American Journal of Epidemiology Copyright © 1991 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved
Vol. 134, No. 6 Printed in U S.A
Nutritional Epidemiology of Postmenopausal Breast Cancer in Western New York Saxon Graham,1 Rosemary Hellmann,1 James Marshall,1 Jo Freudenheim,1 John Vena,1 Mya Swanson,1 Maria Zielezny,1 Takuma Nemoto,2 Nancy Stubbe,3 and Teresa Raimondo1
The authors studied 439 postmenopausal breast cancer cases, identified in hospitals throughout western New York, with an interview schedule that considered frequency and amount ingested of 172 foods and provided data for an estimate of total calories ingested. These were compared with age-matched controls comprising a random sample of the same communities as the cases. The extensive interviews, requiring 2.0 hours on average to administer, also covered alcohol ingestion, Quetelet index, and a wide variety of reproductive factors. The authors found, as have most investigators over the past 25 years, that risk increased with increases in age at first pregnancy, decreased with increases in numbers of children and pregnancies, and increased in those with history of benign breast disease and in those with female relatives previously affected with breast cancer. Risk adjusted for potential confounders was highest among women with the lowest ingestion of carotene or a substance correlated with its ingestion. Risk was not associated with retinol ingestion. It increased with increases in Quetelet index. Fat intake, whether studied in terms of quantity or the proportion of total calories derived from fat, was not associated with risk of breast cancer. Our analyses of these factors were adjusted for age, education, and the reproductive history traits described above. Am J Epidemiol 1991 ;134:552-66. breast neoplasms; nutrition
index, alcohol ingestion, and various nutrients, such as carotene and fats (5-9). The present inquiry took place in western New York. It echoed the usual findings on reproductive and familial history and discovered a substantial increase in risk with increases in body mass and a higher risk in women with low ingestion of carotene or factors associated with it.
As one of the most frequently occurring tumors in women, breast cancer has been the focus of considerable epidemiologic inquiry in the last two decades. Substantial variations in risk of breast cancer have been found to occur according to characteristics of reproductive history and familial aggregation (1-4). In the field of diet and nutrition, various investigators have discovered variations in risk associated with Quetelet
MATERIALS AND METHODS
Received for publication November 28, 1990, and in final form April 5, 1991. 1 Department of Social and Preventive Medicine, State University of New York at Buffalo, Buffalo, NY. 2 Breast Care Center, Sisters Hospital, Buffalo, NY. 3 Department of Surgery, School of Medicine and Biomedical Sciences, State University of New York at Buffalo, Buffalo, NY. Reprint requests to Dr. Saxon Graham, Department of Social and Preventive Medicine, State University of New York at Buffalo, 2211 Main Street, Buffalo, NY 14214. This publication was made possible by grant CA11535 from the National Cancer Institute, National Institutes of Health, and grant PDT-434 from the American Cancer Society.
The study findings were based upon 439 histologically confirmed, incident cases of postmenopausal cancer of the breast, aged 41-85 years, residing in Erie and Niagara counties (the counties containing Buffalo, Lackawanna, Tonawanda, and Niagara Falls and their rural hinterlands). Subjects were considered postmenopausal if they were 50 years of age and over, had ceased menstruation, or were less than 50 years of age 552
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
Dietary Epidemiology of Breast Cancer with natural menopause or nonfunctional ovaries. Thus, females under age 50 were considered postmenopausal if they had experienced natural menopause, bilateral oophorectomy, or irradiation to the ovaries. Women 50 years of age and over were considered postmenopausal if they no longer were menstruating. Controls were 494 randomly chosen postmenopausal residents of the two counties and frequency matched on age group with cases; those under age 65 were selected from rolls of New York State motor vehicle driver's license holders (90.2 percent of the female population under age 65 in western New York are licensed) (10) and those 65 years and over from the Health Care Finance Administration rolls. Our nurse-interviewers maintained continuous contact with all of the hospitals in the two counties to ascertain as soon as possible when a case was admitted. The interviewers then contacted patients' physicians for permission to ask patients to participate in the inquiry. Over the years of the study, from 1986 to 1989, the interviewers identified 777 eligible patients from hospitals, and for these patients, their physicians gave us permission to interview 561 (72.2 percent). Of those patients permitted by their doctors to participate in the study, 100 (17.8 percent) refused and 22 (3.9 percent) were too ill to be interviewed. Thus, the proportion providing usable interviews (439 patients) of those determined eligible (777 patients) was 56.5 percent. Once selected, controls were sent letters requesting their participation. The interviewer then contacted them by telephone to make appointments for the interview. Of those 1,076 individuals randomly selected to participate as controls, 494 (45.9 percent) agreed to participate. The implications of this will be examined in the discussion. The interview was conducted using a schedule of questions that required approximately 2.0 hours to complete. An attempt was made to estimate the total calories ingested. We are aware that we have fallen short of this aim, but because of the large number of foods covered, because of the continual probing our questionnaire required, and because it was an extensive face-
553
to-face interview, we feel we have a reasonable measure of total calories ingested. The interview schedule itself was developed on the basis of our experience with a number of pretests and, equally important, the experience we have gathered in similar, very detailed interviews with over 5,000 subjects since 1975. It was designed to study frequency and amount of various foods ingested by subjects over a 1-year period starting 2 years before the interview. The amount ingested was assessed by respondents in referring to pictures of portion sizes, reference to standard measures of volume (e.g., cups), and where feasible the number of items or pieces (e.g., eggs, ears of corn, or pieces of chicken). We have conducted a number of assessments of the reliability of our interview process. They have consisted of concurrent interviews of spouses with regard to the dietary behavior of the subject as observed by the spouse (11), reinterviews of cases and controls by telephone subsequent to the face-toface interview, interviews of subjects and controls 3-5 years after the initial interview (12), and reinterviews of individuals first interviewed 20-25 years before regarding their earlier diet as well as their current diet (13). Our studies, like those of other investigators (14, 15), suggest that it would be unwise to attempt to make point estimates of levels of ingestion but that, nevertheless, it is possible to assess associations and to compare groups of cases and controls on their relative level of ingestion of various foods. To convert reported consumption of foods and beverages to quantitative estimates of specific nutrients, we have used tables from the US Department of Agriculture handbooks 8 and 456 (16, 17), together with updates and associated data tapes, as well as tables from Paul and Southgate (18) and Pennington (19,20) dealing with macroand micronutrients and fiber. We were careful to conduct our analyses on not only levels of nutrients, e.g., carotene, tocopherol, and fats, but also individual foods so as to obtain a notion of both nutrition and diet. Our statistical analysis started with examinations of single factors, adjusting for
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
554
Graham et al.
age and education (21) (which has been shown to be an excellent indicator of socioeconomic status), using logistic regression analysis (22). When dietary factors were identified as conferring statistically significant high or low risk, these were examined still more carefully, adjusting for characteristics of reproductive history and familial aggregation that had been identified as altering risk in our preliminary analyses. Next, those dietary factors identified as carrying low or high risk were examined in the context of each other.
RESULTS
Table 1 presents findings on various characteristics that have been associated with a higher risk of breast cancer in many previous studies. For example, as have many others, we found the lowest risk among females with early age at menarche and increasing risk with increasing age at first pregnancy, nulliparity, prior breast lesions, and a first degree relative with breast cancer. Age at first pregnancy is somewhat correlated (r = —0.32) with number of pregnancies; we found that adjusting for number of pregnancies did not change the increased risk associated with late age at first pregnancy. Moreover, parity was not a significant risk factor, whether or not adjusted for age at first pregnancy. As in earlier research, we found some small evidence of an increase in risk with decreases in duration of nursing and increases in lifetime months of menstruation. Although increased lifetime months of menstruation (months between menarche and menopause minus months pregnant and months nursed) were shown in our earlier study to be associated with higher risk of breast cancer (4), and though there appeared to be a doseresponse increase in risk with an increase in lifetime menstrual months in our present study (Arend = 0.05), the odds ratio confidence intervals all included 1.0. In a hospitalbased study, we found a decrease in risk with increases in duration of nursing (4) and, in a subsequent community-based inquiry, we replicated this finding, but only for premen-
opausal women (23). There was a similar tendency in this study of postmenopausal women, but again the confidence intervals embraced 1.0. Table 2 presents mean values of nutrients ingested by cases and random-sample controls. Note that the differences in intake for total fat, saturated fat, or cholesterol were small and could readily have occurred by chance. The same is true for total calories ingested, alcohol, protein, carbohydrate, and calcium. Note especially that there was no significant difference in quantity of caffeine ingested; controls actually consumed more caffeine. On the other hand, cases had a substantially smaller ingestion of dietary fiber, a-tocopherol, folic acid, vitamin C, and carotene. The mean ingestion of total vitamin A was less in cases, but the average ingestion of retinol was very similar in cases and controls. Thus, the estimate for total vitamin A may be influenced by the substantially smaller ingestion of carotene among cases. Our analyses of the average amounts eaten by cases and controls of individual foods showed some significant differences that reflected the nutritional findings described above. Thus, cases ingested less tomato juice, V-8 vegetable juice, tomatoes, corn, asparagus, strawberries, apples, grapefruit, lemons, limes, and olives. They also ate significantly less whole wheat rolls and onion soup. There were no differences between cases and controls in use of vitamin A or C, tocopherol, or carotene supplements. A number of researchers have found an increased risk of breast cancer associated with increases in ingestion of alcohol (24, 25). This was observed particularly in series that excluded older cases of breast cancer. In this inquiry, which dealt solely with postmenopausal breast cancer, we observed no difference for total alcohol ingestion in the period 2 years prior to interview, i.e., ingestion of alcohol in the form of wine, beer, and distilled spirits. Similarly, there was no difference between cases and controls in the amount used of either wine, beer, or distilled beverages, studied separately. There similarly was no difference in drinking habits,
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
Dietary Epidemiology of Breast Cancer
TABLE 1. Risk of breast cancer associated with reproductive and familial traits, adjusting for age and education: western New York, 1986-1989
Age at menarche 9-11 12 13 14-19 Lifetime months menstruated 12-328 329-386 387-438 439-683 Age at first pregnancy 15-20 21-23 24-26 27-42 Never Total no. of pregnancies 0 1 2 3 4 £5
Total months pregnant 0-17 18-27 28-37 38-112 Total months nursed for women who had a live birth 0 0.5-1 1.5-4 5-11 12-89 Ever had problems with breasts (lumps, cysts, and/or fibrocystic disease) No Yes Mother had breast cancer No Yes % of sisters with breast cancer for women who had a sister 0 13-49 50 100
Cases in)
Controls In)
75 127 102 127 431
104 97 138 151 490
1 00 1.80 1.02 1.19
104 106 113 116 439
134 123 121 116 494
1.00 1.12 1.21 1.33
81 101 79 110 67 438
120 131 90 97 55 493
1.00 1.15 1.35 1.80 1.87
67 40 78 83 67 103 438
55 42 102 94 81 120 494
1 00 0.80 0.65 0.73 0.67 0.70
119 101 126 93 439
115 124 144 111 494
1.00 0.79 0.84 0.79
174 50 50 46 48 368
182 52 72 54 67 427
1.00 1.00 0.73 0.89 0.76
326 111 437
397 95 492
1.00 1.38
1.01-1.89
408 31 439
477 17 494
1.00 2.07
1.13-3.81
293 25 16 7 341
347 12 7 3 369
1.00 2.51 2.74 2.99
• OR, odds ratio; 95% Cl, 95% confidence interval, t Excluding those never pregnant.
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
OR*
95% Cl*
1.21-2.68 0.69-1.52 0.81-1.74 Ptrcna = 0.66
0.78-1.62 0.84-1.75 0.92-1.93 Pmnfl = 0 05
0.78-1 69 0.88-2.05 1.19-2 71 1 17-2.97 p,n»« = 0.01t
0.46-1 41 0.41-1 03 0.46-1 17 0 41-1.09 0.45-1.10 Pt™« = 0.39
0 55-1 15 0.59-1 19 0.54-1.16 Pt«x. = 0 33
0.65-1.56 0.48-1.11 0.56-1 40 0 49-1.17 p^na = 0.53
1.24-5.10 1.11-6.77 0.75-11.86 PnM = 0.002
555
556
Graham et al.
TABLE 2.
Mean values of nutrients ingested by cases and controls per day: western New York, 1986-1989
Calories including alcohol Calories without alcohol Alcohol (g) Total fat (g) Saturated fat (g) a-Tocopherol (mg) Vitamin C (mg) Total vitamin A (IU) Retinol (IU) Carotene (IU) Dietary fiber (g) Folic acid (^g) Cholesterol (mg) Protein (g) Carbohydrate (g) Caffeine (mg) Calcium (mg)
Cases (mean ± SD")
Controls (mean ± SD)
P
2,197 ±814 2,115 ±809 9±18 82 ±39 30 ±15 8±4 180 ±82 13,945 ±7,994 3,698 ±2,914 10,246 ±6,987 26 ± 1 0 373 ±132 344 ±179 88 ± 3 3 274 ±103 254 ± 234 909 ± 421
2,218 ±919 2,148 ± 909 8±14 83 ± 4 6 30 ± 1 6 9±5 193 ± 9 4 14,862 ± 8,454 3,740 ±2,814 11,121 ±7,205 28 ± 1 2 392 ±161 343 ±181 91 ±39 277 ±106 264 ± 249 903 ± 432
0.72 0.57 0.29 0.67 0.99 0.04 0.02 0.09 0.82 0.06 0.01 0.05 0.92 0.37 0.65 0.54 0.84
• SD, standard deviation.
10 years before interview, 20 years before, and at age 16. Table 3 exhibits crude odds ratios for various nutritional factors as well as body mass as measured by the Quetelet index (weight (kg)/height (m)2). The estimates are also adjusted by logistic regression for age, education, age at first pregnancy, number of pregnancies, age at menarche, whether there was a first degree female relative with breast cancer, Quetelet index (for nutrients only), and prior benign breast disease. The crude and adjusted odds ratios are almost exactly the same. The carotene section of the table shows that risk of breast cancer was highest among those eating the smallest amounts of carotene. The vitamin C and a-tocopherol sections of table 3 show a similar trend. The risk associated with dietary fiber mirrored the risks associated with micronutrients found in vegetables with a more consistent dose-response, but the risk reduction was not as substantial. Little risk alteration was found to be associated with fat or saturated fat, considered separately, in this series of postmenopausal women as also shown in table 3. All of the risks and all of the confidence intervals hovered around 1.0. The folic acid section of the table shows a relatively flat risk pattern for folic acid. The
only decline in risk is in the highest quartile. Finally, table 3 shows a dose-response increase in risk associated with increases in Quetelet index, and none of the odds ratios embraced 1.0. The trend tests approached significance for carotene and were significant for ascorbic acid, a-tocopherol, dietary fiber, folic acid, and the Quetelet index. Because many of the dietary factors discussed above are found in the same foods, there is the possibility of high intercorrelations among them, making it difficult to assess the risk associated with a given nutrient, adjusting for correlated ones. Table 4 shows that fat, tocopherol, folic acid, and fiber are highly correlated (r > 0.7) with total ingestion of calories. Note, too, that total fat and saturated fat are highly associated with each other and with a-tocopherol. Vitamin C and tocopherol were found to be highly associated with folic acid and dietary fiber. Carotene demonstrated only slight association with a-tocopherol, fat, and calories as well as somewhat larger associations with vitamin C (r = 0.66), folic acid (r = 0.65), and dietary fiber (r = 0.68) among the controls. The correlations among these nutrients for cases were similar to those of controls. The Quetelet index was not associated with any variable in table 4. Table 5 shows risk of breast cancer for
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
TABLE 3. Crude risk of breast cancer associated with Quetelet index and various nutritional traits and risk adjusting for age, education, age at first pregnancy, number of pregnancies, age at menarche, relative with breast cancer, benign breast disease, and Quetelet index: western New York, 1986-1989* Quartiles
Carotene daily (IU) 13,520
OR
103 133 126 132 494
1.00 0.56 0.68 0.56
1.00 0.54 0.69 0.56
1.00 0.81 0.95 0.63
439
115 126 116 137 494
a-Tocopherol daily (mg) 11
123 108 109 99 439
110 128 125 131 494
1.00 0.75 0.78 0.68
Dietary fiber daily (g) 33
116 117 110 96 439
121 115 123 135 494
1.00 1.06 0.93 0.74
Fat daily (g) 99
113 107 110 109 439
123 128 123 120 494
1.00 0.91 0.97 0.99
Saturated fat daily (g) 37
111 106 112 110 439
125 126 124 119 494
1.00 0.95 1.02 1.04
Folic acid daily (/*g) 451
116 110 116 97
121 124 117 132 494
1.00 0.93 1.03 0.77
151 120 110 113 494
1.00 1.51 1.62 1.72
134 97 111 97
122 108 117 92
439 Quetelet index 29
Controls
C)
439 Vitamin C daily (mg) 229
Adjusted
Crude ORt
Cases
94 113 111 121 439
95%Clft
0.37-0.79 0.47-1.00 0.38-0.82 Ptrend = 0.08
1.00 0.84 0.92 0.62
0.58-1.23 0.63-1.34 0.42-0.91 Ptrend = 0.02
1.00 0.77 0.78 0.64
0.53-1.12 0.53-1.13 0.44-0.95 Ptrend = 0.02
1.00 1.05 0.96 0.74
0.72-1.54 0.66-1.39 0.51-1.09 Ptrend = 0.01
1.00 0.95 1.01 0.93
0.65-1.38 0.69-1.49 0.63-1.38 Ptrend = 0.38
1.00 1.05 1.01 1.04
0.72-1.53 0.69-1.48 0.71-1.53 Ptrend = 0.67
1.00 0.88 1.04 0.70
0.60-1.28 0.71-1.51 0.48-1.02 Ptrend = 0.03
1.00 1.48 1.71 1.80
1.02-2.16 1.16-2.51 1.23-2.63 Ptrend = 0.01
* Ail estimates are adjusted for age, education, age at first pregnancy, age at menarche, relative with breast cancer, benign breast disease, and number of pregnancies, and nutritional traits are additionally adjusted for Quetelet index, t OR, odds ratio; 95% Cl, 95% confidence interval. t Because of missing data, multivariate analysis is based on 430 cases and 487 controls.
557 Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
558
Graham et a).
nutrient characteristics associated with increased risk of breast cancer in table 3, with each risk adjusted for age and education as well as ingestion of other related nutrient factors. The carotene section of the table shows that risk of breast cancer was highest among those with the smallest ingestion of carotene, adjusting for age, education, Quetelet index, and vitamin C intake as well as for intake of a-tocopherol, folic acid, and total calories. In all except three instances, the confidence interval did not embrace 1.0. Despite these significant results, the trend tests were not remarkable. Rather, the distribution of risk by exposure level suggests a significant decrease at any exposure over 5,824 IU per day. The vitamin C section of table 5 shows less reduction in risk for ascorbic acid, i.e., that risk decreased with increases in ingestion, but in all but two instances the confidence intervals included 1.0. Results for a-tocopherol in the next section are similar. For none of the micronutrients described above was there a clear dose-response. No noteworthy alteration in risk was shown for variations in ingestion of folic acid. In the case of Quetelet index, however, the risk increased substantially with each increase in the index, adjusted for age, education, and each nutrient trait. In most instances, the risk at the lower bound of the 95 percent confidence interval exceeded 1.0 substantially, and there was a definite increasing trend with level of obesity. The study of nutrient ingestion differences between cases and controls should usually take into consideration total calories ingested. When we examined the relation between risk and ingestion of carotene and other micronutrients, adjusting for calories ingested, we found no change from risks already described. In the interest of making public health recommendations, knowing the risk associated with ingesting various proportions of total calories in the form of fat, carbohydrates, and protein is useful. We examined all of the macronutrients in terms of the proportions of total calories ingested that derive from protein, carbohydrates, and fats and found no relations for any macro-
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
s d in p o o in
°? P do CO CD
cq O
s
d d
3 odd CD CM o a> CD co oi eg in TT p
d d d d dd
N t ^f o ^ o CM O) CO • * CO CO CD O
d d d d d dd
Vol. 134, No. 6 Printed in U S.A
Nutritional Epidemiology of Postmenopausal Breast Cancer in Western New York Saxon Graham,1 Rosemary Hellmann,1 James Marshall,1 Jo Freudenheim,1 John Vena,1 Mya Swanson,1 Maria Zielezny,1 Takuma Nemoto,2 Nancy Stubbe,3 and Teresa Raimondo1
The authors studied 439 postmenopausal breast cancer cases, identified in hospitals throughout western New York, with an interview schedule that considered frequency and amount ingested of 172 foods and provided data for an estimate of total calories ingested. These were compared with age-matched controls comprising a random sample of the same communities as the cases. The extensive interviews, requiring 2.0 hours on average to administer, also covered alcohol ingestion, Quetelet index, and a wide variety of reproductive factors. The authors found, as have most investigators over the past 25 years, that risk increased with increases in age at first pregnancy, decreased with increases in numbers of children and pregnancies, and increased in those with history of benign breast disease and in those with female relatives previously affected with breast cancer. Risk adjusted for potential confounders was highest among women with the lowest ingestion of carotene or a substance correlated with its ingestion. Risk was not associated with retinol ingestion. It increased with increases in Quetelet index. Fat intake, whether studied in terms of quantity or the proportion of total calories derived from fat, was not associated with risk of breast cancer. Our analyses of these factors were adjusted for age, education, and the reproductive history traits described above. Am J Epidemiol 1991 ;134:552-66. breast neoplasms; nutrition
index, alcohol ingestion, and various nutrients, such as carotene and fats (5-9). The present inquiry took place in western New York. It echoed the usual findings on reproductive and familial history and discovered a substantial increase in risk with increases in body mass and a higher risk in women with low ingestion of carotene or factors associated with it.
As one of the most frequently occurring tumors in women, breast cancer has been the focus of considerable epidemiologic inquiry in the last two decades. Substantial variations in risk of breast cancer have been found to occur according to characteristics of reproductive history and familial aggregation (1-4). In the field of diet and nutrition, various investigators have discovered variations in risk associated with Quetelet
MATERIALS AND METHODS
Received for publication November 28, 1990, and in final form April 5, 1991. 1 Department of Social and Preventive Medicine, State University of New York at Buffalo, Buffalo, NY. 2 Breast Care Center, Sisters Hospital, Buffalo, NY. 3 Department of Surgery, School of Medicine and Biomedical Sciences, State University of New York at Buffalo, Buffalo, NY. Reprint requests to Dr. Saxon Graham, Department of Social and Preventive Medicine, State University of New York at Buffalo, 2211 Main Street, Buffalo, NY 14214. This publication was made possible by grant CA11535 from the National Cancer Institute, National Institutes of Health, and grant PDT-434 from the American Cancer Society.
The study findings were based upon 439 histologically confirmed, incident cases of postmenopausal cancer of the breast, aged 41-85 years, residing in Erie and Niagara counties (the counties containing Buffalo, Lackawanna, Tonawanda, and Niagara Falls and their rural hinterlands). Subjects were considered postmenopausal if they were 50 years of age and over, had ceased menstruation, or were less than 50 years of age 552
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
Dietary Epidemiology of Breast Cancer with natural menopause or nonfunctional ovaries. Thus, females under age 50 were considered postmenopausal if they had experienced natural menopause, bilateral oophorectomy, or irradiation to the ovaries. Women 50 years of age and over were considered postmenopausal if they no longer were menstruating. Controls were 494 randomly chosen postmenopausal residents of the two counties and frequency matched on age group with cases; those under age 65 were selected from rolls of New York State motor vehicle driver's license holders (90.2 percent of the female population under age 65 in western New York are licensed) (10) and those 65 years and over from the Health Care Finance Administration rolls. Our nurse-interviewers maintained continuous contact with all of the hospitals in the two counties to ascertain as soon as possible when a case was admitted. The interviewers then contacted patients' physicians for permission to ask patients to participate in the inquiry. Over the years of the study, from 1986 to 1989, the interviewers identified 777 eligible patients from hospitals, and for these patients, their physicians gave us permission to interview 561 (72.2 percent). Of those patients permitted by their doctors to participate in the study, 100 (17.8 percent) refused and 22 (3.9 percent) were too ill to be interviewed. Thus, the proportion providing usable interviews (439 patients) of those determined eligible (777 patients) was 56.5 percent. Once selected, controls were sent letters requesting their participation. The interviewer then contacted them by telephone to make appointments for the interview. Of those 1,076 individuals randomly selected to participate as controls, 494 (45.9 percent) agreed to participate. The implications of this will be examined in the discussion. The interview was conducted using a schedule of questions that required approximately 2.0 hours to complete. An attempt was made to estimate the total calories ingested. We are aware that we have fallen short of this aim, but because of the large number of foods covered, because of the continual probing our questionnaire required, and because it was an extensive face-
553
to-face interview, we feel we have a reasonable measure of total calories ingested. The interview schedule itself was developed on the basis of our experience with a number of pretests and, equally important, the experience we have gathered in similar, very detailed interviews with over 5,000 subjects since 1975. It was designed to study frequency and amount of various foods ingested by subjects over a 1-year period starting 2 years before the interview. The amount ingested was assessed by respondents in referring to pictures of portion sizes, reference to standard measures of volume (e.g., cups), and where feasible the number of items or pieces (e.g., eggs, ears of corn, or pieces of chicken). We have conducted a number of assessments of the reliability of our interview process. They have consisted of concurrent interviews of spouses with regard to the dietary behavior of the subject as observed by the spouse (11), reinterviews of cases and controls by telephone subsequent to the face-toface interview, interviews of subjects and controls 3-5 years after the initial interview (12), and reinterviews of individuals first interviewed 20-25 years before regarding their earlier diet as well as their current diet (13). Our studies, like those of other investigators (14, 15), suggest that it would be unwise to attempt to make point estimates of levels of ingestion but that, nevertheless, it is possible to assess associations and to compare groups of cases and controls on their relative level of ingestion of various foods. To convert reported consumption of foods and beverages to quantitative estimates of specific nutrients, we have used tables from the US Department of Agriculture handbooks 8 and 456 (16, 17), together with updates and associated data tapes, as well as tables from Paul and Southgate (18) and Pennington (19,20) dealing with macroand micronutrients and fiber. We were careful to conduct our analyses on not only levels of nutrients, e.g., carotene, tocopherol, and fats, but also individual foods so as to obtain a notion of both nutrition and diet. Our statistical analysis started with examinations of single factors, adjusting for
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
554
Graham et al.
age and education (21) (which has been shown to be an excellent indicator of socioeconomic status), using logistic regression analysis (22). When dietary factors were identified as conferring statistically significant high or low risk, these were examined still more carefully, adjusting for characteristics of reproductive history and familial aggregation that had been identified as altering risk in our preliminary analyses. Next, those dietary factors identified as carrying low or high risk were examined in the context of each other.
RESULTS
Table 1 presents findings on various characteristics that have been associated with a higher risk of breast cancer in many previous studies. For example, as have many others, we found the lowest risk among females with early age at menarche and increasing risk with increasing age at first pregnancy, nulliparity, prior breast lesions, and a first degree relative with breast cancer. Age at first pregnancy is somewhat correlated (r = —0.32) with number of pregnancies; we found that adjusting for number of pregnancies did not change the increased risk associated with late age at first pregnancy. Moreover, parity was not a significant risk factor, whether or not adjusted for age at first pregnancy. As in earlier research, we found some small evidence of an increase in risk with decreases in duration of nursing and increases in lifetime months of menstruation. Although increased lifetime months of menstruation (months between menarche and menopause minus months pregnant and months nursed) were shown in our earlier study to be associated with higher risk of breast cancer (4), and though there appeared to be a doseresponse increase in risk with an increase in lifetime menstrual months in our present study (Arend = 0.05), the odds ratio confidence intervals all included 1.0. In a hospitalbased study, we found a decrease in risk with increases in duration of nursing (4) and, in a subsequent community-based inquiry, we replicated this finding, but only for premen-
opausal women (23). There was a similar tendency in this study of postmenopausal women, but again the confidence intervals embraced 1.0. Table 2 presents mean values of nutrients ingested by cases and random-sample controls. Note that the differences in intake for total fat, saturated fat, or cholesterol were small and could readily have occurred by chance. The same is true for total calories ingested, alcohol, protein, carbohydrate, and calcium. Note especially that there was no significant difference in quantity of caffeine ingested; controls actually consumed more caffeine. On the other hand, cases had a substantially smaller ingestion of dietary fiber, a-tocopherol, folic acid, vitamin C, and carotene. The mean ingestion of total vitamin A was less in cases, but the average ingestion of retinol was very similar in cases and controls. Thus, the estimate for total vitamin A may be influenced by the substantially smaller ingestion of carotene among cases. Our analyses of the average amounts eaten by cases and controls of individual foods showed some significant differences that reflected the nutritional findings described above. Thus, cases ingested less tomato juice, V-8 vegetable juice, tomatoes, corn, asparagus, strawberries, apples, grapefruit, lemons, limes, and olives. They also ate significantly less whole wheat rolls and onion soup. There were no differences between cases and controls in use of vitamin A or C, tocopherol, or carotene supplements. A number of researchers have found an increased risk of breast cancer associated with increases in ingestion of alcohol (24, 25). This was observed particularly in series that excluded older cases of breast cancer. In this inquiry, which dealt solely with postmenopausal breast cancer, we observed no difference for total alcohol ingestion in the period 2 years prior to interview, i.e., ingestion of alcohol in the form of wine, beer, and distilled spirits. Similarly, there was no difference between cases and controls in the amount used of either wine, beer, or distilled beverages, studied separately. There similarly was no difference in drinking habits,
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
Dietary Epidemiology of Breast Cancer
TABLE 1. Risk of breast cancer associated with reproductive and familial traits, adjusting for age and education: western New York, 1986-1989
Age at menarche 9-11 12 13 14-19 Lifetime months menstruated 12-328 329-386 387-438 439-683 Age at first pregnancy 15-20 21-23 24-26 27-42 Never Total no. of pregnancies 0 1 2 3 4 £5
Total months pregnant 0-17 18-27 28-37 38-112 Total months nursed for women who had a live birth 0 0.5-1 1.5-4 5-11 12-89 Ever had problems with breasts (lumps, cysts, and/or fibrocystic disease) No Yes Mother had breast cancer No Yes % of sisters with breast cancer for women who had a sister 0 13-49 50 100
Cases in)
Controls In)
75 127 102 127 431
104 97 138 151 490
1 00 1.80 1.02 1.19
104 106 113 116 439
134 123 121 116 494
1.00 1.12 1.21 1.33
81 101 79 110 67 438
120 131 90 97 55 493
1.00 1.15 1.35 1.80 1.87
67 40 78 83 67 103 438
55 42 102 94 81 120 494
1 00 0.80 0.65 0.73 0.67 0.70
119 101 126 93 439
115 124 144 111 494
1.00 0.79 0.84 0.79
174 50 50 46 48 368
182 52 72 54 67 427
1.00 1.00 0.73 0.89 0.76
326 111 437
397 95 492
1.00 1.38
1.01-1.89
408 31 439
477 17 494
1.00 2.07
1.13-3.81
293 25 16 7 341
347 12 7 3 369
1.00 2.51 2.74 2.99
• OR, odds ratio; 95% Cl, 95% confidence interval, t Excluding those never pregnant.
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
OR*
95% Cl*
1.21-2.68 0.69-1.52 0.81-1.74 Ptrcna = 0.66
0.78-1.62 0.84-1.75 0.92-1.93 Pmnfl = 0 05
0.78-1 69 0.88-2.05 1.19-2 71 1 17-2.97 p,n»« = 0.01t
0.46-1 41 0.41-1 03 0.46-1 17 0 41-1.09 0.45-1.10 Pt™« = 0.39
0 55-1 15 0.59-1 19 0.54-1.16 Pt«x. = 0 33
0.65-1.56 0.48-1.11 0.56-1 40 0 49-1.17 p^na = 0.53
1.24-5.10 1.11-6.77 0.75-11.86 PnM = 0.002
555
556
Graham et al.
TABLE 2.
Mean values of nutrients ingested by cases and controls per day: western New York, 1986-1989
Calories including alcohol Calories without alcohol Alcohol (g) Total fat (g) Saturated fat (g) a-Tocopherol (mg) Vitamin C (mg) Total vitamin A (IU) Retinol (IU) Carotene (IU) Dietary fiber (g) Folic acid (^g) Cholesterol (mg) Protein (g) Carbohydrate (g) Caffeine (mg) Calcium (mg)
Cases (mean ± SD")
Controls (mean ± SD)
P
2,197 ±814 2,115 ±809 9±18 82 ±39 30 ±15 8±4 180 ±82 13,945 ±7,994 3,698 ±2,914 10,246 ±6,987 26 ± 1 0 373 ±132 344 ±179 88 ± 3 3 274 ±103 254 ± 234 909 ± 421
2,218 ±919 2,148 ± 909 8±14 83 ± 4 6 30 ± 1 6 9±5 193 ± 9 4 14,862 ± 8,454 3,740 ±2,814 11,121 ±7,205 28 ± 1 2 392 ±161 343 ±181 91 ±39 277 ±106 264 ± 249 903 ± 432
0.72 0.57 0.29 0.67 0.99 0.04 0.02 0.09 0.82 0.06 0.01 0.05 0.92 0.37 0.65 0.54 0.84
• SD, standard deviation.
10 years before interview, 20 years before, and at age 16. Table 3 exhibits crude odds ratios for various nutritional factors as well as body mass as measured by the Quetelet index (weight (kg)/height (m)2). The estimates are also adjusted by logistic regression for age, education, age at first pregnancy, number of pregnancies, age at menarche, whether there was a first degree female relative with breast cancer, Quetelet index (for nutrients only), and prior benign breast disease. The crude and adjusted odds ratios are almost exactly the same. The carotene section of the table shows that risk of breast cancer was highest among those eating the smallest amounts of carotene. The vitamin C and a-tocopherol sections of table 3 show a similar trend. The risk associated with dietary fiber mirrored the risks associated with micronutrients found in vegetables with a more consistent dose-response, but the risk reduction was not as substantial. Little risk alteration was found to be associated with fat or saturated fat, considered separately, in this series of postmenopausal women as also shown in table 3. All of the risks and all of the confidence intervals hovered around 1.0. The folic acid section of the table shows a relatively flat risk pattern for folic acid. The
only decline in risk is in the highest quartile. Finally, table 3 shows a dose-response increase in risk associated with increases in Quetelet index, and none of the odds ratios embraced 1.0. The trend tests approached significance for carotene and were significant for ascorbic acid, a-tocopherol, dietary fiber, folic acid, and the Quetelet index. Because many of the dietary factors discussed above are found in the same foods, there is the possibility of high intercorrelations among them, making it difficult to assess the risk associated with a given nutrient, adjusting for correlated ones. Table 4 shows that fat, tocopherol, folic acid, and fiber are highly correlated (r > 0.7) with total ingestion of calories. Note, too, that total fat and saturated fat are highly associated with each other and with a-tocopherol. Vitamin C and tocopherol were found to be highly associated with folic acid and dietary fiber. Carotene demonstrated only slight association with a-tocopherol, fat, and calories as well as somewhat larger associations with vitamin C (r = 0.66), folic acid (r = 0.65), and dietary fiber (r = 0.68) among the controls. The correlations among these nutrients for cases were similar to those of controls. The Quetelet index was not associated with any variable in table 4. Table 5 shows risk of breast cancer for
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
TABLE 3. Crude risk of breast cancer associated with Quetelet index and various nutritional traits and risk adjusting for age, education, age at first pregnancy, number of pregnancies, age at menarche, relative with breast cancer, benign breast disease, and Quetelet index: western New York, 1986-1989* Quartiles
Carotene daily (IU) 13,520
OR
103 133 126 132 494
1.00 0.56 0.68 0.56
1.00 0.54 0.69 0.56
1.00 0.81 0.95 0.63
439
115 126 116 137 494
a-Tocopherol daily (mg) 11
123 108 109 99 439
110 128 125 131 494
1.00 0.75 0.78 0.68
Dietary fiber daily (g) 33
116 117 110 96 439
121 115 123 135 494
1.00 1.06 0.93 0.74
Fat daily (g) 99
113 107 110 109 439
123 128 123 120 494
1.00 0.91 0.97 0.99
Saturated fat daily (g) 37
111 106 112 110 439
125 126 124 119 494
1.00 0.95 1.02 1.04
Folic acid daily (/*g) 451
116 110 116 97
121 124 117 132 494
1.00 0.93 1.03 0.77
151 120 110 113 494
1.00 1.51 1.62 1.72
134 97 111 97
122 108 117 92
439 Quetelet index 29
Controls
C)
439 Vitamin C daily (mg) 229
Adjusted
Crude ORt
Cases
94 113 111 121 439
95%Clft
0.37-0.79 0.47-1.00 0.38-0.82 Ptrend = 0.08
1.00 0.84 0.92 0.62
0.58-1.23 0.63-1.34 0.42-0.91 Ptrend = 0.02
1.00 0.77 0.78 0.64
0.53-1.12 0.53-1.13 0.44-0.95 Ptrend = 0.02
1.00 1.05 0.96 0.74
0.72-1.54 0.66-1.39 0.51-1.09 Ptrend = 0.01
1.00 0.95 1.01 0.93
0.65-1.38 0.69-1.49 0.63-1.38 Ptrend = 0.38
1.00 1.05 1.01 1.04
0.72-1.53 0.69-1.48 0.71-1.53 Ptrend = 0.67
1.00 0.88 1.04 0.70
0.60-1.28 0.71-1.51 0.48-1.02 Ptrend = 0.03
1.00 1.48 1.71 1.80
1.02-2.16 1.16-2.51 1.23-2.63 Ptrend = 0.01
* Ail estimates are adjusted for age, education, age at first pregnancy, age at menarche, relative with breast cancer, benign breast disease, and number of pregnancies, and nutritional traits are additionally adjusted for Quetelet index, t OR, odds ratio; 95% Cl, 95% confidence interval. t Because of missing data, multivariate analysis is based on 430 cases and 487 controls.
557 Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
558
Graham et a).
nutrient characteristics associated with increased risk of breast cancer in table 3, with each risk adjusted for age and education as well as ingestion of other related nutrient factors. The carotene section of the table shows that risk of breast cancer was highest among those with the smallest ingestion of carotene, adjusting for age, education, Quetelet index, and vitamin C intake as well as for intake of a-tocopherol, folic acid, and total calories. In all except three instances, the confidence interval did not embrace 1.0. Despite these significant results, the trend tests were not remarkable. Rather, the distribution of risk by exposure level suggests a significant decrease at any exposure over 5,824 IU per day. The vitamin C section of table 5 shows less reduction in risk for ascorbic acid, i.e., that risk decreased with increases in ingestion, but in all but two instances the confidence intervals included 1.0. Results for a-tocopherol in the next section are similar. For none of the micronutrients described above was there a clear dose-response. No noteworthy alteration in risk was shown for variations in ingestion of folic acid. In the case of Quetelet index, however, the risk increased substantially with each increase in the index, adjusted for age, education, and each nutrient trait. In most instances, the risk at the lower bound of the 95 percent confidence interval exceeded 1.0 substantially, and there was a definite increasing trend with level of obesity. The study of nutrient ingestion differences between cases and controls should usually take into consideration total calories ingested. When we examined the relation between risk and ingestion of carotene and other micronutrients, adjusting for calories ingested, we found no change from risks already described. In the interest of making public health recommendations, knowing the risk associated with ingesting various proportions of total calories in the form of fat, carbohydrates, and protein is useful. We examined all of the macronutrients in terms of the proportions of total calories ingested that derive from protein, carbohydrates, and fats and found no relations for any macro-
Downloaded from https://academic.oup.com/aje/article-abstract/134/6/552/196644 by Sussex University user on 07 May 2018
s d in p o o in
°? P do CO CD
cq O
s
d d
3 odd CD CM o a> CD co oi eg in TT p
d d d d dd
N t ^f o ^ o CM O) CO • * CO CO CD O
d d d d d dd
