Longitudinal trajectory of vascular age indices and cardiovascular risk factors: a repeated-measures analysis

Study design and setting

The current research was conducted as part of a larger prospective cohort study conducted and managed by the National Institute of Health and Nutrition (NIHN) since 2007. This was a multisite cohort study conducted among healthy individuals in Tokyo, the capital of Japan, and Okayama Prefecture, a rural region in Japan, to provide the knowledge to prevent lifestyle-related diseases. Individuals with terminal diseases were excluded. The research details have been described elsewhere^21,22,23. A total of 760 individuals (504 in Tokyo, 256 in Okayama Prefecture) participated in this study between 2007 and 2018 (Table S1). These participants were recruited during specific health checkups conducted by the NIHN or the Okayama Southern Institute of Health.

This study was conducted according to the guidelines laid down in the 1964 Declaration of Helsinki and all procedures involving research study participants were approved by the Research Ethics Committee of the National Institutes of Biomedical Innovation, Health and Nutrition (approval no: kenei102-01). Written informed consent was obtained from all participants. In reporting this study, we have followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines²⁴.

Study sample

From the group that completed the baseline survey (n = 760), we excluded those participants who lacked follow-up data (n = 60), those for whom age and sex data were missing (n = 1), and participants for whom BVAI assessments were not performed (n = 2). The sample included 697 Japanese adults aged 26–85 years who completed the baseline examination as well as at least two follow-up assessments of BVAIs and lifestyle risk factors. The investigations were conducted annually using the same survey methodology and content, and the participants were followed up for a maximum of 12 years. Data for the following three samples were used (for sensitivity analysis). First, we performed an analysis in a sample of 690 individuals (a maximum of 3636 measurements) for whom all BVAI data were available. This sample was termed the full analysis set (FAS). Second, we included a sample of 678 individuals (2943 measurements) for whom complete data on all nine BVAIs were obtained through in-person testing. This sample was called the BVAI complete case (BCC). Third, we included 648 individuals (2633 measurements) for whom complete data on all BVAI and lifestyle risk factors could be assessed. This sample was called the complete case (CC).

Assessment of biological vascular aging indicators

To assess functional and structural BVAIs, systolic blood pressure (SBP), ankle-brachial index (ABI), heart rate (HR), common carotid diastolic diameter (DD), carotid artery mean blood velocity (MBV), blood flow (BF), common carotid intima-media thickness (IMT), carotid-femoral pulse wave velocity (PWV), and vascular aging index (VI) were assessed in the morning after an overnight fasting period of 10 h or more. Details of the assessment methods have been described^21,25. SBP, ABI, HR, and PWV were measured noninvasively using a vascular testing device (Model BP203RPE II, from PWV/ABI; OMRON Colin Medical Instruments, Tokyo, Japan). While the participants were at rest in the supine position, cuffs were placed on both arms and ankles, electrocardiogram electrodes on both wrists, a cardiac sound sensor on the left sternal border, and a tonometer on the common carotid and femoral artery. A multi-element tonometry sensor (CAP-350 and FAP-350; Colin Medical Technology, Komaki, Japan) was pressed perpendicularly against the wall of the carotid artery and the femoral artery to simultaneously record pulse waveforms of the common carotid and femoral arteries to calculate the carotid-femoral PWV.

The DD, IMT, and MBV were measured using ultrasound devices (Vivid i; GE Medical Systems, USA, and model 180 Plus; Sonosite, USA). While the participants were at rest in the supine position, an ultrasound device with a high-frequency linear array probe at 10 MHz was used to image the longitudinal common carotid artery in B-mode, and the images were recorded as a video. Longitudinal images of the common carotid artery were analyzed using image analysis software (Image J, National Institutes of Health, USA). The mean DD and IMT was calculated from the images using five frames at showing the end diastolic diameter of the left ventricle per cardiac cycle. DD was defined as the distance between near and far lumen-intima interfaces. IMT was defined as the distance between the lumen-intima interface and the medial-adventitial interface²⁶. No participants had an IMT ≥ 1.5 mm which is defined as grade 1 plaque²⁷. The MBV in the common carotid artery was measured using the above ultrasound device and Doppler ultrasonography. The BF²⁸ and VI¹⁰ of the carotid artery were calculated using the equations from previous studies as follows: BF (mL/min) = MBV (cm/s) × π × DD² (cm²) × 60; VI = log_e(1.09) × 10IMT (mm) + log_e(1.14) × PWV (cm/s).

Assessment of covariates

In this study, previously reported factors associated with BVAIs were analyzed in a comprehensive manner³. Participants wore light clothing and their body weight was measured using a digital scale (BC-600, TANITA Corp., Tokyo, Japan). Body mass index (BMI) was calculated as weight divided by height squared (kg/m²). We calculated the waist/hip ratio as the abdominal obesity index by measuring the waist (at the height of the navel) and the hip circumference (at the largest bulge perpendicular to the long axis of the trunk). The trunk flexibility was measured using a sit-and-reach digital instrument (T.K.K.5112; Takei Scientific Instruments Co. Ltd., Niigata, Japan). Regarding leg strength, unidirectional lower limb extension strength was measured using a multi-joint leg extension apparatus (Anaeropress 3500; Combi Co., Tokyo, Japan). The grip strength was measured using a Smedley hand dynamometer (Grip-D TKK5101, Takei Scientific Instruments, Niigata, Japan). Measurements were performed twice on each hand; the highest value for each hand was used for analysis. The following biochemical parameters were measured: red blood cell count, white blood cell count, platelet count, hemoglobin, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol (LDL-C), triglycerides, hemoglobin A1c (HbA1c), homeostasis model assessment of insulin resistance (HOMA-IR), aspartate aminotransferase (AST), alanine aminotransferase (ALT), and γ-glutamyl transpeptidase (γ-GTP). As an objective measure of physical activity, step counts were measured using a previously validated triaxial accelerometer (Actimarker EW4800, Panasonic, Osaka, Japan)²⁹. The n-3/n-6 fatty acid ratio and the intake of saturated fatty acids (SFA), alcohol, salt, sugar, meat, fruits and vegetables (FV), and pulses were assessed using a dietary questionnaire that was validated against the dietary record method^30,31. We calculated the food and nutrient intake per 1000 kcal using the density method to adjust for the energy intake³². Questionnaires were administered to assess demographic information, smoking status, family history of heart disease, presence of comorbidities, and sleep status. Comorbidities were classified as a comorbidity score based on the total number of comorbidities, out of 10, that were present in the participants—hypertension, dyslipidemia, diabetes, ischemic heart disease, other heart diseases, cerebrovascular disease, renal failure, cancer, osteoporosis, and mood disorders^22,23.

Statistical analysis

Regarding participant characteristics, continuous variables are presented as mean (standard deviation), while categorical variables are presented as number (percentage). The missing covariate values (see Supplementary Material) were supplemented with five datasets using multivariate imputation by chained equation in R Statistical Software³³. All missing values were treated as missing at random.

We identified the longitudinal trajectories of BVAIs using latent growth curve models (LGCMs) and latent class growth models from repeated BVAI measurement data (FAS dataset). These analyses were performed using the STATA macro TRAJ³⁴. After stratifying by sex, the overall mean trajectory of BVAIs was estimated using LGCMs (cubic splines).

The results of the cross-sectional and repeated longitudinal analyses were compared using the BCC dataset to identify the between- and within-person trajectory of BVAIs³⁵. This analysis was stratified by age group (≤ 39 years, 40–49 years, 50–59 years, 60–69 years, and ≥ 70 years) assuming heterogenous age-related trajectories of BVAIs³⁶. The CA-related trajectories of nine BVAIs were assessed using univariate panel data regression analysis. The results are presented as regression coefficients and 95% confidence intervals by changes per year of CA. Furthermore, to assess the correlation coefficients of cross-sectional and longitudinal repeated analyses between CA and BVAIs³⁶, a longitudinal analysis was performed using Repeated Measures Correlation (rmcorr) by R Statistical Software³⁷, and a cross-sectional analysis was performed using the Pearson correlation coefficient.

To assess the parallel changes in factors associated with changes over time in the longitudinal trajectories of nine BVAIs, we performed a multivariate regression analysis of the random effect panel data (using baseline covariate data and longitudinal trajectories)^22,23. The variables mentioned above were used as covariates for the multivariate analysis model. Variables with a variance inflation factor (VIF) ≤ 10 were used in the model to avoid multicollinearity³⁸. If the VIF was > 10, variables with the highest predicted value were maintained in the model (Table S2). Results are presented as regression coefficients and 95% confidence intervals per unit of the relevant variable. Sensitivity analysis was similarly performed using three datasets (FAS, BCC, and CC).

Statistical significance was established at a two-tailed P < 0.05 (z score, ≤  − 1.96 or ≥ 1.96). All analyses were performed using STATA MP, version 15.0 (StataCorp LP, College Station, TX, USA) or R software 3.4.3 (R Core Team, Vienna, Austria).