The interaction between early-life body size and physical activity on risk of breast cancer

Abstract

While early-life body leanness is associated with increased breast cancer risk, early-life physical activity may protect against breast cancer. We examined whether the excess risk among lean girls is modified by their levels of prior, concurrent, or future physical activity. We conducted an analysis among 74,723 women in the Nurses’ Health Study II (follow-up 1997–2011). Participants recalled their body size at ages 5, 10, and 20 years in 1989 using a 9-level pictogram (level 1: most lean). In 1997, they reported adolescent levels of physical activity (ages 12–13 and 14–17 years). Cox proportional hazards models estimated the overall association of body size with breast cancer risk and assessed interactions of adolescent physical activity with body size at three different age periods (5–10, 10–20, and 20 years), adjusting for early-life and adult risk factors for breast cancer. Regardless of levels of adolescent physical activity, early-life body leanness (level 1–2 vs. 4.5+) was significantly associated with higher breast cancer risk. The association was slightly attenuated among those who were active (60+ MET-hr/wk) during adolescence compared to those who were inactive (<30 MET-hr/wk) (body size at ages 5–10 years: hazard ratio=1.37, 95% confidence interval=1.04–1.81 vs. 1.66, 1.29–2.12), but the interaction was not significant (p=0.72). The results were similar for body size at three different age periods. Being lean during early life is a risk factor for breast cancer among both inactive and active girls. Adolescent physical activity did not significantly modify the association, although some interaction cannot be excluded.

INTRODUCTION

Evidence consistently supports that early-life body fatness is associated with decreased breast cancer risk in both premenopausal ^1–5 and postmenopausal ^{1, 2, 4–7} women. In our prior analysis in the Nurses’ Health Study (NHS) and NHSII cohorts we observed that childhood and adolescent body fatness was significantly associated with a 20–50% decreased breast cancer risk regardless of menopausal status ⁴. The association did not change after adjustment for adult body mass index (BMI) ⁴ suggesting an independent, lifelong protection from breast cancer. Although most studies were conducted among Caucasian women ^{3–6, 8, 9}, the inverse association was confirmed in African American ^{2, 10}, Hispanic ¹¹, and Asian women ^{1, 7}. This finding contrasts with adult obesity, which increases the risk of postmenopausal breast cancer ¹². The hypothesized underlying mechanisms are increased anovulation ^{13, 14}, altered circulating levels of hormones (e.g., estrogen, IGF-1) ^{15, 16}, altered development of breast tissue (e.g., breast density) ^{17, 18}, and decelerated adolescent growth ¹⁹ in heavier girls.

While early-life body leanness may be a risk factor for breast cancer, some evidence suggests a modest inverse association between early-life physical activity and later breast cancer risk ²⁰. In a recent analysis in the NHSII, we observed that physical activity (highest vs. lowest quintiles) at ages 14–17 years was associated with a 15% lower risk of premenopausal breast cancer ²¹. Thus, we hypothesized that an excess risk associated with early-life body leanness may be attenuated among women who engaged in high levels of activity during their adolescence. An interaction between these two factors is biologically plausible because they share potential mechanisms such as change in IGF-1 and sex hormones levels ^{15, 16, 22}, age at menarche ^23–25, and menstrual cycle patterns ^{23, 26}. Furthermore, high levels of physical activity may help maintain a lean body weight, which we hypothesized may not increase the risk among girls who became lean as a result of physical activity. In contrast, being lean despite being inactive may result from a carcinogenic hormonal profile (e.g., high androgen).

To our knowledge, only one study to date has examined the potential interaction between early-life body size and physical activity on breast cancer risk (p-interaction=0.02) ²⁷. However, this study was a retrospective, case-control study where participants recalled their body size and activity levels after their diagnosis of breast cancer. In a prospective analysis within the NHSII cohort, we considered three different questions when assessing the interactions between body size and physical activity on the risk of breast cancer, with respect to the timing of each. First, we assessed an interaction between body size and ‘future’ physical activity to address the question of whether a girl with a given body size could alter her risk profile by engaging in high levels of activity immediately afterward. Next, we assessed an interaction between ‘concurrent’ body size and physical activity. Finally, we assessed physical activity ‘prior’ to body size to address the question of risk among lean girls who may have been lean as a result of physical activity.

MATERIALS AND METHODS

Study population

The NHSII is an ongoing cohort study that began in 1989, including 116,430 female registered nurses aged 25–42 years. An initial self-administered questionnaire was mailed to collect information on health behaviors, lifestyle and reproductive factors, and medical histories. Biennial follow-up questionnaires updated information on known and suspected risk factors for breast cancer, as well as new disease diagnoses. Response rates are >90% based on cumulative person-time of follow-up.

The follow-up of this analysis began in 1997 when adolescent physical activity was assessed. We excluded women who died or had a diagnosis of cancer, other than nonmelanoma skin cancer, prior to 1997, or did not report their early-life body size or physical activity.

This study was approved by the Committee on the Use of Human Subjects in Research at Brigham and Women’s Hospital (Boston, Massachusetts). Completion of the self-administered questionnaire was considered to imply informed consent.

Assessment of body size

In 1989, participants were asked to recall their body size at ages 5, 10, and 20 years using a 9-level pictogram (Figure 1) ²⁸. According to a validation study conducted within the Third Harvard Growth study, a Boston-area longitudinal study of schoolchildren initiated in 1922, recalled body size using this pictogram by older women at ages 71–76 years (n=100) has a good correlation with their BMI measured while they were at ages 5 (Pearson r=0.60), 10 (r=0.65), and 20 years (r=0.66) ²⁹. We averaged body size at ages 5 and 10 years and ages 10 and 20 years to obtain estimates of childhood and adolescent body size, respectively. Higher levels (>5) of body size were collapsed because there were fewer women in those levels. For stratification analyses, the levels with a similar risk (i.e., level 1 and 2) were collapsed to increase power.

Figure 1.

A 9-level figure drawing used to assess body fatness at ages 5, 10, 20 years in the Nurses’ Health Study II (Reproduced from Stunkard et al. 1983 ²⁸).

Assessment of physical activity

In 1997, participants were asked to recall their average hours per week spent on strenuous (e.g. running, swimming laps) and moderate (e.g. hiking, casual cycling) recreational activities, and walking (to and from school or work) during grades 7–8 (ages 12–13) and 9–12 (ages 14–17). Within a subsample of 160 NHSII participants, the 4-year reproducibility of recalled activities using this questionnaire was good (r=0.64 for total activity) ³⁰.

We calculated participants’ metabolic equivalent (MET)-hour/week of total physical activity by multiplying the number of hours per week of each activity with its corresponding average MET values (strenuous = 7 METs; moderate = 4.5 METs; walking=3 METs) ³¹ and then summing the values from all activities. We averaged total activity (sum of recreational activities and walking) at ages 12–13 and 14–17 years to obtain an estimate of adolescent activity. Because we previously observed the strongest association with breast cancer by activity at ages 14–17 years ²¹, we conducted a sensitivity analysis using only physical activity at ages 14–17 years.

Ascertainment of breast cancer cases

Breast cancer cases diagnosed through May 31, 2011 were identified by self-administered biennial questionnaire and the National Death Index, and confirmed by reviewing medical records. Information on tumor characteristics was abstracted from the pathology reports. The analysis included both premenopausal and postmenopausal cases and both invasive and in situ cases because the association with early-life body size did not vary by these characteristics ⁴. A total of 2,641 cases were identified during 1997–2011 follow-up (622 in situ and 1,769 invasive cases confirmed by medical records; 1,683 estrogen receptor-positive and 381 estrogen receptor-negative tumors). Among those with medical records, 99% of reported cases were confirmed; therefore, we included 250 breast cancer cases for whom the medical records were missing and only verbally confirmed by telephone.

Covariate assessment

Height, adolescent alcohol intake, adolescent smoking status, age at menarche, menstrual length and pattern at ages 18–22 years, and BMI at age 18 were reported once via the 1989 baseline questionnaire. Age, parity/age at first birth, personal history of benign breast disease, oral contraceptive use, smoking, menopausal status, and current BMI were updated biennially and family history of breast cancer was updated every 4 years since 1989. Birth weight was assessed in 1991 and adolescent physical activity in 1997. Adult alcohol intake was reported in 1991 and updated every 4 years. Adult physical activity was reported in 1989, 1991, 1997, 2001, 2005 and 2009.

Statistical analysis

Each woman contributed person-time from the return date of her questionnaire in 1997 until a diagnosis of cancer except nonmelanoma skin cancer, death, or the end of follow-up on June 1, 2011, whichever came earlier. Cox proportional hazards models were used to estimate hazard ratios (HR) and their 95% confidence intervals (CI) for categories of body size. We performed tests for trend by modeling body size as a continuous variable (level 1–9).

Multivariate-adjusted models included both early-life and adult risk factors for breast cancer. Body size at different age periods during early-life were not simultaneously adjusted in the model because they were strongly correlated with each other (r = 0.5–0.8). Covariates that were reported more than once during the follow up were treated as time-varying covariates. Because age at menarche, menstrual pattern at ages 18–22 years, BMI at age 18 (for body size at ages 5–10 years only), and current BMI could be intermediate variables through which early-life body size is associated with breast cancer, we adjusted for them only in secondary analyses.

To assess whether the association varied by the level of adolescent physical activity, we stratified the participants by approximate tertiles (inactive, moderately active, active: <30, 30–59.9, 60+ MET-hr/wk). To assess the joint effect of body size and physical activity, we cross-classified by both body size and physical activity and estimated the HR using the most overweight and inactive (<30 MET-hr/wk) group as a reference. The same analyses were performed on: 1) childhood body size and adolescent (‘future’) physical activity, 2) adolescent body size and adolescent (‘concurrent’) physical activity, and 3) body size at age 20 years and adolescent (‘prior’) physical activity. Likelihood ratio tests compared models with and without interaction terms to determine significance of the interactions. We repeated the analyses after restricting to premenopausal women (1,364 cases) because adolescent activity is suggestively more strongly associated with the risk of premenopausal breast cancer ²¹. For these analyses, women contributed person-time only while they were premenopausal.

Lastly, we used a competing risks Cox proportional hazards model ^{32, 33}, using a data augmentation method ³⁴, to assess whether the interaction differed by estrogen receptor (ER) status of tumor. A separate observation was created for each subject for each type of tumor (ER+/−) and stratified the data on event type, allowing for estimation of separate associations with each type of tumor. Likelihood ratio tests compared models that assumed different associations for each type of tumor with models that assumed the same association. All statistical tests were two-sided with 5% type I error and performed using SAS software version 9.

RESULTS

The mean age at the beginning of follow-up was 42 years for the 74,723 women who were included in our analysis. Compared with women who were overweight during childhood, those who were lean were more likely to be older at menarche, engage in higher levels of physical activity, and have lower BMI later in life (Table 1). These women were also less likely to have weighed 3.9 kg or more at birth and have smoked during adolescence and adulthood, and more likely to have a personal history of benign breast disease.

Table 1.

Baseline characteristics of study population in 1997 by average childhood (ages 5–10 years) self-reported body size in the Nurses’ Health Study II (N=74,723)^a

	Average body size at ages 5–10 years (Levelb)
	1	1.5–2	2.5–3	3.5–4	4.5+ c
N	12,609	22,846	18,960	12,759	7,549
	Mean (SD) or Percentage
Age (years)	42.5 (4.8)	41.8 (4.7)	42.0 (4.6)	42.3 (4.6)	42.5 (4.6)
Height (inches)	64.9 (2.6)	64.9 (2.6)	64.8 (2.6)	64.9 (2.6)	65.1 (2.6)
Birth weight ≥ 3.9 kg, %	10.0	11.5	13.9	16.2	18.7
Age at menarche (years)	12.8 (1.4)	12.6 (1.4)	12.3 (1.4)	12.1 (1.4)	12.0 (1.4)
Menstrual cycle pattern at ages 18–22 years
- Irregular or no periods, %	24.5	23.3	23.9	24.4	26.5
Time to menstrual regularity since menarche
- 5+ years, %	23.3	21.8	21.7	22.4	24.8
BMI at age 18 years (kg/m2)	19.4 (2.1)	20.2 (2.3)	21.5 (2.8)	22.7 (3.3)	24.4 (4.7)
Total physical activity at ages 12–17 years (MET-hr/wk)	56.3 (37.9)	54.5 (35.9)	51.9 (35.2)	48.4 (33.9)	46.0 (33.8)
Alcohol drinkers at ages 15–17 years, %	22.8	23.0	23.0	23.6	23.2
Alcohol intake at ages 15–17 years among drinkers (g/day)	4.6 (6.3)	4.2(5.4)	4.3(5.4)	4.7(6.5)	4.9(7.5)
Total caloric intake during high school, kcal/d	2465 (580)	2488 (566)	2493 (557)	2486 (573)	2505 (561)
Smoking at ages 10–20 years, %	31.0	30.1	31.4	34.1	38.0
Current BMI (kg/m2)	23.8 (4.2)	24.6 (4.8)	26.4 (5.9)	28.1 (6.7)	30.0 (7.7)
Current total physical activity (MET-hr/wk)	19.1 (24.7)	18.7 (22.3)	18.2 (22.2)	18.3 (22.5)	17.5 (21.8)
Current alcohol intake (g/day)	3.5 (6.6)	3.5 (6.4)	3.5 (6.6)	3.5 (6.9)	3.3 (6.7)
Current smoking status
- Never, %	66.7	68.0	66.6	63.8	59.6
- Past, %	22.9	23.4	24.4	25.4	27.2
- Current, %	10.3	8.6	9.0	10.8	13.2
Ever use of oral contraceptives
- Never, %	11.7	12.9	14.0	14.2	15.6
- Past users, %	79.1	78.4	77.9	77.5	76.6
- Current users, %	9.1	8.7	8.1	8.3	7.7
Parous, %	80.7	82.4	81.1	78.9	73.1
Parity among parous women	2.2 (0.9)	2.3 (0.9)	2.3 (0.9)	2.2 (0.9)	2.2 (0.9)
Age at first birth among parous women (years)	26.3 (4.6)	26.5 (4.6)	26.5 (4.6)	26.5 (4.6)	26.5 (4.7)
Premenopausal women, %	88.2	89.2	89.7	89.4	88.7
First-degree family history of breast cancer, %	9.6	9.7	9.3	9.3	9.5
Personal history of benign breast disease, %	46.9	45.5	44.7	43.4	43.3

^aValues are standardized to the age distribution of the study population. Abbreviations: SD: standard deviation; kg: kilograms; g: grams; m: meters; BMI: body mass index; MET-hr/wk: metabolic equivalent of task-hour/week; kcal/d: kilocalorie per day.

^bParticipants recalled their body size at ages 5 and 10 years using a 9-level pictogram (level 1: most lean; level 9: most overweight). We averaged body size at ages 5 and 10 years to obtain an estimate of childhood body size.

^cHigher levels (4.5 to 9) were collapsed because there were fewer women in those levels.

Body size at ages 5 to 20 years were all significantly inversely associated with breast cancer risk (Table 2). Women who were lean (level 1 vs. 4.5+) during their childhood had a 55% higher risk compared to the women who were overweight, adjusting for adolescent and adult risk factors for breast cancer (HR=1.55, 95% CI=1.31–1.83; p-trend<0.0001). The estimates were similar to those from age-adjusted models. The association became slightly stronger when menstrual characteristics were added to the model; the change was primarily driven by age at menarche. Adjustment for current adult BMI only slightly attenuated the association (HR=1.48, 95% CI=1.24–1.75) and adjustment for BMI at age 18 somewhat attenuated the association (HR=1.38, 95% CI=1.15–1.65).

Table 2.

Hazard ratios of breast cancer according to self-reported early-life body size in the Nurses’ Health Study II, 1997–2011

	Average body size (level)a					Per 1-unit decrease in body size	p-trendf
	1	1.5 – 2	2.5 – 3	3.5 – 4	4.5+
	Hazard Ratio (95% Confidence Interval)
AGES 5–10 YEARS
No. of cases	503	895	643	404	196
Person-years	168,232	305,585	253,570	170,207	100,788
Age-adjusted	1.53 (1.30, 1.81)	1.57 (1.34, 1.83)	1.35 (1.15, 1.58)	1.24 (1.04, 1.47)	1.00 (Ref)	1.10 (1.07, 1.14)	<0.0001
MVb -adjusted	1.55 (1.31, 1.83)	1.56 (1.33, 1.82)	1.35 (1.15, 1.58)	1.24 (1.04, 1.47)	1.00 (Ref)	1.11 (1.07, 1.14)	<0.0001
MVb without physical activityc	1.53 (1.30, 1.81)	1.54 (1.32, 1.80)	1.34 (1.14, 1.57)	1.24 (1.04, 1.47)	1.00 (Ref)	1.10 (1.07, 1.14)	<0.0001
MVb + Menstrual characteristicsd	1.63 (1.38, 1.93)	1.61 (1.38, 1.89)	1.38 (1.17, 1.62)	1.25 (1.05, 1.48)	1.00 (Ref)	1.12 (1.08, 1.16)	<0.0001
MVb + Current BMIe	1.48 (1.24, 1.75)	1.49 (1.27, 1.75)	1.31 (1.11, 1.54)	1.22 (1.03, 1.45)	1.00 (Ref)	1.09 (1.06, 1.13)	<0.0001
MVb + BMI at age 18	1.38 (1.15, 1.65)	1.41 (1.19, 1.67)	1.26 (1.07, 1.49)	1.19 (1.00, 1.41)	1.00 (Ref)	1.07 (1.03, 1.11)	0.0002
AGES 10–20 YEARS
No. of cases	110	768	935	590	238
Person-years	33,311	253,995	348,090	233,982	129,004
Age-adjusted	1.64 (1.31, 2.06)	1.63 (1.41, 1.88)	1.49 (1.29, 1.72)	1.37 (1.18, 1.60)	1.00 (Ref)	1.14 (1.10, 1.18)	<0.0001
MVb –adjusted	1.64 (1.30, 2.06)	1.63 (1.40, 1.89)	1.49 (1.29, 1.72)	1.38 (1.19, 1.61)	1.00 (Ref)	1.14 (1.09, 1.18)	<0.0001
MVb without physical activityc	1.62 (1.29, 2.03)	1.60 (1.38, 1.86)	1.48 (1.28, 1.70)	1.38 (1.18, 1.60)	1.00 (Ref)	1.13 (1.09, 1.17)	<0.0001
MVb + Menstrual characteristicsd	1.74 (1.38, 2.19)	1.70 (1.47, 1.98)	1.53 (1.32, 1.77)	1.40 (1.20, 1.62)	1.00 (Ref)	1.15 (1.11, 1.20)	<0.0001
MVb + current BMIe	1.56 (1.23, 1.97)	1.56 (1.33, 1.82)	1.44 (1.24, 1.67)	1.36 (1.17, 1.58)	1.00 (Ref)	1.12 (1.08, 1.17)	<0.0001

	Body size (level)a					Per 1-unit decrease in body size	p-trendf
	1	2	3	4	5+
	Hazard Ratio (95% Confidence Interval)
AGE 20 YEARS
No. of cases	135	778	996	526	206
Person-years	42,612	258,124	378,017	210,564	109,065
Age-adjusted	1.53 (1.23, 1.90)	1.55 (1.33, 1.81)	1.39 (1.20, 1.62)	1.32 (1.12, 1.55)	1.00 (Ref)	1.10 (1.07, 1.14)	<0.0001
MVb –adjusted	1.54 (1.23, 1.91)	1.56 (1.34, 1.82)	1.40 (1.21, 1.64)	1.33 (1.13, 1.57)	1.00 (Ref)	1.10 (1.07, 1.14)	<0.0001
MVb without physical activityc	1.52 (1.22, 1.89)	1.54 (1.32, 1.80)	1.39 (1.19, 1.62)	1.33 (1.13, 1.56)	1.00 (Ref)	1.10 (1.06, 1.14)	<0.0001
MVb + Menstrual characteristicsd	1.61 (1.29, 2.00)	1.61 (1.38, 1.88)	1.43 (1.23, 1.66)	1.34 (1.14, 1.58)	1.00 (Ref)	1.11 (1.08, 1.16)	<0.0001
MVb + current BMIe	1.44 (1.15, 1.81)	1.47 (1.25, 1.74)	1.34 (1.15, 1.57)	1.30 (1.10, 1.53)	1.00 (Ref)	1.09 (1.05, 1.13)	<0.0001

^aParticipants recalled their body size at ages 5, 10, and 20 years using a 9-level pictogram (level 1: most lean; level 9: most overweight). Higher levels (4.5+ or 5+) were collapsed because there were fewer women in those levels.

^bMultivariate (MV)-adjusted model includes age (years, continuous), average alcohol intake between ages 15–18 years (g/day, continuous), average physical activity between ages 12–17 years (MET-hr/wk, continuous), birthweight (<5.5/5.5–6.9/7–8.4/8.5+), height (inches, continuous), parity/age at first birth (nulliparous/1–2 at < 25 years/1–2 at 25–29 years/1–2 at 30+ years/3+ at <25 years/3+ at 25+ years), ever oral contraceptive use (never/past/current), smoking status (never/past with adolescent smoking/past with no adolescent smoking/current with adolescent smoking/current with no adolescent smoking), adult alcohol intake (g/d, continuous), family history of breast cancer (yes/no), personal history of benign breast disease (yes/no), adult physical activity (MET-hr/wk, continuous), menopausal status/postmenopausal hormone use (premenopausal/postmenopausal never user/postmenopausal past user/postmenopausal current user/dubious), and age at menopause (year, continuous).

^cAdolescent physical activity (average MET-hr/wk at ages 12–17 years, continuous) and adult physical activity (MET-hr/wk, continuous).

^dMenstrual characteristics includes age at menarche (<12/12/13/14/15+ years), menstrual pattern at ages 18–22 years (regular/irregular or no periods), time to menstrual regularity since menarche (<1/1–4/5+ years or never).

^eCurrent adult BMI (kg/m², continuous)

^fp-trend was estimated using a Wald test for a continuous variable of body size (level 1–9).

Within each stratum of adolescent activity, early-life body leanness was significantly associated with higher breast cancer risk (Table 3). Multivariate-adjusted HR for women who were lean (level 1–2 vs. 4.5+) during their childhood were 1.66 (95% CI=1.29–2.12) among inactive women, 1.59 (95% CI=1.22–2.08) among moderately active women, and 1.37 (95% CI=1.04–1.81) among active women. Similar patterns were observed for each time comparison of body size and physical activity, with slightly attenuated positive associations for body leanness among women who engaged in higher levels of activity during adolescence. However, the interactions were not statistically significant.

Table 3.

Hazard ratios of breast cancer according to self-reported body size, stratified by adolescent physical activity (average between ages 12–17 years) in the Nurses’ Health Study II, 1997–2011

		Average body size (level)a				Per 1-unit decrease in body size
		1–2	2.5–3	3.5–4	4.5+
	Multivariate hazard ratio (95% confidence interval)b
Stratified by ‘future’ physical activity:
Average childhood body size (ages 5–10 years) and average adolescent physical activity (ages 12–17 years)
All women	Cases HR 95% CI	1,398 1.55 (1.33, 1.81)	643 1.35 (1.15, 1.58)	404 1.24 (1.04, 1.47)	196 1.00 (Ref)	1.11 (1.07, 1.14)
Inactive women: <30 MET-hr/wk	Cases HR 95% CI	431 1.66 (1.29, 2.12)	216 1.45 (1.11, 1.89)	145 1.22 (0.92, 1.62)	76 1.00 (Ref)	1.12 (1.06, 1.19)
Moderately active women: 30–59.9 MET-hr/wk	Cases HR 95% CI	479 1.59 (1.22, 2.08)	229 1.40 (1.06, 1.86)	149 1.34 (1.00, 1.81)	63 1.00 (Ref)	1.11 (1.05, 1.18)
Active women: 60+ MET-hr/wk	Cases HR 95% CI	488 1.37 (1.04, 1.81)	198 1.18 (0.88, 1.59)	110 1.09 (0.79, 1.51)	57 1.00 (Ref)	1.08 (1.02, 1.15)
	p-interactionc = 0.72
Stratified by ‘concurrent’ physical activity:
Average adolescent body size (ages 10–20 years) and average adolescent physical activity (ages 12–17 years)
All women	Cases HR 95% CI	878 1.63 (1.41, 1.88)	935 1.49 (1.29, 1.72)	590 1.38 (1.19, 1.61)	238 1.00 (Ref)	1.14 (1.09, 1.18)
Inactive women: <30 MET-hr/wk	Cases HR 95% CI	264 1.74 (1.37, 2.21)	294 1.67 (1.32, 2.11)	215 1.52 (1.19, 1.94)	95 1.00 (Ref)	1.16 (1.09, 1.23)
Moderately active women: 30–59.9 MET-hr/wk	Cases HR 95% CI	314 1.59 (1.24, 2.04)	323 1.40 (1.09, 1.79)	202 1.26 (0.97, 1.63)	81 1.00 (Ref)	1.14 (1.07, 1.21)
Active women: 60+ MET-hr/wk	Cases HR 95% CI	300 1.52 (1.15, 2.00)	318 1.41 (1.07, 1.86)	173 1.38 (1.03, 1.85)	62 1.00 (Ref)	1.11 (1.03, 1.18)
	p-interactionc = 0.72

		Body size (level)a				Per 1-unit decrease in body size
		1–2	3	4	5+
	Multivariate hazard ratio (95% confidence interval)b
Stratified by ‘prior’ physical activity:
Body size at age 20 years and average adolescent physical activity (ages 12–17 years)
All women	Cases HR 95% CI	913 1.56 (1.34, 1.82)	996 1.40 (1.21, 1.64)	526 1.33 (1.13, 1.57)	206 1.00 (Ref)	1.10 (1.07, 1.14)
Inactive women: <30 MET-hr/wk	Cases HR 95% CI	271 1.73 (1.33, 2.25)	330 1.72 (1.33, 2.22)	193 1.52 (1.16, 1.99)	74 1.00 (Ref)	1.12 (1.06, 1.19)
Moderately active women: 30–59.9 MET-hr/wk	Cases HR 95% CI	331 1.39 (1.08, 1.79)	337 1.17 (0.91, 1.51)	176 1.11 (0.85, 1.46)	76 1.00 (Ref)	1.10 (1.04, 1.17)
Active women: 60+ MET-hr/wk	Cases HR 95% CI	311 1.48 (1.11, 1.98)	329 1.32 (0.99, 1.76)	157 1.37 (1.00, 1.86)	56 1.00 (Ref)	1.08 (1.01, 1.15)
	p-interactionc = 0.64

^aParticipants recalled their body size at ages 5, 10, and 20 years using a 9-level pictogram (level 1: most lean; level 9: most overweight). Higher levels (4.5+ or 5+) were collapsed because there were fewer women in those levels.

^bAdjusted for age (years, continuous), average alcohol intake between ages 15–18 years (g/day, continuous), average physical activity between ages 12–17 years (MET-hr/wk, continuous), birthweight (<5.5/5.5–6.9/7–8.4/8.5+), height (inches, continuous), parity/age at first birth (nulliparous/1–2 at < 25 years/1–2 at 25–29 years/1–2 at 30+ years/3+ at <25 years/3+ at 25+ years), ever oral contraceptive use (never/past/current), smoking status (never/past with adolescent smoking/past with no adolescent smoking/current with adolescent smoking/current with no adolescent smoking), adult alcohol intake (g/d, continuous), family history of breast cancer (yes/no), personal history of benign breast disease (yes/no), adult physical activity (MET-hr/wk, continuous), menopausal status/postmenopausal hormone use (premenopausal/postmenopausal never user/postmenopausal past user/postmenopausal current user/dubious), and age at menopause (year, continuous).

^cp-interaction was estimated using a likelihood ratio test comparing models with and without a interaction term (between a continuous body size level 1–9 and a continuous physical activity MET-hr/wk)

When we cross-classified body size and physical activity, the risk in women who were lean and active in early-life was only suggestively lower compared to women who were lean and inactive (future activity: HR=1.55, 95% CI=1.15–2.09 vs. HR=1.66, 95% CI=1.30–2.13; concurrent activity: HR=1.61, 95% CI=1.21–2.15 vs. HR=1.72, 95% CI=1.36–2.17; prior activity: HR=1.69, 95% CI=1.24–2.30 vs. HR=1.73, 95% CI=1.34–2.25) and their confidence intervals overlapped considerably (Figure 2). The results did not differ by ER status of tumors (p-heterogeneity_{future activity}=0.47, p-heterogeneity_{concurrent activity}=0.40, p-heterogeneity_{prior activity}=0.58) and were consistent when we restricted the analyses to premenopausal women or using activity (tertile/quintile) at ages 14–17 years only.

Figure 2.

Hazard ratios^a of breast cancer according to self-reported adolescent physical activity (average between ages 12–17 years) and body size^b at three different age periods (ages 5–10, 10–20, and 20 years) in the Nurses’ Health Study II, 1997–2011 (reference = body size 4.5+ or 5+ level, physical activity <30 MET-hr/wk)

*Statistically significant (p<0.05)

^aAdjusted for age (years, continuous), average alcohol intake between ages 15–18 years (g/day, continuous), average physical activity between ages 12–17 years (MET-hr/wk, continuous), birthweight (<5.5/5.5–6.9/7–8.4/8.5+), height (inches, continuous), parity/age at first birth (nulliparous/1–2 at < 25 years/1–2 at 25–29 years/1–2 at 30+ years/3+ at <25 years/3+ at 25+ years), ever oral contraceptive use (never/past/current), smoking status (never/past with adolescent smoking/past with no adolescent smoking/current with adolescent smoking/current with no adolescent smoking), adult alcohol intake (g/d, continuous), family history of breast cancer (yes/no), personal history of benign breast disease (yes/no), adult physical activity (MET-hr/wk, continuous), menopausal status/postmenopausal hormone use (premenopausal/postmenopausal never user/postmenopausal past user/postmenopausal current user/dubious), and age at menopause (year, continuous).

^bParticipants recalled their body size at ages 5, 10, and 20 years using a 9-level pictogram (level 1: most lean; level 9: most overweight). Higher levels (4.5+ or 5+) were collapsed because there were fewer women in those levels.

DISCUSSION

In this large prospective study, body leanness during childhood to early adulthood was associated with a significantly higher breast cancer risk, regardless of the level of adolescent physical activity. The association was independent of early-life and adult risk factors for breast cancer and adult BMI. Although the associations were slightly attenuated among women who were active compared to those who were inactive, adolescent physical activity did not significantly modify the association. The results did not vary by ER status of tumors.

Our findings of a positive association between early-life body leanness and breast cancer risk are consistent with those from previous studies ^{1–4, 8, 9, 11} in which lean girls had a 10–70% increased risk compared with overweight girls. The magnitude of the association may vary among studies due to variations in the range of BMI, sample size, length of follow-up, and population characteristics (e.g., ethnicity, age). For example, in our study of primarily Caucasian women, the corresponding median BMI at age 18 (the most proximate age available, r=0.69) for women in the lowest and highest somatotype groups at age 20 were 18 and 26 kg/m², respectively. In a Danish study (3,340 cases), a more narrow contrast of BMI at age 14 (median 16 vs. 22 kg/m²) was associated with a 20% higher risk ⁸. Another study in African American women (1,062 cases) reported a 32–47% higher risk in women with BMI <20 vs. ≥25 kg/m² at age 18 ². Most studies, including ours, did not include women who were extremely lean (such as those in the settings of extreme caloric restriction) during early life; little is known about the relationship in extremely lean girls although some studies observed reduced risk in women who experienced severe famine ³⁵ or anorexia nervosa ³⁶ during adolescence.

While the biology underlying the association between early-life body size and breast cancer risk is not clear, multiple mechanisms, including both estrogen and non-estrogen pathways, are likely to be involved as the association is significant for both ER+ and ER− tumors ⁴. First, overweight girls are more likely to experience irregular menstrual cycles ²³ and anovulation ^{13, 14}, despite having earlier menarche ²³. However, these factors are unlikely to explain the strong association of early-life body size because adjustment for menstrual pattern between ages 14–22 years did not alter our results. Second, although biomarker studies have generally not observed an association between childhood body size and the levels of adolescent ³⁷ and adult ¹⁶ endogenous sex hormones, we cannot exclude the possibility that the breast tissue-specific response to sex hormones (e.g., estrogen sensitivity) may vary by early-life body size. Third, early-life body leanness may contribute to a lifelong set point of IGF-1. In the NHS and NHSII, we observed 14% higher adult levels of plasma IGF-1 among women who were lean (vs. overweight) at young ages ¹⁵, and adult IGF-1 levels have been associated with a higher risk of breast cancer ³⁸. However, genetically-determined IGF-1 levels do not seem to play a major role because no association was observed between childhood body size and SNPs in the IGF-1 and IGFBP-1/-3 gene regions ³⁹. Lastly, early-life body size has been inversely associated with mammographic density ^{17, 18} and the risk of benign breast disease ^{40, 41}. Given the strong association between mammographic density and breast cancer, breast density may play a role in the association between early-life body size and breast cancer risk; however, adjustment for mammographic density did not substantially alter the association in our prior results ⁴². Evidence from mathematical models ⁴³ and epidemiological studies ⁴⁴ supports that breast tissues are particularly susceptible to carcinogens during early life, especially between menarche and the first childbirth ⁴⁵. For instance, overweight girls may have earlier differentiation and altered development of breast tissue structures ⁴⁶ that decreases susceptibility to breast cancer throughout their lives; differentiated cells become less likely to proliferate or transform into tumor cells ^{43, 45}.

A lack of significant interaction in our study suggests that the modest association of adolescent physical activity with breast cancer risk may not be strong enough to counteract the robust association of early-life body size. While early-life physical activity and body size share several potential mechanisms, it is possible that early-life body size acts upon breast carcinogenesis in a way that is not modifiable by physical activity. For example, early-life body size is inversely associated with mammographic density ^{17, 18} while physical activity is not ⁴⁷. If early-life body leanness increases risk primarily through higher breast density, engaging in high levels of activity during early life may not modify the association between early-life body size and breast cancer risk. However, we observed a slightly attenuated association of early-life body size among women who were physically active during adolescence. Our results contrast with those from the only prior study to date, which has suggested a significantly stronger association of BMI at age 18 among postmenopausal women who engaged in higher levels of activity at ages 14–22 year ²⁷. Possible recall bias in their retrospective study may explain the contrast, as suggested by their stronger inverse association of adolescent physical activity (≥364 vs. 0 times/year: OR=0.55, 95% CI=0.39–0.78) and weaker positive association of early-life body leanness (BMI <18.6 vs. >21.8 kg/m²: OR=1.09, 95% CI=0.97–1.22) compared to ours.

Given some limitations of our study, we cannot exclude the possibility of some interaction between early-life body size and physical activity. Because participants recalled their early-life body size and adolescent physical activity when they were ages 25–50 years, our exposure data are inevitably imperfect. Physical activity is particularly difficult to measure accurately using questionnaires as there are many components including intensity, frequency, and timing that may be differentially related to breast cancer risk. Our physical activity data also did not include non-recreational activity such as household and occupational activities. Given we focused on adolescence, levels of these activities were likely minimal in our study population; however, failure to meet this assumption would result in measurement error in physical activity data. In our analysis where participants recalled body size and physical activity prior to breast cancer diagnosis, the resulting measurement error is most likely independent of outcome (i.e., non-differential) and may have attenuated the association and interaction. For early-life body size, the association was similar in other studies that used measured BMI ^{8, 9} supporting the validity of our recalled body size data. Despite these limitations, this study included a long follow-up and adjustment for both early-life and adult risk factors for breast cancer.

In summary, while early-life body leanness was significantly associated with a higher risk of breast cancer, adolescent physical activity did not significantly modify the association, although some interaction cannot be excluded. Further study is required to elucidate the mechanisms underlying this association and to identify strategies to mitigate the excess risk of breast cancer in lean girls. Despite the higher risk of breast cancer observed among lean girls, lifetime obesity and physical inactivity increase the risk of other types of cancer and various other chronic diseases (e.g., cardiovascular, endocrine, pulmonary, renal, gastrointestinal, musculoskeletal) ⁴⁸. Further, adult body fatness increases postmenopausal breast cancer risk, and girls who are overweight are likely to remain overweight as adults ⁴⁹. Therefore, public health recommendations should still support early life maintenance of a healthy weight and physical activity.

Novelty and impact of the work.

Early-life body size is inversely associated with both premenopausal and postmenopausal breast cancer risk, suggesting an excess risk in lean girls. In our prospective analysis with a large sample size, we observed lean girls were at higher risk, regardless of the level of adolescent physical activity. However, the association was slightly, though not significantly, attenuated among the most active girls.