Case-control study of anthropometric measures and breast cancer risk

Authors


Abstract

A population-based case-control study of 1,233 incident breast cancer cases and 1,241 controls was conducted in Alberta between 1995 and 1997 to examine the influence of anthropometric factors on the risk of breast cancer using several newly derived variables. Data on current height, weight and waist and hip circumference were collected by interviewers using standardized methods. Respondents recalled their body weight at each decade from age 20 to the referent year. Several variables were estimated, and unconditional logistic regression was used to estimate odds ratios (ORs) and 95% confidence intervals (95% CIs). No statistically significant associations for any of the estimated variables with breast cancer risk for premenopausal women (462 cases, 475 controls) were found. The results for postmenopausal women (771 cases, 762 controls) in the highest vs. lowest quartiles were, for waist circumference, OR = 1.30 (95% CI 0.97–1.73); waist–hip ratio, OR = 1.43 (95% CI 1.07–1.93); weight gain since age 20, OR = 1.35 (05% CI 1.01–1.81); difference between maximum and minimum weights over adult lifetime, OR = 1.56 (95% CI 1.16–2.08); and the reference weight minus the minimum weight since age 20, OR = 1.47 (95% CI 1.10–1.97). Statistically significant trends in risk were observed for these variables. Effect modification with hormone replacement therapy use was found for most variables assessed for postmenopausal women, with much stronger associations found among never-users compared to ever-users. We found strong evidence that waist–hip ratio and weight gained over lifetime, as assessed by different variables, are postmenopausal breast cancer risk factors. These effects were independent of dietary intake and lifetime total physical activity. © 2002 Wiley-Liss, Inc.

Evidence for an association between anthropometric factors and breast cancer risk is accumulating rapidly; however, the results from previous epidemiologic studies have been inconsistent and often limited by the self-reporting methods that have been used.1 Modifiable lifestyle risk factors, including dietary intake and physical activity, are integral for breast cancer. Obesity and central adiposity, as measured by weight, weight gain over lifetime, body mass, and waist and hip circumference are potentially important breast cancer risk factors but may also be part of a causal pathway between physical activity, dietary intake and breast cancer risk. The present study was conducted to explore the interrelations between these risk factors and to delineate the roles of each in breast cancer etiology. The association between physical activity and breast cancer from this data set has been examined previously.2, 3, 4 This report presents the results for anthropometric risk factors. Some of the previously identified gaps in knowledge about the association between anthropometric risk factors and breast cancer1 are specifically addressed here, including the independent effect of weight at different time points in life and the interaction of anthropometric factors with other breast cancer risk factors.

MATERIAL AND METHODS

Our study methods have been described previously.2 In brief, we conducted a population-based case-control study in Alberta among 1,239 (78.3%) of 1,582 eligible, incident, histologically confirmed in situ and invasive primary cases of breast cancer identified between 1995 and 1997 directly from the Alberta Cancer Registry, a population-based cancer registry. Cases were eligible for the study if they were residents of Alberta, were 80 years of age or younger, were able to speak English and complete an in-person interview and had no other previous cancer diagnosis. Female controls were identified through random-digit dialing using the Waksberg method and sampled concurrently with cases.5 Controls were frequency-matched to cases by age (±5 years) and place of residence (urban/rural). In addition to the eligibility criteria already listed for cases, controls had to be free of any cancer diagnosis excluding nonmelanoma skin cancer.

Case physicians were sent letters informing them of the study and seeking permission to contact the patients. If the physicians did not decline permission for patient contact within 2 weeks of the letters being sent to them, cases were sent a study package. The package included a cover letter describing the study, a consent form and 2 recall calendars developed specifically for the study. Interviewers contacted cases by telephone and set up an in-person interview, which occurred generally in the respondents' homes. Cases and controls provided informed consent at the time of the interview. Interviews were completed, on average, 14.8 weeks after breast cancer diagnosis for cases and 9.1 weeks after being identified through random-digit dialing for controls.

Interviews were completed with 1,241 women (82% of the eligible and available women who had agreed to receive the study package). The overall response rate for controls was 56.4% (1,241 interviews completed out of 2,197 eligible and available controls). The final data set for analysis consisted of 1,233 cases and 1,237 controls after removing study subjects who had too many missing or questionable data for several of the main risk factors under consideration.

Data Collection

Interviewers used standardized methods and calibrated weigh scales to measure current height, weight and waist and hip circumferences. Interviewers received extensive training in the measurement of these anthropometric indices before the study began, and continual monitoring of their interviews was used to ensure standardization of the methods. Weight at each decade from age 20 to the most recent decade before the reference age was recalled by the respondents. Interviewers had been trained in cognitive interviewing methods6, 7 for this project, and recall calendars were sent to respondents before their interviews to assist in the recall of past information.

Other breast cancer risk factors assessed at the interview included menstrual and reproductive history, hormone use history, mammography history, personal history of breast disease, breast biopsies, personal history of cancer, family history of cancer, lifetime physical activity patterns, dietary intake during the reference year, alcohol consumption, smoking habits, demographic characteristics and current and past anthropometric measurements. During the interview, all data were collected up to the time of diagnosis in cases and a comparable time for controls. This date was referred to as the reference date. For anthropometric measurements taken at the time of the interview, the timing for these measures was referred to as the interview date. Diet during the reference year (i.e., year preceding diagnosis in cases and comparable year period for controls) was assessed with the NCI Block food-frequency questionnaire.8

Women were classified as “postmenopausal” if they stated that they were postmenopausal (n = 644) or that they had stopped menstruating for over a year by the reference date and their age was 50 years or more (n = 201). Women who had had a hysterectomy were considered postmenopausal if their hysterectomy occurred after menopause (n = 40), if they had a bilateral oophorectomy before menopause (n = 242), if they had symptoms of menopause that occurred after their hysterectomy (n = 278) or if their current age was 55 years or older (n = 351). A total of 1,539 women were postmenopausal (771 cases, 762 controls) and 939 were premenopausal (462 cases, 475 controls).

Variables that were estimated for this study included body mass index [weight (kg)/height (m2)], waist–hip ratio [waist circumference (cm)/hip circumference (cm)], weight gained since age 20 (current weight at interview minus recalled weight at age 20), the sum of weight change between decades (sum of absolute weight changes between each decade from weight at age 20 until the reference date, i.e., date at diagnosis, with recalled weight up to age 60 reported), the difference between the maximum and minimum adult lifetime weights (using recalled weights at each decade and the reference weight) and reference weight minus minimum weight since age 20 (using recalled weights).

Statistical Analysis

Since the association between anthropometric factors and breast cancer risk is modified by menopausal status, all analyses were stratified by menopausal status at the time of diagnosis. Descriptive analyses were conducted to examine any case-control differences for the estimated anthropometric variables. These risk factors were then categorized into quartiles according to the distribution of the variables among the controls within each menopausal stratum, and unconditional logistic regression modeling was used to estimate the odds ratios (ORs) associated with breast cancer for these factors. Age-adjusted and multivariable-adjusted models were used with a full examination of confounding by other established and putative risk factors. The variables considered as confounders included age, marital status, educational level, ethnicity, first-degree family history of breast cancer, whether menstruation had ever ceased for reasons besides pregnancy, irregular menstrual cycles, ever oral contraceptive use, duration of oral contraceptive use, parity and gravidity, ever breast-feeding, ever hormone replacement therapy use, duration of hormone replacement therapy use, type of hormone replacement therapy, history of benign breast disease, previous benign breast biopsy, ever alcohol consumption, ever cigarette smoking, current cigarette smoking, total pack years of smoking, total caloric intake, daily dietary fat intake and average total lifetime physical activity. Final models were adjusted for age, education level achieved (in quintiles), ever use of hormone replacement therapy, history of benign breast disease, first-degree family history of breast cancer, current cigarette smoking, ever alcohol consumption, total caloric intake and average total lifetime physical activity. All other variables were eliminated as they did not influence the overall fit of the logistic regression model. The possibility of effect modification was also considered in these data by further stratifying the results by factors that were considered a priori to be possibly important modifiers of the association between anthropometric factors and breast cancer. These included first-degree family history of breast cancer, use of hormone replacement therapy, parity, alcohol use and smoking.

RESULTS

Descriptive analyses were stratified by case-control status and within menopausal strata. Only 1 statistically significant difference was observed between premenopausal cases and controls for the anthropometric variables examined (Table I). Among premenopausal women, controls had slightly greater hip circumferences than cases. For postmenopausal women, there was evidence that cases had more central adiposity than controls since they had higher waist circumferences and higher waist–hip ratios. Postmenopausal cases, compared to controls, also experienced more weight gain since age 20, more weight gained throughout their lifetime across decades and a greater difference between lifetime minimum weight and weight at the reference date. Of these variables, differences were statistically significant for waist circumference, waist–hip ratio and most of the derived anthropometric factors including sum of weight change between decades, difference between minimum and maximum weights over adult lifetime and reference weight minus minimum weight.

Table I. Mean (± SD) Values for Anthropometric Factors, by Menopausal Status, Alberta, 1995–1997, n = 2,470
Risk factorPremenopausal women (n = 937)Postmenopausal women (n = 1,533)
Cases (n = 462)Controls (n = 475)pCases (n = 771)Controls (n = 762)p
Current height (m)1.64 (±0.06)1.63 (±0.06)0.411.61 (±0.06)1.61 (±0.06)0.6
Current weight (kg)70.15 (±14.46)71.21 (±14.67)0.2773.12 (±15.82)72.56 (±14.70)0.48
Current waist (cm)79.88 (±12.37)80.45 (±12.45)0.4886.24 (±13.71)84.82 (±12.75)0.04
Current hip (cm)103.01 (±10.66)104.60 (±11.39)0.03106.74 (±12.36)106.90 (±11.81)0.8
Body mass index (kg/m2)26.14 (±5.09)26.67 (±5.35)0.1228.22 (±5.95)28.05 (±5.37)0.57
Waist–hip ratio0.77 (±0.06)0.77 (±0.06)0.160.81 (±0.07)0.79 (±0.06)
Weight at age 20 (kg)55.27 (±8.00)55.37 (±8.03)0.8454.73 (±8.05)55.30 (±8.12)0.17
Weight gain since age 20 (kg)14.88 (±12.55)15.84 (±12.86)0.2518.41 (±14.55)17.20 (13.40)0.1
Sum of weight change between decades (kg)16.29 (±13.11)16.40 (±12.00)0.921.90 (±14.59)19.86 (±12.65)0.004
Difference between maximum and minimum weights over adult lifetime (kg)14.62 (±10.80)14.76 (±10.93)0.8419.15 (±11.98)17.23 (±10.95)0.001
Reference weight minus minimum weight since age 20 (kg)13.46 (±11.02)13.60 (±11.22)0.8417.06 (±12.50)15.47 (±11.62)0.01

Multivariate analyses of the anthropometric characteristics measured at interview revealed no statistically significant associations with current height, weight, waist and hip circumferences, body mass index or waist–hip ratios for premenopausal women (Table II). Increased breast cancer risk was clearly observed for postmenopausal women in the highest vs. the lowest quartile of waist–hip ratio (≥0.81 vs. <0.72) (Table II). For these women, the risk was 1.43 [95% confidence interval (CI) 1.07–1.93], and a statistically significant trend of increasing risk with increasing waist–hip ratio was observed across the quartiles (p = 0.0006). An elevated, borderline significant, increased risk was also seen for current waist circumference for women in the highest vs. lowest quartile (≥86.8 vs. <71.5) (OR = 1.30, 95% CI 0.97–1.73).

Table II. OR (95% CI) for Anthropometric Factors Measured at Interview, by Menopausal Status, Alberta, n = 2,470
Quartile cut pointsPremenopausal women (n = 937)1Quartile cut pointsPostmenopausal women (n = 1,533)
CasesControlsAge-adjustedMultivariable-adjusted2CasesControlsAge-adjustedMultivariable-adjusted2
  • 1

    Sample sizes vary slightly because of missing data for some variables.

  • 2

    Adjusted for current age, total caloric intake, total lifetime physical activity, educational level (in quintiles), ever use of hormone replacement therapy, ever diagnosed with benign breast disease, first-degree family history of breast cancer, ever alcohol consumption, current smoker.

  • 3

    The continuous form of these anthropometric measurements was used to test for trend.

Height (m)
 <1.5910611711<1.5720518811
 ≥1.59–<1.631071181.00 (0.69–1.45)1.11 (0.75–1.64)≥1.57–<1.611921880.94 (0.71–1.25)0.96 (0.72–1.28)
 ≥1.63–<1.671181211.08 (0.75–1.55)1.21 (0.82–1.77)≥1.61–≤1.651441940.69 (0.51–0.93)0.71 (0.53–0.97)
 ≥1.671311191.21 (0.84–1.74)1.28 (0.87–1.87)≥1.652301921.11 (0.84–1.48)1.15 (0.86–1.54)
 ptrend30.440.290.510.32
Weight (kg)
 <61.212911611<61.819218911
 ≥61.2–<68.31131210.84 (0.59–1.21)0.86 (0.59–1.26)≥61.8–<70.31721920.89 (0.66–1.18)0.92 (0.68–1.24)
 ≥68.3–<78.41141190.87 (0.60–1.24)0.97 (0.66–1.42)≥70.3–<81.11961891.03 (0.77–1.36)1.09 (0.81–1.46)
 ≥78.41061190.81 (0.56–1.16)0.81 (0.55–1.19)≥81.12111921.09 (0.82–1.45)1.11 (0.83–1.49)
 ptrend30.290.40.430.35
Body mass index (kg/m2)
 <23.114511811<24.120619011
 ≥23.1–<25.71021190.70 (0.49–1.00)0.75 (0.52–1.10)≥24.1–<27.31791910.87 (0.65–1.15)0.93 (0.69–1.24)
 ≥25.7–<29.21131190.77 (0.54–1.11)0.81 (0.55–1.17)≥27.3–<31.31871900.91 (0.68–1.20)0.94 (0.70–1.26)
 ≥29.21021190.70 (0.49–1.01)0.69 (0.47–1.02)≥31.31991910.96 (0.73–1.27)0.99 (0.74–1.32)
 ptrend30.130.170.550.55
Waist circumference (cm)
 <71.513311711<75.617518811
 ≥71.5–<78.51081170.82 (0.57–1.17)0.84 (0.58–1.22)≥75.6–<82.81591910.87 (0.65–1.17)0.89 (0.66–1.20)
 ≥78.5–<86.81051200.78 (0.54–1.12)0.79 (0.54–1.16)≥82.8–<91.51871861.05 (0.79–1.40)1.06 (0.79–1.42)
 ≥86.81131180.85 (0.59–1.22)0.89 (0.61–1.31)≥91.52421931.31 (0.99–1.73)1.30 (0.97–1.73)
 ptrend30.520.640.040.07
Hip circumference (cm)
 <96.813511711<98.519618811
 ≥96.8–<103.01071190.79 (0.55–1.13)0.82 (0.57–1.20)≥98.5–<104.71741880.88 (0.66–1.17)0.88 (0.65–1.18)
 ≥103.0–<110.01181180.87 (0.61–1.25)0.95 (0.65–1.38)≥104.7–<113.31941890.97 (0.74–1.29)0.99 (0.74–1.33)
 ≥110.0991180.73 (0.51–1.06)0.76 (0.52–1.11)≥113.31991920.98 (0.74–1.30)1.00 (0.74–1.33)
 ptrend30.030.050.820.88
Waist–hip ratio
 <0.7211611811<0.7515518911
 ≥0.72–<0.761041180.90 (0.62–1.30)0.92 (0.63–1.35)≥0.75–<0.791531890.96 (0.71–1.30)0.86 (0.63–1.17)
 ≥0.76–<0.811031180.90 (0.62–1.29)0.92 (0.63–1.35)≥0.79–<0.832001891.26 (0.94–1.68)1.21 (0.90–1.63)
 ≥0.811361181.19 (0.83–1.69)1.22 (0.84–1.79)≥0.832551901.59 (1.20–2.11)1.43 (1.07–1.93)
ptrend30.140.120.00010.0006

While none of the derived variables for the anthropometric factors that assessed weight gain over lifetime was associated with an increased risk of breast cancer among premenopausal women, all of these variables were clearly associated with an increased risk for postmenopausal women (Table III). For postmenopausal women who were in the highest quartile of weight gained since age 20 vs. the lowest quartile (≥25 vs. <7.8 kg), the risk was 1.35 (95% CI 1.01–1.81). A similar magnitude of increased risk was found when the sum of weight change between decades over the respondents' lifetimes was examined. For women who had gained ≥25.9 kg since age 20 compared to women who gained ≤10.9 kg, the risk was 1.32 (95% CI 0.99–1.76). When the difference between the maximum and minimum weights over adult lifetime was modeled, a large increased risk was observed. Women who had a 22.7 kg difference between maximum and minimum adult weights experienced a risk of 1.56 (95% CI 1.16–2.08) compared to women with a <9.1 kg weight difference. Finally, when comparing the weight during the reference year and the minimum weight since age 20, women who increased their weight by ≥22.7 kg during this time period compared to women with <6.8 kg weight difference, the risk was 1.47 (95% CI 1.10–1.97). A statistically significant linear trend of increasing risk across the quartiles for each of these derived anthropometric variables was found.

Table III. OR (95% CI) for Derived Anthropometric Factors, by Menopausal Status, Alberta, n = 2,470
Quartile cut pointsPremenopausal women (n = 937)1Quartile cut pointsPostmenopausal women (n = 1,533)
CasesControlsAge-adjustedMultivariable-adjusted2CasesControlsAge-adjustedMultivariable-adjusted2
  • 1

    Sample sizes vary slightly because of missing data for some variables.

  • 2

    Adjusted for current age, total caloric intake, total lifetime physical activity, educational level (in quintiles), ever use of hormone replacement therapy, ever diagnosed with benign breast disease, first-degree family history of breast cancer, ever alcohol consumption, current smoker.

  • 3

    The continuous form of these anthropometric measurements was used to test for trend.

Weight at age 20 (kg)
 <49.9869411<49.917414711
 ≥49.9–<54.41121141.08 (0.73–1.59)1.11 (0.73–1.67)≥49.9–<54.41972000.84 (0.62–1.12)0.82 (0.61–1.11)
 ≥54.4–<59.01421231.26 (0.87–1.85)1.35 (0.91–2.02)≥54.4–<59.01941870.88 (0.66–1.19)0.86 (0.64–1.17)
 ≥59.01221440.92 (0.63–1.35)1.02 (0.68–1.52)≥59.02022230.77 (0.58–1.03)0.76 (0.57–1.03)
 ptrend30.820.760.170.10
Weight gain since age 20 (kg)
 <7.7214311811<7.8018118911
 ≥7.72–<13.8921190.64 (0.44–0.92)0.69 (0.47–1.01)≥7.80–<15.71731890.97 (0.73–1.29)1.02 (0.75–1.37)
 ≥13.8–<22.01131190.79 (0.55–1.13)0.84 (0.58–1.21)≥15.7–<25.01821891.02 (0.77–1.36)1.08 (0.80–1.45)
 ≥22.01141190.80 (0.56–1.14)0.79 (0.54–1.15)≥25.02311901.29 (0.98–1.71)1.35 (1.01–1.81)
 ptrend30.280.240.080.05
Sum of weight change between decades (kg)
 <8.6212711811<10.917818911
 ≥8.62–<13.61011150.82 (0.57–1.19)0.85 (0.58–1.24)≥10.9–<18.11491870.84 (0.63–1.13)0.83 (0.61–1.12)
 ≥13.6–<21.31151210.90 (0.62–1.29)0.95 (0.65–1.39)≥18.1–<25.92001911.11 (0.83–1.47)1.10 (0.82–1.47)
 ≥21.31171190.93 (0.64–1.34)0.89 (0.60–1.31)≥25.92361891.32 (1.00–1.74)1.32 (0.99–1.76)
ptrend30.960.740.0040.004
Difference between maximum and minimum weights over adult lifetime (kg)
 <6.8112011311<9.0716118511
 ≥6.81–<12.31111220.86 (0.60–1.25)0.90 (0.62–1.32)≥9.07–<15.41611950.96 (0.71–1.29)0.94 (0.69–1.28)
 ≥12.3–<20.01151200.91 (0.63–1.33)0.95 (0.65–1.40)≥15.4–<22.71841821.18 (0.88–1.58)1.21 (0.89–1.64)
 ≥20.01161200.92 (0.64–1.34)0.92 (0.62–1.37)≥22.72651991.55 (1.17–2.06)1.56 (1.16–2.08)
 ptrend30.910.630.00090.0007
Reference weight minus minimum weight since age 20 (kg)
 <6.3512311711<6.8017218611
 ≥6.35–<11.31141110.98 (0.68–1.41)1.01 (0.69–1.48)≥6.80–<13.61411750.89 (0.66–1.21)0.93 (0.68–1.27)
 ≥11.3–<18.21041260.79 (0.55–1.15)0.90 (0.62–1.33)≥13.6–<22.72102041.14 (0.86–1.52)1.20 (0.90–1.61)
 ≥18.21191190.96 (0.67–1.39)0.94 (0.63–1.38)≥22.72461921.43 (1.08–1.89)1.47 (1.10–1.97)
 ptrend30.910.710.0080.004

Effect modification by other factors was considered in these data. The only factor that was consistently found to be an effect modifier was hormone replacement therapy use; the data for the variables for which an interaction was observed among postmenopausal women are presented in Table IV. A much stronger association was found among never-users of hormone replacement therapy compared to ever-users for current weight, waist and hip circumference, waist–hip ratio, sum of weight change between decades, difference between maximum and minimum weights over adult lifetime and reference weight minus minimum weight since age 20. Indeed, risk estimates for ever-users were close to or less than 1.0 for all of these factors, while for never-users the risks ranged from 1.5 to 2.6. Furthermore, tests for linear trends were statistically significant for all of these variables with the exception of hip circumference, which was borderline significant.

Table IV. OR (95% CI) for Postmenopausal Women by Use of Hormone Replacement Therapy, alberta, n = 1,533
Quartile cut pointsNever-users of hormone replacement therapy (n = 672)1Quartile cut pointsEver-users of hormone replacement therapy (n = 861)
CasesControlsAge-adjustedMultivariable-adjusted2CasesControlsAge-adjustedMultivariable-adjusted2
  • 1

    Sample sizes will vary slightly across variables because of missing data for some variables.

  • 2

    Adjusted for current age, total caloric intake, total lifetime physical activity, educational level (in quintiles), ever use of hormone replacement therapy, ever diagnosed with benign breast disease, first-degree family history of breast cancer, ever alcohol consumption, current smoker.

  • 3

    The continuous form of these anthropometric measurements was used to test for trend.

Current weight (kg)
 <61.8837611<61.911011311
 ≥61.8–<69.571780.85 (0.54–1.33)0.81 (0.51–1.30)≥61.9–<70.6941120.86 (0.59–1.26)0.95 (0.64–1.41)
 ≥69.5–<81.698761.18 (0.77–1.82)1.26 (0.81–1.98)≥70.6–<80.71051140.94 (0.65–1.36)0.97 (0.66–1.43)
 ≥81.6112781.34 (0.88–2.06)1.47 (0.94–2.31)≥80.7981150.87 (0.59–1.26)0.88 (0.59–1.30)
 ptrend30.220.050.760.63
Current waist circumference (cm)
 <75.9707611<75.510911211
 ≥75.9–<82.857760.78 (0.49–1.25)0.76 (0.47–1.24)≥75.5–<82.7971120.88 (0.60–1.28)0.91 (0.62–1.35)
 ≥82.8–<95.6132771.79 (1.17–2.74)1.81 (1.16–2.81)≥82.7–<90.6901150.79 (0.54–1.16)0.79 (0.53–1.17)
 ≥95.6101771.37 (0.89–2.12)1.54 (0.97–2.43)≥90.61071130.96 (0.66–.39)0.92 (0.63–1.36)
 ptrend30.0060.0010.780.52
Current hip circumference (cm)
 <98.5807611<98.511611211
 ≥98.5–<104.675750.95 (0.61–1.49)0.94 (0.59–1.50)≥98.5–<105.11051140.87 (0.60–1.26)0.90 (0.61–1.31)
 ≥104.6–<113.496771.17 (0.76–1.80)1.26 (0.81–1.97)≥105.1–<113.3921130.77 (0.53–1.12)0.77 (0.52–1.14)
 ≥113.4109771.34 (0.88–2.05)1.46 (0.94–2.27)≥113.3901130.75 (0.52–1.10)0.76 (0.51–1.13)
 ptrend30.270.090.090.08
Waist–hip ratio
 <0.76657611<0.749611311
 ≥0.76–<0.7963760.94 (0.59–1.50)0.87 (0.54–1.42)≥0.74–<0.78831130.85 (0.57–1.25)0.80 (0.53–1.19)
 ≥0.79–<0.8483761.24 (0.79–1.94)1.31 (0.82–2.09)≥0.78–<0.831071131.10 (0.76–1.61)1.05 (0.71–1.56)
 ≥0.84149772.19 (1.43–3.36)2.30 (1.47–3.60)≥0.831171131.20 (0.83–1.75)1.09 (0.74–1.60)
 ptrend30.001<0.00010.10.29
Weight at age 20 (kg)
 <50.81057511<49.9989711
 ≥50.8–<54.467610.79 (0.50–1.25)0.77 (0.48–1.24)≥49.9–<54.41011140.87 (0.59–1.29)0.88 (0.59–1.30)
 ≥54.4–<59.4126921.00 (0.67–1.49)0.97 (0.64–1.47)≥54.4–<59.0981200.80 (0.55–1.18)0.80 (0.54–1.19)
 ≥59.464760.62 (0.40–0.96)0.62 (0.39–0.99)≥59.01081220.87 (0.60–1.28)0.87 (0.59–1.29)
 ptrend30.160.140.460.38
Sum of weight change between decades (kg)
 <9.98627511<11.311111111
 ≥9.98–<16.364761.02 (0.64–1.62)1.06 (0.65–1.72)≥11.3–<18.2901140.78 (0.53–1.13)0.76 (0.52–1.13)
 ≥16.3–<25.0103761.65 (1.06–2.57)1.75 (1.10–2.78)≥18.2–<26.3961140.82 (0.56–1.20)0.79 (0.54–1.17)
 ≥25.0130762.09 (1.35–3.22)2.39 (1.51–3.78)≥26.31071140.92 (0.63–1.33)0.90 (0.61–1.32)
 ptrend30.0002<0.00010.790.88
Difference between maximum and minimum weights over adult lifetime (kg)
 <8.16597511<9.8011511311
 ≥8.16–<15.065781.08 (0.67–1.74)1.13 (0.69–1.86)≥9.80–<15.9791130.69 (0.47–1.01)0.69 (0.46–1.02)
 ≥15.0–<22.7101731.83 (1.16–2.89)1.94 (1.20–3.13)≥15.9–<22.7871100.77 (0.53–1.13)0.81 (0.55–1.20)
 ≥22.7139812.28 (1.47–3.53)2.59 (1.63–4.12)≥22.71261181.04 (0.72–1.50)1.02 (0.70–1.49)
 ptrend3<0.0001<0.00010.640.66
Reference weight minus minimum weight since age 20 (kg)
 <6.35687511<6.8110011111
 ≥6.35–<13.660730.96 (0.60–1.54)1.07 (0.66–1.76)≥6.81–<13.6891150.85 (0.58–1.25)0.85 (0.57–1.27)
 ≥13.6–<22.7101791.50 (0.97–2.34)1.63 (1.03–2.60)≥13.6–<22.71051121.03 (0.70–1.51)1.10 (0.75–1.63)
 ≥22.7134772.06 (1.34–3.17)2.39 (1.51–3.77)≥22.71121151.07 (0.73–1.56)1.08 (0.73–1.59)
 ptrend30.006<0.00010.960.88
Body mass index (kg/m2)
 <24.4937711<23.911411311
 ≥24.4–<27.371770.76 (0.49–1.19)0.79 (0.50–1.25)≥23.9–<27.31081140.94 (0.65–1.36)1.06 (0.72–1.56)
 ≥27.3–<31.8102771.09 (0.71–1.66)1.12 (0.72–1.75)≥27.3–<31.0891130.78 (0.53–1.14)0.83 (0.56–1.24)
 ≥31.898771.06 (0.69–1.62)1.18 (0.76–1.85)≥31.0961140.83 (0.57–1.21)0.85 (0.57–1.26)
 ptrend30.320.100.700.52

No effect modification by smoking and alcohol use was found, and only a couple of anthropometric factors were modified by family history of breast cancer and parity. Premenopausal women with no family history who were in the highest quartile of current weight had a reduced risk of breast cancer (0.69, 95% CI 0.45–1.04), while the heaviest women with a family history had a 3-fold increased risk (3.10, 95% CI 0.88–10.9). Postmenopausal women in the highest quartile of current height without a family history had no increased breast cancer risk (1.00, 95% CI 0.73–1.38), but the tallest women with a family history had a statistically significant doubling in risk (2.42, 95% CI 1.11–5.29). For nulliparous premenopausal women, the risks for current height and body mass index were also notably different from those observed for parous women. Nulliparous premenopausal women in the highest quartile of height had a decreased breast cancer risk (0.33, 95% CI 0.10–1.14), while parous, tall women had an increased risk (1.47, 95% CI 0.97–2.22). Opposite associations were found for body mass index: nulliparous women in the highest quartile experienced much higher risk (3.50, 95% CI 1.00–12.3), while parous women with high body mass index had a greatly reduced risk (0.65, 95% CI 0.43–0.99).

To examine the impact of weight in early adulthood on the risks of breast cancer associated with these anthropometric factors, we reanalyzed all of the associations by controlling for weight at age 20. Risks for current weight and body mass index among premenopausal women were lower when adjustment for weight at age 20 was made. For women in the highest quartile of body mass index, the risk was 0.65 (95% CI 0.43–0.98); and for women in the highest quartile of current weight, the risk was 0.74 (95% CI 0.48–1.15). The opposite effect of adjustment for weight at age 20 was observed for postmenopausal women. Postmenopausal women in the highest quartile of current weight had a risk of 1.25 (95% CI 0.91–1.73), and the trend across the quartiles approached statistical significance (p = 0.07). For women in the highest quartile of body mass index, the risk was 1.05 (95% CI 0.78–1.43), for waist circumference the risk was 1.42 (95% CI 1.05–1.91) and a statistically significant trend across quartiles (p = 0.01) was found. Similarly, the risks for current waist circumference and for waist–hip ratio also increased, with ORs, respectively, of 1.42 (95% CI 1.05–1.91, ptrend = 0.01) and 1.50 (95% CI 1.11–2.01, ptrend = 0.0002). As expected, controlling for weight at age 20 attenuated somewhat the risks associated with weight gain since age 20. Postmenopausal women in the highest quartile of weight gain had a risk of 1.29 (95% CI 0.96–1.74), and the trend across quartiles was no longer statistically significant (p = 0.08).

We also examined the possibility that age at diagnosis could influence risk of breast cancer associated with body mass index. When we stratified the results by age at diagnosis (<50, 50–65 and >65 years), among postmenopausal women in the highest quartile of body mass index the risks increased somewhat, though the precision around these risks was substantially decreased. For women under 50 at diagnosis, the risk was 0.54 (95% CI 0.14–2.08), for women 50–65 years at diagnosis the risk was 1.02 (95% CI 0.66–1.56) and for women over 65 years at diagnosis the risk was 1.10 (95% CI 0.70–1.72).

We also examined the independent effects of weight at different time points in life vs. weight gain over lifetime, thereby addressing another identified gap in knowledge in this field.1 The risk of breast cancer increased across the quartiles of body weight with each successive decade between ages 20 and 60 among premenopausal women, but these risks never achieved statistical significance. For postmenopausal women, the risks among women in the highest quartile of weight compared to the lowest were decreased at age 20 (0.75, 95% CI 0.55–1.01). With each subsequent decade, women in the highest quartile of weight had increasing risks of breast cancer, with risks of 0.85 (95% CI 0.64–1.12), 0.91 (95% CI 0.68–1.21), 1.10 (95% CI 0.84–1.45) and 1.22 (95% CI 0.84–1.79) for ages 30, 40, 50 and 60, respectively. Hence, there was a tendency for the risks to increase with increasing age. From these decade-specific risk estimates, it appears that weight gained after age 50 confers an increased risk, with even greater risk at higher weights at age 60. These results are in agreement with the findings for weight gained over lifetime.

Finally, we examined if weight loss was related to breast cancer risk by categorizing the data for the variables weight gained since age 20 and reference weight minus minimum weight since age 20 into quintiles to include a category for weight loss as the bottom quintile and then all other categories of weight gain in the next 4 quintiles. The second quintile was set as the referent category since the sample size for the weight loss group was too small to allow for stable and precise estimations of risk. For the variable weight gain since age 20, there were 68 premenopausal women and 60 postmenopausal women who had values <0. For the variable reference weight minus minimum weight since age 20, there were 19 premenopausal women and 60 postmenopausal women who had values <0. No associations were found between weight loss and breast cancer risk for either of these variables (data not shown). Point estimates were around the null value; however, CIs were sometimes quite wide given the size of these samples. The associations for the other quintiles were essentially the same as had already been found when the data were categorized into quartiles as previously presented.

DISCUSSION

In our study, postmenopausal breast cancer risk was increased with greater waist–hip ratio and higher levels of weight gained throughout lifetime as assessed by several different derived variables of weight change and weight gain. Although no statistically significant associations were found for any of the anthropometric risk factors and premenopausal breast cancer, some indication of decreased risk for women with the highest body mass index, weight and waist and hip circumferences was observed. Effect modification by hormone replacement therapy use was also clearly evident among postmenopausal women since a >2-fold increased risk for most of these anthropometric factors existed for never-users of hormone therapy. Before discussing these results, the limitations of this investigation will be addressed.

Although this was a large population-based case-control study that included population controls, a possible selection bias was introduced because of the lower response rate among controls. To address this issue, controls were compared to a sample of female Albertans aged 20 or above (n = 6,378) who were included in the National Population Health Survey (NPHS) in the 1996–1997 survey, conducted by Statistics Canada.9 The control population was similar to that sampled in the NPHS;2 hence, it was considered to be representative of the base population of women in Alberta from which cases were sampled. If the controls had been a healthier cohort of women than the base population at risk for breast cancer, the risk estimates would have been biased away from the null and more inflated values would have been found. Since our population of controls was actually heavier than the base population, it is likely that the effects we observed are underestimated.

Three types of error or bias in the exposure measurement were possible in our study. First was a random error attributable to memory difficulties when respondents were asked to report their weight per decade since age 20. The effect of such misclassification would have been to decrease the ability of the study to demonstrate an effect of weight on breast cancer risk (nondifferential misclassification bias). The second and third types of error are recall bias and social desirability bias. Recall bias was unlikely in our study since the same methods were used for cases and controls and cases were not given a particular reason to ruminate on their weight gain since age 20. Social desirability bias can occur in studies investigating associations of disease with height and weight that use self-reported measures since overweight subjects tend to underestimate and the short and underweight tend to overestimate.10 This type of bias was reduced to a minimum during the interview by employing strict quality control of interviewing methods, including neutral and nonjudgmental interviewing.

Since multiple subgroups were used in this study, it is possible that some of the findings are related to the multiple comparisons and may be attributable to chance. Such a chance finding would be most likely for comparisons in which there was no a priori hypothesis to be tested in making the subgroup comparison; however, there were few such subgroups in our study since the analyses were based on results from previous research studies.

The strengths of our investigation are also worth mentioning. We included a large population-based sample, used standardized objective rather than solely self-report measures and undertook a complete assessment of confounding and effect modification. Our results are corroborated by most previous investigations of anthropometric factors and breast cancer risk.1 We had more data available on weight changes across lifetime, thereby permitting analysis of numerous derived variables of weight change beyond what has been previously reported.

We found that increasing height was associated with a slight increased risk of breast cancer for both pre- and postmenopausal women, as have previous investigations.1 Attained height may be influenced by childhood and adult nutrition,11 genetic predisposition,12 prenatal exposures13 and IGF levels.14

Previous research has shown that increased weight, assessed by either weight or body mass index, decreases breast cancer risk before menopause but increases risk after menopause.1 We did not find a strong association with weight or body mass index either before or after menopause.

Our data suggest that the risk associated with a higher body mass index increases with a later age at diagnosis. Comparable results have been found in previous studies,15, 16, 17, 18 suggesting that an orderly increase in breast cancer risk is found for higher body mass indices with each decade of increasing age at diagnosis from ages 40 to 70.

From the additional analyses conducted by controlling for weight at age 20, it was found that weight in early adulthood might be an additional important determinant of risk of breast cancer in later years. The magnitude of the associations became stronger with adjustment for weight in early adulthood. A clear biologic mechanism for how obesity in early life may influence breast cancer risk throughout life does not yet exist. Hence, these analyses need to be replicated in other studies before definitive conclusions about the role of early weight on breast cancer risk can be made.

Clear associations with abdominal adiposity were found. Central adiposity, assessed by waist circumference or waist–hip ratio, has been positively associated with postmenopausal breast cancer but has shown no impact on premenopausal breast cancer in other studies.1 The reversal in risk associated with adiposity that occurs at menopause has been attributed to the change in the source of endogenous estrogens that occurs at menopause.1

One of the gaps in knowledge that we addressed is the independent effect of weight gain separate from dietary intake and physical activity. We included a complete measurement of lifetime total physical activity and dietary intake during the reference year. Both of these independent risk factors for breast cancer are components of the proposed biologic model explaining the association between anthropometric risk factors and breast cancer. Analyses were performed that did and did not control for these risk factors. Only slight differences (<2%) in point estimates and CIs were found when these factors were included in the models. We chose to keep the fully adjusted models in the tables since there is a biologic rationale for adjusting for these factors even though they are not major confounders of the association. In adjusting for these factors in the multivariate analyses, the observed associations for breast cancer with weight gain, waist–hip ratio and height were clearly not attributable to diet and physical activity.

Clear evidence for effect modification by hormone replacement therapy use was found in these data for most of the anthropometric factors examined. A nearly 50% increased risk was observed for postmenopausal women in the highest quartiles of current weight and waist and hip circumference who were never hormone replacement therapy users. Even higher, statistically significant risks, ranging from 2.3 to 2.6, were observed among women in the highest quartile for waist–hip ratio and the derived variables for weight gain over lifetime for the never-users of hormone replacement. No elevation in risk was observed for any of these anthropometric risk factors among the ever-users of hormone replacement. These results corroborate the findings of Huang et al.19 from the Nurses Health Study I cohort, in which the dual effects of hormone replacement therapy use and weight gain through life were first examined. Other investigators16, 20–22 have also found an attenuation of excess risk among menopausal hormone users. These results may be attributable to the higher estrogen levels obtained from hormone therapy than those derived from adipose tissue.23, 24

From the preliminary analysis of these data for weight loss, it does not appear that weight loss over lifetime is associated with an increased or decreased risk of breast cancer. However, since our study did not have adequate power to examine this question, future studies designed specifically to include larger cohorts of women who experience weight loss are needed.

Several biologic mechanisms have been postulated that implicate endogenous sex hormones, insulin, growth factors, genetic factors and fat tissue storage in the causal pathway between anthropometric factors and breast cancer risk.1 An overall model of the biologic mechanisms involved combines overnutrition, obesity, low physical activity, chronic changes in endocrine secretion of steroid hormones and reduced production of sex hormone–binding globulin by the liver to increase breast cancer risk.25 In brief, endogenous sex hormone levels may decrease with increasing body mass index in premenopausal women but increase in postmenopausal women.26 This reversal in endogenous estrogen level could explain the differing relation between obesity and breast cancer before and after menopause.27 Nutritionally induced hyperinsulinemia and insulin resistance may be fundamental metabolic changes that result in breast cancer development.25 Obesity, especially central adiposity, may increase breast cancer risk by increasing concentrations of insulin, glucose or triglycerides.28, 29 IGF increases premenopausal breast cancer risk,30, 31 and indirect evidence exists that IGF-I levels are positively correlated with height32 and that reductions in body weight decrease IGF-I levels.33 However, several population-based studies have found that IGF-I levels are either not associated or inversely associated with body weight or body mass index.34 Hence, the role of IGF-I in the association between anthropometric risk factors and breast cancer remains unclear. Women with more fat tissue may be at increased risk because fat tissue has the capacity to store and release carcinogens.35

Although further clarification is needed about the associations and underlying biologic mechanisms that might be operative between anthropometric factors and breast cancer risk, our results corroborate and strengthen the evidence from previous research that avoiding weight gain throughout life is a means of reducing postmenopausal breast cancer risk, particularly among never-users of hormone replacement therapy.

Acknowledgements

CMF was supported by a National Health Research Scholar Award from Health Canada (6609-1929-48) and by a Canadian Institutes of Health Research New Investigator Award. KSC was supported by an Investigator Award from the Canadian Institutes of Health Research, and his research program is supported by the National Cancer Institute of Canada (NCIC) with funds from the Canadian Cancer Society (CCS) through the CCS/NCIC Sociobehavioral Cancer Research Network (grant 010282). We acknowledge the contribution of Ms. K. Douglas-England for study coordination; Ms. Z. Mah, Ms. L. Godard, Ms. M. Belic and Ms. D. Mandziuk for assistance with the research project; Ms. L. Alexander, Ms. S. Chow, Ms. P. Cooke, Ms. S. Cooper, Ms. L. Davison, Ms. M. Dickson, Ms. C. Lavis, Ms. D. Mandziuk, Ms. H. Park, Ms. J. Parrotta and Ms. N. Slot for interviewing the study participants; Ms. V. Hudson for data cleaning; Ms. V. Stagg and Ms. X. Chen for data processing and assistance with the statistical analysis; and Dr. C. Maxwell for statistical analysis of relevant data from the NPHS.

Ancillary