Study design and participants

The CHNS is an ongoing longitudinal survey designed to investigate health and nutritional status in Chinese populations. Its study design and procedures have been described previously^{(Reference Popkin, Du and Zhai11)}. The CHNS completed rounds in 1989, 1991, 1993, 1997, 2000, 2004, 2006, 2009, 2011, 2015 and 2018. In the current analyses, we used all rounds of CHNS survey data. Our sample included adults who were over 18 years old in eight surveys between 1989 and 2009. We further analysed the relationship between the BP trajectories of this population and the subsequent HTN risks from 2009 to 2018. By the end of 2009, the CHNS had 27 441 participants. We excluded 7525 participants whose BP measurements met the diagnostic criteria for HTN, who reported having HTN, or who were taking antihypertensive drugs in or prior to 2009; 8448 participants who were younger than 18 years in the first survey; 197 participants without BP information; 5956 people who participated in only one or two surveys and 1879 participants without a follow-up visit after 2009 (Fig. 1). The number of visits between 1989 and 2009 that included BP measurements ranged from three to eight per participant: three visits, n 691; four visits, n 645; five visits, n 618; six visits, n 397; seven visits, n 607 and eight visits, n 478 (median, five visits). Our study included 3436 participants and 18 198 observations. The Institutional Review Board of the University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA (No. 07-1963), and the Institutional Review Committee of the China National Institute of Nutrition and Food Safety at the Chinese Center for Disease Control and Prevention, Beijing, China, approved the survey protocols and instruments and the process for obtaining informed consent. All participants provided written informed consent prior to the surveys.

Fig. 1.

Flow chart of the participants included in the current analysis.

Blood pressure assessment

Following standard protocol, trained surveyors measured BP in a quiet room on the right arm using mercury sphygmomanometers with appropriate cuff sizes. They collected three measurements at 5-min intervals after the participant rested seated with feet on the floor and arm supported at heart level for 10 min. Participants were told to avoid strenuous activities, caffeinated beverages, long-term exposure to excessively high or low temperatures and medications that affect BP for 1 h before the measurement. Also they were instructed to stop smoking, empty their bladders and maintain mental relaxation for 15 min before the measurement. Before each measurement, the investigator checked and calibrated the sphygmomanometer. If the investigator found a large difference in a participant’s measurements (a difference >5 mm of mercury (mmHg) between any two measurements), the investigator asked the respondent to rest quietly for 5 min and then measured again until achieving three stable measurements. We used the mean of the three stable measurements in our analyses. SBP is the point at which the trained investigator hears the first of two or more Korotkoff sounds. The disappearance of Korotkoff sounds defines diastolic blood pressure (DBP)^{(Reference Chobanian, Bakris and Black12)}. We defined adult HTN according to the 2018 Chinese guidelines cut point for the management of HTN (SBP/DBP ≥ 140/90 mmHg), taking BP medication, or a self-reported history of HTN⁽¹³⁾.

Modifiable factors

Trained personnel conducted detailed in-person interviews using a structured questionnaire to collect information on demographic characteristics, dietary habits, lifestyle and physical condition. We used a combination of three consecutive 24-h dietary recalls (one weekend day and two weekdays) and a household inventory to assess individual consumption. We defined current smoking as a positive answer to the question, ‘Do you still smoke cigarettes or a pipe?’ We defined alcohol consumption as the participant drank beer or other alcohol last year. We defined physical activity (PA) as the combination of occupational, recreational, commuter and home activities, excluding 1989 because that survey collected home and occupational activities only. We determined the metabolic values of various activities according to the Compendium of PA^{(Reference Ainsworth, Haskell and Herrmann14)}. The PA questionnaire asked the respondents about the intensity, frequency and duration of each activity. We applied the equation PA = PA intensity (metabolic equivalent of task (MET)) × PA time (hours per week). Each survey measured height without shoes to the nearest 0·1 cm using a portable SECA stadiometer (SECA). Each survey measured weight without shoes and wearing light clothing to the nearest 0·1 kg using a calibrated beam scale. We calculated BMI as weight in kg/m2. We measured waist circumference (WC) in cm midway between the lowest rib margin and the top of the iliac crest.

Biochemical analyses in 2009

The CHNS collected blood samples for the first time in 2009; therefore, this analysis uses the diabetes-related biomarkers and lipid-related indexes measured in 2009 as covariates in the relationship between BP trajectories and HTN risk. The 2009 survey collected a 12 ml blood sample after overnight fasting. We analysed all samples in a national central laboratory in Beijing (medical laboratory accreditation certificate ISO 15189:2007) under strict quality control. We measured glucose in the serum with a glucose oxidase phenol 4-amnioantipyrine peroxidase kit (Randox) in a Hitachi 7600 automated analyser (Hitachi Inc.). We measured glycated haemoglobin in whole blood by HPLC with an automated glycohaemoglobin analyser (model HLC-723G7, Tosoh)^{(Reference Batis, Mendez and Gordon-Larsen15)}. We identified type 2 diabetes in self-reports of a history of a diabetes diagnosis, fasting blood glucose ≥ 7·0 mmol/l, glycated haemoglobin ≥ 40·0 mmol/mole (6·5 %) and/or receiving treatment for diabetes. We measured total cholesterol, TAG and LDL-cholesterol using the glycerol-phosphate oxidase method and the polyethylene glycol-modified enzyme method, respectively, by determiner regents (Kyowa Medex Co. Ltd.). We measured all lipids on a Hitachi 7600 automated analyzer^{(Reference Yan, Li and Li16,Reference Cheng, Wang and Wang17)} .

Statistical analyses

We performed all statistical analyses using SAS 9.4 (SAS Institute Inc.). We considered a two-sided P value < 0·05 as statistically significant. We used group-based trajectory modelling to identify distinctive groups of individual BP trajectories adjusted by sex. With Proc Traj, a statistical analysis software macro, we estimated the model parameters^{(Reference Jones, Nagin and Roeder18)} using a censored normal model appropriate for continuous normally distributed data^{(Reference Nagin19)}. We used age as a timescale for the trajectories. We considered linear, quadratic and cubic terms of age and evaluated them based on their significance levels. We grouped participants into five classes of SBP trajectories and four classes of DBP trajectories and modelled these groups’ changes in BP with a variety of different order polynomials. We used statistically rigorous criteria to determine best fit. (1) With the lowest Bayesian information criterion, we used the magnitude of difference in Bayesian information criterion (percentage change) to choose between more complex (with one additional specified trajectory group) and simpler models; (2) we included at least 2·0 % of the sample population in each trajectory class and (3) we ascertained within each group values of average posterior probability of membership in which values >0·7 indicate adequate internal reliability^{(Reference Andruff, Carraro and Thompson20)}. Once we determined BP trajectories, we created a nominal categorical variable to describe the trajectory class of each participant then used that in multiple logistic regression and Cox multivariate regression models.

We used multiple logistic regression models to analyse the modifiable influencing factors of the BP trajectories. We analysed population baseline characteristics in the whole model as influencing factors, including inherently invariable factors (age, gender, urban or rural status, north or south region, education and job) and modifiable behavioural factors (alcohol consumption; cigarette smoking; tea drinking; PA; fruit, vegetable and salt consumption; BMI and WC).

We used Cox multivariate regression models with age as the timescale to estimate associations between BP trajectories and risk of HTN. We calculated hazard ratios and 95 % CI. Entry time was the participant’s age in 2009, and exit time was the age at which the participant was diagnosed with HTN (the age was determined by the survey questions, ‘Have you ever been diagnosed with high blood pressure?’ and ‘How long have you had high blood pressure?’) was lost to follow-up or was censored at the end of the follow-up period, whichever came first. We adjusted models for age, gender, urban or rural status, north or south region, BP levels at the last survey (2009), alcohol consumption, cigarette smoking, PA, salt consumption, BMI, WC, total cholesterol, TAG, LDL-cholesterol and diabetes status in 2009. In the sensitivity analysis, we excluded 252 measurements with obesity (BMI ≥ 28 kg/m2).