International Pooled Analysis of Leisure-Time Physical Activity and Premenopausal Breast Cancer in Women From 19 Cohorts

Abstract

PURPOSE

There is strong evidence that leisure-time physical activity is protective against postmenopausal breast cancer risk but the association with premenopausal breast cancer is less clear. The purpose of this study was to examine the association of physical activity with the risk of developing premenopausal breast cancer.

METHODS

We pooled individual-level data on self-reported leisure-time physical activity across 19 cohort studies comprising 547,601 premenopausal women, with 10,231 incident cases of breast cancer. Multivariable Cox regression was used to estimate hazard ratios (HRs) and 95% CIs for associations of leisure-time physical activity with breast cancer incidence. HRs for high versus low levels of activity were based on a comparison of risk at the 90th versus 10th percentiles of activity. We assessed the linearity of the relationship and examined subtype-specific associations and effect modification across strata of breast cancer risk factors, including adiposity.

RESULTS

Over a median 11.5 years of follow-up (IQR, 8.0-16.1 years), high versus low levels of leisure-time physical activity were associated with a 6% (HR, 0.94 [95% CI, 0.89 to 0.99]) and a 10% (HR, 0.90 [95% CI, 0.85 to 0.95]) reduction in breast cancer risk, before and after adjustment for BMI, respectively. Tests of nonlinearity suggested an approximately linear relationship (P_nonlinearity = .94). The inverse association was particularly strong for human epidermal growth factor receptor 2–enriched breast cancer (HR, 0.57 [95% CI, 0.39 to 0.84]; P_het = .07). Associations did not vary significantly across strata of breast cancer risk factors, including subgroups of adiposity.

CONCLUSION

This large, pooled analysis of cohort studies adds to evidence that engagement in higher levels of leisure-time physical activity may lead to reduced premenopausal breast cancer risk.

INTRODUCTION

Physical activity is an important modifiable behavior, with higher levels of activity strongly associated with reduced risk of all-cause mortality,¹ cardiovascular disease,² and some cancer types.³ Although previous studies have established an inverse association of leisure-time physical activity with postmenopausal breast cancer risk, it is unclear whether the same is true for premenopausal breast cancer, which represents the leading cancer diagnosis among reproductive-age women worldwide.^4,5

CONTEXT

• Key Objective

• Despite the strong evidence of a beneficial effect of physical activity on some cancer types, including postmenopausal breast cancer, less is known about the role of physical activity in reducing premenopausal breast cancer risk. This study sought to investigate this subject in a large international, pooled cohort study.

• Knowledge Generated

• Risk of premenopausal breast cancer was reduced in those with higher levels of leisure-time physical activity, and this inverse association strengthened after adjustment for BMI. The association was approximately linear in its dose-response relationship, with a stronger subtype-specific effect on human epidermal growth factor receptor 2–positive breast cancer.

• Relevance (I. Cheng)

• Leisure-time physical activity may serve as a possible modifiable risk factor for premenopausal breast cancer that may be targeted for the prevention of this disease.*

  *Relevance section written by JCO Associate Editor Iona Cheng, PhD, MPH.

Previous prospective cohort studies have reported inconclusive findings on the association between physical activity and premenopausal breast cancer risk,^6-16 due in part to the low numbers of breast cancer cases during follow-up, and hence, low statistical power. By contrast, meta-analyses^17-22 have been better powered but limited by high levels of heterogeneity across studies, likely because of differences in questionnaire designs, types of physical activity measured (eg, leisure-time, occupational), and modeling assumptions regarding important potential confounders and mediators such as BMI. Moreover, previous studies have examined potential effect modifiers of the physical activity-breast cancer association, including age, BMI, family history of breast cancer, race and ethnicity, and parity; however, among premenopausal women, the evidence for any such factors is limited and inconsistent.^18,19,23,24 There is, however, stronger evidence that no association exists among premenopausal women with higher BMI.¹⁹ Additionally, investigation of subtype-specific associations is lacking and may provide greater insight into potential mechanisms.

To address the limitations of previous studies, we assessed the association between self-reported leisure-time physical activity and premenopausal breast cancer risk in a pooled analysis of 19 prospective cohort studies of 547,601 women, with 10,231 incident cases of breast cancer. We used data from studies that measured leisure-time physical activity, harmonized across studies to cohort-specific percentiles, and estimated the relative risk across a large contrast of the highest activity levels (90th percentile) versus the lowest levels (10th percentile). Leisure-time activity was of particular interest as it is potentially modifiable and targetable through cancer prevention efforts. Using individual-level participant data, we assessed the linearity of the relationships between activity levels and cancer risk, the associations within breast cancer subtypes, and the evidence of effect modification by strata of breast cancer risk factors.

METHODS

Data

We used data from 19 of the 22 cohort studies in the Premenopausal Breast Cancer Collaborative Group²⁵ that had available data on participants’ self-reported leisure-time physical activity data. Individual-level data were pooled across studies from North America (n = 10), Europe (n = 6), Asia (n = 2), and Australia (n = 1).

This secondary analysis was approved by the relevant institutional review boards, and women provided informed consent to participate in the original studies. Full details of the study cohorts are provided in the Data Supplement (Table S1, online only).

Leisure-Time Physical Activity Assessment

Leisure-time physical activities are defined as those performed at an individual's discretion and are not required as essential activities of daily living; examples include sports, exercise, and recreational walking. Within each study, metabolic equivalents of task (METs, the ratio of the energy cost of an activity to the energy expenditure at rest) were derived from questionnaire information and used to quantify physical activity.

Each cohort assessed physical activity either by recording information on overall weekly participation in moderate- and vigorous-intensity activities, or on discrete activities such as walking, running, and cycling. Current guidelines define moderate activity as three to six METs, and vigorous activity as six or more METs.²⁶ The intensity of discrete activities was converted to METs as per the 2011 Compendium.²⁷

Seventeen of the 19 cohorts assessed time per week participating in moderate and vigorous leisure-time physical activities, enabling calculation of total leisure-time MET-hours per week. One cohort included only vigorous-intensity activity,²⁸ and one cohort evaluated frequency of moderate-to vigorous-intensity activities, but not time spent.²⁹ Validation studies for the exposures of each study are described in the Data Supplement (Table S2).

Following the approach of Moore et al,³⁰ leisure-time physical activity MET levels were harmonized by converting them to cohort-specific percentiles, ranging from 0 (relatively low activity) to 100 (relatively high activity).

Outcomes

The main end point was diagnosis with a first invasive or in situ premenopausal breast cancer after exposure assessment. We also conducted analyses of invasive and in situ outcomes separately, as well as by estrogen receptor (ER) and progesterone receptor status (PR), and by clinicopathologic surrogate definitions of intrinsic breast cancer subtype.³¹

Follow-up for breast cancer began with the baseline (enrollment) questionnaire, or at the first questionnaire for which leisure-time physical activity data were recorded. As in previous reports from this collaboration,^32,33 observation time continued until breast cancer diagnosis, with censoring at menopause, last follow-up, death, or age 55 years, whichever occurred first.

Statistical Analysis

We used Cox proportional hazards models to calculate hazard ratios (HRs) and 95% CIs for the association between quintiles of leisure-time physical activity and breast cancer, with attained age as the underlying time scale.

In addition, HRs comparing the 90th and 10th percentiles of cohort-specific distributions were computed as e^90β–10β, where β is the log HR from the linear model for the continuous physical activity percentile.³⁰

We first generated cohort-specific HRs and then obtained a pooled estimate across studies using a random-effects meta-analysis, with weights assigned on the basis of the inverse of the study-specific variance. Because we observed minimal heterogeneity between cohorts with the Cochran Q test and I² statistic,³⁴ we pooled individual-level data across cohorts for all subsequent analyses.

As in previous analyses,^33,35 to include the case-cohort study (Canadian Study of Diet, Lifestyle, and Health) in the pooled analysis, we included Barlow weights³⁶ corresponding to a sampling fraction of 5.0% as an offset in the model.

Models were adjusted for cohort, year of birth, age at menarche (<12, 12-13, ≥14 years, missing), parity (0, 1, 2, ≥3 births, missing), oral contraceptive use (ever, never, missing), first-degree family history of breast cancer (yes, no/missing), smoking status (ever, never, missing), and race and ethnicity (White, Black, Asian, Hispanic, other/missing). We performed further analyses adjusting for alcohol consumption, excluding 10 cohorts where these data were unavailable. Covariate information was updated over follow-up, where available.

We investigated the impact of BMI, time updated where information was available, on findings by presenting results from all models with and without adjustment for BMI (<18.5, 18.5-25, 25-30, 30-35, >35 kg/m²).

We assessed effect modification by attained age (25-34, 35-44, ≥45 years), BMI, age at menarche, parity, oral contraceptive use, first-degree family history of breast cancer, smoking status, birth cohort, and race and ethnicity. P values for each interaction were obtained using log-likelihood ratio tests, which were declared significant if P < .05.

To further assess the relationship between leisure-time physical activity and premenopausal breast cancer risk, we examined HRs across quintiles of leisure-time physical activity. Additionally, we tested for evidence of nonlinearity by applying a likelihood ratio test to three-knot restricted cubic spline models with knots at the 10th, 50th, and 90th percentiles of leisure-time physical activity.

For subtype-specific analyses, we censored at breast cancer events that were not the subtype of interest. Additionally, we used a data augmentation method³⁷ with the Wald test³⁸ to test for heterogeneity (denoted P_het) in the associations with leisure-time physical activity across breast cancer subtypes.

All statistical analyses were conducted using STATA 16.0 (College Station, TX).

Ethics Approval and Informed Consent

Individual study protocols were approved by the relevant institutional review boards and obtained informed consent from participants.

RESULTS

Our pooled data set contained information from 547,601 women, whose median age at exposure assessment was 41.3 years (IQR, 35.2-45.8 years). During the 5,113,522 years of follow-up time (median, 11.5 years; IQR, 8.0-16.1 years), 10,231 in situ or invasive breast cancer cases were diagnosed. Descriptive characteristics for the women included in the analysis are shown in Table 1. Distributions of age and breast cancer risk factors such as BMI were broadly similar across cohorts; however, likely reflecting the varying questionnaire designs, the distribution of leisure-time physical activity showed substantial variation (Data Supplement, Table S1). In view of this, we performed pooled analyses using cohort-specific percentiles and quintiles of physical activity.

TABLE 1.

Characteristics at Study Entry of Women Included in Analyses, by Quintiles of Leisure-Time Physical Activity

Characteristic	Participants (N = 547,601), No.	Person-Years of Follow-Up	Quintile of Leisure-Time Physical Activity					Breast Cancer Cases (n = 10,231), No.
			Q1 (least active; n = 109,528), %	Q2 (n = 109,525), %	Q3 (n = 109,521), %	Q4 (n = 109,515), %	Q5 (most active; n = 109,512), %
BMI (at entry), kg/m2
<18.5	14,327	154,961	3	3	2	2	3	323
18.5-24.9	342,377	3,296,922	57	60	63	65	67	6,977
25-29.9	123,683	1,070,977	23	23	23	22	21	1,972
30-34.9	43,426	376,692	10	9	8	7	6	657
≥35	23,788	213,970	7	5	4	3	3	302
Age at entry, years
<25	15,847	201,027	3	3	3	3	4	75
25-34	119,817	1,907,318	21	22	22	22	23	2,497
35-44	255,193	2,453,678	47	46	47	48	45	5,702
≥45	156,744	551,499	30	30	29	27	28	1,957
Age at menarche, years
7-11	114,826	1,106,948	22	21	21	21	21	2,273
12-13	292,018	2,785,266	53	54	53	54	53	5,614
≥14	128,555	1,129,639	23	23	24	23	25	2,148
Missing or no periods	12,202	91,669	2	2	2	2	2	196
Parity
0	102,331	805,357	16	17	18	20	22	1,404
1	102,506	855,620	20	20	19	18	17	1,668
2	182,461	1,791,735	33	34	34	34	31	3,909
≥3	115,097	1,065,575	21	21	21	21	22	2,242
Not known	45,206	595,736	10	8	8	7	8	1,008
Family history of breast cancer
Yes	63,380	645,068	12	12	11	12	11	2,328
No	357,285	3,511,005	64	64	65	67	68	6,132
Not known	126,936	957,449	25	25	24	22	21	1,771
Oral contraceptive use
Ever	418,147	3,888,483	76	77	77	76	75	7,362
Never	103,696	1,054,549	19	18	18	19	20	2,449
Not known	25,758	170,490	5	5	5	5	5	420
Smoking status
Never	207,345	1,816,693	39	38	37	38	38	4,003
Former/current	323,624	3,206,299	58	59	60	59	59	5,995
Not known	16,632	90,530	3	3	3	3	3	233
Race/ethnicity
Black	43,355	477,088	8	8	8	8	8	811
Asian	14,287	107,251	3	3	2	2	2	185
White	325,452	3,219,638	58	59	60	60	60	6,632
Hispanic	11,400	125,370	2	2	2	2	2	242
Other/not known	153,107	1,184,174	28	28	28	28	28	2,361
Birth cohort
Before 1940	30,723	97,863	6	6	6	5	6	400
1940-1949	156,147	971,137	29	29	28	29	28	2,864
1950-1959	214,595	2,209,444	39	39	40	40	39	4,386
1960-1969	105,535	1,408,527	19	19	19	19	20	2,299
1970-1979	32,683	360,908	6	6	6	6	6	260
1980 or later	7,918	65,643	1	1	1	2	2	22

Leisure-time physical activity was inversely associated with BMI at exposure assessment (P < .001), where mean BMI was 25.3 kg/m² (standard deviation [SD], 5.5 kg/m²), 24.5 kg/m² (SD, 4.7 kg/m²), and 23.9 kg/m² (SD, 4.3 kg/m²) across the first (least active), third, and fifth (most active) quintiles of leisure-time physical activity, respectively. Lower levels of leisure-time physical activity were also associated with having a first-degree family history of breast cancer and ever smoking (Data Supplement, Table S3).

On the basis of a linear trend model for physical activity in the pooled data, a higher level of leisure-time physical activity (90th percentile) compared with a lower level of activity (10th percentile) was associated with a 6% (HR, 0.94 [95% CI, 0.89 to 0.99]; P = .02) reduced breast cancer risk (Fig 1). Further adjusting for BMI as a covariate increased the strength slightly (HR, 0.90 [95% CI, 0.85 to 0.95]; P < .001). Random-effects meta-analysis provided no evidence of significant heterogeneity in the association between studies (I² = 0.0%; Q value = 12.4; P = .83; Data Supplement, Tables S4 and S5). Cohort-specific associations are presented in the Data Supplement (Fig S1). A leave-one-out analysis provided no evidence that associations were overly influenced by a single contributing cohort (Data Supplement, Table S6).

FIG 1.

HRs for breast cancer in relation to leisure-time physical activity from the analysis (A) without and (B) with adjustment for BMI, respectively. HRs are with reference to the 10th percentile of leisure-time physical activity. Shaded areas represent 95% CIs. Three knots are positioned at the 10th, 50th, and 90th percentiles of leisure-time physical activity. HR, hazard ratio.

When leisure-time physical activity was modeled as a categorical variable (ie, defined using cohort-specific quintiles), we observed a graded decrease in breast cancer risk with increasing levels of physical activity (Table 2). Using cubic splines, we observed no evidence that the association between leisure-time physical activity and breast cancer was nonlinear (P_nonlinearity = .94), and HR estimates from the cubic splines approximately paralleled those from the categorical analysis on the basis of quintiles (Fig 1).

TABLE 2.

HRs for Invasive and In Situ Breast Cancer in Relation to Quintiles of Leisure-Time Physical Activity, in a Pooled Analysis of 19 Cohorts

Variable	Quintile of Leisure-Time Physical Activity					P for Linear Trend
	1 (least active)	2	3	4	5 (most active)
Participants, No.	109,528	109,525	109,521	109,515	109,512
Breast cancer cases, No.	2,103	2,078	2,056	2,001	1,993
Person-years	1,004,328	1,011,280	1,017,657	1,035,960	1,044,297
HR (95% CI)
Fully adjusteda	1.00 (ref)	0.98 (0.92 to 1.04)	0.97 (0.91 to 1.03)	0.94 (0.88 to 0.99)	0.94 (0.88 to 1.00)	.02
Further adjusted for BMI	1.00 (ref)	0.97 (0.92 to 1.03)	0.95 (0.90 to 1.01)	0.92 (0.86 to 0.97)	0.91 (0.85 to 0.97)	<.001

Abbreviations: HR, hazard ratio; ref, reference.

^a Adjusted for cohort, year of birth, age at menarche, parity, age at first birth, oral contraceptive use, family history of breast cancer, smoking status, and race/ethnicity.

A sensitivity analysis was performed that included further adjustment for alcohol consumption, using a reduced set of 222,950 women across nine cohorts. Within this set of cohorts, we observed that higher (90th percentile) compared with lower (10th percentile) levels of leisure-time physical activity were associated with a 7% (HR, 0.93 [95% CI, 0.84 to 1.02]) and an 8% (HR, 0.92 [95% CI, 0.83 to 1.02]) reduced risk of breast cancer, respectively, for results unadjusted and adjusted for alcohol consumption (Data Supplement, Table S7).

We found no evidence of reverse causation since there was a relatively minor difference in the estimated HR when we excluded person-time occurring within the first 2 years of follow-up. After these exclusions, the estimated risk reductions were 5% (HR, 0.95 [95% CI, 0.90 to 1.01]) and 10% (0.90 [95% CI, 0.85 to 0.96]) before and after adjustment for BMI, respectively (Data Supplement, Table S8).

We performed further analyses assessing whether leisure-time physical activity might differentially affect breast cancer subtypes (Fig 2; Data Supplement, Table S9). Incident cases included 8,165 invasive and 1,971 in situ breast cancers, and the strength of association with leisure-time physical activity was consistent (P_het = .89) for both invasive (HR, 0.94 [95% CI, 0.89 to 1.00]) and in situ (HR, 0.95 [95% CI, 0.84 to 1.07]) breast cancer. ER status was known for 6,964 cases within 16 cohorts, and we observed a slightly stronger magnitude of association for ER-positive (HR, 0.94 [95% CI, 0.87 to 1.01]) compared with ER-negative (HR, 0.98 [95% CI, 0.86 to 1.13]) breast cancer (P_het = .52). We performed further analyses by intrinsic subtype, which were defined in 4,135 cases across 12 of the 19 cohorts. We found some evidence (P_het = .07) of differences in effect size across luminal A, luminal B, human epidermal growth factor receptor 2 (HER2)–enriched, and triple-negative breast cancers. Notably, within the nonluminal subtypes (ER-negative and PR-negative), associations differed markedly by HER2 status. We observed a much larger effect size estimate for HER2-enriched breast cancer, where higher (90th percentile) compared with lower (10th percentile) levels of activity were associated with an estimated 43% reduction in HER2-enriched breast cancer risk (HR, 0.57 [95% CI, 0.39 to 0.84]; P < .001). The association with HER2-enriched breast cancer remained similar in magnitude after adjustment for BMI. By contrast, there was no association with triple-negative breast cancer (HR, 1.08 [95% CI, 0.83 to 1.40]; P = .58), which was little changed by adjustment for BMI.

FIG 2.

HRs for breast cancer for a higher (90th percentile) versus lower (10th percentile) level of leisure-time physical activity, by breast cancer subtype. ^aAdjusted for attained age (implicit time scale in Cox regression model), cohort, year of birth, age at menarche, parity, oral contraceptive use, first-degree family history of breast cancer, smoking status, and race/ethnicity. ^bAdjusted for covariates in the footnote a plus BMI. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; HR, hazard ratio; PR, progesterone receptor.

In stratified analyses, we found no clear pattern in the strength of association between leisure-time physical activity and breast cancer risk across strata of BMI (P_interaction = .33), although we did observe both a significant reduction in risk of 32% (HR, 0.68 [95% CI, 0.51 to 0.92]; P = .01) and 9% (HR, 0.91 [95% CI, 0.85 to 0.87]; P = .005) for underweight (<18.5 kg/m²) and normal-weight (18.5-24.9 kg/m²) women, respectively. Additionally, within strata of race and ethnicity, we found some evidence of a stronger reduction in risk for Hispanic women (HR, 0.75 [95% CI, 0.58 to 1.01]; P = .06) and White women (HR, 0.94 [95% CI, 0.88 to 1.00]; P = .05), compared with a null association for Black women (HR, 1.06 [95% CI, 0.87 to 1.28]; P = .56). There was no evidence of interaction effects for other breast cancer risk factors, including smoking status and family history of breast cancer (Fig 3; Data Supplement, Table S10).

FIG 3.

HRs for breast cancer for a higher (90th percentile) versus lower (10th percentile) level of leisure-time physical activity, by selected subgroups. ^aAdjusted for attained age (implicit time scale in Cox regression model), cohort, year of birth, age at menarche, parity, oral contraceptive use, first-degree family history of breast cancer, smoking status, and race/ethnicity. ^bAdjusted for covariates in the footnote a plus BMI. HR, hazard ratio.

DISCUSSION

In this pooled analysis of over 540,000 women from 19 prospective cohort studies, we found an inverse association between leisure-time physical activity and premenopausal breast cancer risk. We estimated a reduction in risk of 6%, which further strengthened to 10% after adjustment for BMI. This association was robust to adjustment for a range of breast cancer risk factors and lifestyle behaviors, suggesting minimal influence of residual confounding.

Our findings are broadly consistent with two recent meta-analyses that have assessed the association between leisure-time physical activity and premenopausal breast cancer risk. A report from the Continuous Update Project²² included 12 cohort studies, of which four cohorts contributed to our pooled analysis, in a meta-analysis of around 4,000 cases, and estimated a 7% reduction in risk of premenopausal breast cancer when comparing high versus low levels of leisure-time physical activity. Another study by Neilson et al¹⁹ included separately both 13 cohort studies and 36 case-control studies and estimated a stronger risk reduction of 20% when comparing high versus low levels of leisure-time activity, which attenuated slightly to 17% when restricted only to cohort studies. By contrast, results from individual cohort studies have been inconsistent in both the strength and significance of reported associations, likely owing to low power, given the small numbers of cases in each. It was notable that only one study (Nurses' Health Study I) within our analysis of 19 studies displayed an individually significant finding at P < .05 (unadjusted for BMI), demonstrating the gains in power in our pooled study.

We also examined the interaction between physical activity and BMI in relation to breast cancer risk. Previous studies have provided some evidence of BMI as an effect modifier, with a review by Neilson et al¹⁹ on the basis of seven cohort studies of premenopausal women finding no association among overweight and obese women (relative risk, 0.99), while observing a 15% reduced risk among normal-weight and underweight women. Our results appeared broadly consistent with this, as we observed a strong reduction in risk of 32% associated with physical activity in underweight women along with a significant reduction in risk of 9% in normal-weight women, with weaker evidence in overweight and obese women. A combined role of physical activity, BMI, and energy balance may in part explain the stronger association observed among underweight women.^7,39 For instance, higher physical activity levels, particularly when combined with inadequate energy intake,⁴⁰ could affect menstrual function,^41,42 increase likelihood of anovulatory cycles,⁴³ and reduce hormonal exposures that influence breast cancer risk.

We found that the association with risk of breast cancer remained regardless of whether leisure-time physical activity was modeled in terms of continuous percentiles, or categories on the basis of quintiles, which suggested a graded decrease in risk with increasing activity. Moreover, a spline analysis suggested an approximately linear inverse relationship between increasing activity levels and breast cancer risk. This could explain why previous individual cohort studies^22,44 and a recent Mendelian randomization analysis have consistently observed the strongest risk reductions at high levels of moderate- to vigorous-intensity activity.

In further analyses by subtype, we found no statistical evidence of effect heterogeneity by tumor invasiveness or combinations of ER and PR status; however, given the potentially limited statistical power, it was unclear whether this finding suggested that a protective effect of physical activity generalizes across most breast tumors. Notably, we observed a stronger inverse association with HER2-enriched breast cancer, which displayed a 43% reduction in risk, and this could lend further support to the role of estrogen-independent pathways in premenopausal women, as suggested previously.¹⁹ Further work is required to validate these findings and clarify potential subtype-specific associations.

There is substantial evidence for multiple mechanisms,⁴⁵ and although exact mechanisms are currently unclear, several pathways linking physical activity to premenopausal breast cancer risk have been proposed.^46,47 Some evidence suggests that physical activity causes reduced levels of sex steroid hormones such as estrogens and androgens,^48-50 which could mediate an effect on breast cancer risk. Additionally, the impact of physical activity on decreasing fasting insulin levels and insulin/insulin-like growth factor signaling could also drive breast cancer risk,⁵¹ including through a direct effect on tumor growth.⁵² Higher insulin levels may also reduce sex hormone–binding globulin levels and thereby increase bioavailability of estrogens and androgens,⁵³ suggesting another possible mechanism. The anti-inflammatory effects of physical activity may also play a role.⁵⁴

The strengths of our study include the prospective nature of contributing studies, the large number of cases, and the inclusion of many studies across multiple countries, which reduces the influence of study-specific effects and improves the robustness of findings. Additionally, the detailed covariate information and multiple follow-up rounds for several contributing studies enabled inclusion of time-updated covariates. The pooled cohort design also avoids many of the limitations present in existing meta-analyses, such as methodologic differences across contributing studies and potential publication bias.

There are several limitations to note. First, there was substantial heterogeneity in the distribution of leisure-time MET-hours/week across cohorts, which is likely to be primarily attributable to varying questionnaire designs. Although we sought to address this issue by ranking participants' activity levels within each study, we were unable to use absolute METs as a metric for analysis or draw conclusions about recommended levels of physical activity. Second, we restricted our analysis to a single, harmonized measure of leisure-time physical activity, and we were unable to further assess associations with type, amount, frequency, and intensity of activities, as well as occupational and total activity, because of limited and inconsistent reporting of these measures across studies. Third, we were limited by our reliance on self-reported physical activity measures. Although questionnaire-based measures of physical activity have been reported to be valid, particularly when ranking activity levels, misclassification can still result from factors such as social desirability bias. Finally, the findings may be subject to unmeasured and residual confounding.

In summary, this international study of over 540,000 women suggested a small protective effect of leisure-time physical activity on premenopausal breast cancer risk, providing evidence to support breast cancer prevention efforts. Our findings also supported a linear dose-response relationship for this association, while a potentially stronger relationship with HER2-enriched breast cancer could indicate subtype-specific effects.

DATA SHARING STATEMENT

The data that support the findings of this study are not publicly available because of privacy or ethical restrictions. Study data may be shared upon reasonable request to senior authors.

AUTHOR CONTRIBUTIONS

Conception and design: Iain R. Timmins, Michael E. Jones, A. Heather Eliassen, Xiao-Ou Shu, Elisabete Weiderpass, Walter C. Willett, Alicja Wolk, Dale P. Sandler, Anthony J. Swerdlow, Minouk J. Schoemaker

Financial support: Dale P. Sandler

Administrative support: Hazel B. Nichols

Provision of study materials or patients: Michael E. Jones, Kimberly A. Bertrand, Yu Chen, Jessica Clague DeHart, Laure Dossus, A. Heather Eliassen, Richard S. Houlston, Cari M. Kitahara, Woon-Puay Koh, Martha S. Linet, Brigid M. Lynch, Anne M. May, Roger L. Milne, Julie R. Palmer, Thomas E. Rohan, Maria-Jose Sánchez, Karen Steindorf, Celine M. Vachon, Lars J. Vatten, Walter C. Willett, Alicja Wolk, Hazel B. Nichols, Dale P. Sandler

Collection and assembly of data: Iain R. Timmins, Michael E. Jones, Laura Baglietto, Kimberly A. Bertrand, Kristen D. Brantley, Yu Chen, Jessica Clague DeHart, Tess V. Clendenen, Laure Dossus, A. Heather Eliassen, Agnès Fournier, Susan E. Hankinson, Richard S. Houlston, Victoria A. Kirsh, Cari M. Kitahara, Woon-Puay Koh, Martha S. Linet, Brigid M. Lynch, Roger L. Milne, Julie R. Palmer, Fulvio Ricceri, Maria-Jose Sánchez, Karen Steindorf, Malin Sund, Celine M. Vachon, Lars J. Vatten, Kala Visvanathan, Elisabete Weiderpass, Walter C. Willett, Alicja Wolk, Jian-Min Yuan, Wei Zheng, Hazel B. Nichols, Dale P. Sandler, Anthony J. Swerdlow, Minouk J. Schoemaker

Data analysis and interpretation: Iain R. Timmins, Michael E. Jones, Dagfinn Aune, Kristen D. Brantley, A. Heather Eliassen, Olivia Fletcher, Niclas Håkansson, Susan E. Hankinson, Richard S. Houlston, Corinne E. Joshu, Cari M. Kitahara, Hannah Lui Park, Anne M. May, Roger L. Milne, Julie R. Palmer, Thomas E. Rohan, Kathryn J. Ruddy, Maria-Jose Sánchez, Karl Smith-Byrne, Karen Steindorf, Lars J. Vatten, Kala Visvanathan, Elisabete Weiderpass, Walter C. Willett, Alicja Wolk, Wei Zheng, Anthony J. Swerdlow

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

International Pooled Analysis of Leisure-Time Physical Activity and Premenopausal Breast Cancer in Women From 19 Cohorts

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).