The association between blood glucose levels and lipids in the general adult population: results from NHANES (2005–2016)

Abstract

Objective

Although abnormal lipid metabolism is one of the major risk factors for diabetes, the correlation between lipids and glucose is rarely discussed in the general population. The differences in lipid-glucose correlations across gender and ethnicity have been even more rarely studied. We examined the association between fasting blood glucose (FBG) and lipids, including triglycerides (TG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), and apolipoprotein B (ApoB), using 6,093 participants aged 20 years or older from the National Health and Nutrition Examination Survey (NHANES).

Methods

Analyses were performed using multiple logistic regression and generalised additive models.

Results

When other confounders were considered, we found that fasting glucose was positively correlated with triglycerides and negatively correlated with HDL-C, whereas total cholesterol, LDL-C cholesterol, and fasting glucose were related to each other in a U-curve fashion, with inflection points of 5.17 mmol/L and 2.3 mmol/L, respectively.This relationship persisted in subgroups of different sexes and races. A positive correlation was found between fasting glucose and ApoB, but subgroup analyses revealed that this relationship was not correlated across gender and race.

Conclusion

In the general population, fasting blood glucose levels were positively correlated with TG, negatively correlated with HDL-C, and U-shaped with total cholesterol and LDL-C. The likelihood of developing diabetes was 40% higher when LDL-C was greater than 2.3 mmol/L than in patients with LDL-C less than 2.3 mmol/L.

Introduction

Diabetes mellitus is a metabolic disease characterized by elevated blood glucose, and poor glycemic control predisposes to a variety of critical diseases [1]. Diabetic patients are usually accompanied by other diseases or characteristics, such as obesity or overweight, hypertension, high serum low-density lipoprotein (LDL) cholesterol, and low serum high-density lipoprotein (HDL) cholesterol. As one of the main sources of energy in the human body, there is a close relationship between blood glucose and blood lipids, which are essential for maintaining life activities, so the relationship between blood lipids and blood glucose has been a topic of great interest in the field of clinical nutrition.

Several studies [2,3,4] have shown that dyslipidemia may promote the onset and progression of diabetes and increase the risk of cardiovascular disease in patients. The hyperglycemic state of diabetic patients is usually accompanied by hyperlipidemia, while lipid-lowering with statins may increase blood glucose [5], thus the relationship between blood glucose and lipids is complex. The current view is that lipotoxicity-induced insulin resistance due to abnormally elevated lipids is an important cause of hyperlipidemia leading to hyperglycemia [6]. The interplay between hyperlipidemia and diabetes makes the regulation of lipids and blood glucose particularly important. Lipids can affect the stability of blood glucose, while fluctuations in blood glucose may influence the development of hyperlipidemia [7, 8].

However, most of the current studies have been conducted on diabetic patients and there was a lack of studies on the correlation between lipids and blood glucose in the general population. Therefore, we would like to study in detail the correlation between different lipid indices and fasting blood glucose levels in the general population, which will help to understand the co-morbid mechanisms of lipid-glucose related diseases, better manage lipid levels in dysglycaemic populations, reduce complications, and establish a theoretical basis for clinical decision-making that proper regulation of lipid and blood glucose levels contributes to the public prevention of glyco-lipid dysfunctions. This study evaluated the correlation between lipids and fasting glucose using a representative sample of adults from the National Health and Nutrition Examination Survey (NHANES) in the U.S. The aim of this study is to provide early warning of diabetes in the general population by monitoring lipid levels, and to guide the general population to better control their lipids and blood glucose.

Materials and methods

Study population

NHANES is a representative study that employs a complex and multi-stage probability sampling strategy to collect dietary and health data from the US population. The study is based on a total of six cycles of data from 2005 to 2016 [9,10,11,12], which cover information such as triglycerides (TG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), and apolipoprotein B (APOB) and glucose measurement information as well as OGTT testing data. However, it is important to note that some of the tests in NHANES are limited to adults over 20 years of age, and therefore our study was limited to this age group. After excluding those participants with missing routine blood test data, lipid data, follow-up data, and weighting data (n = 28,087), 6,093 participants were finally included in our study.

Data collections and definition

The exposure variables examined in this study were lipid levels, specifically TG, TC, LDL-C, HDL-C, and APOB, which were tested by the Roche/Hitachi Cobas 6000 analyzer series. When considering these lipid metrics, NHANES was only performed on participants who fasted for at least 8.5 h or more. The outcome variable was fasting blood glucose level. A range of covariates were considered in the analyses to control for the effects of potential confounders, including age, sex, race, household poverty-to-income ratio (PIR), smoking habits, alcohol consumption, education, history of hypertension and heart disease, history of diabetes mellitus (DM), body mass index (BMI), waist circumference, and systemic immunoinflammatory (SII) index.The SII index was calculated as: The SII index (× 10⁹/L) = neutrophil count (× 10⁹/L)/lymphocyte count (× 10⁹/L) × platelet count (× 10⁹/L) [13].

In addition we additionally counted the plasma atherogenic index (AIP) and triglyceride-glucose (TyG) index in these participants, and the AIP was obtained by Log10 (TG (mg/dL)/HDL-C (mg/dL)) [14]. In contrast, TyG was calculated as ln(fasting TG (mg/dL) × FBG (mg/dL)/2) [15].

More detailed information about the factors involved in this study can be obtained by consulting the NHANES (https://www.cdc.gov/nchs/nhanes/index.htm) website.

Statistical analysis

To explore differences between the data, we performed weighted and variance estimation analyses. Weighted multiple logistic regression models were used to assess the correlation between lipids and fasting glucose. To determine differences between groups, we used weighted chi-square tests for categorical variables and weighted linear regression models for continuous variables. Multiple regression was used for subgroup analysis. Smooth curve fitting and generalised additive models were used to analyse the non-linear relationship between fasting glucose and lipids. In the nonlinear model, a recursive algorithm was used to calculate the turning points in the relationship between lipids and fasting glucose, followed by a segmented linear regression model on either side of the turning points. All analyses were performed using the R package and EmpowerStats (http://www.empowerstats.com). Significance level α = 0.05, the type I error rate was controlled using the Holm-Bonferroni step-down procedure for multiple comparisons [16]. P-values less than 0.05 were considered statistically significant.

Results

Baseline characteristics of study participants based on fasting blood glucose

The design, sample, and exclusion criteria for this study are shown in Fig. 1. A total of 6,093 participants aged 20 years and older were included in our analyses, and Table 1 shows the baseline characteristics of participants based on fasting blood glucose quartiles. As we can see, there were significant differences in baseline characteristics between subgroups of fasting blood glucose quartiles, with the exception of the income-poverty ratio.

Across all participants, fasting glucose quartiles were positively associated with age, BMI, waist circumference, Apo B, triglycerides, insulin, AIP, and TyG (all P-value less than 0.001). The Q4 group had the highest prevalence of angina, hypertension and diabetes compared to other groups. In addition, males were more prevalent in the Q4 group. The Q1 group had higher education levels and HDL-C levels compared to the Q2, Q3 and Q4 groups.

Association between blood lipids and fasting blood glucose

Tables 2, 3, 4, 5 and 6 reflect the results of the multiple regression analyses. In the uncorrected model, TG and ApoB were positively correlated with fasting glucose (TG: β = 0.562,95% CI: 0.508,0.616) (ApoB: β = 0.980,95% CI: 0.812,1.148) while HDL-C was negatively correlated with FBG (β=-0.692,95% CI: -0.791, − 0.593). These positive correlations remained significant in models 2 and 3 after controlling for confounders. It is difficult to visualise the correlation between LDL-C or TC and FBG from the table. When we analyzed the relationship between TC and fasting blood glucose as a continuous variable, we found that there was no significant positive or negative correlation between the two(P = 0.3635), so we stratified TC and found that after adjusting for the covariates, there was a significant negative correlation between TC and fasting blood glucose when the TC level was 4.84–5.52(β= -0.247, 95% CI: -0.364,-0.129), However, through the test of trend, we found that, when we divided TC into three equal parts, the fasting blood glucose of the strata trends were not consistent(P for trend = 0.064). This suggests that there may not be a linear relationship between TC and fasting blood glucose, or it is not reasonable to divide TC into three equal parts, so we need to further study the relationship between TC and fasting blood glucose by other statistical methods. Therefore, we plotted correlation-fitted smoothed curves for the associations between lipids and FBG. Figures 2, 3 and 4 show generalised additive models with smoothed functions illustrating the linear associations between FBG and TG, HDL-C, and ApoB. As can be seen in Figs. 5 and 6, FBG first decreases with increasing TC and LDL-C and then gradually increases. The threshold effect of TC and FBG was analysed using a two-segmented linear regression model, and in Table 7, the inflection point was found to be 5.17 mmol/L (TC < 5.17 mmol/L:β= -0.19,95% CI: -0.27, -0.10; TC > 5.71 mmol/L: β = 0.21,95% CI: 0.11,0.30). In Table 8, the inflection point of the U-shaped curve for the association between LDL-C and FBG was 2.3 mmol/L (LDL-C < 2.3 mmol/L: β = -0.64,95% CI: -0.84,-0.43; LDL-C > 2.3 mmol/L: β = 0.10,95% CI: 0.03,0.17). In contrast, a multivariate logistic regression model found that patients with LDL-C > 2.3 mmol/L had a 1.40-fold higher risk of diabetes than those with LDL-C ≤ 2.3 mmol/L (odds ratio = 1.40, 95% CI: 1.07,1.83).

Subgroup analyses based on gender and race and history of diabetes mellitus

In subgroup analyses based on gender and race, the positive correlation between TG and FBG remained for men (β = 0.187,95% CI: 0.109,0.266) and women (β = 0.480,95% CI: 0.395, 0.565), as well as for blacks (β = 0.563,95% CI: 0.347,0.778), non-Hispanic whites (β = 0.284,95% CI: 0.205,0.363) and Other/Multiracial (β = 0.274,95% CI: 0.158,0.391), but not among Mexican-Americans and Other Hispanics. As shown in Fig. 7, for Mexican Americans and other Hispanics, the association between TG and FBG was an inverted U-shaped curve, whereas in whites, the association was an S-shaped curve. In men and in all races, HDL-C was negatively associated with FBG almost universally, and in women the association between HDL-C and FBG was an L-shaped curve. In subgroup analyses stratified by sex and race, the correlation between ApoB and FBG was absent. After distinguishing whether diabetes was present or not, we found that the effects of TG and HDL-C on blood glucose were consistent, but TC and LDL-C were negatively correlated in non-diabetic patients, whereas they were positively correlated in diabetic patients.

Discussion

This population-based cross-sectional study was designed to systematically assess the relationship between fasting glucose and lipids in the general adult population. The results of this study showed that among the many lipid markers, triglycerides and HDL-C had a strong correlative relationship with fasting glucose. This relationship remained significant after adjusting for confounders. However, our study found a non-linear relationship between TC and FBG with an inflection point of 5.17 mmol/L, which is close to the upper limit of normal for serum total cholesterol [17].The inflection point of the non-linear relationship between LDL-C and FBG was 2.3 mmol/L.Moreover, people with LDL-C > 2.3 mmol/L were at a higher risk of developing diabetes.

In recent years, the incidence of hyperlipidaemia has continued to rise. Hyperlipidaemia is an important risk factor for cardiovascular disease and is also closely associated with insulin resistance and diabetic nephropathy [18, 19]. Existing studies have shown that dyslipidaemia is associated with blood glucose levels, age and gender [20]. A cross-sectional study of 13,093 diabetic patients in Jilin, China, showed [21] that fasting glucose was negatively correlated with HDL-C and positively correlated with triglycerides. In addition, a prospective longitudinal cohort study including 2,085 members of the general population found [22] that elevated triglyceride levels could lead to an increased incidence of type 2 diabetes. This is consistent with the results of the correlation between triglycerides and fasting blood glucose in this study. In the Uyghur hypertensive population [23], serum HDL-C was positively correlated with fasting blood glucose levels. The heterogeneity between these studies may be due to study design, sample size, ethnic differences and other confounding factors. Admittedly, we found through subgroup analysis that there were differences in the effect of lipids on blood glucose among different ethnic groups, different genders and Whether diabetes is present, which may be related to dietary habits and genetic characteristics.

In our study, triglycerides and HDL were linearly correlated with fasting glucose in the general adult population. When total cholesterol was within the normal reference range (< 5.20 mmol/L), it was negatively correlated with fasting glucose. In contrast, LDL-C was correlated with fasting glucose in a U-shaped curve with an inflection point of 2.3 mmol/L. We noted that in the general adult population, the higher the TG, the higher the fasting glucose; the higher the HDL-C, the lower the fasting glucose; the higher the TC, the lower the fasting glucose in the non-hypercholesterolemic population; and at a TC > 5.17 mmol/L, the fasting glucose increased as the TC increased. At LDL-C < 2.3 mmol/L, higher LDL-C was associated with lower fasting glucose, whereas at LDL-C > 2.3 mmol/L, fasting glucose increased with increasing LDL-C.

Considering these correlations, lipid levels may provide an early warning of the prevalence of diabetes in the general population and suggest that we should keep lipid levels within reasonable limits, thus guiding interventions and avoiding a single emphasis on lipid-lowering therapy in the population.

As we used a nationally representative sample, our findings were highly correlated with the overall population. We also examined the correlation between different LDL-C segments and the prevalence of diabetes and found that the likelihood of developing diabetes was 40% higher when LDL-C was greater than 2.3 mmol/L. However, there are many factors that affect blood glucose and there may be other confounding factors that were not adjusted for in our study, and there is also a need to consider the possibility of selectivity bias that could bias the experimental results. Therefore, further basic mechanistic investigations and prospective studies with large sample sizes are needed to understand the precise mechanisms by which lipids affect blood glucose.

Table 1 Fasting blood glucose quartile weighted characteristics of the study population

Table 2 The association between triglyceride(mmol/L) and fasting blood glucose (mmol/L)

Table 3 The association between total cholesterol (mmol/L) and fasting blood glucose (mmol/L)

Table 4 The association between low density lipoprotein cholesterol (mmol/L) and fasting blood glucose (mmol/L)

Table 5 The association between high density lipoprotein cholesterol(mmol/L) and fasting blood glucose (mmol/L)

Table 6 The association between apolipoprotein B (mmol/L) and fasting blood glucose (mmol/L)

Table 7 Association of blood lipids and fasting blood glucose

Table 8 Threshold effect analysis of total cholesterol on fasting blood glucose using the two-piecewise linear regression model

Table 9 Threshold effect analysis of LDL-C on fasting blood glucose using the two-piecewise linear regression model

Fig. 1

Flow chart of study participants

Fig. 2

The association between triglyceride and Fasting blood glucose. Every black dot on the left graph represents a sample. On the righr graph, the solid red line depicts the smooth curve fit between the variables, while the blue bands represent the 95% confidence interval from the fit. Age, sex, race/ethnicity, alcohol intake, smoking behavior, education, income poverty ratio, Body Mass Index, waist circumference, systemic immune inflammation, history of hypertension, history of heart diseases, history of diabetes mellitusn use were adjusted

Fig. 3

The association between high density lipoprotein cholesterol and Fasting blood glucose. Every black dot on the left graph represents a sample. On the righr graph, the solid red line depicts the smooth curve fit between the variables, while the blue bands represent the 95% confidence interval from the fit. Age, sex, race/ethnicity, alcohol intake, smoking behavior, education, income poverty ratio, Body Mass Index, waist circumference, systemic immune inflammation, history of hypertension, history of heart diseases, history of diabetes mellitusn use were adjusted

Fig. 4

The association between apolipoprotein B and Fasting blood glucose. Every black dot on the left graph represents a sample. On the righr graph, the solid red line depicts the smooth curve fit between the variables, while the blue bands represent the 95% confidence interval from the fit. Age, sex, race/ethnicity, alcohol intake, smoking behavior, education, income poverty ratio, Body Mass Index, waist circumference, systemic immune inflammation, history of hypertension, history of heart diseases, history of diabetes mellitusn use were adjusted

Fig. 5

The association between total cholesterol and Fasting blood glucose. Every black dot on the left graph represents a sample. On the righr graph, the solid red line depicts the smooth curve fit between the variables, while the blue bands represent the 95% confidence interval from the fit. Age, sex, race/ethnicity, alcohol intake, smoking behavior, education, income poverty ratio, Body Mass Index, waist circumference, systemic immune inflammation, history of hypertension, history of heart diseases, history of diabetes mellitusn use were adjusted

Fig. 6

The association between low density lipoprotein cholesterol and Fasting blood glucose. Every black dot on the left graph represents a sample. On the righr graph, the solid red line depicts the smooth curve fit between the variables, while the blue bands represent the 95% confidence interval from the fit. Age, sex, race/ethnicity, alcohol intake, smoking behavior, education, income poverty ratio, Body Mass Index, waist circumference, systemic immune inflammation, history of hypertension, history of heart diseases, history of diabetes mellitusn use were adjusted

Fig. 7

The association between triglyceride and Fasting blood glucose stratified by race/ethnicity. Age, sex, alcohol intake, smoking behavior, education, income poverty ratio, Body Mass Index, waist circumference, systemic immune inflammation, history of hypertension, history of heart diseases, history of diabetes mellitusn use were adjusted