Changes in triglycerides and high-density lipoprotein cholesterol may precede peripheral insulin resistance, with 2-h insulin partially mediating this unidirectional relationship: a prospective cohort study

The participants were from the Harbin People’s Health Study (HPHS). The HPHS was launched by the Centers for Disease Control and Prevention and the Public Health School in Harbin in 2008 [15]. It covered five urban administrate regions of Harbin. Each region were divided into three strata based on their financial situation and one or two neighborhood committees were chosen from 15 communities that were randomly selected from each stratum in each administrate region by performing a stratified multistage random cluster sampling design. A total of 8940 subjects, aged 20–74 years, were recruited for the study. Subjects were eligible to participate in the study if they had no history of postmenopausal hormone therapy, malignancy, thyroid dysfunction, renal calculi, or corticosteroid or calcitriol use. A total of 4515 subjects (approximately 50.5% of the total subjects) were selected by completely randomized sampling method to participate in a follow-up survey due to the financial limitations of this study. In 2012, 4158 subjects completed the first in-person follow-up survey, for a response rate of 92.1%. The basic information did not differ significantly between the original cohort and the selected cohort. After the exclusion of 398 subjects who were taking hypoglycemic drugs or insulin injections, and 435 subjects who were receiving dyslipidemia treatment at either the baseline or the follow-up survey, 3325 subjects were included in this analysis, with an average follow-up period of 4.2 years. These 3325 subjects included 633 subjects with type 2 diabetes who were not receiving any treatment and 1887 subjects with dyslipidemia who were not receiving dyslipidemia medication at either baseline or the follow-up survey. The study protocols were approved by the Ethics Committee of Harbin Medical University, and written informed consent was provided by all subjects. The methods in this study were in accordance with the approved guidelines.

Detailed in-person interviews were administered by trained personnel using a structured questionnaire to collect information on demographic characteristics, dietary habits, lifestyles, physical condition and anthropometric characteristics. Current smokers were defined as those who smoked at least 100 cigarettes in a lifetime or smoked every day or currently smoked some days. Current drinkers were defined as those who consumed ≥1 alcoholic drink each month in the 12 months prior to the survey. Regular exercise was defined as any kind of recreational or sport physical activity other than walking for work or life performed at least 30 min for 3 or more days per week.

Anthropometric measurements, including height, weight, and waist circumference, were obtained by well-trained examiners, with the participants wearing light, thin clothing and no shoes. Body weight and height were measured to the nearest 0.1 kg and 0.1 cm, respectively. Body mass index (BMI) was calculated as weight (kg) divided by the square of the height in meters (m²). An oral glucose tolerance test was performed, according to the World Health Organization guidelines, for each subject.

Fasting serum lipids, including total cholesterol (TCHO), TG, low-density lipoprotein cholesterol (LDL-C), HDL-C, and fasting and 2-h serum glucose were measured using an automatic biochemistry analyzer (Hitachi 7100, Tokyo, Japan). Fasting and 2-h serum insulin was measured by immunofluorescence method (TOSOH automated enzyme immunoassay (EIA) analyzer AIA-2000ST). Gutt index was calculated as an indicator of peripheral IR, which is based on glucose uptake rates, metabolic clearance rates and mean serum insulin by the following equation: [75,000 + (fasting glucose − 2-h glucose) × 0.19 × body weight]/(120 × log [(fasting insulin + 2-h insulin)/2] × [(fasting glucose + 2-h glucose)/2]) [16]. HOMA models were used to estimate hepatic insulin resistance (HOMA-IR) and beta cell function (HOMA-%β) with HOMA2 calculator updated by the University of Oxford in 2004 [17].

Diabetes was identified by self-reports of a history of diabetes diagnosis, fasting blood glucose ≥7.0 mmol/L, and/or 2-h glucose ≥11.1 mmol/L, and/or receiving treatment for diabetes. Dyslipidemia was identified by self-reports of a dyslipidemia diagnosis history, and/or hypercholesterolemia (fasting TCHO ≥6.22 mmol/L, and/or fasting LDL-C ≥4.14 mmol/L), and/or hypertriglyceridemia (fasting serum TG ≥2.26 mmol/L), and/or low HDL-C [fasting serum HDL-C <1.04 mmol/L (male), fasting serum HDL-C <1.29 mmol/L (female)], and/or receiving treatment for dyslipidemia.

All statistical analyses were performed using R 2.15.3 (http://​www.​r-project.​org/​) and LISREL 8.52. A two-sided P < 0.05 was considered statistically significant. Blood lipids, HOMA-IR, HOMA-%β, Gutt index and insulin were log-transformed to improve the normality of the distribution.

Generalized linear models were performed to test differences in continuous variables between gender and menopause status and calculate covariate-adjusted mean yearly rates of change in Blood lipids, HOMA-IR, HOMA-%β, Gutt index and insulin during the follow-up period. Univariate and multivariate linear regression models were used to determine which type of baseline blood lipids independently predicted future HOMA-IR, HOMA-%β, Gutt index and insulin.

Longitudinal changes in blood lipids, HOMA-IR, HOMA-%β, Gutt index and insulin measured at two time points can be modeled using a cross-lagged panel design. In this modeling approach, each variable in the model is regressed on all of the variables that precede it in time. A simplified, conceptual version of the model used in this analysis is presented in Fig. 1. The path coefficient with β₁ describes the effect of baseline TG or HDL-C on the subsequent Gutt index, and the path coefficient with β₂ describes the effect of the baseline Gutt index on the subsequent TG or HDL-C. Prior to the cross-lagged path analysis, the baseline and follow-up biochemical indices were adjusted for age, gender, alcohol consumption, smoking, regular exercise, BMI, and caloric intake using a regression residual analysis and then were standardized by Z-transformation (mean = 0, standard deviation, 1). Pearson correlation coefficients of the Z-transformed quantitative variables of biochemical indices at baseline and follow-up were calculated. The cross-lagged path coefficients (β₁ and β₂) were estimated simultaneously based on the correlation matrix using the maximum likelihood method in LISREL version 8.52. The percentile confidence interval of cross-lagged path coefficient was estimated using bootstrap simulation for the cross-lagged model. The validity of model fitting was indicated by the root mean square residual (RMR) and comparative fitness index (CFI) [18, 19]. RMR < 0.05 and CFI > 90 indicate relatively good fit to the observed data. The temporal relationships of blood lipids with Gutt index, HOMA-models and 2-h insulin were examined in separate models. The difference between β₁ and β₂ derived from the standardized variables was tested using Fisher’s Z-test [14]. Although the significance of individual β₁ or β₂ suggests a directional relationship, a significant difference between β₁ and β₂ provides stronger evidence for a temporal relationship in the model.

Once the temporal relationships of these biochemical indices had been established, a causal mediation model was constructed to examine whether there are any mediating effects existing in these temporal relationships using R package mediation.