Temporal relationships between blood glucose, lipids and BMI, and their impacts on atherosclerosis: a prospective cohort study

Characteristics of the subjects

With baPWV=1400 cm/s as the cut-off point, subjects were divided into low and high AS risk groups, the average baPWV were 1238.79 cm/s and 1671.16 cm/s, respectively. Baseline and follow-up characteristics of the subjects with different AS risk are shown in table 1. Compared with the subjects with low AS risk, the subjects with high AS risk were older, and had higher baseline and follow-up FBG, 2-h PG, TC, TG, LDL-c, BMI, WC and lower baseline HDL-c. There were higher percentage of men and smokers in subjects with high AS risk.

Logistic regression analysis between blood parameters, BMI and AS risk

Table 2 showed the associations between baseline and follow-up blood parameters, BMI and AS risk, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake. Baseline and follow-up FBG, 2-h PG, TC, TG, LDL-c and BMI were all positively associated with AS risk, while baseline and follow-up HDL-c were reversely associated with AS risk.

Cross-lagged path analysis of FBG, 2-h PG, TC, TG, HDL-c, LDL-c and BMI

Figure 1 presented cross-lagged path analysis of the associations between FBG, 2-h PG, TC, TG, HDL-c, LDL-c and BMI. After adjusting for age, sex, drinking, smoking, labour intensity and total energy intake, the path coefficients from baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c to the follow-up BMI (β
₁=0.478 for FBG, β
₁=0.260 for 2-h PG, β
₁=0.230 for TC, β
₁=0.377 for TG, β
₁=−3.593 for HDL-c, β
₁=0.534 for LDL-c; all p <0.001) were significantly greater than the path coefficients from baseline BMI to the follow-up FBG, 2-h PG, TC, TG, HDL-c and LDL-c (β
₂=0.035 for FBG, β
₂=0.125 for 2-h PG, β
₂=0.013 for TC, β
₂=0.091 for TG, β
₂=−0.042 for HDL-c, β
₂=0.024 for LDL-c; all p <0.001), with all p <0.001 for the differences between β
₁ and β
₂ as shown in online supplemental table S1.

Figure 1

Cross-lagged path analysis of the associations between (A) FBG, (B) 2-h PG, (C) TC, (D) TG, (E) HDL-c, (F) LDL-c and BMI, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake. β
₁ represents cross-lagged path coefficients from baseline FBG, 2-h PG, TC, TG, HDL-c or LDL-c to follow-up BMI; β
₂ represents cross-lagged path coefficients from baseline BMI to follow-up FBG, 2-h PG, TC, TG, HDL-c or LDL-c; r
₁ represents synchronous correlations; r
₂ and r
₃ represent tracking correlations; and R
² represents variance explained. ^†P <0.001 for coefficients being different from 0. ^§Difference between β
₁ and β
₂ for being different from 0. BMI, body mass index; FBG, fasting blood glucose; HDL-c, high-density lipoprotein cholesterol; LDL-c, low-density lipoprotein cholesterol; TC, total cholesterol; TG, triglyceride; 2-h PG, 2-hour postprandial glucose.

Mediation analysis

Figure 2 showed mediation analysis of the effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk via follow-up BMI, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake. The total effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk measured as standardised regression coefficient (β
_tot=0.033 for FBG, β
_tot=0.018 for 2-h PG, β
_tot=0.046 TC, β
_tot=0.030 for TG, β
_tot=−0.132 for HDL-c, β
_tot=0.053 for LDL-c, all p <0.001) were estimated without follow-up BMI in the models. The direct effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk were all statistically significant (β
_Dir=0.022 for FBG, β
_Dir=0.013 for 2-h PG, β
_Dir=0.040 for TC, β
_Dir=0.023 for TG, β
_Dir=−0.069 for HDL-c, and β
_Dir=0.042 for LDL-c, all p <0.001). The β
₁ and β
₂ were used to calculate the indirect effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c (β
_ind=0.011 for FBG, β
_ind=0.005 for 2-h PG, β
_ind=0.006 for TC, β
_ind=0.007 for TG, β
_ind=−0.063 for HDL-C, β
_ind=0.011 for LDL-c; all p <0.001) on AS risk via follow-up BMI, respectively. The percentages of the indirect effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk via follow-up BMI were estimated at 33.3%, 27.8%, 13.0%, 23.3%, 47.7% and 20.8%, respectively.

Figure 2

Mediation analysis of the effects of baseline (A) FPG, (B) 2-h PG, (C) TC, (D) TG, (E) HDL-c or (F) LDL-c on atherosclerosis (AS) risk via follow-up BMI, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake. Left and right brachial ankle pulse wave velocity (baPWV) was measured, and the average value was used for analysis, and baPWV was dichotomised at a cut-off point of baPWV=1400 cm/s, representing different AS risk (baPWV ≤1400 cm/s was low AS risk, baPWV >1400 cm/s was high AS risk).^†P <0.05 for coefficients being different from 0. BMI, body mass index; FBG, fasting blood glucose; HDL-c, high-density lipoprotein cholesterol; LDL-c, low-density lipoprotein cholesterol; TC, total cholesterol; TG, triglyceride; 2-h PG, 2-hour postprandial glucose.

Multiple mediation analysis

From the above results, we found that increased FBG, 2-h PG, TC, TG and LDL-c could increase AS risk via increasing BMI, while increased HDL-c could decrease AS risk via decreasing BMI. To compare the indirect effects intensity of baseline blood parameters positively associated with AS risk, a multiple mediation analysis was performed (figure 3). The standardised regression coefficient of total effect of baseline FBG, 2-h PG, TC, TG and LDL-c on AS risk without follow-up BMI in the models was 0.138, and p <0.001. The standardised regression coefficient of direct effects of FBG, 2-h PG, TC, TG and LDL-c was 0.005, 0.009, 0.002, 0.022 and 0.042 (all p <0.001), respectively. The standardised regression coefficient of indirect effects of baseline FBG, 2-h PG, TC, TG and LDL-c on AS risk via follow-up BMI was 0.005, 0.003, 0.017, 0.009 and 0.024 (all p <0.001), respectively. The percentages of the indirect effects of baseline FBG, 2-h PG, TC, TG and LDL-c on AS risk via follow-up BMI were 3.6%, 2.2%, 12.3%, 6.5% and 17.4%, respectively. Therefore, the indirect effect intensity of baseline blood parameters on AS risk via follow-up BMI from strong to weak was LDL-c>TC>TG>FBG>2-h PG.

Figure 3

Multiple mediation analysis of the effects of baseline FBG, 2-h PG, TC, TG and LDL-c on atherosclerosis (AS) risk via follow-up BMI, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake. Left and right brachial ankle pulse wave velocity (baPWV) was measured, and the average value was used for analysis, and baPWV was dichotomised at a cut-off point of baPWV=1400 cm/s, representing different AS risk (baPWV ≤1400 cm/s was low AS risk, baPWV >1400 cm/s was high AS risk).^†P <0.05 for coefficients being different from 0. BMI, body mass index; FPG, fasting plasma glucose; LDL-c, low-density lipoprotein cholesterol; TC, total cholesterol; TG, triglyceride; 2-h PG, 2-hour postprandial glucose.

Sensitivity analysis

Four sensitivity analyses were conducted in the study. The first sensitivity analysis obtained 95% percentile bootstrap CIs to evaluate sensitivity to the distributions of cross-lagged path coefficients shown in figure 1. The results indicated that both β
₁ (the path coefficients from baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c to the follow-up BMI) and β
₂ (from baseline BMI to the follow-up FBG, 2-h PG, TC, TG, HDL-c and LDL-c) were statistically significant, and further verified that β
₁ were significantly greater than β
₂.

The second sensitivity analysis verified yearly variations in FBG, 2-h PG, TC, TG, HDL-c, LDL-c and BMI according to quartiles of their baseline values to validate the results of cross-lagged path analyses, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake (online supplemental figure S2). The yearly variation in BMI significantly increased with the increase of quartiles of baseline FBG, 2-h PG, TC, TG, LDL-c and the decrease of quartiles of baseline HDL-c (p _{for trend}<0.001 for FBG, p _{for trend} =0.002 for 2-h PG, p _{for trend} =0.007 for TC, p _{for trend} =0.007 for TG, p _{for trend} =0.005 for HDL-c, p _{for trend} =0.009 for LDL-c); however, the variations in FBG, 2-h PG, TC, TG, HDL-c and LDL-c did not show significant trend with the variation of quartiles of baseline BMI (p _{for trend} =0.554 for FBG, p _{for trend} =0.725 for 2-h PG, p _{for trend} =0.207 for TC, p _{for trend} =0.624 for TG, p _{for trend} =0.635 for HDL-c, p _{for trend} =0.863 for LDL-c). These results were consistent with the unidirectional relationships from baseline FBG, 2-h PG, TC, TG, HDL-c, LDL-c to follow-up BMI shown in figure 1.

The third sensitivity analysis was performed to analyse the impacts of the associations between FBG, 2-h PG, TC, TG, HDL-c, LDL-c and WC on AS risk using mediation analysis, to validate the impacts of the relationships between FBG, 2-h PG, TC, TG, HDL-c, LDL-c and BMI on AS risk, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake (online supplemental figure S3). The standardised regression coefficient of total effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk without follow-up WC in the models was 0.033, 0.018, 0.045, 0.030, –0.131 and 0.052 (all p <0.001), respectively. The standardised regression coefficient of direct effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk was 0.024, 0.013, 0.038, 0.023, –0.076 and 0.040 (all p <0.001), respectively. The standardised regression coefficient of indirect effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk via follow-up WC was 0.009, 0.005, 0.007, 0.007, –0.055 and 0.012 (all p <0.001), respectively. The percentages of the indirect effects of baseline FBG, 2-h PG, TC, TG, HDL-c and LDL-c on AS risk via follow-up WC were estimated at 27.3%, 27.8%, 15.6%, 23.3%, 42.0% and 23.1%, respectively. These results indicated that increased FBG, 2-h PG, TC, TG and LDL-c could increase AS risk via increasing WC, while increased HDL-c could decrease AS risk via decreasing WC, which were consistent with the results shown in figure 2.

The fourth sensitivity analysis was performed to compare the effect intensity of the associations between FBG, 2-h PG, TC, TG, LDL-c and WC on AS risk using multiple mediation, to validate the effect intensity of the relationships between FBG, 2-h PG, TC, TG, LDL-c and BMI on AS risk, adjusted for age, sex, drinking, smoking, labour intensity and total energy intake (online supplemental figure S4). The standardised regression coefficient of total effect of baseline FBG, 2-h PG, TC, TG and LDL-c on AS risk without follow-up WC in the models was 0.137, and p <0.001. The standardised regression coefficient of direct effects of FBG, 2-h PG, TC, TG and LDL-c on AS risk was 0.007, 0.009, 0.004, 0.023 and 0.043 (all p <0.001), respectively. The standardised regression coefficient of indirect effects of baseline FBG, 2-h PG, TC, TG and LDL-c on AS risk via follow-up WC was 0.003, 0.002, 0.015, 0.008 and 0.023 (all p <0.001), respectively. The percentages of the indirect effects of baseline FBG, 2-h PG, TC, TG and LDL-c on AS risk via follow-up WC were 2.2%, 1.5%, 10.9%, 5.8% and 16.8%, respectively. Therefore, the indirect effect intensity of baseline blood parameters positively associated with the risk of AS from strong to weak was LDL-c>TC>TG>FBG>2-h PG. These results were consistent with the results shown in figure 3.