Association between atherogenic index of plasma control level and incident cardiovascular disease in middle-aged and elderly Chinese individuals with abnormal glucose metabolism

Our study is based on data from CHARLS, which is a nationally representative longitudinal survey of persons aged 45 years in China. Supported by the multistage probability sampling method, the CHARLS baseline survey was conducted between 2011 and 2012 (wave 1). To date, CHARLS has made public three waves of follow-up data (wave 2 in 2013, wave 3 in 2015 and wave 4 in 2018) [18].

In this study, 11,847 participants who underwent a complete blood count test were included in the analysis. We excluded 8,484 participants for reasons including: (1) lack of diagnostic markers for abnormal glucose metabolism; (2) No complete blood cell test was performed on the wave 3; (3) Lack of TG or HDL-C level measurement; (4) under 45 years of age; (5) People who self-reported having CVD events before 2015. A total of 6,821 participants were finally included (Fig. 1).

CHARLS datasets are available for download at the CHARLS home website (http://​charls.​pku.​edu.​cn/​en). The Biomedical Ethics Review Board of Peking University gave its ethical approval for the gathering of CHARLS data (IRB00001052-11015), and all participants signed an informed consent form.

Abnormal glucose metabolism including diabetes and prediabetes. According to American Diabetes Association (ADA) diabetes diagnostic criteria [19], prediabetes is defined as fasting plasma glucose (FPG) at 100–125 mg/dl or glycosylated hemoglobin A1c (HbA1c) at 5.7–6.4%; Diabetes with FPG ≥ 125 mg/dl, HbA1c ≥ 6.5%, self-reported medical history, or use of antidiabetic medications [20].

CVD was the primary outcome event of the study, and heart disease and stroke were secondary outcome events. The presence of heart disease is determined by the question “Did your doctor tell you that you have been diagnosed with a heart attack, angina pectoris, coronary heart disease, heart failure, or other heart problem?” Similarly, the occurrence of stroke is ascertained through the question “Did your doctor tell you that you were diagnosed with a stroke?”. CVD is defined as self-reporting heart disease and/or stroke.

This study investigated AIP changes as exposures in Wave1 and Wave3 participants using k-means cluster analysis in a logistic regression equation where AIP was calculated as lg (TG/HDL-C). In addition, blood measurements from 2011 to 2012 were covered in Wave1, and measurements from 2015 were published in Wave3. Therefore, our cumulative AIP was determined by the expression: (AIP2012 + AIP2015)/2* time (2015 − 2012) [9].

Based on previous studies, we included potential confounding covariates at baseline. Demographic covariates included age, gender (“Male” and “Female”), education level (“Primary school or lower” and “Secondary school or higher”), current married (“Current marital” and “Others”), Hukou (government household registration system) (“Agriculture” and “Others”). Health behavior covariates included smoking status (“Yes”, “No”), drinking status (“Yes”, “No”), systolic blood pressure (SBP), diastolic blood pressure (DBP) and body mass index (BMI). Laboratory examination included fasting blood glucose (FBG), TG, blood urea nitrogen (BUN), total-cholesterol (TC), and HDL-C, Low-Density Lipoprotein Cholesterol (LDL-C), C-reactive protein (CRP), HbA1c and Uric acid (UA). There are some data missing in the CHARLS database, and the missing number and proportion of each variable are shown in Table S1. In order to ensure the accuracy of the research results, Template method (R Package “VIM”) [21]and multiple interpolation method (R Package “mice”) [22]were used to fill them.

K-means clustering is a rule-based method for determining the distance between data items(Fig. 2A). The advantages of this approach are simplicity and scalability. In our study, when the number of clusters is equal to 5. K-means clustering works best [22, 23]: For class 1, AIP ranged from 0.016 to 0.037, representing the best AIP level control; For class 2, AIP ranged from 0.665 to 0.366, representing the better AIP level control; For class 3, AIP ranged from 0.231 to 0.325, representing an increase in AIP levels from a low level to a high level with poor control; For class 4, AIP ranged from 0.467 to 0.694, representing an increase in AIP levels from lower levels to higher levels, with worse control; For class 5, AIP ranged from 1.063 to 0.860, representing the worst AIP level of control (Fig. 2B).

The mean and standard deviation (SD) of continuous data, and percentage of classified data are used to describe basic features. Continuous variables were compared by Kruskal-Wallis rank sum test, and categorical variables were compared by Chi-square test or Fisher exact probability test. Hereafter, we used multivariable logistic regression analysis to establish three models to examine the relationship between AIP and CVD. The crude model was not adjusted for any variables. In model I, the data was adjusted for age, gender. And in model II, the results were adjusted for age, gender, education, marital status, residence location, smoking status, drinking status, SBP, DBP and BMI. The results from the logistic regression analysis are presented as odds ratios (ORs) and 95% confidence intervals (CIs). In addition, restricted cubic splines can reflect nonlinearity between variables. So, we used the restricted cubic spline model to test the correlation between cumulative AIP and CVD, and chose four knots at 20th, 40th, 60th and 80th. We further stratified analyses according to current married, age (Q1, Q2, Q3), gender, education level (primary school or lower, secondary school or higher), Hukou (agriculture, others), SBP (Q1, Q2, Q3), DBP (Q1, Q2, Q3), BMI (Q1, Q2, Q3), drinking status, current smoking status to examine whether potential confounding variables modulated the potential association between changes in AIP and CVD, and to test their interactions. CVD was the primary outcome, and heart failure and stroke were the secondary outcomes. In addition, we explored the association between changes in AIP levels and two secondary outcomes of heart disease and stroke using the same methodology as the CVD outcome. In all calculations, a two-sided P-value < 0.05 was required for statistical significance. All analyses were performed in R version 4.1.0 and EmpowerStats version 4.1 (www.​empowerstats.​com; X&Y Solutions Inc.).

In this study, 2.659 participants with abnormal glucose metabolism were included, with an average age of 58.34 ± 8.34 years, 47.09% males. The Cumulative AIP of 1.22 ± 0.90, 567 participants eventually developed CVD (398 participants had heart disease and 169 had stroke) after a mean follow-up of 3 years. Compared to Class 1, participants in the other classes had a lower prevalence of current smoking and drinking; higher levels of SBP, DBP, BMI, FBG, TG, TC, HbA1c and UA, but lower levels of BUN and HDL-C (Table 1). In addition, we also provided a baseline description of the missing variables before interpolation, which is similar to the above results (Table S2).

The logistic regression analyses are presented in Table 2 for the association between different classes of AIP and CVD. We construct three covariate models with different adjustments to reflect this relationship. After adjusting for various confounding factors, comparing to class 1, the ORs (95% CIs) for incident CVD were 1.31 (0.90, 1.90) for Class 2, 1.38 (0.99, 1.93) for Class 3, 1.46 (1.01, 2.10) for Class 4, and 1.56 (1.03, 2.37) for Class 5. Meanwhile class 4 and class 5 show that AIP has a significant correlation with high CVD incidence. In our secondary outcomes analysis, class 2–5 exhibited a significantly elevated risk of stroke compared to class 1. For heart disease, although the P-value did not reach significance, there was a progressive increase in risk observed for all classes except class 4(Table S3, Table S4).

In the restricted cubic spline regression models shown in Fig. 3, the cumulative AIP and CVD risk are linear. The risk of CVD was increasing with each increase in the cumulative AIP above 1.12 (OR 1, 95% CI, 0.98, 1.01). However, in class 1 and class 2, the risk of CVD decreasing with each increase in the cumulative AIP; The values of class 4 OR were all less than 1, and both class 3 and class 5 were greater than 1.

We performed a subgroup analysis to explore the association of the AIP control level with incident CVD events stratified by potential risk factors. As shown in Table S5, subgroup analysis based on class1, class3-5 showed a higher risk of CVD in agricultural residence location (53 < Age < 61 years) and smoking patients compared to other subgroups.

No interaction was found between AIP control level and subgroup variables. For the two secondary outcomes of heart disease and stroke, the subgroup analysis was similar to that of CVD. It is worth noting that there is interaction between the two potential variables of gender and household registration in the stroke event (Table S6, Table S7).