Height, weight, weight change, and postmenopausal breast cancer risk: the Netherlands Cohort Study Cancer

Transcription

1 Cancer Causes and Control, 1997, 8, pp Height, weight, weight change, and postmenopausal breast cancer risk: the Netherlands Cohort Study Cancer Causes and Control. Vol Piet A. van den Brandt, Miranda J.M. Dirx, Cécile M. Ronckers, Peggy van den Hoogen, and R. Alexandra Goldbohm (Received 18 March 1996; accepted in revised form 29 July 1996) The association between several anthropometric indices and breast cancer risk was evaluated within the Netherlands Cohort Study on diet and cancer, which began in 1986 and is conducted among 62,573 women aged 55 to 69 years at baseline. After 4.3 years of follow-up, data on 626 women with incident breast cancer were available with complete information on height and weight at baseline. In multivariate case-cohort analyses, a significantly positive association between adult height and breast cancer was found (P trend < 0.001). Compared with women with height 155 cm, the rate ratios of breast cancer for women with heights up to 160, 165, 170, 175, and 175 cm were 1.22, 1.19, 1.44, 1.77, and 2.06, respectively. For weight at baseline, the significant positive association with breast cancer observed in age-adjusted analysis disappeared in multivariate analysis with adjustment for height and other confounders. For body mass index (BMI) (wt[kg]/ht[m] 2 ) at baseline, no association was observed with breast cancer in multivariate analysis; compared with women with a BMI less than 23, the RR for women with a BMI of 30 or more was 0.98 with P trend = Weight and BMI at age 20 showed weak inverse associations with breast cancer risk. For gain in weight or BMI between age 20 and cohort baseline age, inconsistent increases in risk were found, with no significant trends. These data support a positive association between height and breast cancer risk among postmenopausal women. Further study is needed to evaluate the role of early diet and breast cancer in this population, and its relationship to height. Cancer Causes and Control 1997, 8, Key words: Breast neoplasms, body height, body weight, Netherlands, weight gain, women. Introduction Many studies have been conducted on the relationship between height, weight, and breast cancer risk. In 1974, a positive association between height and weight and age-adjusted breast cancer risk among Dutch postmenopausal women was observed. 1 No association was found with body mass index (BMI) (wt[kg]/ht[m] 2 ). This was followed by a large number of studies on pre- and postmenopausal breast cancer, revealing that the associations vary by menopausal status. In a recent review of Hunter and Willett, 2 results of the prospective studies as well as large case-control studies (more than 500 cases) have been summarized. Regarding height, they concluded that a modest positive association with breast cancer was found in most case-control studies, but not all; positive associations were found predominantly for postmenopausal breast cancer. Seven cohort studies all showed Drs Van den Brandt and Goldbohm, and Mss Dirx, Ronckers, and Van den Hoogen are with the Department of Epidemiology, Maastricht University, Maastricht, the Netherlands. Dr Goldbohm is also with the Department of Consumer Research and Epidemiology, TNO Nutrition and Food Research Institute, Zeist, the Netherlands. Address correspondence to Dr Van den Brandt, Department of Epidemiology, Maastricht University, P.O. Box 616, NL-6200 MD Maastricht, the Netherlands. This project is supported by the Dutch Cancer Society Rapid Science Publishers Cancer Causes and Control. Vol

2 P.A. van den Brandt et al a positive association with height, both in pre- and postmenopausal women. The reported relative risks in the cohort studies varied between 1.1 and 2.4. Although most of the reviewed case-control and cohort studies showed a positive association between BMI and postmenopausal breast cancer risk, the relative risks found in cohort studies are much closer to the null value than those of case-control studies. For premenopausal women, most cohort studies showed an inverse association with BMI, while case-control studies show both inverse and positive relationships. 2 In a recent meta-analysis on this topic, 3 it was concluded that cohort studies overall showed stronger inverse associations of BMI with premenopausal breast cancer than case-control studies; adjustment for more confounders than age produced relative risks closer to one, however. There is no reason to assume that underadjustment is less important for postmenopausal breast cancer. While adjustment for confounders seems to be important in studying the association between anthropometric indices and breast cancer, most of the cohort studies conducted so far employed only very limited adjustment. 1,4-7 Among these studies are also those which produced the strongest associations with height and/or BMI. It is therefore worthwhile to evaluate these associations in cohort studies which have additional information on potential confounders. It has also been argued 8,9 that it is the change in (relative) weight during adulthood which is the important anthropometric determinant of breast cancer. Recently, it was suggested that adult weight gain is important, particularly in women with low body mass index in early adulthood. 10 We examined the association between height and (relative) weight and postmenopausal breast cancer risk in the Netherlands Cohort Study (NLCS) on diet and cancer, where information on many potential confounders is available; we also evaluated whether dynamic anthropometric measures such as (relative) weight gain during adulthood are predictors of breast cancer risk. Materials and methods The Netherlands Cohort Study on diet and cancer started in September 1986 when 62,573 Dutch women aged 55 to 69 years were enrolled in the cohort. All women were presumably postmenopausal. At baseline, cohort members completed a self-administered questionnaire on usual dietary intake and potential confounders such as reproductive history, anthropometry, smoking habits, education, and family history of cancer. For data processing and analysis, the case-cohort approach is used; the cases are enumerated for the entire cohort (numerator information of incidence rates), while the accumulated person-years (PY) of the entire cohort are estimated using a subcohort sample (providing the denominator information). Following this approach, a subcohort of 3,500 subjects (1,688 men, 1,812 women) was sampled randomly from the cohort after the baseline exposure measurement. The subcohort has been followed-up biennially for vital status information in order to estimate the accumulated PYs in the cohort. Incident cancer cases occurring in the entire cohort have been identified by record linkage to cancer registries and a pathology register. Further details on the design of the NLCS and the method of cancer follow-up can be found in previous publications. 11,12 The present analysis is restricted to cancer incidence in the 4.3-year follow-up period from September 1986 to December The completeness of cancer follow-up was estimated to be at least 96 percent. 13 In these 4.3 years of follow-up, 762 breast cancer cases were detected in the full cohort of 62,573 women. After excluding selfreported prevalent cancer cases other than skin cancer (n = 85), cases without microscopic confirmation of diagnosis (n = 3), and carcinoma in situ (n = 24), 650 incident cases with microscopically confirmed, primary invasive breast carcinoma were available for analysis. After excluding self-reported prevalent cancer cases other than skin cancer from the female subcohort as well, 1,716 women remained in this group. In the 4.3-year follow-up period, no subcohort members were lost to follow-up. Measurement of anthropometric data and data analysis Self-reported information on height (in cm) and weight at baseline (in kg), as well as weight at age 20 (in kg) was obtained using the questionnaire. Questionnaire data of all cases and subcohort members were key-entered twice and processed in a manner blinded with respect to case/subcohort status in order to minimize observer bias in coding and interpretation of the data. Women with incomplete anthropometric data were excluded, leaving 626 cases and 1,652 subcohort members available for analyses regarding anthropometry at baseline. For analysis regarding weight at age 20, 574 cases and 1,528 subcohort members with complete data remained available. The mean values of the anthropometric variables were compared between cases and subcohort members. Other factors were considered as potential confounders if they were associated with breast cancer risk and clearly associated with the anthropometric variables. Previous analyses 14,15 have indicated that age, history of benign breast disease, family history of breast cancer, age at menarche, age at menopause, parity, age at first birth, and alcohol intake were related to disease risk in this cohort. Across the various levels of each of these factors, mean values of the anthropometric variables were compared and tested using analysis of variance. This was followed by case-cohort analyses 16 of the anthropometric variables and risk of breast cancer, based on the assumption that 40 Cancer Causes and Control. Vol

3 Anthropometry and breast cancer survival times were exponentially distributed. 15 In these age-adjusted and multivariate case-cohort analyses with categorized or continuous anthropometric variables, rate ratios, 95 percent confidence intervals (CI) and tests for trend in the rate ratios (RR) were estimated, which were corrected for the additional variance introduced by the subcohort sampling. The analyses were carried out using the GLIM-statistical package. 17,18 Two-sided P-values are reported throughout this paper. For (relative) weight, analyses also were conducted after excluding cases occurring in the first year of follow-up. Results Among the subcohort women, the mean (± standard deviation [SD]) values for height, weight and BMI at baseline were (± 6.2) cm, 68.6 (± 10.2) kg, and 25.2 (± 3.5) kg/m 2. For the breast cancer cases, the mean (± SD) values for height, weight, and BMI were (± 6.5 cm), 69.7 (± 10.3) kg, and 25.3 (± 3.4) kg/m 2, respectively. The analysis of variance showed that of the mentioned factors related to disease risk age, benign breast disease, age at menarche, parity, age at first birth, and alcohol intake were associated significantly with any of the anthropometric variables (data not shown). The final multivariate models included: age (continuous); age at menarche ( 12, 13-14, 15 years); parity (nulliparous, 1-2, 3 children); age at first birth (< 25, 25 years); and alcohol intake (0, , 30 g/day); and depending on the analysis height and/or BMI at baseline (continuous). Benign breast disease was not included because its addition did not materially change the estimated RRs. Table 1 shows the results of the age-adjusted and multivariate-adjusted analyses, of height and breast cancer risk. In the age-adjusted analysis a monotonic consistent increase in risk was seen with increasing height, with a highly significant trend test (P < 0.001). Compared with women 155 cm, the RR for women 175 cm was When height was entered as a continuous variable into the model, the RR for an increment of 10 cm was estimated at 1.32 (CI = ). The RRs were changed slightly in multivariate analysis, but the association remained highly significant (test for trend, P < 0.001), with the RR comparing the two mentioned extreme categories being 2.06 (CI = ). Entered as a continuous variable, the RR per increment of 10 cm was 1.35 (CI = ) in the multivariate model. Addition of the BMI to this model resulted in similar estimates; therefore, we presented the simpler model only. Weight at baseline was associated significantly positively with breast cancer risk in age-adjusted analyses (P trend = 0.02) (Table 2). However, multivariate adjustment led to different conclusions. In a multivariate model with the mentioned covariates but without height, there was still a significant positive association with weight (P trend = 0.01). When height was additionally included in the model, the association was weakened and no longer significant (P trend = 0.23). Further, the dose-response relationship was inconsistent. Entered as a continuous variable, the RR per increment in weight of 10 kg was 1.09 (CI = ) in the latter multivariate model (Table 2). Exclusion of cases diagnosed in the first year of follow-up resulted in an RR of The BMI at baseline was not associated with risk of postmenopausal breast cancer in age-adjusted or multivariate models. Table 1. Rate ratio (RR) of breast cancer according to height; Netherlands Cohort Study, Anthropometric variable Age-adjusted Multivariate-adjusted Cases Personyears RR (CI) a Cases Personyears RR b Height (cm), categorical (mean) c 155 (152.2) d d (157.7) ( ) ( ) (161.9) 147 1, ( ) 124 1, ( ) (166.8) 205 2, ( ) 185 2, ( ) (171.4) 129 1, ( ) 112 1, ( ) 175 (176.7) ( ) ( ) Test for trend: P value < < Height, continuous 10 cm increment 1.32 ( ) 1.35 ( ) a CI = 95% confidence interval. b The model included terms for age, age at menarche, parity, age at first birth, and alcohol intake. c Numbers in parentheses, mean in subcohort women. d Reference group. (CI) a Cancer Causes and Control. Vol

4 P.A. van den Brandt et al Table 2. Rate ratio (RR) of breast cancer according to weight and BMI a at baseline; Netherlands Cohort Study, Anthropometric variable Age-adjusted Multivariate-adjusted Cases Personyears RR (CI) b Cases Personyears RR Weight (kg), categorical (mean) c 59 (54.4) 73 1, d d,e (61.7) 117 1, ( ) 107 1, ( ) (66.6) 153 1, ( ) 136 1, ( ) (71.4) 93 1, ( ) 79 1, ( ) (76.2) ( ) ( ) 80 (85.9) 107 1, ( ) ( ) Test for trend: P value Weight, continuous 10 kg increment 1.12 ( ) 1.09 ( ) (CI) b BMI (kg/m 2 ), categorical (mean) c 22.9 (21.2) 164 1, d 149 1, d,f (24.0) 163 1, ( ) 147 1, ( ) (26.0) 133 1, ( ) 114 1, ( ) (28.3) 112 1, ( ) ( ) 30 (32.6) ( ) ( ) Test for trend: P value BMI, continuous 8 kg/m 2 increment 1.08 ( ) 1.11 ( ) a BMI = body mass index (wt[kg]/ht[m] 2 ). b CI = 95% confidence interval. c Numbers in parentheses, mean in subcohort women. d Reference group. e The model included terms for age, age at menarche, parity, age at first birth, alcohol intake, and height. f The model included all of the above terms, except height. Women with a BMI of 30 or more had an RR of 0.98 (CI = ) compared with women with a BMI less than 23 kg/m 2 (P trend = 0.46) in multivariate analysis (Table 2). When BMI was entered as continuous variable, the RR per increment of 8 kg/m 2 was 1.11 (CI = ). When cases diagnosed in the first year were excluded, an RR of 1.11 was found. For both weight and BMI, separate analyses in the three five-year age categories (55-59, 60-64, years) did not indicate that the association with breast cancer was strongest among older women (data not shown). For weight at age 20 and BMI at age 20, the mean (± SD) values among subcohort women were (± 7.91) kg, and (± 2.73) kg/m 2, respectively. The mean (± SD) difference between weight at baseline and weight at age 20 was (± 10.13) kg; for BMI, the mean (± SD) difference was 3.71 (± 3.74) kg/m 2. Results regarding weight and BMI at age 20 and breast cancer risk are presented in Table 3. Weight at age 20 was not associated significantly with breast cancer risk. The nonsignificant positive association in age-adjusted analysis was reversed to a nonsignificant inverse association after multivariate adjustment. For BMI at age 20, inverse associations with breast cancer risk were found in both analyses. When BMI at age 20 was entered as categorical variable, the test for trend was significant (P trend = 0.03) in the multivariate analysis, whereas entering it as continuous variable did not result in a significant association (Table 3). The results for weight change between 20 years of age and baseline are shown in Table 4. In both age-adjusted analysis and multivariate analyses, weight change was not associated significantly with breast cancer risk (P trend in multivariate analysis = 0.13). The trend in RRs also was inconsistent: the highest weight-gain category showed a borderline significantly increased RR of 1.57 compared with the reference category (0-4.9 kg gain), but for the second highest weight-gain category, an RR of 0.99 was observed. With regard to change in BMI between age 20 and baseline, all RRs for the body-mass-gain categories were above one, and two were increased significantly in the multivariate analysis. However, there was no clear increasing trend in risk with increasing gain in body mass, and the test for trend was not significant (Table 4). Following the recent approach by Barnes-Josiah et al, Cancer Causes and Control. Vol

5 Anthropometry and breast cancer Table 3. Rate ratio (RR) of breast cancer according to weight and BMI a at age 20; Netherlands Cohort Study, Anthropometric variable Age-adjusted Multivariate-adjusted Cases Personyears RR (CI) b Cases Personyears RR Weight age 20 (kg), categorical (mean) c 49 (46.3) d d,e (51.2) 112 1, ( ) 97 1, ( ) (56.7) 130 1, ( ) 112 1, ( ) (61.0) 151 1, ( ) 134 1, ( ) 65 (69.2) 129 1, ( ) 109 1, ( ) Test for trend: P value Weight age 20, continuous 10 kg increment 1.04 ( ) 0.94 ( ) (CI) b BMI age 20, categorical (mean) c 19.9 (18.4) 176 1, e 159 1, e,f (20.5) ( ) ( ) (22.0) 167 1, ( ) 153 1, ( ) (24.3) 113 1, ( ) 95 1, ( ) 27 (29.0) ( ) ( ) Test for trend: P value BMI age 20, continuous 8 kg increment 0.84 ( ) 0.79 ( ) a BMI = body mass index (wt[kg]/ht[m] 2 ). b CI = 95% confidence interval. c Numbers in parentheses, mean in subcohort women. d Reference group. e The model included terms for age, age at menarche, parity, age at first birth, alcohol intake, and height. f The model included all of the above terms, except height. we also analyzed the relationship between weight change and breast cancer across categories of BMI at age 20 (Table 5). For these analyses, weight change was grouped into four categories (up to 4.9; 5-9.9; ; and 20 or more kg gained) and BMI at age 20 was dichotomized using 21 kg/m 2 as cutoff point. The mean (± SD) weight gain in the subgroups of women with early BMI below and above 21 kg/m 2 was (± 8.71) and 7.52 (± 9.94) kg/m 2, respectively. In both subgroups defined by the BMI at age 20, a weak positive association was observed between weight change at breast cancer. None of the RR estimates or tests for trend were significant, however. The table also shows that the effect of weight gain is not stronger in those women who had a BMI at age 20 of less than 21 kg/m 2 compared with women who had a higher early BMI. Discussion In this cohort study, a significant positive association between self-reported height and postmenopausal breast cancer incidence was found after controlling for potential confounders. For weight and BMI at baseline, no association was observed; in particular, the multivariateadjusted positive association between weight and breast cancer disappeared after additional control for height. Weight and BMI at age 20 were associated, if at all, weakly inversely with breast cancer risk. Change in (relative) weight between age 20 and age at baseline were associated inconsistently positively with breast cancer risk, with no significant tests for trend. Mean values for self-reported height and weight at baseline of the women in the NLCS cohort were comparable to those of a representative sample of Dutch women of 50 to 69 years of age measured in The reported mean height and weight of the latter group (n = 3,377) were cm an 68.9 kg, respectively, compared with cm and 68.6 for NLCS subcohort women. The potential for selection bias in the NLCS is low considering the high completeness of cancer follow-up and PYs follow-up. Given the prospective design of the study, it is also unlikely that information bias can explain the results. Regarding confounding, it was noted that the moderately strong association between height and breast Cancer Causes and Control. Vol

6 P.A. van den Brandt et al Table 4. Rate ratio (RR) of breast cancer according to change in weight and change in BMI a between age 20 and baseline: Netherlands Cohort Study, Anthropometric variable Age-adjusted Multivariate-adjusted Cases Weight change, categorical (mean) c Personyears RR (CI) b Cases Personyears -34-<0 (-6.5) ( ) ( ) (2.4) d d,e (6.8) 113 1, ( ) 104 1, ( ) (11.6) 137 1, ( ) 120 1, ( ) (16.5) ( ) ( ) (21.6) ( ) ( ) +25 (30.7) ( ) ( ) Test for trend: P value 0.27 f 0.13 f RR (CI) b BMI change, categorical (mean) c < 0 (-2.5) ( ) ( ) (1.2) 92 1, d 83 1, d,g (3.1) 155 1, ( ) 142 1, ( ) (4.9) 117 1, ( ) 102 1, ( ) (6.9) ( ) ( ) (8.8) ( ) ( ) (12.3) ( ) ( ) Test for trend: P value 0.33 f 0.09 f a BMI = body mass index (wt[kg]/ht[m] 2 ). b CI = 95% confidence interval. c Numbers in parentheses, mean in subcohort women. d Reference group. e The model included terms for age, age at menarche, parity, age at first birth, alcohol intake, and height. f Test for trend applies to weight gain categories only. g The model included all of the above terms, except height. Table 5. Rate ratio (RR) of breast cancer according to BMI a at age 20 and weight change between age 20 and baseline: Netherlands Cohort Study, BMI age 20 Weight change (kg) age 20-baseline ( d ) No. of cases Person-years RR b (CI) c Trend (P-value) < 21 < +5.0 (1.2) e (7.0) ( ) (13.9) 101 1, ( ) (25.4) ( ) 21 <+5.0 (-2.2) 90 1, e (6.6) ( ) (13.2) ( ) (25.6) ( ) a BMI = body mass index (wt[kg]/ht[m] 2 ). b The model included terms for age, age at menarche, parity, age at first birth, alcohol intake, and height. c CI = 95% confidence interval. d Numbers in parentheses, mean in subcohort women. e Reference group. 44 Cancer Causes and Control. Vol

7 Anthropometry and breast cancer cancer risk seen in age-adjusted analysis was relatively insensitive to additional adjustment for other breast cancer risk factors. It cannot be excluded that residual confounding is still present, but, in the current multivariate analysis, the potential confounding by the great majority of known breast-cancer risk factors was considered. Regarding (relative) weight, the follow-up period of 4.3 years may still be rather short, but analyses without cases occurring in the first year of follow-up produced similar results as were obtained for the total group. It will still be interesting, nevertheless, to reevaluate these relationships after a longer follow-up period with more time between weight assessment and diagnosis. Also, more data will be available then to evaluate further whether (relative) weight is a predictor of breast cancer in older postmenopausal women. The observed positive association with height is in agreement with most other prospective studies on postmenopausal breast cancer. 1,4-6,20 The association in the NLCS appears equally strong as in an earlier Dutch cohort study 1 or in Norwegian cohorts, 5,6 suggesting that residual confounding is a less likely explanation for the results in these earlier studies. The RR of breast cancer for the category of tallest cf the smallest women was 2.06 in the NLCS. This is higher than in US cohorts which included mostly White women. 20,21 However, US studies where women with a greater chance of undernutrition during childhood or adolescence were overrepresented, also show stronger associations with height. 22,23 Palmer et al, 23 for example, observed a positive association with height in Black US women but not in White US women. Attained height might be an indicator of childhood energy intake in situations where there is enough variation in energy intake. 24 Interest in the association with height recently has been intensified because of the hypothesized potential influence of early (childhood) diet on breast cancer. 25 The observed stronger associations found in the Norwegian cohorts also have been attributed to energy restriction during the peripubertal years 26 due to food scarcity in World War II. In the Netherlands, energy restriction during World War II (particularly during the Hungerwinter ) may have resulted in stunted growth and greater variation in attained adult height. 27 This energy restriction was concentrated in certain regions of the Netherlands; 28 we are investigating currently whether this also might have affected breast cancer risk and whether attained height plays a role in this respect. Thus, the protective effect of energy restriction on breast cancer observed in rodents 29 has been suggested 24 to explain the positive association with height. Recently, Stoll 30 has indicated that better nutrition accelerates growth hormone release which, in turn, increases levels of insulin-like growth factor (IGF). The adolescent growth spurt involves stimulation by growth hormone, IGF and sex steroids, and Stoll hypothesizes that the combination of IGF and sex steroids results in mitogenic effect on developing mammary tissue in adolescence and a concomitant increased risk of epithelial atypia and carcinogenesis. 30 Dorgan et al 31 recently have suggested, based on a cross-sectional study, that height might influence breast cancer risk through its positive association with follicular-phase plasma-estradiol. It is expected that the renewed attention for height and breast cancer soon will result in more data on the potential associated hormonal pathways. For weight at baseline, the significant positive association observed initially in our study disappeared once height was controlled for. Similar to what Brinton et al 32 already noted in their study, the overriding effect of height and breast cancer risk is also reflected in the absence of an association between BMI and breast cancer in the NLCS-cohort. Our observations on risk associated with BMI at baseline are in agreement with many other cohort studies indicating no or only a weak positive association among postmenopausal women 1,4,7,20,21,33 although significant positive associations have been reported. 5 It would be interesting to evaluate further why so many casecontrol studies on this issue have shown positive associations with postmenopausal breast cancer while cohort studies generally have not. Most case-control studies have examined the relation between weight assessed close to diagnosis and found positive associations, whereas in cohort studies, the interval between weight assessment and diagnosis is usually much longer. 9 Interestingly, in another Dutch cohort study, conducted within a mammography screening project, a positive association was reported initially between baseline BMI and the prevalence of postmenopausal breast cancer (as detected at first screening), 34 whereas no association existed with incident cases detected afterwards. 33 When we limited the analysis of the NLCS data to the first year of follow-up, the association between BMI and breast cancer risk was similar to that estimated for the total group. Hypotheses regarding fat distribution and breast cancer 9,21 could not be evaluated in our study because there is no information on regional adiposity. Our results on a slight inverse association between weight or BMI at age 20 and breast cancer risk are in agreement with most other studies 10,20,21,32,35,37 although positive associations also have been reported. 35 Based on observations from several studies, 8,10,32,36 it has been hypothesized 9 that adult weight gain is a strong predictor of postmenopausal breast cancer risk. In our data, there was no consistent increase in breast cancer risk across (relative) weight gain categories and the trend tests were not significant. Similar results regarding the inconsistent shape of the dose-response relationship have been Cancer Causes and Control. Vol

8 P.A. van den Brandt et al reported by London et al 20 and, in fact, also in the study of Brinton et al, 32 where most analyses revealed an increased risk for the category representing the smallest weight gain but no large further increase in relative risk thereafter. Our analyses did not show any significant association with weight gain when stratified by early BMI, although there was some suggestion of a weak positive association in both subgroups. Contrary to what Barnes- Josiah et al 10 observed, we found no stronger association with weight gain in the women with a low early BMI. Although the association between weight gain and breast cancer risk is weak in the NLCS cohort, the higher average weight gain observed in the low early BMI group (14.0 cf 7.5 kg) might explain part of the slight inverse association between BMI at age 20 and breast cancer. It also may be that the timing of weight and weight gain needs to be investigated further. 8,32 Ballard-Barbash et al, 8 for example, used the difference between lowest weight since age 18 and highest weight ever in their analysis. This latter weight is not necessarily equal to the baseline weight which is often used in cohort studies. We could not evaluate further the timing of weight (gain) because we had no information on weight at other ages than age 20 and at baseline. In conclusion, this study confirms the importance of height as an independent predictor of postmenopausal breast cancer risk. In the light of emerging hypotheses on the potential influence of childhood or adolescent diet (and its reflection in attained height in situations of energy restriction 2 ) on breast cancer risk, further investigation of the role of early diet on breast cancer is warranted. Acknowledgments We are indebted to the participants of this study and further thank the regional cancer registries, the Dutch national database of pathology (PALGA), and the National Health Care Information Center for providing incidence data; A. Volovics for statistical advice; E. Dorant, S. van de Crommert, H. Brants, W. van Dijk, P. Florax, M. Moll, J. Nelissen, and A. Pisters for assistance; H. van Montfort, R. Schmeitz, T. van Montfort, and M. de Leeuw for programming and statistical assistance. References 1. De Waard F, Baanders-Van Halewijn EA. A prospective study in general practice on breast cancer risk in postmenopausal women. Int J Cancer 1974; 14: Hunter DJ, Willett WC. Diet, body size, and breast cancer. Epidemiol Rev 1993; 15: Ursin G, Longnecker MP, Haile RW, Greenland S. A metaanalysis of body mass index and risk of premenopausal breast cancer. Epidemiology 1995; 6: Törnberg SA, Holm LE, Carstensen JM. Breast cancer risk in relation to serum cholesterol, serum beta-lipoprotein, height, weight, and blood pressure. Acta Oncol 1988; 27: Tretli S. Height and weight in relation to breast cancer morbidity and mortality. A prospective study of 570,000 women in Norway. Int J Cancer 1989; 44: Vatten LJ, Kvinnsland S. Body height and risk of breast cancer. A prospective study of 23,831 Norwegian women. Br J Cancer 1990; 61: Vatten LJ, Kvinnsland S. Body mass index and risk of breast cancer: a prospective study of 23,826 Norwegian women. Int J Cancer 1990; 45: Ballard-Barbash R, Schatzkin A, Taylor PR, Kahle LL. Association of change in body mass with breast cancer. Cancer Res 1990; 50: Ballard-Barbash R. Anthropometry and breast cancer. Body size a moving target. Cancer 1994; 74: Barnes-Josiah D, Potter JD, Sellers TA, Himes JH. Early body size and subsequent weight gain as predictors of breast cancer incidence (Iowa, United States). Cancer Causes Control 1995; 6: Van den Brandt PA, Goldbohm RA, van t Veer P, Volovics A, Hermus RJJ, Sturmans F. A large-scale prospective cohort study on diet and cancer in The Netherlands. J Clin Epidemiol 1990; 43: Van den Brandt PA, Schouten LJ, Goldbohm RA, Dorant E, Hunen PMH. Development of a record linkage protocol for use in the Dutch cancer registry for epidemiological research. Int J Epidemiol 1990; 19: Goldbohm RA, Van den Brandt PA, Dorant E. Estimation of the coverage of Dutch municipalities by cancer registries and PALGA based on hospital discharge data. Tijdschr Soc Gezondheidsz 1994; 72: Van den Brandt PA, van t Veer P, Goldbohm RA, et al. A prospective cohort study on dietary fat and the risk of postmenopausal breast cancer. Cancer Res 1993; 53: Van den Brandt PA, Goldbohm RA, van t Veer P. Alcohol and breast cancer: results from the Netherlands Cohort Study. Am J Epidemiol 1995; 141: Self SG, Prentice RL. Asymptotic distribution theory and efficiency results for case-cohort studies. Ann Stat 1988; 16: Baker RJ. Glim 3.77 Reference Manual. Oxford, UK: Numerical Algorithms Group, Aitkin M, Anderson D, Francis B, Hinde J. Statistical Modelling in GLIM. Oxford: Oxford University Press, Netherlands Central Bureau of Statistics. Statistical Yearbook of the Netherlands The Hague, Netherlands: SDU Publishers, London SJ, Colditz GA, Stampfer MJ, Willett WC, Rosner B, Speizer FE. Prospective study of relative weight, height, and risk of breast cancer. JAMA 1989; 262: Folsom AR, Kaye SA, Prineas RJ, Potter JD, Gapstur SM, Wallace RB. Increased incidence of carcinoma of the breast associated with abdominal adiposity in postmenopausal women. Am J Epidemiol 1990; 131: Swanson CA, Jones DY, Schatzkin A, Brinton LA, Ziegler RG. Breast cancer risk assessed by anthropometry in the NHANES I epidemiological follow-up study. Cancer Res 1988; 48: Palmer JR, Rosenberg L, Harlap S, et al. Adult height and risk of breast cancer among US black women. Am J Epidemiol 1995; 141: Willett WC. Nutritional Epidemiology. New York, NY (USA): Oxford University Press, De Waard F, Trichopoulos D. A unifying concept of the aetiology of breast cancer. Int J Cancer 1988; 41: Cancer Causes and Control. Vol

9 Anthropometry and breast cancer 26. Vatten LJ, Kvinnsland S. Prospective study of height, body mass index and risk of breast cancer. Acta Oncol 1992; 31: Van Noord PAH, Arias-Careaga S. The Dutch-Famine : lasting effects on adult height. Am J Epidemiol 1995; 141: S Stein Z, Susser M, Sanger G, Marolla F. Famine and Human Development. The Dutch Hungerwinter of Oxford, UK: Oxford University Press, Albanes D. Total calories, body weight, and tumor incidence in mice. Cancer Res 1987; 47: Stoll BA. Does extra height justify a higher risk of breast cancer? Ann Oncol 1992; 3: Dorgan JF, Reichman ME, Judd JT, et al. The relation of body size to plasma of estrogens and androgens in premenopausal women (Maryland, United States). Cancer Causes Control 1995; 6: Brinton LA, Swanson CA. Height and weight at various ages and risk of breast cancer. Ann Epidemiol 1992; 2: Den Tonkelaar I, Seidell JC, Collette HJA, De Waard F. A prospective study on obesity and subcutaneous fat patterning in relation to breast cancer in postmenopausal women participating in the DOM-project. Br J Cancer 1994; 69: Den Tonkelaar I, Seidell JC, Collette HJA, De Waard F. A prospective study on obesity and subcutaneous fat patterning in relation to breast cancer in postmenopausal women participating in the DOM-project. Cancer 1992; 69: Le Marchand L, Kolonel LN, Earle ME, Mi MP. Body size at different periods of life and breast cancer risk. Am J Epidemiol 1988; 128: Radimer K, Siskind V, Bain C, Schofield F. Relation between anthropometric indicators and risk of breast cancer among Australian women. Am J Epidemiol 1993; 138: Chy SY, Lee NC, Wingo PA, Senie RT, Greenberg RS, Peterson HB. The relationship between body mass and breast cancer among women enrolled in the Cancer and Steroid Hormone Study. J Clin Epidemiol 1991; 44: Cancer Causes and Control. Vol

10