plasma glucose ≥ 7 mmol/L, and/or self-reported during follow-up. As this is a retrospective cohort study, it reduced the possibility of selection bias and observation bias.

Statistical analysis
The continuous variable of the normal distribution was represented by mean with standard deviation (SD), and the continuous variable of non-normal distribution was replaced by median with interquartile range (IQR). Categorical variables were expressed as frequencies and percentages. For the handling of missing values, missing continuous variables were replaced with a mean or median depending on the distribution. Missing categorical variables could be a new categorical group. Strati ed by TyG-BMI index quartiles, statistical differences of the groups were described with one-way ANOVA (normal distribution), Kruskal Wallis H (skewed distribution) test, and chi-square test (categorical variables). To explore the relationship between TyG-BMI and incident DM, univariate and multivariate Cox proportional hazard models were applied and adjusted hazard ratios (HRs) with 95% con dence intervals (CIs). According to the recommendation of the STROBE statement, We used three models: crude model; model I adjust for: Age, Gender, SBP, DBP, Smoking Status, Drinking Status, Family History; model II adjust for: Age, Gender, SBP, DBP, Smoking Status, Drinking Status, Family History, TC, HDL, LDL, ALT, AST, Scr. Sensitivity analysis was used to ensure the robustness of data analysis. The TyG-BMI was converted into a categorical variable and calculated the P for trend to perform the linear trend tests. We used the Cox proportional hazards regression model with cubic spline functions to identify non-linear relationships. In addition, If there was obvious in a smoothed curve, the recursive method automatically calculates the in ection point. The associations of TyG-BMI with incident diabetes in subgroups were also studied using a strati ed linear regression model and likelihood ratio test to nd modi cations and interactions. The subgroups were classi ed by age (20 to < 30, 30 to < 40, 40 to < 50, 50 to < 60, 60 to < 70, ≥70), gender (male vs. female), HDL( low, middle, high, not recorded), LDL( low, middle, high, not recorded), SBP(< 140, ≥ 140 ), DBP(< 90, ≥ 90), Smoking status(current smoker, ever smoker, never smoker, not recorded), Drinking status(current drinker, ever drinker, never drinker, not recorded), Family history of diabetes(no, yes). Survival estimates and cumulative event rates were compared using the Kaplan-Meier method by using the time-to-rst event for each endpoint.
All of the analyses were performed with the statistical software package R (http://www.R-project.org, The R Foundation) and Empower-Stats (http://www.empowerstats.com, X&Y Solutions, Inc., Boston, MA). Signi cance was accepted at a two-tailed P < 0.05.

Result
Baseline characteristics of the study participants Of the 211,833 subjects recruited in the former study, 204,978 participants were included in the current analysis (described in Fig. 1). The mean age of the population was 42.2 ± 12.7 years old, and 45.17% of participants were women. The mean year of follow-up was 3.1 ± 0.9 years, and 4093 participants happened diabetes during follow-up. The mean TyG-BMI was 158.7 ± 32.4, and the mean FPG, BMI, TG were 88.6 ± 10.9 mg/dl, 23.3 ± 3.3 kg/m 2 and 24.2 ± 18.6 mg/dl respectively. Individuals in the highest TyG-BMI group (Q4) were generally older than those in the lowest TyG-BMI group (Q1) and had higher BMI, FPG, TG, SBP, DBP, TC, ALT, SCR values. What's more, with the increase of TyG-BMI value, the incidence of diabetes increased gradually. (Q1: 0.23% vs. Q2: 0.62% vs. Q3: 1.74% vs. Q4: 5.35%). Compared with the Q1 group, the Q4 group had lower HDL levels, higher AST and LDL levels, higher rates of smoking, drinking, family history. As shown in Table 1.

Univariate Analysis
The results of the univariate analysis were shown in Table 2. Table 2 showed that men were more likely to develop diabetes than women, and age, BMI, FPG, TG, TyG-BMI, SBP, DBP, TC, LDL, ALT, AST, SCR, smoking, drinking, and family history were all positively associated with incident diabetes. In Fig. 2, the cumulative risk of incident diabetes Kaplan Meier curves strati ed by TyG-BMI showed that the cumulative risk of diabetes increased gradually with increasing TyG-BMI. There was a signi cant difference in diabetes risk between the TyG-BMI quartile groups (log-rank test P < 0.0001).

The Multivariate Analysis Of Tyg-bmi With Dm Risk
To evaluate group differences in the association between TyG-BMI and incident diabetes, we applied Cox proportional hazards models, and Table 3 showed the unadjusted and adjusted models. In crude model, TyG-BMI had a positive correlation with diabetes incidence (HR = 1.026, 95% con dence interval ( the risk of diabetes in the Q4 group increased by 10.261 times compared to the Q1 group, and the trend in the quartile was signi cant (trend P < 0.00001).  Fig. 3). The Results Of Subgroup Analyses  Metabolic disorders, such as obesity, hyperglycemia, hypertension, dyslipidemia, and hyperinsulinemia, are the pathological basis of cardiovascular and cerebrovascular diseases and diabetes. In developing type 2 diabetes, decreased β cell function and insulin resistance are the main events [12]. Adipose tissue is a complex and highly active metabolic and endocrine organ [13]. Elevated blood glucose and lipid levels have toxic effects on beta cells and interfere with normal glucose metabolism. Insulin resistance triggers hyperinsulinemia, and hyperinsulinemia, in turn, causes insulin resistance [2]. It is generally accepted that insulin resistance is closely related to the risk of type 2 diabetes. Clinically, the gold standard of insulin resistance is the glucose clamp test, which is inconvenient and expensive. Although homeostasis model assessment of insulin resistance (HOMA-IR) has a wide range of clinical applications, its application is limited due to its relatively high cost and low repeatability. Simental-Mendía et al. [3] Proposed the concept of the TyG index, which indicated that the TyG index could be used as an alternative index of insulin resistance in healthy subjects. Many studies have shown that the TyG index is a good alternative marker of insulin resistance. [4,[14][15][16][17] Compared with the HOMA-IR index, the TyG index has higher sensitivity in recognizing insulin resistance [15]. TyG index is regular and easy to get, including FBG and TG, which have been associated with diabetes risk [8,9,14,18]. In addition to the TyG index, the relationship between obesity and diabetes is also well documented. The prevalence of type 2 diabetes mellitus is rapidly increasing, in parallel with the current obesity epidemic. The incidence rate of type 2 diabetes is lower in non-obese patients [19]. BMI is a simple, economical, and useful indicator of general obesity. A cross-sectional study of the Taiwan population shows that TyG-BMI (a combination of TyG index and BMI) is an effective marker for early recognition of insulin resistance [3]. A recent study involving 511 individuals indicated that TyG-BMI was a stronger predictor of IR than TyG-WC [3]. In a Nigerian cross-sectional study [20], in all 473 participants, TyG-BMI shows larger AUC for metabolic syndrome detection (0.838, 95% CI: 0.802-0.870) than TyG index (0.796, 95% CI: 0.757-0.831). After adjusting for gender, age, smoking, SBP and DBP, only the TyG index and TyG-BMI signi cantly predicted metabolic syndrome in men [20]. However, a Korean retrospective study involving 11,149 people showed that TyG-BMI was superior to the other parameters(TyG index, TyG-WC, TyG-WHtR) in predicting insulin resistance [21]. These conclusions are similar to ours. Firstly, consistent with previous studies, TyG-BMI was positively correlated with the incidence of diabetes. Then we further analyzed and discovered the curvilinear relationship. After adjusting for confounding factors in our study, the association between TyG-BMI and incident diabetes was nonlinear by using two piecewise linear regression models. The in ection point of GAM was 232.416 after adjusting for potential confounding factors (age, gender, SBP, DBP, TC, HDL, LDL, ALT, AST, SCR, smoking, SBP). We found that the in ection point had a stronger relationship on the left side of the in ection point. Therefore, controlling TyG-BMI is more valuable for reducing the risk of diabetes under the in ection point.
Our study has some strength: (1).To our knowledge, this is the rst study to assess the association between TyG-BMI and incident diabetes in the Chinese population; (2). Compared with other researches, our sample size is relatively larger, which can better represent the Chinese people; (3). This study was a retrospective cohort study, which reduced selection bias and observation bias; (4). We found the nonlinear relationship and made a deeper discussion, and there are also more confounding factors for adjustment; (5). To make the results more robust, TyG-BMI was treated both as a continuous and categorical variable.
The study also has some potential limitations: (1). The data was from the Rich Healthcare Group in China, representing the Chinese population, and couldn't be extended to other races and particular groups like pregnant women and children. (2). This research was based on a secondary analysis of published data, variables were limited to the original study's data, some other important variables such as hip circumference, medication history, hemoglobin A1C, physical activity, dietary factors were not included.
(3). The incidence of diabetes may underestimate because of the study's diabetes de nition, which did not conduct a 2-hour oral glucose tolerance test. But for such a large cohort, It is a vast project to improve participants' oral glucose tolerance test. (4). The study did not differentiate diabetes types. But these conclusions may be more applicable to type 2 diabetes which accounts for approximately 90ariablesdiabetes patients. (5). According to TyG-BMI, we only measured it at baseline, not measured over time. In the future, we can consider more variables and a longer follow-up with a more re ned method.

Conclusion
The association between TyG-BMI and incident diabetes is positive and nonlinear after correcting the   Kaplan-Meier event-free survival curve. Kaplan-Meier analysis of incident of diabetes based on TyG-BMI quartiles (logrank, P < 0.0001).