First, the majority of the first-generation studies between physical activity and breast cancer focussed on leisure time physical activity (exercise) and did not always consider the influence of other types of physical activity. Second, the current research focus has shifted towards breast cancer risk reduction in population subgroups and critical time periods in life during which the impact of physical activity may vary. Findings to date are mostly inconclusive (Friedenreich and Cust, 2008). And finally, the role of physical activity at work has been primarily studied in industrialised countries. Less is known about the impact of physical activity in economically less developed countries, where occupational activities may comprise a relatively larger proportion of energy expenditure in daily life (Monda et al, 2008).

To comprehensively assess the role of physical activity, including occupational and non-occupational activities, in breast cancer risk, we examined data from a large cohort of Chinese women, of whom over 99% had held a job outside the home. We sought to evaluate the associations with specific types of physical activity, and to explore whether breast cancer risk varies by age periods of exposure.

Breast cancer risk tended to be inversely associated with the highest level of adult non-occupational physical activities (Table 2). The associations with the exception of adult exercise, however, were not statistically significant and were reduced to near null after adjustment for potential confounding factors, particularly educational attainment. Adolescent exercise was also not associated with breast cancer risk after adjusting for confounding factors (data not shown).

After stratification for menopausal status, a significant reduction in breast cancer risk associated with adult exercise was mainly confined in post-menopausal women. Compared with women who exercised little, post-menopausal women who exercised at or above the current recommended level of 8 MET h per week per year had a 30% lower risk (HR=0.73, 95% CI=0.57, 0.92) of breast cancer compared with women reporting lower levels of exercise (<8 MET h per week per year). A corresponding reduction in risk for this level of exercise was not observed among pre-menopausal women (HR=1.25, 95% CI=0.77, 2.01). Post-menopausal women who exercised at levels that were more than twice the recommended minimum levels did not appear to have any additional reduction in risk (Figure 1). Stratification on BMI showed a decreased risk of breast cancer associated with adult exercise at or above the currently recommended levels in women with a BMI of 23.73kgm⁻² or greater (HR=0.66, 95% CI=0.49, 0.88), but not with a BMI below this level (HR=1.02, 95% CI=0.75–1.40). For walking, cycling, and household chores and adolescent exercise, stratification by menopausal status or BMI did not result in any consistent differences in the strength or direction of the associations by BMI for the exposures examined.

On average, women were employed for 29 years in 2.3 jobs. Jobs with high, medium, and low energy expenditure comprised 7%, 52%, and 41% of the total of 2096414 years worked, respectively. In contrast, 39%, 42%, and 19% were spent in jobs with low, medium, and high sitting time. Increasing lifetime occupational sitting time and decreasing occupational energy expenditure were significantly associated with breast cancer risk when adjusting for age only (Table 3). The average, cumulative, and duration indices for sitting time and energy expenditure showed similar significant trends of decreasing risk with increasing physical activity. The associations for sitting time were attenuated after further adjusting for other potential confounding factors, but the trends remained significant for average sitting time, duration in jobs with low sitting time, and duration in jobs with high energy expenditure. Similarly to non-occupational physical activity, the attenuation in risk after adjustment for potential confounders could be attributed mainly to educational attainment. No clear differences were observed between pre- and post-menopausal women or women with high and low BMI (data not shown).

We also examined the joint effect of occupational sitting time by reported exercise participation in adulthood in post-menopausal women (Table 4). Compared with women who were both occupationally inactive and had inadequate exercise, breast cancer risk was ~30% lower among women who had either active jobs or adequate exercise, but having both an active job and adequate exercise did not confer further reduction in risk. No statistically significant interaction was observed.

Recent Physical Activity Guidelines for Americans (Committee, 2008) recommended that adults accumulate ~8–17 MET h per week of moderate-vigorous physical activity to prevent chronic disease. We examined the dose–response for adult exercise and observed a reduction in risk within the recommended range in post-menopausal women only. Exercise at levels beyond the recommended range (i.e., >8 MET h per week per year) did not appear to confer additional benefit in our population. This was also observed by the Women's Health Initiative Observational study, a cohort of almost 70000 post-menopausal women aged 50–79 years (McTiernan et al, 2003). Interestingly, the exercise effect appeared to be confined to post-menopausal women with higher BMI levels. This finding is in contrast to earlier studies which have found effects mostly in leaner women from the United States and Europe (Thune et al, 1997; McTiernan et al, 2003). It may be that the relatively low BMI levels in this population (i.e., median 23.73kgm⁻²) partially accounts for the present finding, although we have reported an inverse association for exercise among women with lower and higher BMI levels in a previous case–control study from Shanghai (Matthews et al, 2001a). Notably, we did not find evidence for effect modification by BMI on our occupational physical activity exposures in the present study. Our null results for exercise among pre-menopausal women are consistent with studies indicating a weak or little association among younger women (Friedenreich and Cust, 2008). We do not believe that this null finding is due completely to a lack of statistical power since 212 of our cases were pre-menopausal and the HRs remained close to 1.0 in each exposure level.

Our results of a borderline significant decrease in breast cancer risk for women with high levels of occupational physical activity are consistent in the direction and magnitude of the association with several other prospective studies that have reported a 15–25% lower risk with occupational activity (Thune et al, 1997; Moradi et al, 1999; Dirx et al, 2001; Rintala et al, 2002, 2003; Mertens et al, 2006). The results of the joint effect analyses of occupational physical activity and adult exercise showed a lower risk with active jobs among women who did not exercise and suggested that exercise is particularly important for women who held sedentary jobs. However, having both an active job and exercise did not confer an additional reduction in risk. We considered developing a measure of total activity in the present study, reflecting both occupational and non-occupational exposures, but the variation in the length of the time frame for the last job reported in relation to our non-occupational exposures made logical combinations of the two exposures difficult. Future studies with measurements for occupational and non-occupational physical activity that can be integrated more effectively needed to better evaluate the joint effects of these two important exposures on risk for breast cancer.

In addition to occupational activity and exercise, non-exercise activities such as housework and walking and cycling for transportation are also recommended as a means of increasing overall activity levels. In contrast to studies that have reported inverse associations between breast cancer and higher levels of household activities (Friedenreich et al, 2001; Tehard et al, 2006; Lahmann et al, 2007) or walking (Suzuki et al, 2008), we did not find evidence of an inverse association in this population. At least two other prospective studies (Patel et al, 2003; Tehard et al, 2006) also did not observe this relation. The null findings may have resulted from measurement error in our assessment of non-exercise activity (Ferrari et al, 2007). While our evaluation of the physical activity questionnaire was generally supportive of its validity, reproducibility of non-exercise physical activities tended to be lower than for exercise participation, suggesting they may have been more difficult to recall, or that they may have been more variable over time (Matthews et al, 2003). In addition, the assessment of our non-exercise physical activities only reflected the past year of exposure, while our assessment of exercise and occupational physical activity reflected multiple years of exposure, which may provide a better estimate of long-term exposure for occupational activity. Finally, it is possible that the variability in household and transportation activity was low in this population.

To our knowledge, our study is the first population-based prospective cohort study that collected lifetime occupational histories. This has resulted in a more comprehensive measure of long-term occupational exposure than in previous cohort studies, which have confined their exposure assessment to the current or main job or to a specific time period. In addition, questionnaire-based estimates were available for several sources of non-occupational physical activity, allowing for more thorough assessment of the role of physical activity. Other advantages include the high response rate, the availability of detailed information on potential confounding variables and a large proportion of women who had held a job (over 99%), and long (median 26 years) work histories. In addition, most previous studies have focussed on European and North American populations. Our study also suggests an inverse effect in a population that differs in several important respects from these populations: Asian race, lower BMI, few cohort members report a family history of breast cancer, and breast cancer incidence is low. There are also some limitations to our study. Information on non-occupational physical activity was self-reported and limited to the past year (for daily activities) or past 5 years (exercise). We only gathered information about three or fewer exercise activities, which could result in an underestimation of exercise. However, only 1.6% of the women reported three or more exercise activities. This information may be susceptible to systematic reporting errors (Adams et al, 2005) and misclassification due to the natural variability of behaviours over time (Wareham et al, 2000; Matthews et al, 2001b). In addition, a limitation to using job codes for the assessment of occupational physical activity is that a single activity level is linked to the specific job title, and we cannot account for changes in activity levels within a specific job class. Therefore, we may have underestimated the activity levels for job types that have experienced reductions in activity within the job class. The most likely effect of this kind of exposure misclassification (assuming the inverse relationship between activity and breast cancer is true) would be an underestimation of the actual effect. The classification system should reflect a great deal of the modernisation that has taken place in China in recent decades. In addition, there is also a potential for selection bias as a result of women with functional limitations to avoid physically demanding jobs resulting in exposure misclassification.

It has been suggested that there might be critical time periods during which physical activity may be particularly important, since physical activity has been hypothesised to affect breast cancer risk through changes in menstrual characteristics, body size, serum hormone levels, or immune function (Gammon et al, 1998). Two previous occupational studies did not show obvious differences for timing of exposure (D'Avanzo et al, 1996; Levi et al, 1999), while two others suggested that only occupational physical activity during women's reproductive years was associated with decreased breast cancer (Moradi et al, 2000; Peplonska et al, 2008). The lack of an effect by critical time periods in our study may be explained by the fact that the time span over which women in our cohort had worked (roughly age 18–50 years) did not or only partially cover adolescence and the perimenopause for most women. In addition, because of the relatively small number of jobs held by the women in this study (median 2) there may not be sufficient variation in occupational physical activity within study subjects for a meaningful evaluation of the effect of time period.