An Epidemiologic Investigation of Physical Activity and Breast Cancer Risk in Africa

This article is featured in Highlights of This Issue, p. 2619

As a modifiable lifestyle factor, physical activity (PA) has been linked to decreased breast cancer risk, particularly among postmenopausal women (1–8). It was estimated that the protective effect of PA on breast cancer ranges from 15% to 30% among premenopausal women, and 20% to 80% among postmenopausal women, contrasting the most active versus the least active group (9–12). However, most of the studies were conducted in developed countries, with little knowledge on PA level and its impact on breast cancer risk among indigenous African populations, where breast cancer has low incidence, high mortality, and higher proportion of triple-negative subtype (13, 14).

Guthold and colleagues compared prevalence of PA among 22 African countries, and reported large variations by country (lowest in Mali and highest in Mozambique; ref. 15). Few studies had measured PA in details (domain, type, frequency, and intensity) in Africa other than the Sub-Saharan Africa Activity Questionnaire (SSAAQ; ref. 16). To obtain data on PA in indigenous African women and evaluate its association with breast cancer risk, we conducted a study in three African countries: Nigeria, Cameroon, and Uganda.

The African Breast Cancer Study (ABCS) was originally started in Nigeria in 1998 and was expanded to Uganda and Cameroon in 2011 using the same questionnaires and protocol. The study setting and design of the Nigeria site were described in detail elsewhere (17, 18). Participants from the three study sites were recruited as described below: In Nigeria, breast cancer cases were identified through the surgical oncology and radiotherapy units of the University College Hospital (UCH) in Ibadan, Nigeria which serves a population of 3 million and is the main referral center for other hospitals and thus treats the majority of the breast cancer cases in the region. All consecutive females who were 18 years or older, black of African descent, capable of providing informed consent, and had a histologic or clinical diagnosis of invasive breast cancer were eligible. Controls were females ages 18 years or older, absent of breast cancer, and able to give informed consent. In Nigeria, during the period of case enrollment, several communities were randomly selected by ballot from the list of all the communities in catchment area of UCH obtained from census. These communities were selected according to the ethnic and socioeconomical composition of the communities so they can roughly represent UCH patients. Field interviewers randomly approached households in these communities and invited eligible women to visit community centers for the study. We also enrolled hospital-based controls through general internal medicine's outpatient clinic and ophthalmology clinic in the UCH. Women were enrolled during their waiting in clinics and they were unselected for their medical conditions. As the characteristics of hospital-based and community controls were similar, they were pooled in the further analysis.

In Uganda, breast cancer cases were identified at the breast and endocrine unit in department of surgery of the Mulago Hospital, Makerere University in Kampala, Uganda. Mulago Hospital is a national referral hospital in Uganda and serves the 1.3 million residents in Kampala. Controls were randomly recruited from the general outpatient clinics and surgical ward admissions at Mulago Hospital, frequency-matched to cases for age (within 5-year-age category) and ethnicity. In Cameroon, breast cancer cases were enrolled at the department of medical oncology of Yaounde General Hospital in Yaounde, Cameroon, which serves a population of 2.5 million. Controls were randomly recruited from the clinics of general medicine and obstetrics and gynecology departments at the same hospital, frequency-matched to cases for age (within 5-year-age category) and ethnicity. At the Kampala and Yaoundé sites, hospital controls were unselected for their medical conditions. During the recruitment, both patients and community controls were enthusiastic to participate in the ABCS, with a response rate of >90%. The study protocol was reviewed by the institutional review boards of the three sites and the University of Chicago (Chicago, IL).

After obtaining informed consent, trained nurse interviewers administered a structured questionnaire, measured height and weight, and obtained blood samples. In addition to demographics, anthropometry, menstrual and reproductive history, medical history, and family history, a new module of PA was designed and incorporated into the questionnaire since 2011, when the Uganda and Cameroon study centers were established.

After consulting with native investigators in Africa, we developed the PA module to assess the type, frequency, intensity, and duration of various activities in the past year or the year before diagnosis. The PA questionnaire contains three parts: housework PA, occupational PA, and leisure PA. For housework PA, participant was interviewed about how often they conducted housework (months per year, days per week, and hours per day), and then daily duration of their activity performed at various levels of intensity (light, moderate, and vigorous). Occupational PA was assessed in a similar structure for each paid or unpaid job that they worked for over one month in the past year. For both domains, average metabolic equivalent of task (MET) was assigned for the three intensity levels (1.5 MET for light, 2.5 MET for moderate, and 6 MET for vigorous PA). To assess leisure PA, participants were asked to report all activities they did at least six times over the past year for leisure purposes, and we determined the intensity level of each activity according to the Compendium of Physical Activity by Ainsworth and colleagues (19, 20). MET indicates that the intensity of PA and each activity has an assigned value, abstracted from the Compendium of Physical Activities by Ainsworth and colleagues (19, 20); for example, quiet sitting as 1 MET, walking as 4 MET, and jogging as 7 MET. On the basis of these questions, we calculated subscores of housework PA, occupational PA, and leisure PA, as well as a total PA score estimating daily energy expenditure (MET-hours/day). A PA score was calculated by multiplying frequency (number of hours/day) by intensity (MET), and then summarizing the scores by type of PA. By levels of intensity, PA scores were also summarized as light PA, separated from moderate and vigorous PA. Potential confounders include age at diagnosis or interview, education, body mass index (BMI), menopausal status, marital status, parity, age at menarche, occupation, and study site.

Between cases and controls, continuous demographic variables such as age and BMI were compared by t tests, while PA scores were compared by the Wilcoxon rank-sum test. Categorical variables were compared between cases and controls by χ² tests. PA scores that were over 120 MET-hours/day (e.g., vigorous activity at 6 MET for 20 hours per day) were considered as outliers and excluded from the analysis. Continuous PA scores were categorized into quartiles of all participants. Kruskal–Wallis rank test and linear regression were used to explore factors related to PA scores in controls. Logistic regression was used to examine the relationship between PA scores and case–control status. We examine each potential confounder by removing it from the full model and considered a variable as a confounder if the change of point estimate of PA was more than 10%. The final model was adjusted for age (10-year interval categories), BMI (<18.5, 18.5–24.99, 25–29.99, and 30+ kg/m²), study sites (dummies for Cameroon, Nigeria, and Uganda), education (none, primary school, secondary school, vocational/some college, and bachelor degree or above), and postmenopausal status (premenopausal and postmenopausal). Using likelihood ratio test, we tested menopausal status, BMI, and study site as potential effect measure modifiers; and stratified the analysis by BMI and study site, respectively. As the primary analysis showed that PA score varies greatly by country, additional analysis was conducted by stratifying study site. All statistical analysis was conducted using STATA 12.1. A P value of <0.05 was considered as statistically significant.

Between January 2011 and August 2013, 558 cases and 1,014 controls were enrolled into the study and the PA questionnaires were administered. The mean age of cases was 45.7 years, ranging from 18 years to 87 years. Cases were significantly older than controls, had higher education level, and lower BMI (Table 1). Most patients were incident cases, with 75% of patients being interviewed within 4 months of diagnosis.

Among controls, the overall PA score averaged at 17.8 MET-hours/day, mainly composed by housework PA (an average of 7.0 MET-hours/day) and occupational PA (an average of 10.3 MET-hours/day), with low level of leisure PA (an average of 0.5 MET-hours/day). In general, PA scores were the highest in the Uganda study site (total PA average 23.9 MET-hours/day), lower in the Cameroon site (total PA average 21.7 MET-hours/day), and the lowest in the Nigerian site (13.0 MET-hours/day).

Table 2 shows the PA levels among controls. The PA level was the highest in women ages 40 to 50 years old, and the lowest in women ages 70 years or above. Women with education tended to be more physically active, compared with those with no education; those who had a bachelor's degree or above had higher housework PA and occupational PA, on average, compared with women with other education backgrounds. Farmers had the highest overall PA, averaging at 25.5 MET-hour/day. Professionals reported the highest occupational PA (average 13.5 MET-hour/day), even slightly above the farmers, possibly because farmers attributed some of their work-related activities as housework activities. Housewives reported the highest level of housework (average 15.7 MET-hour/day), followed by farmers (average 12.4 MET-hour/day). Farmers and housewives were more likely to engage vigorous PA compared with other occupations. PA scores differed greatly by country, and were significantly lower in Nigeria than in Cameroon and Uganda.

We found that the association between prediagnostic PA and breast cancer risk was not modified by menopausal status (P_interaction = 0.81), and thus we combined pre- and postmenopausal women in the following analysis. Using multivariable logistic regression model (Table 3), breast cancer risk was inversely associated with the total PA score with a dose–response trend (P_trend < 0.001). Compared with the least active quartile, the most active quartile was associated with about 60% reduced risk of breast cancer (OR = 0.39, 95% CI, 0.27–0.58). This trend of PA protecting against breast cancer was highly significant among women with BMI < 25 kg/m² (P_trend < 0.001) and those with BMI between 25 kg/m² and 30 kg/m² (P_trend = 0.003), but insignificant among women with BMI over 30 kg/m² (P_trend = 0.5).

As PA levels strongly differ by country, we performed additional analysis stratified by study site (Table 3). The quartiles of PA scores were categorized on the basis of the distribution of PA for the entire study samples. These results confirmed the inverse associations between PA level and breast cancer risk, with a significant dose response in Nigeria (P_trend < 0.001) and Uganda (P_trend = 0.04). Although there was no apparent dose-response trend in Cameroon (P_trend = 0.2), women in the top three quartile had lower risk of breast cancer than women in the least active quartile.

Light and moderate intensity of PA were both significantly associated with breast cancer risk in a dose-responsive pattern (P_trend < 0.001; Table 4). Vigorous PA was significantly associated with reduced breast cancer risk (approximately 50%), but OR estimates for top and middle tertiles were similar. In addition, we found all domains of PA (housework, occupational, and leisure) were associated with reduced breast cancer risk (Table 5).

With a concern that tumor stage may disturb PA level in one year before the diagnosis, we conducted additional analysis among 102 patients who have complete pathology data. We found that there was no association between PA level and tumor stage.

Using data collected from three African countries, we conducted a multicenter case–control study on prediagnostic PA in relation to breast cancer risk. In Nigeria, Cameroon, and Uganda, we observed that African women's PA mainly consists of housework activities and work-related activities, with little activities conducted for recreational purposes. This is in line with the literature, where researchers found that work-related activities were the most important contributor to overall PA (21) while leisure-time physical activities were the least performed activity type in African countries (15). However, there was another study that reported considerable amount of leisure PA in Cameroon (22), using the SSAAQ, which was modified for the PA module in ABCS.

Our study showed that overall PA was significantly associated with reduced breast cancer risk in both pre- and postmenopausal women. Women who engaged in 25 or more MET-hour/day of total activity had approximately 60% lower risk of breast cancer than the least active women (less than 5 MET-hour/day). The activity level of 25 MET-hours/day translates to about 6 hours of brisk walking per day. Our results also suggested that conducting a large amount of vigorous PA may not further reduce the risk of breast cancer. Some epidemiologic studies suggested that PA is more protective against breast cancer in postmenopausal women than in premenopausal women (1–6), but the findings were not consistent (23, 24). Our study adds to the literature by being one of the few studies on PA and breast cancer risk conducted in the understudied African populations. Only a handful of case–control studies have reported results in African American populations (23, 25, 26).

Interestingly, we found that the association between PA and breast cancer risk varied by body weight as represented by BMI. Higher level of PA does not appear protective against breast cancer risk among obese women; yet for leaner women with BMI < 30 kg/m², there was a strong inverse association. There is much evidence linking weight status and weight change to breast cancer risk, suggesting a higher risk associated with obesity and weight gain in postmenopausal women (27, 28). One likely mechanism underlying exercise and reduced breast cancer risk was consequential body fat reduction leading to reduced substrate for production of estrogen from androgen in fat tissue (29). However, little is known if the inverse association between PA and breast cancer differs by weight, and further studies are needed to confirm this heterogeneous effect.

We observed that even light-intensity PA was associated with lowered breast cancer risk. Many studies reported inverse associations between moderate-to-vigorous PA and reduced breast cancer risk (12, 30, 31), whereas light-intensity PA is less studied because light-intensity PA is typically harder to capture by questionnaire. We found that all domains of PA were associated with reduced breast cancer risk, and this is consistent with other studies that examined household activity (3–5), occupational activity (3, 23), and leisure-time activity (5, 24).

Compared with American women (26), the indigenous African women in our study seemed to have much higher PA level (measured by MET-hour in both studies). There are considerable variations between study sites, where Nigerian women had lower PA scores compared with women from Uganda and Cameroon. This could be a real variation due to cultural and geographic differences by country, but this is also possibly due to systematic bias during questionnaire administration. We adjusted study site in all regression models, and further stratified analysis by study site. We found that the association between PA and breast cancer risk existed in all the three study sites.

Our study has several limitations. We relied on self-reported PA, and the PA module has not been validated by objective measures such as accelerometry. The calculated MET-hours/day was lower but comparable with the validated SSAAQ (16). Our PA module in questionnaire was designed to ask habitual PA, during previous year before interview or diagnosis of breast cancer. However, the cases may underreport their activity level due to stress with breast cancer diagnosis though we did not find activity level varied by tumor stage. Also, PA has been hypothesized to lower breast cancer incidence through several hormone-related mechanisms (32, 33), and some studies found that the association differs by hormone receptor status (34). However, hormone receptor status was only available for a subset of the enrolled cases, and thus we cannot perform this analysis. In addition, the control recruitments were not the same for the three study centers, and not all controls were community based, but hospital based. Hospital-based control participants may have lower PA due to other medical conditions, and therefore the association observed between prediagnostic PA and breast cancer risk may be underestimated. Since the Nigeria center was established a decade earlier, the overall study sample was mostly from Nigeria, with limited generalization toward African populations.

In conclusion, preliminary results of the ABCS study support the hypothesis that PA reduces breast cancer incidence, in both pre- and postmenopausal indigenous African women. The PA levels are traditionally high in Africa, due to large amount of housework and agricultural occupations that require heavy labor. It is possible that the labor-intensive PA patterns contribute to the lower incidence rates in Africa compared with other continents. However, because the case–control design has limitations in making a causal inference, further investigations with larger sample sizes are needed to validate our findings.

No potential conflicts of interest were disclosed.

Conception and design: N. Hou, T. Ogundiran, A. Gakwaya, D. Huo

Development of methodology: N. Hou, T. Ogundiran, I. Morhason-Bello, A. Gakwaya, D. Huo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Ndom, J. Jombwe, T. Ogundiran, A. Ademola, O. Ojengbede, A. Gakwaya, D. Huo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Hou, P. Ndom, I. Morhason-Bello, D. Huo

Writing, review, and/or revision of the manuscript: N. Hou, P. Ndom, T. Ogundiran, I. Morhason-Bello, O. Ojengbede, A. Gakwaya, D. Huo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Jombwe, O. Ojengbede, A. Gakwaya

Study supervision: J. Jombwe, T. Ogundiran, A. Ademola, I. Morhason-Bello, O. Ojengbede, A. Gakwaya, D. Huo

The authors thank Dr. Olofunmilayo I. Olopade for her help in securing funding and insightful comments.

This work was partially supported by Avon Foundation (to D. Huo) and Susan G. Komen for the Cure (to D. Huo).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.