Introduction  

Prevention and control of non-communicable diseases are a contemporary global public health priority. At present, 40 million deaths per year, which accounts for nearly 70% of total deaths globally, are attributable to non-communicable diseases.1 2 Moreover, the number of deaths due to non-communicable diseases, such as cardiovascular disease (CVD),3 hypertension,4 diabetes5 and cancer,6 has increased dramatically over the past few decades, although age-standardised CVD and cancer rates as well as systolic blood pressure levels7 have declined.8 9 However, trends in these disease traits have varied across different populations, particularly with less favourable changes observed in East Asian populations compared with Western populations. For example, the prevalence of diabetes10 has increased more rapidly, while the age-standardised prevalence of CVD3 and systolic blood pressure levels7 have fallen less steeply in East Asians in comparison with Westerners.

In addition, adults in East Asia tend to have higher prevalence of physical inactivity,11 which is one of the four target behaviours (including unhealthy diet, tobacco use and harmful use of alcohol) that have been set as the global focus to reduce the risk of non-communicable diseases.12 The beneficial impacts of increased physical activity on various non-communicable outcomes have been demonstrated by numerous previous investigations. However, the majority of previous research has been predicated on evidence from Western populations, thereby limiting its application to other populations including East Asians. As such, little is currently known about levels of physical activity including exercise in relation to non-communicable diseases in East Asian populations as compared with Western populations.13 Another critical gap in the existing literature is the use of data measured only at a single point in time (ie, baseline), in which case physical activity or exercise levels are assumed to remain constant over time. This methodology, therefore, precludes the fact that individuals’ physical activity or exercise levels change with time, and hence may increase the potential for regression dilution.14 Furthermore, it is well known that temporal changes occur in other traditional behavioural and metabolic risk factors for non-communicable diseases, such as adiposity levels,15 16 smoking,17 glucose levels18 and total cholesterol levels,19 exhibiting different patterns of changes between East Asian and Western populations. Nevertheless, no previous research of East Asians or Westerners took into account changes in these risk markers in understanding the relationships between physical activity and non-communicable diseases. Moreover, the dose–response relationship between physical activity and various non-communicable disease outcomes has remained unclear in East Asians. Therefore, the purpose of this research was to explore the dose–response relationships between exercise frequency and various types of incident non-communicable diseases, such as myocardial infarction, stroke, hypertension, type 2 diabetes and site-specific cancers, using a large-scale prospective cohort of South Korean adults with multiple repeated measures of exercise frequency and other risk markers.

Methods

Results

Of an initial sample of 512 190 individuals, 74 931 had missing data on at least one of the model covariates, and 179 405 had self-reported physician-diagnosed heart attack, stroke, hypertension (additionally, systolic ≥140 mm Hg or diastolic ≥90 mm Hg), diabetes (additionally, fasting glucose levels ≥126 mg/dL) or cancer at baseline, respectively. Excluding these individuals resulted in a final sample for analysis of 257 854 individuals (figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

A flow diagram. Note: ‘N’ indicates numbers of total observations (ie, participants who provided repeated measures are treated as separate observations) and ‘n’ indicates numbers of unique participants at baseline. Data without missingness and prevalence of major diseases were used to create final analysis datasets for different incident disease outcomes; while the number of unique participants at baseline is the same for all incident disease outcomes, the total number of observations varied due to the nature of time-updated covariate analyses (ie, censoring of subsequent time-updated covariates when an incident disease case occurs before the end date of repeated measures). Data were obtained from a prospective cohort, which has been established by linking physical examination data of over half a million South Korean adults (2002–2015) with the entire South Korean population’s health insurance system. NHIS, National Health Insurance Service.

Individuals provided up to seven measures of exercise frequency and each confounder (ie, baseline plus six repeated measures). Participants’ characteristics at baseline are summarised in table 1. Online supplementary table 2 summarises participants’ characteristics at each repeat assessment. Individuals in the categories of 1–2, 3–4 or 5–6 times/week of exercise were slightly younger, but showed higher proportions of family history of disease and lower proportions of never smoking or drinking alcohol, compared with those in the categories of none or almost every day of exercise. Across the seven time points (online supplementary figure 1), the proportion of individuals who reported no exercise decreased while the proportion who reported 1–2 or 3–4 times/week of exercise increased; there were no noticeable changes for the categories of 5–6 times/week or almost every day of exercise.

Overall, J-shaped associations were found between exercise frequency and incident myocardial infarction, stroke, hypertension and type 2 diabetes. HRs for these diseases were lowest in the middle categories of exercise frequency (eg, 3–4 or 5–6 times/week) (figure 2). There were no associations for the most frequent exercise category (eg, almost every day) with the incidence of myocardial infarction, stroke and type 2 diabetes.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Associations of exercise frequency with various incident cardiovascular disease outcomes. Cox regression models with age as the underlying timescale were adjusted for sex, body mass index, systolic blood pressure, fasting glucose levels, total cholesterol levels, family history of heart disease, stroke or hypertension (in models for myocardial infarction, stroke and hypertension) or diabetes (in models for type 2 diabetes), smoking status and alcohol consumption. Crude rates are per 100 000 person years. ‘N’ indicates numbers of total observations (ie, participants who provided repeated measures are treated as separate observations) and ‘n’ indicates numbers of unique participants at baseline. Data were obtained from a prospective cohort, which has been established by linking physical examination data of over half a million Korean adults (2002–2015) with the entire South Korean population’s health insurance system.

J-shaped associations were also found for incident stomach, lung, liver and head and neck cancers (figure 3). Higher exercise frequencies (eg, 1–2, 3–4 times/week and almost every day) were associated with lower hazards of incident stomach cancer. No statistical significance was observed for incident colon, rectum, pancreas, kidney, gallbladder and oesophagus cancers. Crude event rates per 100 000 person years in the middle categories of exercise frequency were relatively lower for incident rectum, and oesophagus cancers, but higher for incident pancreas, kidney and gallbladder cancers. Cox regression models with no adjustment for confounders (online supplementary figure 2) and a sensitivity analysis (online supplementary figure 3) in which incident cases occurring in the first 2 years of follow-up were removed both revealed nearly identical patterns of associations as the main analyses.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Associations of exercise frequency with various incident cancer outcomes. Cox regression models with age as the underlying timescale were adjusted for sex, body mass index, systolic blood pressure, fasting glucose levels, total cholesterol levels, family history of cancer, smoking status and alcohol consumption. Crude rates are per 100 000 person years. ‘N’ indicates numbers of total observations (ie, participants who provided repeated measures are treated as separate observations) and ‘n’ indicates numbers of unique participants at baseline. Data were obtained from a prospective cohort, which has been established by linking physical examination data of over half a million South Korean adults (2002–2015) with the entire South Korean population’s health insurance system.

Figure 4 shows comparisons of results that showed statistical significance for multiplicative interaction terms between exercise frequency and each incident disease outcome. Strong J-shaped associations for incident hypertension were identified at each level of body mass index. J-shaped associations of exercise frequency with incident hypertension were strong only in the more favourable levels of smoking (eg, never, previously) and alcohol consumption (eg, never, 2–3 times/month, 1–2 times/week); no or weak associations were identified in the most harmful level of smoking (eg, current smokers) and alcohol consumption (eg, ≥3 times/week). J-shaped associations were evident at all levels of family history of CVD and sex for incident hypertension, and sex for incident type 2 diabetes. Exercise frequency was associated with incident lung cancer in non-obese individuals, but there was no evidence of association in obese individuals. All comparisons stratified by each potential effect modifier are presented in online supplementary figures 4 and 5.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Results from assessment of effect modification of sex, body mass index (BMI), smoking, alcohol consumption and family history of disease in the associations between exercise frequency and each incident disease outcome. Only associations for which the multiplicative interaction terms were statistically significant are presented. Cox regression models with age as the underlying timescale were adjusted for sex (not in models for effect modification by sex), BMI (not in models for effect modification by BMI), systolic blood pressure, fasting glucose levels, total cholesterol levels, family history (not in models for effect modification by family history of respective disease) of heart disease/stroke/hypertension (in models for myocardial infarction, stroke and hypertension), diabetes (in models for type 2 diabetes) or cancer (in models for each cancer), smoking status (not in models for effect modification by smoking status) and alcohol consumption (not in models for effect modification by alcohol consumption). Crude rates are per 100 000 person years. ‘N’ indicates numbers of total observations (ie, participants who provided repeated measures are treated as separate observations) and ‘n’ indicates numbers of unique participants at baseline. P values for multiplicative interactions with exercise frequency are as follows; outcome—hypertension: BMI (0.049), smoking (<0.001), alcohol consumption (<0.001), family history of CVD (0.029) and sex (<0.001); outcome—lung cancer: BMI (0.016), and outcome—type 2 diabetes: sex (0.012). Data were obtained from a prospective cohort, which has been established by linking physical examination data of over half a million South Korean adults (2002–2015) with the entire South Korean population’s health insurance system.

Discussion

This is the first investigation examining the prospective associations of exercise with various incident non-communicable disease outcomes using multiple repeated measures of covariates in East Asian populations. We identified J-shaped associations of sweat-inducing exercise with incident myocardial infarction, stroke, hypertension, type 2 diabetes, stomach, lung, liver and head and neck cancers, with the greatest benefits being observed in the middle categories of exercise frequency (eg, 3–4 or 5–6 times/week): 1–2 times/week for head and neck cancer. These findings provide two important clinical and public health implications. First, prevention and management of non-communicable diseases in East Asians may benefit considerably from employing an exercise promotion approach in the context of combined non-communicable disease prevention. Mechanism research indicates that CVD and type 2 diabetes have similar biological pathways relating to exercise,22 23 so an integrated prevention approach can be applied to control and manage these two diseases at a minimum.5 Moreover, regular participation in exercise can induce favourable changes in intermediate cardiometabolic risk markers,24 which are important predictors of typical non-communicable diseases. Hence, promoting exercise has great potential to act as an integrative behavioural strategy for preventing and controlling various non-communicable diseases simultaneously in East Asian populations.

Second, individuals who engage in exercise 3–4 or 5–6 times/week, rather than every day, may be able to reduce their risk of developing myocardial infarction, stroke, hypertension, type 2 diabetes, stomach, lung and liver cancers: 1–2 times/week for head and neck cancer. Similar J-shaped associations between high intensity exercise (eg, running) and CVD risk have also been reported in previous cohort studies of Western25 26 and Japanese adults.27 Nevertheless, the present study as well as previous research25–27 found that the risk of developing cardiovascular events in individuals who had the highest level of exercise was not noticeably higher compared with those who had the lowest level of exercise. No previous research in East Asians has found such J-shaped relationships between exercise or physical activity and other incident disease outcomes such as hypertension,28–32 diabetes33–40 and different type of cancers.41–47 However, previous meta-analyses of cohort studies comprising predominantly Westerners found leisure-time physical activity to have curvilinear (but not J-shaped) associations with the incidence of type 2 diabetes,48 and linear associations with the incidence of hypertension49 and various site-specific cancers (liver, lung, head and neck, kidney, colon, rectal, bladder, gastric cardia, breast, endometrial, myeloid leukaemia, myeloma, oesophageal adenocarcinoma).50 While additional research is needed to confirm the J-shaped associations of exercise with various incident diseases in other samples of East Asians, findings of this research provide a strong rationale for the development and implementation of public health policies and clinical trials aimed at promoting a moderate level of sweat-causing exercise to minimise the risk of myocardial infarction, stroke, hypertension, type 2 diabetes, stomach, lung, liver and head and neck cancers.

Another finding of this research is that associations of sweat-inducing exercise with hypertension were modified by body mass index, smoking, alcohol consumption, family history of CVD and sex: lung cancer by body mass index and type 2 diabetes by sex. Notably, exercise frequency was not associated with hypertension in individuals who are smokers or drinking alcohol ≥3 times/week (except for 3–4 times/week of exercise). This observation provides some evidence that the harmful impacts of smoking or binge drinking on hypertension51–53 may not be offset completely by exercise. This, in turn, appears to advocate for the need for implementing a combined hypertension prevention strategy targeting promotion of exercise in conjunction with smoking cessation and reductions in alcohol consumption in East Asians.13 For lung cancer, the null associations in individuals with body mass index ≥25 may be indicative of potential residual confounding through reported bias in smoking behaviours. Nonetheless, there was little evidence for effect modification for other disease comparisons, highlighting the importance of promoting exercise for the prevention of various non-communicable diseases in individuals at different levels of body mass index, smoking, alcohol consumption, family history of disease and sex.

This study has several notable strengths. First, we used data from a large prospective cohort study in which exercise and other risk markers were assessed on multiple occasions (up to seven times). Nearly 84% and 5% of the full participants provided one and six repeated measures of all covariates, respectively. Compelling evidence indicates that the risk of regression dilution can be reduced using repeated measures of exposure and confounders.14 Moreover, we examined the dose–response relationship of exercise frequency with a wide variety of specific types of incident non-communicable disease outcomes simultaneously using inpatient and outpatient diagnosis data as well as mortality data. The large sample size (n=257 854) is another strength.

This study has some limitations. Findings of this study may not be generalisable to adult populations of other ethnic origins. Due to the observational nature of this research, no strong causal inference can be drawn about the exercise-incident disease relationships. In addition, the accuracy of hospital admission records is uncertain, although the accuracy of death records from Statistics Korea was found to be 92% in previous research.53 No information about medication use was available in the cohort data, so we could not use it as a potential confounder and another condition when defining disease status (eg, hypertension, type 2 diabetes) at both baseline and follow-up. Furthermore, no exercise duration was assessed; hence, inference was made purely based on exercise frequency. Moreover, ICD-10 codes for sex-specific cancers (eg, prostate and breast cancers) were masked due to the data management policy set forth by the NHIS, so it was not possible to examine such cancers in the present study. The lack of data on diet, which is another behavioural risk marker for non-communicable diseases,12 is another limitation. Moreover, a sizeable proportion (n=74 931; 14.6%) of individuals were excluded due to the missing information on the covariates. Another limitation is that the measurement methods to assess the covariates were not standardised across the different medical institutes participating in the NHIS-HEALS cohort.

Conclusion

Individuals who engaged in sweat-inducing exercise around 3–6 times/week (as opposed to every day) generally had the lowest risk of developing myocardial infarction, stroke, hypertension, type 2 diabetes, stomach, lung, liver and head and neck cancers. These findings were generally applicable to different subpopulations as stratified by body mass index, smoking, alcohol consumption, family history of disease and sex. Public health and lifestyle interventions should promote a moderate level of sweat-inducing exercise as a behavioural strategy for prevention and control of non-communicable diseases in a wider population of East Asians.