2.1 Study population

The participants were from an ongoing cohort study in Jiading District, Shanghai, China. The study design and detailed methods have been described previously.^{8, 9} Briefly, registered residents from two communities in Jiading District were invited to attend a comprehensive health examination, and 10 375 residents aged 40 years or older were recruited between March and August 2010. All the participants completed a standard questionnaire and a set of clinical measurements. A standard 75-g oral glucose tolerance test (OGTT) was performed in participants who did not already receive glucose-lowering treatment. Fasting and 2-h post-load plasma glucose levels were measured. During August 2014 and May 2015, all participants were invited to attend a follow-up visit and received the OGTT again for a reevaluation of glucose levels. In the current study, we excluded participants who had (a) preexisting diabetes (n = 1872); (b) missing values of 0 or 2-h glucose levels (n = 26); and (c) lipid-lowering drugs (n = 21) at baseline. We further excluded participants who (a) died during follow-up (n = 181); (b) did not participate in the follow-up visit (n = 2919); and (c) had missing values of 0 or 2-h glucose levels at follow-up (n = 82). Details of the selection of the study participants are shown in Figure 1. Ultimately, 5274 participants were included in the current analysis. Those who were included and those who did not participate in the follow-up visit had similar demographic and metabolic characteristics, as well as baseline glucose levels.

2.2 Ethics statement

The Institutional Review Board of Ruijn Hospital affiliated with Shanghai Jiao Tong University School of Medicine approved the study protocol, according to the principles of the Declaration of Helsinki. All the participants provided written informed consent before data collection.

2.3 Data collection

Face-to-face interviews were performed at the baseline and the follow-up visit by trained study personnel using a standard questionnaire that collected information on sociodemographic characteristics, lifestyle factors, medical history, and current medication use. Information on intensity, duration, and frequency of physical activity was obtained using the short form of the International Physical Activity Questionnaire.¹⁰ Current smokers or drinkers were defined as those who had smoked cigarettes or consumed alcohol regularly in the past 6 months.

Anthropometric measurements, including body height and weight, were performed using calibrated instruments according to a standard protocol. Body mass index (BMI) was calculated as body weight in kilograms divided by height in meters squared. Blood pressure was measured, after at least 5 min of sitting rest, on the nondominant arm of each participant using an automated device (OMRON Model HEM-752 FUZZY; Omron Co., Dalian, China). Three consecutive measurements with 1 min intervals were recorded. The mean of the three measurements was calculated and used for analysis.

All participants underwent a standard OGTT after an overnight fast of at least 10 h, and venous blood samples were obtained at 0 and 2 h during the test. Blood glucose levels were measured with an autoanalyzer (Modular P800; Roche, Basel, Switzerland) using the glucose oxidation method. The levels of total cholesterol, high-density lipoprotein cholesterol (HDL-C), LDL-C, and triglycerides were measured with an autoanalyzer (Modular E170; Roche, Basel, Switzerland) using the chemiluminescence method.

2.4 Definition

According to the 1999 World Health Organization criteria, diabetes was defined as a fasting plasma glucose (FPG) level ≥ 126 mg/dL, and/or a 2-h post-load plasma glucose (PPG) level ≥ 200 mg/dL, and/or the use of any glucose-lowering medications. In those without diabetes, impaired fasting glucose was defined as FPG ≥ 110 mg/dL and < 126 mg/dL and impaired glucose tolerance was defined as PPG ≥ 140 mg/dL and < 200 mg/dL. Hypertension was defined as a systolic blood pressure (BP) ≥ 140 mm Hg, and/or diastolic BP ≥ 90 mm Hg, and/or taking any BP-lowering medications. Dyslipidemia was defined as total cholesterol ≥ 240 mg/dL, and/or LDL-C ≥ 160 mg/dL, and/or triglycerides ≥ 200 mg/dL, and/or HDL-C ≤ 40 mg/dL, and/or taking any lipid-lowering medications.¹¹ History of CVD was defined as a self-reported previous diagnosis of stroke, myocardial infarction, or other coronary heart disease. A family history of diabetes was defined as the presence of at least one first-degree family member diagnosed with diabetes. Ideal physical activity was defined as ≥ 150 min/week moderate intensity or ≥ 75 min/week vigorous intensity or ≥ 150 min/week moderate + vigorous intensity activity.¹²

2.5 Statistical analysis

Baseline characteristics of the study population are demonstrated according to LDL-C levels. Continuous variables are presented as means ± SDs for normally distributed variables or as medians (interquartile ranges) for skewed variables. Categorical variables are presented as numbers and proportions.

We examined the pattern of association between LDL-C levels and incident diabetes using restricted cubic splines with knots at the 5th, 50th, and 95th percentiles of the distribution of LDL-C levels. The LDL-C level was also fit as a continuous variable to estimate the risk of developing diabetes in association with a 1-SD increase in LDL-C levels. For all regression analyses, we used three models. Model 1 was adjusted for age and sex. Model 2 was adjusted for age, sex, educational level (≥ 9 years of education, yes or no), physical activity (ideal level, yes or no), current smoking (yes or no), current drinking (yes or no), family history of diabetes (yes or no), hypertension (yes or no), BMI, triglycerides, and HDL-C. Model 3 was further adjusted for homeostatic model assessment for insulin resistance (HOMA-IR) based on model 2. The association between a 1-SD increase in LDL-C levels and incident diabetes was examined in subgroups of age (< and ≥ 60 years), sex, BMI (< and ≥ 28 kg/m²), hypertension (with and without), family history of diabetes (with and without), and LDL-C levels (< and ≥ 130 mg/dL). Several sensitivity analyses were conducted by excluding participants with baseline diseases.

All statistical analyses were performed using the SAS software (version 9.4; SAS Institute, Cary, NC, USA). A 2-sided p value of less than .05 was considered statistically significant.

3.1 Basic characteristics of the study population

Study participants' characteristics are presented in Table 1 according to quartiles of serum LDL-C concentrations at baseline. Those with higher serum LDL-C levels were more likely to be older, female, not currently smoking or drinking, and had completed less education (all p <.0001). Higher serum LDL-C levels were also associated with worse metabolic profiles, including higher BMI, BP, FPG, PPG, and HOMA-IR (all p < .0001).

3.2 Baseline LDL-C levels and incident diabetes

During a median follow-up of 4.4 years, 652 participants (12.4%) developed diabetes. The cumulative proportion of incident diabetes increased significantly across LDL-C quartiles. After adjustment for age and sex, there was a significant and positive association between serum LDL-C levels and incidence of diabetes (model 1, odds ratio [OR] for quartiles 2 vs. 1: 1.42, 95% confidence interval [CI] 1.10–1.82; OR for quartiles 3 vs. 1: 1.50, 95% CI 1.17–1.92; OR for quartiles 4 vs. 1: 1.61, 95% CI 1.26–2.06; all p <.01) (Table 2). The results remained significant after additional adjustment for education level, current smoking, current drinking, physical activity, BMI, hypertension, triglycerides, and HDL-C based on model 1. Similarly, the results were essentially unchanged after accounting HOMA-IR (model 3).

The cubic spline representing the shape of the association between serum LDL-C concentrations and the incidence risk of diabetes demonstrated a nearly linear association (Figure 2). The risk of diabetes increased 12% in association with per 1-SD increment in LDL-C levels (OR 1.12, 95% CI 1.03–1.22, p =.01110) (Table 2).

3.3 Subgroup analysis

The association between LDL-C levels and risks of incident diabetes was similar for subgroups of age, sex, BMI, baseline hypertension status, and LDL-C levels (all p values for interaction >.05). However, LDL-C levels showed a significantly stronger (p <.0001 for interaction) association with the incidence of diabetes in those without a family history of diabetes than in those with a family history of diabetes. Specifically, even in participants with normal LDL-C levels (<130 mg/dL), a 1-SD increase in LDL-C levels was significantly associated with a higher risk of diabetes development (OR 1.24, 95% CI 1.02–1.51) (Figure 3).

3.4 Sensitivity analysis

We conducted several sensitivity analyses to assess the robustness of our findings. The risk of incident diabetes associated with 1-SD increase in LDL-C levels was 1.14 (1.04–1.24) after excluding participants who had a CVD history at baseline, 1.10 (1.00–1.21) after excluding participants who had impaired fasting glucose at baseline, 1.15 (1.01–1.32) after excluding participants who had impaired glucose tolerance at baseline, and 1.12 (1.03–1.23) after excluding participants who were using lipid-lowering medications at follow-up, due to the concern that statins might increase glucose levels. Therefore, findings from sensitivity analyses were similar to findings from the main analysis.