Data analyzed in this study were obtained from the China Health and Nutrition Survey (CHNS), which was a 26-years national cohort study from 1989 to 2015 and covered nine provinces. A multistage random stratified cluster sampling method was used to select nine provinces from the 31 provinces of China to make the study representative across geography, economic development, and health indicators. Then, four counties were selected in each province using stratified random sampling by income. Villages or neighborhoods within those four counties were sampled randomly. Participants might enter into the CHNS in different waves. The main study population remained fixed between waves 1989 and 1993. Since the wave of 1997, a few new households were added to the survey to replace lost-to-follow-up participants. Meanwhile, an additional province of Heilongjiang was added. Since wave 2011, three megacities of Beijing, Chongqing, and Shanghai were added. In the CHNS, information on household survey, health and nutrition survey, physical activity, aging , body image, mass media behaviors and practices, marraige status, community survey, food market survey, and health and family planning facility surveys were collected. Especially, individual data on demographics, health-related behaviors, health history, dietary intake, blood pressure, and body composition were collected using a validated questionnaire. The details of analysis goals, survey design, and data collection in the CHNS were found in the literature published elsewhere [15].

All subjects entering into the CHNS before wave 2015 and aged 18 years or over at baseline were the targeted population. Participants with any of the following items would be excluded: i. hypertension at baseline, ii. missing data, and iii. implausible outlying data (e.g., a weight >300 kg or <20 kg, a waist circumference (WC) > 200 cm or <20 cm).

Obesity patterns were defined using general and abdominal obesity, which were calculated using body mass index (BMI) and WC, respectively. BMI was calculated as weight in kilograms divided by height squared in meters, both of which were measured according to the standardized protocol. Participants with a BMI ≥28 kg/m² were considered to have general obesity [16]. Otherwise, participants with a BMI <28 kg/m² were identified as normal weight or overweight. Similarly, males with a WC ≥ 85 cm or females with a WC ≥ 80 cm were considered to have abdominal obesity [16]. Obesity patterns were defined as follows: G-/A- indicated participants without neither general obesity nor abdominal obesity, G+/A- or G-/A+ indicated participants with either general obesity or abdominal obesity, and G+/A+ indicated participants with both general and abdominal obesity.

The type of physical exercises in this study included martial arts, track and field, gymnastics, ping pong, badminton or volleyball, and soccer or basketball. Martial arts included Kung Fu and Tai Chi. Track and field included walking and running. And gymnastics included dancing and acrobatics. Additionally, sedentary behavior included TV or computer usage and reading or writing. Each type of physical exercise was collected using a validated questionnaire via a similar question: Do you participate in this activity? 0 indicates no, 1 indicates yes, and 9 indicates unknown [17]. In the final analysis, the option of “unknown” was removed. Since physical exercise types involved in the CHNS were limited to only those mentioned above, other physical exercises, such as cycling and swimming, were not analyzed in this study.

Similarly, each participant was asked their dietary preferences using a similar question: How much do you like this food? [18] The options included five items, which ranked in order from 1 indicating dislike very much to 5 indicating like very much. If participants did not eat this food, the number of 9 was assigned. Dietary preferences were dichotomized as follows: “dislike” included options 1, 2, and 9, and “like” included options 3, 4, and 5.

The first wave of the CHNS was conducted in 1989, and the subsequent follow-up waves were conducted in 1991, 1993, 1997, 2000, 2004, 2006, 2009, 2011, and 2015, respectively. Since participants were not all recruited in 1989, the follow-up times of all participants were different. When participants entered into the CHNS, they were interviewed for blood pressure in each wave. If participants were firstly diagnosed with hypertension in a wave, the exact time when firstly diagnosed with hypertension was further checked. Each participant was required to rest for 10 min in the seated position before measurement of blood pressure. Suitable cuff sizes were chosen according to the upper arm circumference. A standard mercury sphygmomanometer was used to measure diastolic blood pressure (DBP) and systemic blood pressure (SBP), which were indicated by the first and fifth Korotkoff sounds, respectively [19]. These measurements were repeated three times, and the averages were used in the final analysis. If participants with an SBP ≥ 140 mmHg or/and a DBP ≥ 90 mmHg, use of antihypertensive medications, or a self-reported diagnosis with hypertension, they were identified as having hypertension [19].

Continuous variables, such as age, BMI, and WC, were normal distribution. Therefore, means and standard deviations were employed to describe their distributions. Frequencies and constituent ratios were used to express categorical variables. Phi coefficient was used to assess the correlations across physical exercise types and food preferences. To correct the competitive risk of death, Cox regression with Fine-Gray competing risk model was employed to obtain hazard ratios (HRs) and 95% confidence intervals (CIs) on the associations of specific type of physical exercises, dietary preferences, and obesity patterns with the incidence of hypertension. New-onset hypertension was considered an end-event. Age-scale from birth year to the occurrence of hypertension, death, loss to follow-up, or the end of this study, whichever came first, was considered as the time variable. The proportional hazards assumption was confirmed to hold in all Cox regressions, and Cox regression was further stratified by living area. Furthermore, the interactions between sex and exposure variables on the incidence of hypertension were examined. All analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC, United States). A two-tailed p ≤ 0.05 indicated the statistical significance.

There were 10,713 participants included in the final analysis. The number of subjects with hypertension was 4,346, accounting for 40.57%. The proportions of men and women were 43.57 and 56.43%, respectively. The average age was 44.40 ± 15.42 years, and the averages of BMI and WC were 23.41 ± 3.66 kg/m² and 81.54 ± 10.85 cm, respectively. (Table 1).

Correlations across physical exercise types or food preferences were performed. Supplementary Table S1 shows that there were strong correlations between martial arts and gymnastics (Phi = 0.543), between gymnastics and soccer or basketball (Phi = 0.662), and between gymnastics and badminton or volleyball (Phi = 0.658). And there was a very strong correlation between soccer or basketball and badminton or volleyball (Phi = 0.982). Supplementary Table S2 shows that a strong correlation was observed only between fast food and salty snack food (Phi = 0.583).

Table 2 shows that there were significant interactions between sex and soccer or basketball, badminton or volleyball, TV or computer usage, fast food, soft/sugared drinks, salty snack food, and vegetables (HR: 0.848, 0.857, 0.716, 0.756, 0.841, 0.822, and 1.847; 95% CI: 0.701–0.996, 0.710–1.004, 0.572–0.859, 0.586–0.925, 0.715–0.966, 0.677–0.966, and 1.384–2.309; p = 0.029, 0.039, <0.001, 0.001, 0.007, 0.008, and 0.009, respectively).

Table 3 shows the associations of each specific type of physical exercise with the incidence of hypertension. In total population, martial arts, gymnastics, and ping pong were associated with a lower risk of hypertension (HR: 0.792, 0.884, and 0.855; and 95% CI: 0.743–0.845, 0.825–0.948, and 0.767–0.953; and p < 0.001, = 0.001, and 0.005, respectively). However, TV or computer usage was associated with a higher risk of hypertension (HR: 1.418; and 95% CI: 1.315–1.529, p < 0.001). There were no significant associations of track and field, soccer or basketball, badminton or volleyball, and reading or writing with hypertension. The results of urban areas were in line with those in the total population. However, significant associations were observed in martial arts (p < 0.001), track and field (p = 0.041), gymnastics (p = 0.039), and TV or computer usage (p < 0.001) in rural areas. The Kaplan-Meier curves of specific type of physical exercise with the incidence of hypertension are shown in Supplementary Figure S1. And the full adjustment model of the associations of the specific types of physical exercise with the incidence of hypertension is displayed in Supplementary Table S3.

The associations of dietary preferences with the risk of hypertension are displayed in Table 4. Subjects with dietary preferences of fast food, soft/sugared drinks, salty snack food, and fruit had a higher incidence of hypertension (HR: 1.381, 1.233, 1.225, and 1.155; and 95% CI: 1.269–1.504, 1.157–1.314, 1.139–1.316, and 1.021–1.307, respectively). However, a dietary preference for vegetables was not associated with the development of hypertension (p = 0.127). When stratified by region, the results were comparable with those in the total population, except for preference for fruits in rural areas (p = 0.260). The Kaplan-Meier curves of dietary preferences with the incidence of hypertension are shown in Supplementary Figure S2. The full adjustment model of the associations of dietary preferences with the incidence of hypertension is displayed in Supplementary Table S4.

Compared to G-/A-, G+/A- or G-/A+, and G+/A+ could increase the incidence of hypertension. Furthermore, the effect of G+/A+ on hypertension was stronger than G+/A- or G-/A+ (HR: 1.803 vs 1.290; and 95% CI: 1.624–2.003 vs. 1.141–1.459) in the total population, as shown in Table 5. In both rural and urban areas, G+/A- or G-/A+ and G+/A+ were associated with a higher risk of hypertension. Similarly, subjects with G+/A+ had a higher risk of hypertension than those with G+/A- or G-/A+. The Kaplan-Meier curve of obesity patterns with the incidence of hypertension is shown in Supplementary Figure S2. On the other hand, when obesity patterns were divided into four categories, G+/A-, G-/A+, and G+/A+ were linked to a higher incidence of hypertension (HR: 1.505, 1.100, and 1.768; and 95% CI: 1.306–1.734, 1.001–1.208, and 1.646–1.898, respectively) compared to G-/A-. And the full adjustment model of the associations of obesity patterns with the incidence of hypertension is displayed in Supplementary Table S4.

Supplementary Table S5 shows that TV or computer usage mostly contributed to the incidence of hypertension (HR: 1.411; 95% CI: 1.307–1.524; and p < 0.001) when all physical exercise types were analyzed simultaneously. When dietary preferences were analyzed simultaneously, fast food was the most promising predictor of incident hypertension (HR: 1.263; 95% CI: 1.140–1.398; and p < 0.001). When physical exercise types, dietary preferences, and obesity patterns were analyzed simultaneously, an obesity pattern of G+/A+ was the most promising predictor of incident hypertension (HR: 1.845; 95% CI: 1.715–1.985; and p < 0.001).

This study has the advantage of a large-scale representative sample of the Chinese population. Therefore, the results of this study were convictive. The population in this study included a geographically diverse population, which made the results generalizable to the greater Chinese population. Furthermore, the associations of obesity patterns and specific types of physical exercise with the risk of hypertension were rarely investigated, especially in the Chinese population. Therefore, this study was expected to provide additional evidence and insights for the prevention of hypertension. However, there were also some limitations in this study. Firstly, since the salt intake of participants in the CHNS was unavailable, the dietary-related information failed to be adjusted. Secondly, the population analyzed in this study is limited to the ethnically Chinese. Researchers should be cautious in extrapolating the conclusion to other ethnicities. Thirdly, the data on physical exercise and dietary preferences were collected using a self-reported questionnaire. There might be measurement bias. Fourthly, since dietary intakes were not available in the CHNS, dietary preferences but not dietary intake were used to evaluate the association of diet with the risk of hypertension.

Martial arts, gymnastics, and ping pong were associated with a lower risk of hypertension. Dietary preferences of fast food, soft/sugared drinks, and salty snack food, as well as TV or computer usage, increased the risk of hypertension. Obese subjects were more likely to develop hypertension. Furthermore, the obesity pattern of both general and abdominal obesity was the most promising predictor of incident hypertension. The findings of this study have important public health implications regarding unhealthy dietary preferences and which type of physical exercise benefited cardiovascular health.