BODY MASS INDEX IN RELATION TO OVARIAN CANCER: AMULTI-CENTRE NESTED CASE-CONTROL STUDY

Transcription

1 Int. J. Cancer: 99, (2002) 2002 Wiley-Liss, Inc. DOI /ijc BODY MASS INDEX IN RELATION TO OVARIAN CANCER: AMULTI-CENTRE NESTED CASE-CONTROL STUDY Publication of the International Union Against Cancer Annekatrin LUKANOVA 1,2 *,Paolo TONIOLO 3 5,Eva LUNDIN 2,6,Andrea MICHELI 7,Arslan AKHMEDKHANOV 3 5,Paola MUTI 8, Anne ZELENIUCH-JACQUOTTE 4,5,Carine BIESSY 1,Per LENNER 9,Vittorio KROGH 7,Franco BERRINO 7,Goran HALLMANS 2,Elio RIBOLI 1 and Rudolf KAAKS 1 1 International Agency for Research on Cancer, Lyon, France 2 Department of Public Health and Clinical Medicine/Nutritional Research, Umeå University Hospital, Sweden 3 Department of Obstetrics and Gynecology, New York University School of Medicine, New York, NY, USA 4 Kaplan Comprehensive Cancer Center, New York University School of Medicine, New York, NY, USA 5 Department of Environmental Medicine, New York University School of Medicine, New York, NY, USA 6 Department of Medical Biosciences/Pathology, University of Umeå, Umeå, Sweden 7 Unit of Epidemiology, Istituto Nazionale Tumori, Milan, Italy 8 Department of Social and Preventive Medicine, State University of New York at Buffalo, School of Medicine and Biomedical Sciences, Buffalo, NY, USA 9 Department of Oncology, Umeå University Hospital, Umeå, Sweden The incidence of ovarian cancer is up to 10 times higher in Western countries than in rural Asia and Africa. One common consequence of awestern lifestyle is the development of excessive body weight and obesity. Amulti-centre prospective study was conducted to investigate the association between body mass index (BMI) and ovarian cancer risk. A case-controlstudywasnestedwithin3prospectivecohortsin New York (USA), Umeå (Sweden) and Milan (Italy). Information on anthropometry, demographic characteristics, medical history and lifestyle was obtained at the time of subjects recruitmentineachcohort.womendiagnosedwith primary, invasive epithelial ovarian cancer from the 3cohorts (n 122) diagnosed 12 months or later after recruitment into the respective cohort served as case subjects. For each case subject, 2control subjects that matched the case subject on cohort, menopausal status, age and date of recruitment were randomly identified. Data were analyzed by conditional logistic regression. There was an inverse association between BMI and ovarian cancer risk. For increasing quartiles of BMI above the lowest, the ORs were 0.62 ( ), 0.59 ( ) and 0.46 ( ), p Analyses limited to women diagnosed 3 or more years after recruitment into the cohorts did not alter these findings. When obese women (BMI > 30) were compared to lean women(bmi<23),theinverseassociationbecamestronger, with an OR of 0.38 ( ), p<0.02. There was some evidence of direct association of ovarian cancer with height, which was limited to cancers diagnosed before age 55. Our data suggest that increasing body weight may confer aprotection against ovarian cancer Wiley-Liss, Inc. Keywords:bodymassindex;height;anthropometry;ovariancancer; prospective study; cohort Age-standardised incidence rates of ovarian cancer are up to 10 times higher in Western, industrialized countries than in rural Africa or Asia. 1 Migration studies have shown that women of Chinese and Japanese descent who reside in the U.S. tend to have higher incidence rates than their Asian counterparts and that in women years of age, the incidence of ovarian cancer is similar in U.S.-born Chinese women and U.S. white women. 2 These observations suggest that environmental factors related to a Western lifestyle may be associated with the increase in ovarian cancer risk. Acommon consequence of Western lifestyle, which has been shown to increase risk of some female cancers, such as postmenopausal breast and endometrial cancers, is the development of excessive body weight and obesity. Increase in body weight relative to body height, as reflected by an increased body mass index (BMI), may alter cancer risk through changes in endogenous hormone metabolism. 3 5 The role of agreater body weightisconsiderablyclearerinrelationtobreastandendometrial cancers than to ovarian cancer, for which the majority of studies report non-significant associations. 5 8 Arecent extensive review, including 29 reports on the association of body weight/bmi with ovariancancerrisk,concludedthatthereisadirect,howeverweak, association of increased BMI with ovarian cancer based on the results of cohort [summary OR 1.2 ( )] and populationbasedcase-controlstudies[summaryor1.4( )]. 9 Hospitalbasedcase-controlstudiesshowedheterogeneousresults;however, some of the differences in the results might reflect specific characteristics of the control groups included, e.g., ahigher proportion of overweight women in the control series due to obesity-related conditions such as cardiovascular disease and diabetes. 9 In this paper, we report the results of the relationship of BMI with ovarian cancer risk in amulti-centered case-control study nested within 3cohort studies in New York, USA, Umeå, Sweden and Milan, Italy. MATERIAL AND METHODS The current study is part of alarger collaborative project between 3 cohort studies to investigate the relationship between circulating hormone levels and ovarian cancer risk. Abbreviations: BMI, body mass index; CI, confidence interval; FSH, follicle-stimulating hormone; GLM, generalized linear model; HRT, hormone replacement therapy; NSHDS, Northern Sweden Health and Disease Cohort; NYUWHS, New York University Women s Health Study; OC, oral contraceptives; OR, odds ratio; ORDET, hormones and diet in the etiology of breast cancer study; WHR, waist-to-hip ratio. The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute. Grant sponsor: US National Cancer Institute; Grant numbers: R01 CA81200, R01 CA81188, R01 CA *Correspondence to: Department of Nutrition and Cancer, IARC, 150 Cours Albert Thomas Lyon, France. Fax: , lukanova@iarc.fr Received 5 October 2001; Revised 25 January 3002; Accepted 30 January 2002 DOI /ijc Publishedonline29March2002inWileyInterScience( wiley.com).

2 604 LUKANOVA ET AL. Study cohorts The collaborating cohorts have been described in detail previously These are the New York University Women s Health Study (NYUWHS), the Northern Sweden Health and Disease Study (NSHDS) and the Hormones and Diet in the Etiology of Breast Cancer Study (ODRET). A summary of selected characteristics of these cohorts is shown in Table I. At the time of recruitment in each cohort study, subjects were asked to complete a self-administered questionnaire to collect demographic, lifestyle (smoking, education and physical activity) and medical information (diabetes, hypertension, breast disease and family history of cancer). At recruitment, measurements of body height and weight were taken for all participants of the ORDET study and for 77% of the NSHDS participants. In the NYUWHS cohort and the remaining NSHDS participants height and weight at baseline were self-reported. Analysis in a subsample of the NSHDS, for which information on both self-reported and measured anthropometric characteristics were available, showed very high correlation of self-reported with measured height (r 0.97) and less strong with weight (0.75). In addition, waist and hip measurements were available for all ORDET subjects and a small proportion of the NYUWHS subjects [45 (20%) of the subjects included in this study]. The baseline questionnaires in NYUWHS and ORDET cohorts included a detailed section on reproductive history and exogenous hormone use. In the NSHDS, a reproductive history questionnaire has been administered prospectively to 47% of the subjects and a similar questionnaire was sent out to all women, selected to participate in the ovarian cancer study (response rate: 95%). There was a good correspondence between the reproductive information obtained prospectively and that from questionnaires administered, retrospectively. In the NSHDS, reproductive information concerning deceased case subjects (n 12) was obtained from their medical records. Multi-centre nested case-control study Case subjects were all cohort members with primary, invasive epithelial ovarian cancer diagnosed 12 months after recruitment into the respective cohort study, who had not reported previous diagnoses with other cancers, who did not use of exogenous hormones and had data about their height and weight at the time of recruitment. A total of 122 eligible ovarian cancer case subjects were included (Table I). Among those with histological verification (n 105), 51% of tumors were serous (n 54), 11% endometrioid (n 12), 10% mucinous (n 11) and 7% clear cell (n 7), while 20% were classified as carcinoma not-otherwise specified (n 17), or as mixed (n 1) or undifferentiated (n 3) carcinomas. The time between recruitment in the parent cohort and cancer diagnosis ranged from 12 months to 13.3 years, with an average of 5.8 years (median 5.4 years). Mean age at cancer diagnosis was years (median 61.1). Eighty-seven percent of the case subjects (n 106) were diagnosed 2 years after recruitment into the cohort and 76% (n 93) after 3 or more years. For each case subject, 2 control subjects were selected at random among appropriate risk sets. The risk set for a given case subject included all cohort subjects alive, free of cancer and not having had a bilateral ovariectomy at the date of cancer diagnosis of the index case and matching the case on parent cohort study, menopausal status at enrollment, age ( 6 months), date at recruitment ( 3 months) and, for premenopausal subjects, day of the menstrual cycle at blood donation (for the NYUWHS and ORDET subjects). The goodness of the matching for menopausal status was confirmed by follicle-stimulating hormone (FSH) measurements (FSH levels 12.5 IU/ml were considered premenopausal) and control subjects were replaced if necessary. The agematching criteria were somewhat relaxed for the ORDET subjects because of the small number of potential control subjects in the risk sets. Only women who did not report use of hormones at recruitment were included in the NYUWHS and ORDET cohorts. TABLE I SELECTED CHARACTERISTICS OF THE COHORT STUDIES INCLUDED IN THE POOLED ANALYSIS OF BMI AND OVARIAN CANCER Cases/controls Anthropometry Eligibility criteria for enrolment in the parent cohort Last complete follow-up Age at recruitment Baseline cohort Study setting Year of recruitment Cohort 14, December /139 Self-reported No use of hormones or pregnancies in the last 6 months Mamographic screening clinic NYUWHS New York, USA None 1980 present General population 43, September /64 Measured (for 77%) / self-reported NSHDS Umeå, Sweden Healthy volunteers 10, January 97 16/30 Measured No current pregnancy or nursing or hormone use, previous cancer or bilateral ovariectomy ORDET Milan, Italy 1 With histological verification and review of the medical records.

3 BMI AND OVARIAN CANCER 605 NSHDS participants, who were users of exogenous hormones at the time of blood donation, were excluded after the initial selection of cases and controls. The NSHDS and ORDET components of the study included only Caucasian subjects, while among NYUWHS subjects for this study, 5% of the women had classified themselves as Black, 2% were Hispanic, 2% reported other ethnicity and for 22% of the women no data about ethnicity was available. In total 122 case subjects and 233 control subjects were included for this study. The Ethical Committees of New York University School of Medicine, the University of Umeå, the Istituto Nazionale Tumori in Milan and the International Agency for Research on Cancer, in Lyon, France, have reviewed and approved the present study. Statistical analysis An analysis of co-variance was used to investigate subgroup differences adjusting for age at sampling, case-control and menopausal status by the Generalized Linear Model (GLM) procedure of the Statistical Analysis System (SAS)(SAS Institute, Cary, NC). 13 To estimate odds ratios (OR) and their 95% confidence intervals (CI) for disease by quartiles of the anthropometric variables of interest, conditional logistic regression models were constructed using the SAS PHREG procedure. Quartile cut-points were determined according to the distribution of variables in the control group. In addition, all the analyses were performed using cohortspecific cut-off points, again based on the distribution of the controls. Likelihood ratio tests were used to assess linear trends in ORs across quartiles, assigning scores of 1, 2, 3 and 4 to the 4 categories. Linear trends were also calculated on the natural scale of the anthropometric variables. Statistical tests and corresponding p-values were 2-sided and the criterion for statistical significance was set at the 5% level. Interaction was assessed by the use of Wald test. Multivariate logistic regression was used to estimate ORs adjusted for possible confounding factors other than those controlled for by matching. Confounding factors investigated included age at menarche and menopause, full term pregnancies (yes/no/missing), smoking (lifetime non-smoker/current/exsmoker/missing data), OC (yes/no/missing) and hormone replacement therapy (HRT) use. In addition, mutual adjustments of BMI models for height and of height models for BMI were performed. The potential confounding effects of physical activity, education and selected medical conditions (family history of breast cancer, diabetes and hypertension) were also examined. RESULTS Mean age at cohort recruitment was 53.7 for the NYUWHS subjects, 55.4 in the NSHDS and 50.2 for the ORDET subjects. Thirty-four percent of the case and control subjects were premenopausal at recruitment. Overall, case subjects had lower mean weight (65.1 vs kg, p 0.05) and mean BMI than controls (25.0 vs. 26.0, p 0.03), but mean height was similar between the 2 groups (161.4 vs cm.). Despite the similar mean BMI values of the 3 cohorts, mean height was different: cm in the NYUWHS, cm in the NSHDS and cm in the ORDET (p ). There was no difference in mean age at menarche between the cases and the controls, but on average the cases experienced menopause 1 year later than the controls (at age 50.1 vs years, p 0.09). More cases than controls were nulliparous (31% vs. 18%, p 0.01) and among parous women controls more often than cases had given birth to 3 or more children (40% vs. 34%). Age at first full-term pregnancy was similar between the 2 groups. The cases reported use of OC less frequently than the controls (24 vs. 31%) and for a shorter duration, but the number of users was small resulting in a large variation in the data about duration of OC use. Exclusion of the NSHDS subjects, for whom data about hormone use had been collected retrospectively, did not change the results. More cases reported use of HRT than controls, but these results appeared to have been influenced largely by the retrospectively collected information in the NSHDS. Because the retrospectively collected information may have been prone to recall bias or be of better than average quality, these portion of data were not utilized in the final analysis. Conditional logistic regression analysis showed an inverse association between BMI and ovarian cancer risk. The OR was 0.52 ( ) in the highest BMI quartile compared to the lowest (Table II). Adjustments for parity, OC use and tobacco smoking only slightly influenced the point estimates (Table II). These findings remained essentially the same when analyses were restricted to subjects with ovarian cancer whose diagnosis was made 3 or 4 or more years after recruitment into the cohorts (data not shown). The inverse association of BMI with ovarian cancer was driven by the associations in the NYUWHS and ORDET cohorts, while no relationship between BMI and ovarian cancer was less evident in the NSHDS (Table III). However, restriction of the analyses to women whose height and weight was measured (all ORDET and 77% of the NSHDS subjects) also showed an attenuation in ovarian cancer with increasing BMI [OR 0.74 ( ) in the upper tertile of the conditional regression model with no adjustments and OR 0.43 ( ) for the top tertile in the adjusted model]. When cohort-specific quartiles were used, the results were very similar (data not shown). There was no indication that the association between BMI and ovarian cancer risk was modified by menopausal status at recruitment, or age at diagnosis of the case. When obese women (BMI 30) were compared to lean women (BMI 23), the inverse association became stronger (OR 0.37 ( ), p 0.02 after adjustments for parity, OC use and smoking). Waist-hip ratio (WHR) data were available only in the ORDET study and for 20% of the NYUWHS subjects: a total of 32 matched sets. In unadjusted analyses, there was no relationship of categories of WHR with ovarian cancer risk. Adjustment for BMI increased the point estimates to 2.06 ( ) and 1.58 ( ) in the 2 upper tertiles, respectively. Additional adjustments for parity, OC and smoking appeared to somewhat increase the risk estimates, but the number of subjects in these analyses were very small, the confidence intervals remaining very wide. Height was not related to ovarian cancer risk in the full study population. Subsequent analyses indicated that the association between height (as a continuous variable) and ovarian cancer risk was modified by age at diagnosis (p for interaction of age TABLE II RISK OF OVARIAN CANCER BY QUARTILES OF BMI BMI quartile Cases Controls OR 95% CI 1 OR 95% CI ( ) 0.62 ( ) ( ) 0.59 ( ) ( ) 0.46 ( ) p for trend 0.05 (cont. 0.03) 0.03 (cont. 0.03) 1 Conditional logistic regression on case-control pairs matched for study cohort, age and time at recruitment into the study, menopausal status and day of menstrual cycle for premenopausal women. 2 Additional adjustment for parity, OC and smoking.

4 606 LUKANOVA ET AL. TABLE III RISK OF OVARIAN CANCER BY BMI CATEGORIES IN THE THREE COHORTS Cohort study Cases Controls Reference 2nd category 3rd category p for trend NYUWHS ( ) 0.50 ( ) 0.05 (cont. 0.04) ( ) 0.51 ( ) 0.06 (cont. 0.05) NSHDS ( ) 1.02 ( ) 0.97 (cont. 0.94) ( ) 0.87 ( ) 0.84 (cont. 0.73) ORDET ( ) 0.08 (cont. 0.21) ( ) 0.14 (cont. 0.27) 1 Conditional logistic regression model, additionally adjusted for parity, OC-use and smoking. height 0.02). We used age at cancer diagnosis of 55 as the cut-off point for a subgroup analysis as this approach allowed us to discriminate relatively young cases, occurring before or soon after the menopause from those who were older at cancer diagnosis and also provided a sufficiently large number of cases in each subgroup. The increase in ovarian cancer risk in the tertile with the greatest height as compared to the tertile with the lowest height was about 3-fold, but the confidence intervals included unity [2.91 ( )]. Further adjustments for age at menarche increased the point estimates and the association between height and ovarian cancer risk reached statistical significance [OR in the second and top tertiles of 2.08 ( ) and 4.36 ( ), p 0.03]. No association between height and ovarian cancer risk was found in the group of cases diagnosed after age 55 and their controls. The use of cohort-specific cut-off points did not alter the observed results. Mutual adjustments of BMI models for height and of height models for BMI, both as continuous variables or in categories, did not alter the point estimates. In the NYUWHS, a small percentage of the women were from African-American, Hispanic or other non-caucasian ethnic origin. Restricting the analysis only to subjects of Caucasian origin did not influence the results. Among Caucasian women, the association of BMI with ovarian cancer was similar in women who reported to belong to the Jewish religion and in those who did not, however, the numbers were very small. Physical activity, education or selected medical conditions (family history of breast cancer, diabetes and hypertension) did not show association with ovarian cancer risk and adjustment for these factors did not alter the estimated relationships of ovarian cancer risk with BMI or height. DISCUSSION In this multi-center, prospective cohort study ovarian cancer risk decreased with increasing BMI. The protective effect was most evident when the obese women (BMI 30) were compared to the lean women (BMI 23). The findings of our study are in contradiction with the results from the 5 other cohort studies reported in the literature, 3 of which report no association and 2 of which show a very slight increase in ovarian cancer with increasing BMI. 17,18 This study was part of a collaborative effort involving 3 prospective cohorts, in New York (USA), Umeå (Sweden) and Milan (Italy) on endogenous hormones and ovarian cancer risk. An advantage of the case-control design nested in a cohort is that case and control subjects originate from the same, well-defined source population, thereby minimizing the risk of selection biases. In addition, all anthropometric data were recorded at the time of recruitment in the study, which minimizes the inverse causation bias (i.e., changes in body weight resulting from the presence of disease). The persistence of the observed associations after exclusion of case subjects diagnosed within 3 years after recruitment into the study also makes it unlikely that case-control differences were caused by a latent, as yet undiagnosed tumor, that was sufficiently advanced to have effects on metabolism and body weight. Several potential confounders were controlled for in the analyses. For a proportion of the NSHDS subjects, the information on reproductive history was obtained from a retrospective questionnaire or from the medical records of the deceased case subjects. Generally, recall of previous full-term pregnancies has been shown to be reliable and consequently it is unlikely that we have introduced bias when adjusting for parity. 19 The reliability of selfreported age at menarche, age at menopause and use of exogenous hormones may decrease with increased elapsed time since the event occurred; 20 however, the association of these variables with ovarian cancer risk was homogeneous and of a similar magnitude across the 3 cohorts, which makes us confident that the adjustment for confounders was adequate. The observed inverse association of BMI with ovarian cancer risk in this study was driven by the NYUWHS and ORDET data and no association was found in the NSHDS. The mean age at cohort recruitment and the prevalence of established risk factors for ovarian cancer differed somewhat between the 3 cohorts, but there was no obvious explanation for the observed variation in the association of BMI with ovarian cancer risk amongst the 3 cohorts. In the NYUWHS, a substantial number of Jewish women had been recruited (36% of the participants in this study) and as Jewish women are more often than non-jewish women carriers of genes and specific mutations conferring increased susceptibility to development of ovarian cancer, 21 this could have influenced our results. However, restricting the analysis in the NYUWHS cohort to Caucasian women, with or without Jewish ancestry, or adding Jewish ancestry as a co-variate in the model did not provide an explanation for the observed differences of the BMI association with ovarian cancer between the cohorts. On average, NSHDS subjects were taller, but had similar BMI to those of the NYUWHS and ORDET subjects, suggesting that there were differences in the distribution of body mass among the subjects of the 3 cohorts. For given BMI, the NSHDS women appeared to have relatively less adipose tissue compared to NYUWHS and, especially, to ORDET women. However, analysis using cohort-specific cut-off points yielded results very similar to those obtained by using study-wide cut-off points. Some imprecision of the exposure measurement may have been introduced in the study by the self-reported height and weight for the NYUWHS and a portion of the NSHDS subjects. Self-reported height measurements have shown very strong correlation with measured height in the NSHDS, while self-reported weight is less accurate and under-reporting is especially confined to overweight and obese people. 22 Finally, the lack of association of BMI with ovarian cancer risk observed in the NSHDS might be a result of some other specific characteristics of this population, which has not been taken into account in this study. The combined results of this study are in contradiction with the weak direct or null findings of other prospective studies and we can not exclude the possibility that the observed protective effect of BMI on ovarian cancer is a chance finding. Several hypotheses about ovarian carcinogenesis have been proposed, implicating incessant ovulation, 23 gonadotropins 24,25 and steroid hormones 25,26 as key etiological factors. The association of BMI with ovarian cancer may be a result of an effect of BMI on these hormonal factors. The incessant ovulation hypothesis postulates that the risk of epithelial ovarian cancer increases with the number of ovulatory cycles, resulting in an increased exposure of the traumatized epi-

5 BMI AND OVARIAN CANCER 607 thelium of the ruptured follicles to mitotic stimuli and recurrent repair. 23 This hypothesis predicts that any factor that decreases the frequency of ovulation would be protective for ovarian cancer. The decrease in ovarian cancer risk with increasing BMI observed in our data is consistent with this hypothesis, because excessive body weight and obesity, have been associated with an increased number of anovulatory cycles, as well as with anovulation. 27,28 The relationship of obesity with ovarian cancer under the gonadotropin hypothesis is more complex. This hypothesis states that excessive gonadotropin stimulation, directly or indirectly through stimulation of the ovarian steroidogenic activity, results in increased proliferation and malignant transformation of the ovarian epithelium trapped within inclusion cysts formed through repeated invaginations of the epithelium during ovulation. 24,25 Obesity has been associated with decreased gonadotropin levels, especially after the menopause; hence the gonadotropin hypothesis will predict protective effect of increased body weight on ovarian cancer risk. However, obesity is also related to an increase in postmenopausal estrogen levels, which might have the opposite effect and increase the risk of ovarian cancer. Androgens have also been implicated in the development of ovarian cancer. 26,34 In several studies, it has been shown that obesity is associated with increased levels of free testosterone, especially in premenopausal women 30,35 37 and that a preponderance of abdominal fat (as expressed by higher WHR) is more strongly associated with the development of an androgenic hormone profile. 38,39 In the only prospective study on indices of body weight and ovarian cancer, in which adjustments for possible confounders were made, there was no association of BMI with ovarian cancer risk, yet there was more than a 2-fold increase in risk in the highest tertile of WHR. 14 In our study, WHR measurements were available only for a small number of subjects and we observed some increase in ovarian cancer risk with increasing categories of WHR after adjustment for BMI, but the confidence intervals were wide and included unity. Future studies with sufficiently large numbers of case and control subjects to study the effects of android versus gynecoid obesity on ovarian cancer risk would be of interest. There are very few epidemiological studies reporting on the association of height with ovarian cancer risk and the majority of these do not show any association, or only a slight increase, of risk in tall women. 40,41 In our data, there was no association of height with ovarian cancer risk, but there was an indication that the relationship of ovarian cancer with height varied according to age at diagnosis of the cases. In the analysis restricted to women with ovarian cancer diagnosis at a relatively early age ( 55 years), we found a tendency towards an increase in risk with greater height. If real, this finding would concur with similar associations that have been reported between height and breast cancer risk Adult height is determined by longitudinal bone growth rates during childhood and adolescence, which is under the complex control of various hormones, most notably growth hormone and insulin like-growth factor-i, but also androgens and estrogens. 45 A greater height may thus reflect comparatively elevated levels of these hormones during childhood and adolescence, although such an effect may decrease with age. In conclusion, the results of our study suggest that ovarian cancer may be inversely associated with BMI and directly associated with height in younger women. Despite the similarities in the ecological pattern of the incidence rates between ovarian, endometrial and breast cancer, pointing to an effect of western-lifestyle, their relationship with BMI appears to depend on the subtle hormonal mechanisms involved in the organ-specific carcinogenic process. For endometrial and postmenopausal breast cancer, there is a direct association with BMI, while for premenopausal breast and for ovarian cancer there seems to be a relative protective effect of increased body mass. The decrease in ovarian cancer risk with increasing BMI observed in this prospective study can be predicted on the basis of the incessant ovulation hypothesis, which is one of the leading hypotheses for ovarian cancer development. ACKNOWLEDGEMENTS Ms Å. Ãgren and Mr H. Sjodin helped with the management of the Swedish biobank database; Ms Y. Afanasyeva, Ms L. Quinones and Ms D. Masciangelo provided technical assistance in the NYU Women s Health Study and Ms B. Vozar and Mr D. Achaintre contributed to the laboratory analyses and Ms J. Dehedin in preparation of the manuscript. REFERENCES 1. Ferlay J, Parkin DM, Pisani P. GLOBOCAN1: cancer incidence and mortality worldwide. Lyon: IARC, Herrinton LJ, Stanford JL, Schwartz SM, Weiss NS. Ovarian cancer incidence among Asian migrants to the United States and their descendants. J Natl Cancer Inst 1994;86: Kaaks R, Lukanova A. Energy balance and cancer: the role of insulin and insulin-like growth factor-i. Proc Nutr Soc 2001;60: Key TJ, Allen NE, Verkaslo PK, Banks E. Energy balance and cancer: the role of sex hormones. Proc Nutr Soc 2001;60: Weight control and physical activity. 1. vol. 6. Lyon: IARC, Frost C, Coleman MP. Obesity and ovarian cancer. Eur J Cancer 1997;33: Grady D, Ernster VL. Endometrial Cancer. In: Schottenfeld D, Fraumeni FJ, eds. Cancer epidemiology and prevention. New York: Oxford University Press, Henderson BE, Pike MC, Bernstein L, Ross RK. Breast cancer. In: Schottenfeld D, Fraumeni FJ, eds. Cancer epidemiology and prevention. New York: Oxford University Press, Purdie DM, Bain CJ, Webb PM, Whiteman DC, Pirozzo S, Green AC. Body size and ovarian cancer: case-control study and systematic review (Australia). Cancer Causes Control 2001;12: Toniolo P, Bruning PF, Akhmedkhanov A, Bonfrer JM, Koenig KL, Lukanova A, Shore RE, Zeleniuch-Jacquotte A. Serum insulin-like growth factor-i and breast cancer. Int J Cancer 2000;88: Palmqvist R, Hallmans G, Rinaldi S, Biessy C, Stenling R, Riboli E, Kaaks R. Plasma IGF-I, IGF-binding protein-3 and risk of colorectal cancer: a prospective study in northern Sweden. Gut, in press. 12. Muti P, Bradlow HL, Micheli A, Krogh V, Freudenheim JL, Schunemann HJ, Stanulla M, Yang Y, Sepkovic DW, Trevisanm M, Berrino, F. Estrogen metabolism and risk of breast cancer: a prospective study of the 2:16alpha-hydroxyestrone ratio in premenopausal and postmenopausal women. Epidemiology 2000;11: SAS Institute. SAS/STATR user s guide version 6. SAS Manual. Cary, NC: SAS Institute, Mink PJ, Folsom AR, Sellers TA, Kushi LH. Physical activity, waist-to-hip ratio and other risk factors for ovarian cancer: a follow-up study of older women. Epidemiology 1996;7: Moller H, Mellemgaard A, Lindvig K, Olsen JH. Obesity and cancer risk: a Danish record-linkage study. Eur J Cancer 1994;30A: Tornberg SA, Carstensen JM. Relationship between Quetelet s index and cancer of breast and female genital tract in 47,000 women followed for 25 years. Br J Cancer 1994;69: Wolk A, Gridley G, Svensson M, Nyren O, McLaughlin JK, Fraumeni JF, Adam HO. A prospective study of obesity and cancer risk (Sweden). Cancer Causes Control 2001;12: Lew EA, Garfinkel L. Variations in mortality by weight among 750,000 men and women. J Chronic Dis 1979;32: Tomeo CA, Rich-Edwards JW, Michels KB, Berkey CS, Hunter DJ, Frazier AL, Willett WC, Buka SL. Reproducibility and validity of maternal recall of pregnancy-related events. Epidemiology 1999;10: den T. I. Validity and reproducibility of self-reported age at menopause in women participating in the DOM-project. Maturitas 1997; 27: Steinberg KK, Pernarelli JM, Marcus M, Khoury MJ, Schildkraut JM, Marchbanks PA. Increased risk for familial ovarian cancer among Jewish women: a population-based case-control study. Genet Epidemiol 1998;15: Paccaud F, Wietlisbach V, Rickenbach M. Body mass index: comparing mean values and prevalence rates from telephone and examination surveys. Rev Epidemiol Sante Publique 2001;49:33 40.

6 608 LUKANOVA ET AL. 23. Fathalla MF. Incessant ovulation: a factor in ovarian neoplasia? Lancet 1971;2: Gardner W. Hormonal imbalances in tumorigenesis. Cancer Res 1948;8: Cramer DW, Welch WR. Determinants of ovarian cancer risk. II. Inferences regarding pathogenesis. J Natl Cancer Inst 1983;71: Helzlsouer KJ, Alberg AJ, Gordon GB, Longcope C, Bush TL, Hoffman SC, Comstock GW. Serum gonadotropins and steroid hormones and the development of ovarian cancer. JAMA 1995;274: Glass AR, Burman KD, Dahms WT, Boehm TM. Endocrine function in human obesity. Metabolism 1981;30: Shoupe D, Brenner PE, Mishell DR Jr. Effects of nutrition, stress and exercise on reproductive function. In: Lobo RA, Mishell Jr DR, Paulson RJ, Shoupe D, eds. Infertility, contraception and reproductive endocrinology. London: Blackwell Science, Bottiglioni F, de Aloysio D, Nicoletti G, Mauloni M, Mantuano M, Motta R. Physiopathological aspects of body overweight in the female climacteric. Maturitas 1984;5: Kaye SA, Folsom AR, Soler JT, Prineas RJ, Potter JD. Associations of body mass and fat distribution with sex hormone concentrations in postmenopausal women. Int J Epidemiol 1991;20: Malacara JM, Fajardo ME, Nava LE. Gonadotropins at menopause: the influence of obesity, insulin resistance and estrogens. Steroids 2001;66: Bohlke K, Cramer DW, Barbieri RL. Relation of luteinizing hormone levels to body mass index in premenopausal women. Fertil Steril 1998;69: Sheu WH, Lee WJ. Abnormal gonadotropin release and carbohydrate metabolism in morbid obese women. Chin J Physiol 1997;40: Risch HA. Hormonal etiology of epithelial ovarian cancer, with a hypothesis concerning the role of androgens and progesterone. J Natl Cancer Inst 1998;90: Evans DJ, Hoffmann RG, Kalkhoff RK, Kissebah AH. Relationship of androgenic activity to body fat topography, fat cell morphology and metabolic aberrations in premenopausal women. J Clin Endocrinol Metab 1983;57: Wajchenberg BL, Marcondes JA, Mathor MB, Achando SS, Germak OA, Kirschner MA. Free testosterone levels during the menstrual cycle in obese versus normal women. Fertil Steril 1989;51: Zeleniuch-Jacquotte A, Bruning PF, Bonfrer JM, Koenig KL, Shore RE, Kim MY, Pasternack BS, Toniolo P. Relation of serum levels of testosterone and dehydroepiandrosterone sulfate to risk of breast cancer in postmenopausal women. Am J Epidemiol 1997;145: Wabitsch M, Hauner H, Heinze E, Bockmann A, Benz R, Mayer H, Teller W. Body fat distribution and steroid hormone concentrations in obese adolescent girls before and after weight reduction. J Clin Endocrinol Metab 1995;80: Hauner H, Ditschuneit HH, Pal SB, Moncayo R, Pfeiffer EF. Fat distribution, endocrine and metabolic profile in obese women with and without hirsutism. Metabolism 1988;37: Weiss NS, Cook LS, Farrow DC, Rosenblatt KA. Ovarian cancer. In: Schottenfeld D, Fraumeni FJ, eds. Cancer Epidemiology and Prevention. New York: Oxford University Press, 1996: Chen Y, Wu PC, Lang JH, Ge WJ, Hartge P, Brinton LA. Risk factors for epithelial ovarian cancer in Beijing, China. Int J Epidemiol 1992; 21: van den Brandt PA, Dirx MJ, Ronckers CM, van den Hoogan P, Goldbohm RA. Height, weight change and postmenopausal breast cancer risk: The Netherlands Cohort Study. Cancer Causes Control 1997;8: van den Brandt PA, Spiegelman D, Yaun SS, Adami HO, Beeson L, Folsom AR, Fraser G, Goldbohm RA, Graham S, Kushi L, Marshall JR, Miller AB. Pooled analysis of prospective cohort studies on height, weight and breast cancer risk. Am J Epidemiol 2000;152: Palmer JR, Rao RS, Adams-Campbell LL, Rosenberg L. Height and breast cancer risk: results from the Black Women s Health Study (United States). Cancer Causes Control 2001;12: Grumbach MM, Styne DM. Puberty: ontogeny, neuroendocrinology, physiology and disorders. In: Wilson JD, Foster DW, eds. Williams textbook of endocrinology. Philadelphia: W.B. Saunders Company,