From October 2017 to April 2023, all participants of this prospective study were registered at the Metabolic Management Center (MMC) of Beijing Luhe Hospital. The MMC is a national project that aims to manage metabolic patients according to a standardized approach. All MMCs in China have the same facilities structure, databases, and layout, as well as the same routine daily operations, which aims to establish a platform with standardized diagnosis and treatment of metabolic diseases and their long-term follow-up. Patients can get one-stop care to receive a comprehensive series of services from registration, tests, evaluation, prescriptions, to health education. The protocol of the MMC project was published previously [15].

Patients with type 2 diabetes aged 18–80 years were recruited. The recruitment process included blood sample collection, systematic physical examination, and oral questionnaire interviews. Diabetes was defined as having a fasting plasma glucose level of 126.13 mg/dL, a 2-h plasma glucose level of at least 200 mg/dL, an HbA1c level of at least 6.5%, or a self-reported previous diagnosis by health care professionals. In this study, only type 2 diabetes (T2DM) were enrolled.

Participants were excluded from the study if they met any of the following criteria: (1) pregnant or nursing women; (2) suffering from a malignant tumor; (3) experiencing acute complications of diabetes; (4) had visited times less than four times and the follow-up time less than 12 months; (5) had missing data of critical variables.

The study protocol was approved by the Medical Ethics Committee of Beijing Luhe Hospital, Capital Medical University. This study was performed by the Declaration of Helsinki, and all participants provided written informed consent.

Blood samples were obtained in the morning after overnight fasting. Glycated hemoglobin (HbA1c) levels were assayed using the method of high-performance liquid chromatography (HPLC) with a D10 set (Bio-RAD, Hercules, CA, USA). HbA1c was a normal inspection in the MMC program, which was examined every 3–6 months for each participant. The average time for HbA1c measurements in this study was 5.07 ± 2.25 months.

Participants were examined by ultrasonography. Participants in the supine position with the head slightly extended and turned to the opposite direction of the carotid artery being studied. Images were recorded at the internal carotid arteries bilaterally. The maximum carotid IMT readings of the right and left far walls for common, bulb, and internal segments were used for analysis. Considering the study design and statistical methods of this study, only the IMT results of the first and last follow-up were included in the analysis.

Systematic physical examination and oral questionnaire interviews were performed by trained personnel according to the protocol. The questionnaire contains information on demographic characteristics (including age, sex, education level, weight, height), medical history (including duration of diabetes, hypertension history, antihypertensive drugs usage, lipids treatments information.), and lifestyle factors (including cigarette smoking, drinking.). Smoking status was defined as ‘ideal’ if the participants did not smoke or had quit smoking for more than 12 months. Drinking status was recorded as ‘yes’ for participants who drank weekly or almost weekly. Education attainment was categorized as less than high school and high school or more.

Height and body weight were measured with a standard protocol, and body mass index (BMI) was calculated as weight divided by height squared. Moreover, an auto biochemical analyzer measured LDL, HDL, and triglyceride (AU5800, Beckman Coulter, USA).

Continuous variables were described as mean ± standard deviation (SD) or median [interquartile range (IQR)], and categorical variables were described as frequency (%). Logarithmically transformed were required before statistical analysis when data were tested as non-normal distribution. We used the Cochran-Armitage trend test and linear regression analysis to calculate the P values for categorical and continuous variables across the groups.

The coefficients of variation (CV) of HbA1c were calculated as the ration of standard deviation (SD) to the mean. The ARV (average real variability) was calculated as the average of the absolute differences between consecutive HbA1c measurements (Formulate 1) [16].

$$ARV = \frac^ \left| - HbA1c_ } \right|}}$$

(1)

where K is the ordinal number of HbA1c, and N is the total number of HbA1c measurements.

The variation independent of the mean (VIM) was calculated as the SD divided by the mean HbA1c raised to the power of x, where x is obtained from fitting a nonlinear regression model (Formulate 2) [17].

$$VIM = \frac \right)}} \right)^ }}\,where\,k = Mean\,\left( \right)^$$

(2)

Multiple linear regressions were used to explore the association between the variation of HbA1c and the changes of IMT; two models were established and adjusted for different covariates. Model 1 was adjusted by age, duration of diabetes and sex; model 2 was adjusted by ideal smoking, drink status, BMI, education level, SBP, LDL, HDL, and TC plus the covariates in model 1; model 3 was adjusted by oral antidiabetic agents, lipid-lowering agents and insulin plus the covariates in model 2. As a secondary analysis, we performed generalized additive models (GAM) to investigate the association between variation of HbA1c and the changes of IMT in non-linear conditions.

The latent class mixture model (LCMM) was used to identify similar trajectories of HbA1c [3, 12,13,14]. The lcmm package of R was used to execute the procedure, with the number of latent categories ranging from 2 to 5. Each trajectory group was named based on the baseline of HbA1c and the visual change patterns observed of HbA1c during the follow-up period. The optimal model was selected using the Bayesian information criterion (BIC).

We calculated the change value of IMT (ΔIMT) and the change percent of IMT (ΔIMT%) of every participant in the whole follow-up period as the primary outcome. The associations between each HbA1c trajectory pattern and ΔIMT% were examined using covariate-adjusted means (SE) of ΔIMT and ΔIMT%, which were calculated by multiple linear regression analyses adjusted for the covariates mentioned previously.

Finally, a cross-lagged panel model (CLPM) was performed to further verify the bidirectional relationship between IMT and HbA1c, which measured the effect size of baseline HbA1c measurement on subsequent IMT and the effect size of baseline IMT on subsequent HbA1c measurement simultaneously [18,19,20]. The analysis procedure was performed in the lavaan package of R. The CLPM is adjusted by age, sex, duration of diabetes, SBP, ideal smoking, drink status, HDL, LDL, oral antidiabetic agents, lipid-lowering agents and insulin.

Moreover, we also performed sex subgroup analyses to investigate the consistency of the association between variation of HbA1c and the changes of IMT. Multiple linear regressions adjusted all covariables in model 2.

All statistical analyses were performed in R software (version 4.2.2, https://www.r-project.org/).