During a median of 3.1-year follow-up, 2512 participants developed diabetes. Among the 114,854 participants, the average age was 44.1 ± 12.9 years, with 53.3% male and 46.7% female. The mean values of TG/ non-HDL-c,FPG and BMI were 0.4±0.3, 4.9±0.6 mmol/L and 23.3±3.3 kg/m2, respectively. The number of participants with missing data of Scr, DBP, SBP, and LDL-c was 1341(1.1675%), 18(0.0156%),18(0.0156%),and 192(0.1671%), respectively. Among the classification variables, the missing drinking status and smoking data were 84,169(73.283%) and 84,169(73.283%).
Baseline characteristics of participants
Table 1 shows the baseline characteristics of the different TG/non-HDL-C subgroups. According to TG/ non-HDL-c levels (<0.24, 0.24-0.33, 0.33-0.46, ≥0.46) divided the participants into four subgroups. According to the results, participants in the subgroup with the highest TG/ non-HDL-c had higher age, BMI, blood pressure levels (including diastolic and systolic blood pressure), TG, LDL-c, FPG, AST, ALT, SCR, as well as higher rates of never and ever drinking alcohol and smoking. In addition, HDL-c levels were lower in the highest TG/ non-HDL-c group. Family history of diabetes was not statistically significant in different TG/ non-HDL-c groups.
The results of the univariate analysis:
In Table 2, we regard TG/non-HDL-c as a continuous variable. The results of the univariate analysis showed that BMI, age, TC, LDL, FPG, SBP, DBP, were positively correlated with incident diabetes. Participants with a family history of diabetes and current smoking and drinking status had a higher risk of developing diabetes. We also found that gender is a factor in the development of diabetes, with women having a lower risk of diabetes than men.
The results of the relationship between TG/non-HDL-C and incident diabetes
The Cox proportional risk regression model were used to explore the correlation between TG/non-HDL-c and the incidence of diabetes. At the same time, we showed the non-adjusted and three adjusted models in Table 3. In crude model, TG/non-HDL-c was positively correlated with incident diabetes (HR= 13.424, 95% confidence interval (CI): 11.447 ~ 15.743, P <0.00001). In the minimally adjusted model (adjusted BMI, gender, age, SBP, DBP, smoking, family history of diabetes,and drinking status), The positive correlation with incident diabetes was weakened (HR: 4.592, 95%CI: 3.840-5.492; P <0.00001). After adjusting for the full model (adjusted BMI, gender, age, SBP, DBP, FPG, TG, LDL, ALT, AST, BUN, Scr, family history of diabetes, smoking, and drinking status), The results also changed, but the positive correlation remained (HR=2.815, 95%CI: 2.330 to 3.399, P <0.00001). The results showed that for each unit increase in TG/non-HDL-c, the risk of developing diabetes increased by 181.5%.
Then, TG/non-HDL-c ratio was converted into four categorical variables (quartile method). Compared with the lowest quartile (TG/non-HDL-c<0.24) in the full model, the risk of diabetes in the highest quartile (TG/non-HDL-c≥0.46) was increased by 67.1%, and the trend of the four quartiles was found to be significant (P for trend<0.00001).
Figure 2 shows the Kaplan-Meier curve of the incident diabetes-free survival in each quartile. There was a significant difference in the probability of diabetes-free survival among the four quartiles (Log-Rank test, P <0.0001). The higher TG/non-HDL-c quartile had a higher cumulative risk of diabetes.
The result of the two-piecewise linear regression model
In the present study, to more accurately understand the relationship between TG/non-HDL-c and the risk of diabetes, we also made a generalized Additive model (GAM), which showed the potential connection between TG/non-HDL-c and the risk of diabetes was not simply linear. (figure 3) (after adjusted BMI, gender, age, SBP, DBP, Scr, TC, LDL, FPG, drinking status, smoking and and family history of diabetes). After using the two-Piecewise Linear Regression Model, we noted that the inflection point of TG/non-HDL-c was 0.544 (Loglikelihood ratio test P < 0.001). When TG/non-HDL-c value was less than 0.544, indicating a strong positive correlation between TG/ non-HDL-c and the incidence of diabetes (HR: 5.888 CI:(3.959, 8.757) p<0.0001). When TG/non-HDL-c value was greater than 0.544, however, TG/non-HDL-c and diabetes risk tended to flatten (HR: 1.509;95%CI: (1.056, 2.156); p<0.0238) (Table 4).
The results of the subgroup analysis
Subgroup analysis is necessary to evaluate the correlation between TG/non-HDL-c and the risk of diabetes in participants at different levels of each variable or to accurately determine the factors influencing TG/non-HDL-c and the risk of diabetes. We considered gender, age, BMI, age, SBP, DBP, family history of diabetes, drinking status and smoking, as the stratification variables to explore the trend of effect sizes in these variables (Table 5). Among them, the variable with obvious interaction relationship is gender, BMI, FPG, SBP, and drinking status (all P values for interaction < 0.05). and the stronger association were detected among women and the participants with FPG < 6.1 mmol/L(HR=4.060;95%CI:( 3.140, 5.249);P< 0.00001), BMI (≥18.5, < 24 kg/m2) (HR=4.452;95%CI:( 3.110, 6.372) P< 0.0001)、SBP<140mmhg (HR=3.233; 95%CI: (2.582, 4.049); P< 0.0001) and the participants who never drinking(HR=6.467;95%CI: (1.186, 35.281);P< 0.0310) . In addition, a weak association were detected among males and FPG ≥ 6.1 mmol/L(HR=2.208;95%CI:( 1.699, 2.870) P= 0.0008), BMI (>=24,<28kg/m2) (HR=2.705;95%CI:( 2.052, 3.565) P<0.0001) and ≥ 28 kg/m2(HR=1.880;95%CI:(1.302, 2.713) P=0.0008), SBP>=140mmHg(HR=1.950;95%CI:(1.410, 2.698) P<0.0001)、the participants who current drinking(HR=2.274;95%CI:( 1.432, 3.613) P=0.0005).