Long-term exposure to ambient PM2·5, active commuting, and farming activity and cardiovascular disease risk in adults in China: a prospective cohort study

Summary Background Increased physical activity is associated with a reduced risk of cardiovascular disease, but outdoor physical activity can be accompanied by increased inhalation of fine particulate matter (PM2·5). The extent to which long-term exposure to PM2·5 can offset the cardiovascular benefits of physical activity is unknown. We aimed to evaluate whether the associations between active commuting or farming activity and incident risks of cerebrovascular disease and ischaemic heart disease were consistent between populations with different ambient PM2·5 exposures. Methods We did a prospective cohort study using data from people aged 30–79 years without cardiovascular disease at baseline from the China Kadoorie Biobank (CKB). Active commuting and farming activity were assessed at baseline using questionnaires. A high-resolution (1 × 1 km) satellite-based model was used to estimate annual average PM2·5 exposure during the study period. Participants were stratified according to PM2·5 exposure (54 μg/m3 or greater vs less than 54 μg/m3). Hazard ratios (HRs) and 95% CIs for incident cerebrovascular disease and ischaemic heart disease by active commuting and farming activity were estimated using Cox proportional hazard models. Effect modifications by PM2·5 exposure were tested by likelihood ratio tests. Analyses were restricted to the period from Jan 1, 2005, to Dec 31, 2017. Findings Between June 25, 2004, and July 15, 2008, 512 725 people were enrolled in the CKB cohort. 322 399 eligible participants completed the baseline survey and were included in the analysis of active commuting (118 274 non-farmers and 204 125 farmers). Among 204 125 farmers, 2985 reported no farming time and 201 140 were included in the farming activity analysis. During a median follow-up of 11 years, 39 514 cerebrovascular disease cases and 22 313 ischaemic heart disease cases were newly identified. Among non-farmers with exposure to annual average PM2·5 concentrations of less than 54 μg/m3, increased active commuting was associated with lower risks of cerebrovascular disease (highest active commuting vs lowest active commuting HR 0·70, 95% CI 0·65–0·76) and ischaemic heart disease (0·60, 0·54–0·66). However, among non-farmers with exposure to annual average PM2·5 concentrations of 54 μg/m3 or greater, there was no association between active commuting and cerebrovascular disease or ischaemic heart disease. Among farmers with exposure to annual average PM2·5 concentrations of less than 54 μg/m3, increased active commuting (highest active commuting vs lowest active commuting HR 0·77, 95% CI 0·63–0·93) and increased farming activity (highest activity vs lowest activity HR 0·85, 95% CI 0·79–0·92) were both associated with a lower cerebrovascular disease risk. However, among farmers with exposure to annual average PM2·5 concentrations of 54 μg/m3 or greater, increases in active commuting (highest active commuting vs lowest active commuting HR 1·12, 95% CI 1·05–1·19) and farming activity (highest activity vs lowest activity HR 1·18, 95% CI 1·09–1·28) were associated with an elevated cerebrovascular disease risk. The above associations differed significantly between PM2·5 strata (all interaction p values <0·0001). Interpretation For participants with long-term exposure to higher ambient PM2·5 concentrations, the cardiovascular benefits of active commuting and farming activity were significantly attenuated. Higher levels of active commuting and farming activity even increased the cerebrovascular disease risk among farmers with exposure to annual average PM2·5 concentrations of 54 μg/m3 or greater. Funding National Natural Science Foundation of China, National Key Research and Development Program of China, Kadoorie Charitable Foundation, UK Wellcome Trust.


199
Model equation

200
We estimated the associations between physical activity and CVD using Cox proportional hazard

207
In the subgroup analyses, we split the study population (farmers or non-farmers) into two subgroups by 208 the median (or upper tertile in the sensitivity analysis) of PM2.5 exposure. We performed the above 209 stratified Cox models in each subgroup. We used likelihood ratio tests to compare the models with and 210 without the interaction terms of the PM2.5 strata and physical activity.

228
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with 229 restricted cubic splines. Curves within the 95 th percentile of active commuting level were shown.

230
The models were adjusted for sex, age (years), education, household income, occupation, smoking status, alcohol consumption, consumption of fresh vegetables, fresh fruits,

231
and red meat (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time to solid fuels in the past 232 three houses (years), cookstove ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke (hours/day), BMI (kg/m 2 ), WHR, 233 self-rated health status, prevalent hypertension and diabetes, family histories of heart disease and stroke, and the remaining part after deducting the active commuting from

253
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with restricted cubic splines.

254
Curves within the 95 th percentile of active commuting time in each stratum were shown.

255
The models were adjusted for sex, age (years), education, household income, occupation, smoking status, alcohol consumption, consumption of fresh vegetables, fresh fruits, and red meat

262
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with restricted cubic splines.

263
Curves within the 95 th percentile of active commuting level in each stratum were shown.

264
The models were adjusted for sex, age (years), education, household income, occupation, smoking status, alcohol consumption, consumption of fresh vegetables, fresh fruits, and red meat 265 (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time to solid fuels in the past three houses (years), cookstove 266 ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke (hours/day), BMI (kg/m 2 ), WHR, self-rated health status, prevalent hypertension and 267 diabetes, family histories of heart disease and stroke, and the remaining part after deducting the active commuting from the total physical activity (MET-h/d).

270
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with 271 restricted cubic splines. Curves within the 95 th percentile of active commuting level were shown.

272
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of farm, smoking status, alcohol consumption, consumption of fresh 273 vegetables, fresh fruits, and red meat (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time 274 to solid fuels in the past three houses (years), cookstove ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke 275 (hours/day), BMI (kg/m 2 ), WHR, self-rated health status, prevalent hypertension and diabetes, family histories of heart disease and stroke, and the remaining part after

282
MET, metabolic equivalent of task.

283
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of

295
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with 296 restricted cubic splines. Curves within the 95 th percentile of active commuting time in each stratum were shown.

297
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of farm, smoking status, alcohol consumption, consumption of fresh 298 vegetables, fresh fruits, and red meat (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time 299 to solid fuels in the past three houses (years), cookstove ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke 300 (hours/day), BMI (kg/m 2 ), WHR, self-rated health status, prevalent hypertension and diabetes, family histories of heart disease and stroke, and the remaining part after

305
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with 306 restricted cubic splines. Curves within the 95 th percentile of active commuting level in each stratum were shown.

307
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of farm, smoking status, alcohol consumption, consumption of fresh 308 vegetables, fresh fruits, and red meat (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time 309 to solid fuels in the past three houses (years), cookstove ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke 310 (hours/day), BMI (kg/m 2 ), WHR, self-rated health status, prevalent hypertension and diabetes, family histories of heart disease and stroke, and the remaining part after Figure S10. Associations of physical activity from farming with cerebrovascular disease and ischemic heart disease among farmers (n=201,140).

314
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with 315 restricted cubic splines. Curves within the 95 th percentile of farming activity level were shown.

316
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of farm, smoking status, alcohol consumption, consumption of fresh 317 vegetables, fresh fruits, and red meat (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time 318 to solid fuels in the past three houses (years), cookstove ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke 319 (hours/day), BMI (kg/m 2 ), WHR, self-rated health status, prevalent hypertension and diabetes, family histories of heart disease and stroke, and the remaining part after 320 deducting the farming activity from the total physical activity (MET-h/d).

327
MET, metabolic equivalent of task.

328
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of

340
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with 341 restricted cubic splines. Curves within the 95 th percentile of farming time in each stratum were shown.

342
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of farm, smoking status, alcohol consumption, consumption of fresh 343 vegetables, fresh fruits, and red meat (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time 344 to solid fuels in the past three houses (years), cookstove ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke 345 (hours/day), BMI (kg/m 2 ), WHR, self-rated health status, prevalent hypertension and diabetes, family histories of heart disease and stroke, and the remaining part after deducting the farming activity from the total physical activity (MET-h/d).

351
MET, metabolic equivalent of task. Solid lines represent hazard ratios and the ribbons represent 95% confidence intervals from Cox proportional hazard models with 352 restricted cubic splines. Curves within the 95 th percentile of farming activity level in each stratum were shown.

353
The models were adjusted for sex, age (years), education, household income, seasonal work pattern of farm, smoking status, alcohol consumption, consumption of fresh 354 vegetables, fresh fruits, and red meat (days per week), leisure sedentary time (hours per day), fuel types for cooking and heating in baseline house, cumulative exposure time 355 to solid fuels in the past three houses (years), cookstove ventilation in the baseline house, duration of living with a smoker (years), exposure to secondhand smoke 356 (hours/day), BMI (kg/m 2 ), WHR, self-rated health status, prevalent hypertension and diabetes, family histories of heart disease and stroke, and the remaining part after