Int. J . Cancer: 45, 44W44 (1990) 0 1990 Wiley-Liss, Inc.
Publication of the International Union Against Cancer Publication de I’Union lnternationale Contre le Cancer
BODY MASS INDEX AND RISK OF BREAST CANCER. A PROSPECTIVE STUDY OF 23,826 NORWEGIAN WOMEN Lars J.VATTEN’s2*3*4and Stener KVINNSLAND’ lDepartment of Oncology, University Hospital, N-7006 Trondheim;2The Norwegian Cancer Registry, Montebello, N-0310 Oslo 3; in collaboration with 3The National Health Screening Service, P.O. Box 8155 Dep., Oslo I , Norway. The association between body mass index (BMI) and the incidence rate of breast cancer has been examined in 236 cases of breast cancer that developed among 23,826 Norwegian women during I I to I 4 years of follow-up. A t the time of height and weight measurement they were 35 to 51 years of age, and at the end of follow-up their age was between 46 and 63 years. There was an overall age-adjusted incidence rate ratio (IRR) of 0.52 (95% confidence limits, 0.34 and 0.77) for women in the highest quartile of BMI compared to women in the lowest quartile, which was confined to an effect observed among women who were diagnosed at age 50 or earlier (IRR = 0.36). The association with BMI displayed an inverse dose-related trend (x2for trend = 14.22, p < 0.001). The negative trend was particularly pronounced among non-smoking women = 14.63), and no clear trend associated with BMI was observed among women who smoked 10 or more cigarettes per day = 0.41), indicating an interaction between BMI and cigarette smoking interaction = 3.86, p = 0.05). We thus suggest that there is a negative association between body mass index and risk of breast cancer among premenopausal women.
(x2
(x2
(x’
The association between body mass and breast cancer may exhibit a seemingly paradoxical relation. Whereas obesity increases the risk of breast cancer in post-menopausal women (Henderson et al., 1984; Boyle and Leake, 1988), the opposite may be m e for breast cancer occurring in pre-menopausal women (Willett et al., 1985; Lubin et al., 1985; Le Marchand et al., 1988; Kampert et al., 1988; Tretli, 1989). The increased risk associated with obesity after menopause has been attributed to an effect exerted by oestrogens aromatized from androgenic prohormones in adipose tissue (Grodin et a l . , 1973; Siiteri, 1987). It has also been interpreted as a result of an energy-rich diet throughout the woman’s life (Williams and Dickerson, 1987; deWaard and Trichopolous, 1988; Miller, 1986). The age-incidence curve (Pike, 1987) of breast cancer suggests that the ovaries play an important role in determining the incidence of disease, particularly in pre-menopausal women. The protective effect of increased body mass observed in women before menopause, however, may be difficult to reconcile with the present understanding of breast cancer pathogenesis, and cannot easily be given a plausible biological interpretation. One hypothesis, which may comply with breast cancer epidemiology, is that a greater proportion of anovulatory menstrual cycles among obese pre-menopausal women may decrease their risk of developing breast cancer (Key and Pike, 1988). In this prospective study we present epidemiological evidence which supports a negative association between body mass and the risk of breast cancer among pre-menopausal women.
were invited, of whom 24,617 (93.8%) attended screening. The examination included a questionnaire, standardized measurements of height and weight, and a non-fasting blood sample. These procedures have previously been described in detail by Bjartveit et al. (1979; 1983). Measurements of height and weight were performed in conjunction with an X-ray procedure, from which pregnant women were exempted. Thus, height and weight were not registered for 504 participants, and these were excluded from the analysis. To reduce a potential bias due to weight reduction in cancer cases, women who were diagnosed with a malignancy (including breast cancer) prior to or during the calendar year of examination were excluded. These amounted to 287, thus leaving a total number of 23,826 women eligible for analysis. The questionnaire was primarily designed to obtain information on known and suspected risk factors for cardiovascular disease. Hence, there was a lack of information on factors known to predict breast cancer risk, such as age at menarche, age at first full-term pregnancy, family history of breast cancer, and reliable information on exact age at menopause. However, the questionnaire did include detailed history of past and current smoking habits, and various demographic variables. For each participant, standardized measurements of height (in cm) and weight (in kg) were performed. Based on these values, Davenport’s index of body mass (Keys et al., 1972) was computed as weight in grams divided by the squared value of height in centimeters, rendering values that are one-tenth of the more often used Quetelet’s Index. Table I displays the distribution of weight and body mass index (BMI) in the study population according to age at examination and related to disease status at the end of follow-up. Each participant had an official 11-digit personal number which facilitated linkage to the Cancer Registry of Norway. This allowed for identification of every incident case of breast cancer that occurred in the cohort from the time of examination until the end of follow-up (October 1 , 1988). During 11 to 14 years (mean = 11.9) of follow-up, a total of 236 incident cases of breast cancer were diagnosed. Among these, 137 cases had occurred in women under 51 years of age, and 99 in women aged 5 1 or older. The age of 5 1 as a dividing line for allocating breast cancer to ‘‘pre- or post-menopausal’’ groups was arbitrarily chosen. It can only serve as a rough separation between the 2 types. BMI was categorized into quartiles based on individual values for the complete study population. For each person belonging to a certain quartile of BMI, observation years at risk of developing breast cancer were computed as the number of years accumulated from entry into the study until withdrawal in the year of diagnosis, death from another cause, or the end of follow-up. Observation years at risk of developing breast cancer before age 51 were analogously censored, but follow-up
SUBJECTS AND METHODS
From 1974 to 1977 all men and women aged 35 to 51 years living in 3 separate counties in Norway were invited to participate in a health screening examination organized by the National Health Screening Service. A total of 26,252 women
4 T whom ~ reprint requests should be sent, at the Department of Oncology, University Hospital, N-7006 Trondheim, Norway. ~
Received: October 23, 1989 and in revised form December 14, 1989.
441
BODY MASS AND BREAST CANCER RISK
TABLE I - MEAN BODY MASS INDEX AND MEAN WEIGHT IN THE STUDY COHORT. ACCORDWG TO AGE AT HEALTH SCREENING, AND IN BREAST CANCER CASES DIAGNOSED BEFORE AND AFTER TH6 AGE OF 51 M W
BMI k!cm2)
SEM
7.494 7,550 8,782 23,826 137
2.38 2.48 2.55 2.48 2.33
0.004
99
2.53
M a might
SEM
(ks)
~~
35-39 40-44 45-5 1 Total
Breast cancer diagnosed before age 51 Breast cancer diagnosed at age 51 or
0.003 0.03
63.6 65.4 66.7 65.3 62.9
0.13 0.12 0.07 0.75
0.05
66.7
1.20
0.005 0.005
0.12
later
ended when the person reached 51 years. For breast cancer diagnosed at age 51 or later, years at risk were computed from the age of 51 until withdrawal. This procedure allowed comparison of person-time-based incidence rates of breast cancer for each quartile of body mass index, giving overall incidence estimates, and distinguishing risk associated with BMI according to whether diagnosis was made before or after age 51. Incidence rate ratios (IRR) were computed as the rate in a specific quartile of BMI divided by the estimated rate in the lowest (reference) quartile. The precision of the IRR estimates was assessed by 95% confidence limits using the Miettinen test-based method applying Mantel-Haenszel Chi-square statistics (Kleinbaum et al., 1982). The effect of age was adjusted using the direct method for 5-year age categories of person/ years (Rothman, 1986). Assessing the presence of interaction between BMI and cigarette smoking was tested by fitting the (cumulative incidence) data to a multiple logistic model (Kleinbaum et ul., 1982). The maximum likelihood statistic of a model containing a product term between BMI (dichotomized at population median) and cigarette smoking (3 categories) was compared to a model omitting this interaction term, but keeping smoking status as a co-variate in the model.
and indicates that the relation between BMI and risk of b t cancer among women aged 50 or less may be modified by whether they smoked cigarettes or not. To increase the likelihood that a predominant proportion of women in each category of smoking was truly pre-menopausal, we tested the interaction with cigarette smoking only including women who were at risk of developing breast cancer diagnosed at age 48 or younger (data not shown), which c o n f i i e d the interaction (xz = 4.10, p < 0.05). We also examined the effect of cigarette smoking evaluated as the number of years a person had smoked (data not shown). The risk pattern in never-smokers was identical to that detected among non-smokers, and a woman who reported smoking for less than 15 years had a moderate, but not statistically significant decrease in her risk of breast cancer with increasing BMI. In women who had smoked for 15 years or more, no association between BMI and breast cancer was detected. DISCUSSION
Among women in this cohort the overall inverse relation between body mass index and risk of developing breast cancer was confined to women diagnosed before the age of 51. A moderately obese woman (e.g.,167 cm tall and weighing 75 kg) would have a 60%reduced breast cancer risk before age 5 1 compared to a woman of equal height who weighed 57 kg. The inverse relation with BMI displayed a statistically significant linear trend, and the precision of the estimated rate ratios for each quartile of BMI did not indicate that the observed inverse relation can be attributed to chance. There is evidence for an inverse relation between body mass and breast cancer risk among pre-menopausal women (Willett et al.. 1985; Le Marchand et al.. 1988; Tgimberg et al., 1988), although some studies have reported negative results (Adami er al., 1977; Ewertz, 1988). In this study, 23,826 women were followed-up for 11 and 14 years (mean = 11.9 years) constituting more than 280,000 persodyears of observation, Standardized measurements of body height and weight were performed, and incident cases of breast cancer were ascertained through the Norwegian Cancer Registry (Lund, 1981). In certain other studies anthropometric measurements were based on the participant’s own reporting (Willett et ul., 1985; Ewertz, 1988). This may have resulted in RESULTS non-differentid misclassification of height and weight, which The overall age-adjusted incidence rate ratios (IRR)of breast could have led to an underestimate of the relative risk associcancer (Table II) show that there was a decreasing risk of breast ated with BMI. A higher accuracy of BMI measurements may cancer with increasing body mass index. Women in the highest explain why the inverse relation with body mass was stronger quartile of BMI (mean = 3.04 g/cm2) had an IRR of 0.52 in this study compared to other published reports. The lack of information on known risk factors for breast (95% CI, 0.34 and 0.77) relative to women in the lowest quartile (mean = 2.05 g/cm2). This effect was largely con- cancer (Hanis and Henderson, 1987; Verreault et al., 1989) is fined to breast cancer diagnosed among women under 51 a limitation of this study. To confound the result, a factor (IRR = 0.36,95% CI, 0.20 and 0.65), and this relation with should have an independent effect on the risk of disease in the BMI displayed an inverse linear trend (x2 trend = 14.22, absence of the exposure under study, and simultaneously be p < 0.001). Among women diagnosed at age 51 or later there associated with exposure in the data (Rothman, 1986). Consewas no clear relation with BMI (x2trend = 0.41, p = 0.52). quently, body mass index should be related to variables such as We explored whether cigarette smoking may modify the age at menarche, age at first full-term pregnancy and age at negative association with BMI by examining the effect of BMI menopause for confounding from these factors to be anticiover 3 categories of smoking status (Table 111). Among non- pated in the data. We cannot exclude a possible confounding smokers the age-adjusted IRR was 0.20 (95% confidence lim- with any one of these factors, but prior research may justify a its, 0.08 and 0.50) comparing women in the highest and lowest brief discussion of the relation between BMI and age at menquartile of BMI, and the negative association with BMI arche. A large study of Norwegian school-girls showed that early showed a strong dose-related gradient (xz trend = 14.63, p < 0,001). Although there was a negative association with maturers were heavy for their age at the onset of menarche, and BMI among moderate smokers (1-9 cigarettes per day), and those who matured later than the average girl were light for among women who smoked 10 or more cigarettes per day, the their age at this time (Brundtland et al., 1980; Liestbl, 1980), relations were weak, and did not display linear trends. The while evidence indicates that overweight in early maturers presence of a statistical interaction between BMI and cigarette tends to persist into adulthood (Garn el d.,1986; Sherman ef smoking was significant (xz interaction = 3.86, p = 0.05), af.. 1981). If this relation with BMI applies to women in this
442
VATTEN A N D K V l N N S L A N D
TABLE U - INCIDENCE RATE RATIO (IRR) OF BREAST CANCER ACCORDING TO QUARTILES OF BODY MASS INDEX (BMI) FOR (A) ALL CASES, (8) CASES DIAGNOSED BEFORE THE AGE OF 51, AND (C) FOR CASES DIAGNOSED AT AGE 51 OR LATER’ Body mass index (in g/cm2)
Age (at measurement)
12.20 mean: 2.05
2.20-2.39 2.29
A. All cases 35-39
Cases Person/years
4
2.40-2.67 2.52
a2.68 3.04
31 29,748
15 24,983
16 20,262
5 15,898
25 21,112
32 23,139
12 22,913
9 22,579
17 18,472
16 23,005
30 27,823
28 32,247
Chi-square trend
u
Cases Person/years
45-5 1
Cases Person/ years Total 35-51 Cases Person/years Crude IRR Age-adjusted IRR 95% confidence limits B. Cases
Publication of the International Union Against Cancer Publication de I’Union lnternationale Contre le Cancer
BODY MASS INDEX AND RISK OF BREAST CANCER. A PROSPECTIVE STUDY OF 23,826 NORWEGIAN WOMEN Lars J.VATTEN’s2*3*4and Stener KVINNSLAND’ lDepartment of Oncology, University Hospital, N-7006 Trondheim;2The Norwegian Cancer Registry, Montebello, N-0310 Oslo 3; in collaboration with 3The National Health Screening Service, P.O. Box 8155 Dep., Oslo I , Norway. The association between body mass index (BMI) and the incidence rate of breast cancer has been examined in 236 cases of breast cancer that developed among 23,826 Norwegian women during I I to I 4 years of follow-up. A t the time of height and weight measurement they were 35 to 51 years of age, and at the end of follow-up their age was between 46 and 63 years. There was an overall age-adjusted incidence rate ratio (IRR) of 0.52 (95% confidence limits, 0.34 and 0.77) for women in the highest quartile of BMI compared to women in the lowest quartile, which was confined to an effect observed among women who were diagnosed at age 50 or earlier (IRR = 0.36). The association with BMI displayed an inverse dose-related trend (x2for trend = 14.22, p < 0.001). The negative trend was particularly pronounced among non-smoking women = 14.63), and no clear trend associated with BMI was observed among women who smoked 10 or more cigarettes per day = 0.41), indicating an interaction between BMI and cigarette smoking interaction = 3.86, p = 0.05). We thus suggest that there is a negative association between body mass index and risk of breast cancer among premenopausal women.
(x2
(x2
(x’
The association between body mass and breast cancer may exhibit a seemingly paradoxical relation. Whereas obesity increases the risk of breast cancer in post-menopausal women (Henderson et al., 1984; Boyle and Leake, 1988), the opposite may be m e for breast cancer occurring in pre-menopausal women (Willett et al., 1985; Lubin et al., 1985; Le Marchand et al., 1988; Kampert et al., 1988; Tretli, 1989). The increased risk associated with obesity after menopause has been attributed to an effect exerted by oestrogens aromatized from androgenic prohormones in adipose tissue (Grodin et a l . , 1973; Siiteri, 1987). It has also been interpreted as a result of an energy-rich diet throughout the woman’s life (Williams and Dickerson, 1987; deWaard and Trichopolous, 1988; Miller, 1986). The age-incidence curve (Pike, 1987) of breast cancer suggests that the ovaries play an important role in determining the incidence of disease, particularly in pre-menopausal women. The protective effect of increased body mass observed in women before menopause, however, may be difficult to reconcile with the present understanding of breast cancer pathogenesis, and cannot easily be given a plausible biological interpretation. One hypothesis, which may comply with breast cancer epidemiology, is that a greater proportion of anovulatory menstrual cycles among obese pre-menopausal women may decrease their risk of developing breast cancer (Key and Pike, 1988). In this prospective study we present epidemiological evidence which supports a negative association between body mass and the risk of breast cancer among pre-menopausal women.
were invited, of whom 24,617 (93.8%) attended screening. The examination included a questionnaire, standardized measurements of height and weight, and a non-fasting blood sample. These procedures have previously been described in detail by Bjartveit et al. (1979; 1983). Measurements of height and weight were performed in conjunction with an X-ray procedure, from which pregnant women were exempted. Thus, height and weight were not registered for 504 participants, and these were excluded from the analysis. To reduce a potential bias due to weight reduction in cancer cases, women who were diagnosed with a malignancy (including breast cancer) prior to or during the calendar year of examination were excluded. These amounted to 287, thus leaving a total number of 23,826 women eligible for analysis. The questionnaire was primarily designed to obtain information on known and suspected risk factors for cardiovascular disease. Hence, there was a lack of information on factors known to predict breast cancer risk, such as age at menarche, age at first full-term pregnancy, family history of breast cancer, and reliable information on exact age at menopause. However, the questionnaire did include detailed history of past and current smoking habits, and various demographic variables. For each participant, standardized measurements of height (in cm) and weight (in kg) were performed. Based on these values, Davenport’s index of body mass (Keys et al., 1972) was computed as weight in grams divided by the squared value of height in centimeters, rendering values that are one-tenth of the more often used Quetelet’s Index. Table I displays the distribution of weight and body mass index (BMI) in the study population according to age at examination and related to disease status at the end of follow-up. Each participant had an official 11-digit personal number which facilitated linkage to the Cancer Registry of Norway. This allowed for identification of every incident case of breast cancer that occurred in the cohort from the time of examination until the end of follow-up (October 1 , 1988). During 11 to 14 years (mean = 11.9) of follow-up, a total of 236 incident cases of breast cancer were diagnosed. Among these, 137 cases had occurred in women under 51 years of age, and 99 in women aged 5 1 or older. The age of 5 1 as a dividing line for allocating breast cancer to ‘‘pre- or post-menopausal’’ groups was arbitrarily chosen. It can only serve as a rough separation between the 2 types. BMI was categorized into quartiles based on individual values for the complete study population. For each person belonging to a certain quartile of BMI, observation years at risk of developing breast cancer were computed as the number of years accumulated from entry into the study until withdrawal in the year of diagnosis, death from another cause, or the end of follow-up. Observation years at risk of developing breast cancer before age 51 were analogously censored, but follow-up
SUBJECTS AND METHODS
From 1974 to 1977 all men and women aged 35 to 51 years living in 3 separate counties in Norway were invited to participate in a health screening examination organized by the National Health Screening Service. A total of 26,252 women
4 T whom ~ reprint requests should be sent, at the Department of Oncology, University Hospital, N-7006 Trondheim, Norway. ~
Received: October 23, 1989 and in revised form December 14, 1989.
441
BODY MASS AND BREAST CANCER RISK
TABLE I - MEAN BODY MASS INDEX AND MEAN WEIGHT IN THE STUDY COHORT. ACCORDWG TO AGE AT HEALTH SCREENING, AND IN BREAST CANCER CASES DIAGNOSED BEFORE AND AFTER TH6 AGE OF 51 M W
BMI k!cm2)
SEM
7.494 7,550 8,782 23,826 137
2.38 2.48 2.55 2.48 2.33
0.004
99
2.53
M a might
SEM
(ks)
~~
35-39 40-44 45-5 1 Total
Breast cancer diagnosed before age 51 Breast cancer diagnosed at age 51 or
0.003 0.03
63.6 65.4 66.7 65.3 62.9
0.13 0.12 0.07 0.75
0.05
66.7
1.20
0.005 0.005
0.12
later
ended when the person reached 51 years. For breast cancer diagnosed at age 51 or later, years at risk were computed from the age of 51 until withdrawal. This procedure allowed comparison of person-time-based incidence rates of breast cancer for each quartile of body mass index, giving overall incidence estimates, and distinguishing risk associated with BMI according to whether diagnosis was made before or after age 51. Incidence rate ratios (IRR) were computed as the rate in a specific quartile of BMI divided by the estimated rate in the lowest (reference) quartile. The precision of the IRR estimates was assessed by 95% confidence limits using the Miettinen test-based method applying Mantel-Haenszel Chi-square statistics (Kleinbaum et al., 1982). The effect of age was adjusted using the direct method for 5-year age categories of person/ years (Rothman, 1986). Assessing the presence of interaction between BMI and cigarette smoking was tested by fitting the (cumulative incidence) data to a multiple logistic model (Kleinbaum et ul., 1982). The maximum likelihood statistic of a model containing a product term between BMI (dichotomized at population median) and cigarette smoking (3 categories) was compared to a model omitting this interaction term, but keeping smoking status as a co-variate in the model.
and indicates that the relation between BMI and risk of b t cancer among women aged 50 or less may be modified by whether they smoked cigarettes or not. To increase the likelihood that a predominant proportion of women in each category of smoking was truly pre-menopausal, we tested the interaction with cigarette smoking only including women who were at risk of developing breast cancer diagnosed at age 48 or younger (data not shown), which c o n f i i e d the interaction (xz = 4.10, p < 0.05). We also examined the effect of cigarette smoking evaluated as the number of years a person had smoked (data not shown). The risk pattern in never-smokers was identical to that detected among non-smokers, and a woman who reported smoking for less than 15 years had a moderate, but not statistically significant decrease in her risk of breast cancer with increasing BMI. In women who had smoked for 15 years or more, no association between BMI and breast cancer was detected. DISCUSSION
Among women in this cohort the overall inverse relation between body mass index and risk of developing breast cancer was confined to women diagnosed before the age of 51. A moderately obese woman (e.g.,167 cm tall and weighing 75 kg) would have a 60%reduced breast cancer risk before age 5 1 compared to a woman of equal height who weighed 57 kg. The inverse relation with BMI displayed a statistically significant linear trend, and the precision of the estimated rate ratios for each quartile of BMI did not indicate that the observed inverse relation can be attributed to chance. There is evidence for an inverse relation between body mass and breast cancer risk among pre-menopausal women (Willett et al.. 1985; Le Marchand et al.. 1988; Tgimberg et al., 1988), although some studies have reported negative results (Adami er al., 1977; Ewertz, 1988). In this study, 23,826 women were followed-up for 11 and 14 years (mean = 11.9 years) constituting more than 280,000 persodyears of observation, Standardized measurements of body height and weight were performed, and incident cases of breast cancer were ascertained through the Norwegian Cancer Registry (Lund, 1981). In certain other studies anthropometric measurements were based on the participant’s own reporting (Willett et ul., 1985; Ewertz, 1988). This may have resulted in RESULTS non-differentid misclassification of height and weight, which The overall age-adjusted incidence rate ratios (IRR)of breast could have led to an underestimate of the relative risk associcancer (Table II) show that there was a decreasing risk of breast ated with BMI. A higher accuracy of BMI measurements may cancer with increasing body mass index. Women in the highest explain why the inverse relation with body mass was stronger quartile of BMI (mean = 3.04 g/cm2) had an IRR of 0.52 in this study compared to other published reports. The lack of information on known risk factors for breast (95% CI, 0.34 and 0.77) relative to women in the lowest quartile (mean = 2.05 g/cm2). This effect was largely con- cancer (Hanis and Henderson, 1987; Verreault et al., 1989) is fined to breast cancer diagnosed among women under 51 a limitation of this study. To confound the result, a factor (IRR = 0.36,95% CI, 0.20 and 0.65), and this relation with should have an independent effect on the risk of disease in the BMI displayed an inverse linear trend (x2 trend = 14.22, absence of the exposure under study, and simultaneously be p < 0.001). Among women diagnosed at age 51 or later there associated with exposure in the data (Rothman, 1986). Consewas no clear relation with BMI (x2trend = 0.41, p = 0.52). quently, body mass index should be related to variables such as We explored whether cigarette smoking may modify the age at menarche, age at first full-term pregnancy and age at negative association with BMI by examining the effect of BMI menopause for confounding from these factors to be anticiover 3 categories of smoking status (Table 111). Among non- pated in the data. We cannot exclude a possible confounding smokers the age-adjusted IRR was 0.20 (95% confidence lim- with any one of these factors, but prior research may justify a its, 0.08 and 0.50) comparing women in the highest and lowest brief discussion of the relation between BMI and age at menquartile of BMI, and the negative association with BMI arche. A large study of Norwegian school-girls showed that early showed a strong dose-related gradient (xz trend = 14.63, p < 0,001). Although there was a negative association with maturers were heavy for their age at the onset of menarche, and BMI among moderate smokers (1-9 cigarettes per day), and those who matured later than the average girl were light for among women who smoked 10 or more cigarettes per day, the their age at this time (Brundtland et al., 1980; Liestbl, 1980), relations were weak, and did not display linear trends. The while evidence indicates that overweight in early maturers presence of a statistical interaction between BMI and cigarette tends to persist into adulthood (Garn el d.,1986; Sherman ef smoking was significant (xz interaction = 3.86, p = 0.05), af.. 1981). If this relation with BMI applies to women in this
442
VATTEN A N D K V l N N S L A N D
TABLE U - INCIDENCE RATE RATIO (IRR) OF BREAST CANCER ACCORDING TO QUARTILES OF BODY MASS INDEX (BMI) FOR (A) ALL CASES, (8) CASES DIAGNOSED BEFORE THE AGE OF 51, AND (C) FOR CASES DIAGNOSED AT AGE 51 OR LATER’ Body mass index (in g/cm2)
Age (at measurement)
12.20 mean: 2.05
2.20-2.39 2.29
A. All cases 35-39
Cases Person/years
4
2.40-2.67 2.52
a2.68 3.04
31 29,748
15 24,983
16 20,262
5 15,898
25 21,112
32 23,139
12 22,913
9 22,579
17 18,472
16 23,005
30 27,823
28 32,247
Chi-square trend
u
Cases Person/years
45-5 1
Cases Person/ years Total 35-51 Cases Person/years Crude IRR Age-adjusted IRR 95% confidence limits B. Cases
