Ecological, case-control, cohort, and experimental study designs are used to assess relationships between diet and chronic diseases (Howe 1994). Ecological studies examine associations at the population level, which means that confounding factors such as economy and education may produce biased findings. Dietary data are collected by food balance sheets, which describe inaccurately the food consumption of individual subjects. Ecological studies generate hypotheses rather than test associations. Case-control and cohort studies, on the other hand, take into account food consumption and confounding factors at the individual level. Case-control studies have been more common, because they are easier, quicker, and cheaper to carry out than cohort studies. Case-control studies, however, are susceptible to selection and recall biases (Hebert and Miller 1988), and thus cohort studies in which subjects are recruited before the onset of disease are recommended although not always possible. Experimental studies investigate biological mechanisms and relevant timing of exposure. All these four designs are needed to gain deeper understanding of associations between dietary factors and diseases.
Selection bias occurs if exposed subjects participate selectively in a study (Howe 1994). The overall voluntary participation in medical studies has decreased during the last decades in the Western countries. This has been observed also in Finland where response rates have traditionally been quite high. For example in the FINRISK Surveys among Finnish adults, carried out every fifth year since 1972, the response rate has decreased from 91% to 76% for men and 94% to 85% for women living in Kuopio Province (Vartiainen et al. 1994). No data about the dropouts are available in this survey, but factors associated with non-response in the dietary survey among participants in the 1992 FINRISK Survey were young age, loss of spouse, divorce, and less healthy dietary habits (e.g., consumption of fat and vegetables) (Roos 1998). Furthermore, obese men and male smokers as well as less educated women and female heavy drinkers had low response rates. On the other hand, women with a high education and healthy lifestyle are more likely to volunteer in epidemiological studies (Criqui et al. 1978, Rimer et al. 1996). In the Kuopio Breast Cancer Study, the response rate of the population controls was 72%. The reasons for non-response were not assessed in this study. The prevalence of smokers (14%), however, was somewhat lower than in another study carried out in Kuopio province (19%) (Vartiainen et al. 1994). Although the present study included more old women who smoked less, the possibility of a slight selection bias in terms of smoking cannot be excluded.
Selection bias is an important factor also when results are interpreted and generalized to different populations. A noteworthy feature of the study design was the attempt to recruit all breast cancer cases in the catchment area of the hospital. This aim was quite well achieved, which was shown by comparisons between the Finnish Cancer Registry, the clinical and pathological registers of the hospital and the register of the Kuopio Breast Cancer Study. In all, only 15% of all eligible breast cancer cases were missed. It is, however, possible that educated and wealthy women are over-represented in this group of non-participants, and thus difference between the cases and the population controls may have further widened with respect to health-consciousness. This may have theoretically biased the results as follows: if health-conscious subjects (more in population controls) tend to over-report foods that are considered healthy (e.g., fruit), it may increase the effects of protective factors, whereas if they underestimate foods possibly hazardous for health (e.g., fat), it may increase the effects of risk factors.
With certain reservations, the results of this study can be extrapolated to differing groups of women. It has been assumed that a dietary factor related to risk of breast cancer in one population is also probably an apparent risk or protective factor in other populations with a similar level of dietary exposure (Hebert and Miller 1988).
Another important bias related to case-control studies is recall bias, which occurs, for example, if cases report their food consumption differently from controls. This may happen because patients have often thought about their previous exposure in order to find causes for their disease. It has also been shown that subjects who are aware of exposure-disease relationships may over-report or under-report their consumption of certain food items. In the Kuopio Breast Cancer Study, the subjects with potential breast disease were interviewed before the diagnosis was known. However, the possibility of interviewer bias between the subjects referred to the hospital for breast examination and the population controls could not be eliminated.
Recall bias between diet and risk of breast cancer has been examined in only a few studies. The Adventist Health Study (Lindsted and Kuzma 1989, Lindsted and Kuzma 1990) and the Canadian National Breast Screening Study (Friedenreich et al. 1991a, Friedenreich et al. 1991b, Friedenreich et al. 1993) found no recall bias between the diets reported by cases and controls. The Nurses' Health Study found by comparing the 1986 questionnaire to the 1989 questionnaire a non-significant increase in relative risk, from 0.87 to 1.43, between fat intake and breast cancer (Giovannucci et al. 1993). On the other hand, in Sweden, breast cancer patients reported the consumption of meat, snacks, coffee, and tea differently from healthy women (Holmberg et al. 1996). None of these studies examined the possible effects of the threat of disease.
The approach examining recall bias in this study was different from the approach of earlier studies, in which the recall bias has been evaluated by comparing the reported diet before and after the diagnosis of breast cancer. We used two different control groups: population controls and referral controls. Because the diagnosis was not known at the time of interview, it was possible to evaluate whether worry about possible breast cancer changed the reporting of diet by comparing the results obtained from the two control groups.
This study showed that reporting bias due to the threat of breast cancer may change the results for some foods and nutrients. The premenopausal women tended to misreport the consumption of liquid milk products, tea and sugar. The reporting of some fatty acids and vitamins was also different between the two control groups. The postmenopausal women misreported the consumption of milk products (all milk consumed, low-fat milk, sour milk, cream, and cheese). Despite the suspicion that recall bias is related to case-control studies, only one previous study on breast cancer has explicitly shown some recall bias in its data (Holmberg et al. 1996). That Swedish study was conducted in the 1990s, while the other studies on recall bias are from the 1980s when associations between diet and breast cancer had not yet been widely discussed in the media. It is also possible that the subjects under the threat of disease describe their diet as they wish it to be. This hypothesis is supported by a study in which the twin sisters of breast cancer cases reported changes in dietary habits that had not occurred (Richardson et al. 1993).
The enrollment methods of this study between the two control groups were not identical. The population controls were drawn from the Finnish National Population Register, whereas the referral controls comprised women who were referred to further breast examination but were later diagnosed as healthy. The question remains how much the control groups differed from each other. The referral controls were younger, had had more mammography examinations, and they reported more regular self-monitoring of their breasts than did the population controls (Study III). Therefore, the referral controls may have been more worried about their risk of breast cancer. All results, however, were adjusted for the known risk factors for breast cancer. Further adjustment for mammography examinations, self-monitoring, and seasonal variation in participation did not change the results. Nevertheless, the possibility of hidden premalignant breast disease in the referral controls cannot be ruled out. It is also possible that the threat of disease was not exactly the same between the breast cancer cases and the referral controls. No questions concerning this issue were presented in our study and no information on conversations between the first physician at the local health center and the subject was available. These issues should be kept in mind in future studies.
The potential pitfalls of nutrition research, mainly related to dietary assessment methods, are briefly summarized here.
The imperfections of food composition databases add to the inaccuracy of information measured by dietary assessment methods (Hebert and Miller 1988, Kohlmeier and Mendez 1997). The official Finnish National Food Composition Database, which was used in this study, is of very high quality but certainly not perfect. For example, loss of vitamins in cooking could not be considered in the analyses, with some exceptions (potatoes and cooked vegetables) (Ovaskainen et al. 1996).
Another problem of nutrition research is that some subjects (e.g., female, obese and more educated subjects) tend to under-report their food consumption. This tendency has increased during the last decades because of increasing health-consciousness. In the 1992 Dietary Survey of Finnish adults, over 40% of subjects underestimated their energy intake when estimated basal metabolic rate was used as a reference method (Hirvonen et al. 1997). Another study observed that 46% of subjects admitted having altered their diet during the period when they completed a 7-day weighed food record (Macdiarmid and Blundell 1997). Two main reasons for misreporting were the knowledge of unhealthy dietary habits and neglect in recording all food items accurately. Although energy adjustment is used to correct under- and overestimation of FFQs (Willett 1998), it is possible that misclassification may dilute associations between diet and diseases.
Undiagnosed illness may change the appetite and thus alter the subject's reporting of diet. This concerns especially cancers of the digestive tract, whereas the effect is minor in breast cancer, which is often diagnosed at an early stage. In the Kuopio Breast Cancer Study, most of the tumors were diagnosed at an early stage: 60% were axillary-node negative, and only 2% of the patients had stage IV disease (distant metastases) at the time of the diagnosis.
Other factors, such as alcohol consumption, smoking, and education, may correlate with dietary factors, and thus confound associations between diet and diseases (Franceschi 1993). In this study, the results were adjusted for the known risk factors for breast cancer (the most important factors were first-degree family history of breast cancer, history of benign breast disease, age at first full-term pregnancy, and use of oral contraceptives). In most of the models, interpretation of the results was quite similar between the basic model (including age and area of residence) and the multivariate model.
High body mass index did not increase risk of breast cancer in the Kuopio Breast Cancer Study, whereas waist-to-hip ratio was linearly related to risk. Relative risk was 4.6-fold in premenopausal women and 2.6-fold in postmenopausal women categorized into the highest quintile of waist-to-hip ratio compared to the lowest one. High waist-to-hip ratio is an independent predictor of metabolic disturbances in sex-steroid metabolism, glucose metabolism, and insulin-like growth factors (see Ballard-Barbash 1994, Stoll 1996). Besides these factors, behavioral characteristics such as weight cycling, smoking, high alcohol consumption, low physical activity, and high parity may explain a high waist-to-hip ratio (Seidell 1991, Kahn et al. 1997, Han et al. 1998). Our result on waist-to-hip ratio is compatible with earlier findings (Table 8) such as a recent cohort study including 11,663 subjects in the Netherlands, in which the relative risk of breast cancer in women with natural menopause was 2.63 (95% CI 1.09-6.35) for high waist-to-hip ratio compared to women with a low ratio (Kaaks et al. 1998).
In a clinical setting, waist-to-hip ratio can be measured accurately, and thus women with an increased risk of breast cancer are easily identified (Roberts et al. 1997). In Finland, obesity (Pietinen et al. 1996, Lahti-Koski et al. 1999a) and waist-to-hip ratio (preliminary results, Lahti-Koski et al. 1999b) have significantly increased since the 1970s. The National Academy of Sciences (1989) has recommended that waist-to-hip ratio should not exceed 0.80. In this study, only 20-30% of premenopausal and postmenopausal women had a waist-to-hip ratio within the recommended limits.
High body fat percent doubled the risk of postmenopausal breast cancer in this study. This may be explained by an elevated estrogen level in obese postmenopausal women due to the conversion of androgens into estrogen in adipose tissue after the menopause (see Ballard-Barbash 1994).
In the Nurses' Health Study, weight gain seemed to be a potential risk factor in those postmenopausal women whose weight was low at the age of 18 and who had never used postmenopausal estrogen replacement therapy (RR=1.99, 95% CI 1.43-2.76) (Huang et al. 1997). No association between adult weight gain and breast cancer was found in the Kuopio Breast Cancer Study (Table 8). However, the lowest weight in adulthood, causes for gaining weight, and weight cycling were not asked about.
It is well known that height may modestly increase risk of breast cancer (see Hunter and Willett 1996) (Table 8). In this study, the tallest women had a two-fold risk of breast cancer compared to the shortest ones with the analyses adjusted for age at menarche.
Table 8. Comparison of findings in earlier studies and in the Kuopio Breast Cancer Study.
|Summary of earlier studies||Kuopio Breast Cancer Study|
|Saturated fatty acids||-||-|
|Monounsaturated fatty acids||(¯)||-|
|Polyunsaturated fatty acids||-||-|
|n-3 fatty acids||(¯)||¯pre|
|n-6 fatty acids||-||¯pre|
|Body mass index||(post, ¯pre)||-|
|Adult weight gain||()||-|
, ¯ Convincing effect
pre Convincing effect for premenopausal women
post Convincing effect for postmenopausal women
( ) Suggestive effect
- No association
The findings between fat intake and risk of breast cancer in epidemiological studies have been inconclusive (Table 8). In a review of case-control studies, high fat intake was modestly related to an increased risk of breast cancer (Boyd et al. 1993). Because fat intake, however, did not appear to increase the risk in a pooled analysis of seven cohort studies (Hunter et al. 1996), it has been suggested that type of fat may be more important than total amount of fat (Table 8). Different kinds of fatty acids may have diverse effects on carcinogenesis of breast cancer, and thus flatten out the importance of total fat. In this study, no relation to total fat was found (after exclusion of reporting bias), whereas a high intake of n-3 and n-6 polyunsaturated fatty acids (and perhaps monounsaturated fatty acids) was related to a lower risk of breast cancer, especially in premenopausal women. The intakes of saturated fatty acids and trans-fatty acids were not associated with breast cancer. It seems that the types of fats considered beneficial against other major diseases, such as cardiovascular diseases and diabetes, may also decrease the occurrence of breast cancer.
A diet rich in olive oil has been related to the low breast cancer incidence in the Mediterranean countries, and monounsaturated fatty acids have been suggested to play a central role in this relationship. Rapeseed oil, which is the most consumed oil in Finland, consists of plenty of monounsaturated fatty acids as well as a-linoleic acid (n-3 series fatty acid). In Sweden, where fatty acid composition of the diet is quite similar, monounsaturated fatty acids decreased and polyunsaturated fatty acids increased the risk of breast cancer in a cohort study of 674 breast cancer cases aged 40 to 76 (Wolk et al. 1998). Although oil consumption has become more common in Finland since the 1980s, mean consumption is still low, only 3.5 g per day (National Public Health Institute 1998). It is therefore possible that oil does not have an effect of its own but acts as an indicator of oil consumers' healthy lifestyles. It should also be noted that intakes of fatty acids are difficult to assess without biochemical indicators because dozens of new soft margarine products with various fatty acid compositions have been developed during the last years. It is difficult to separate the intakes of monounsaturated and polyunsaturated fatty acids due to reporting inaccuracies and due to difficulties in keeping food composition databases up-to-date with the fast product development.
Some studies have associated dairy products, such as fermented milk, cheese, and yogurt (Le et al. 1986, van't Veer et al. 1989a), with decreased risk of breast cancer. By altering the intestinal bacterial activity, these products may influence the formation and function of estrogenic compounds. Furthermore, micro-organisms provided by cultured dairy product may stimulate immunological activity in the host. It is also possible that the conjugated linoleic acid (CLA) found especially in dairy products and meat may prevent the development of mammary tumors (Ip and Scimeca 1997), even in small quantities (under 1% in the diet) (Ip et al. 1995). CLA, which is produced from linoleic acid by microflora in the rumen (Kepler et al. 1966), may be one explanation for the inverse association between milk consumption and breast cancer found in a Finnish cohort study (Knekt et al. 1996). Human studies specifically focusing on CLA and breast cancer have not yet been published. In the Kuopio Breast Cancer Study, high consumption of milk products tended to decrease postmenopausal breast cancer. However, high milk consumption increased the risk of breast cancer in premenopausal women. No relationship was found for sour milk, either in the premenopausal or postmenopausal women.
Lignans and isoflavonoids are food components that are converted to biologically active hormone-like substances by intestinal microflora and may have anticarcinogenic effects (Adlercreutz et al. 1992). Rye bread, which is the most important contributor of fiber to the Finnish diet, also contains high amounts of lignans (enterolactone). However, no association was found between rye products and risk of breast cancer in this study, although the mean daily consumption of rye product varied from 34 g to 120 g in quintiles. Misclassification of rye products was high, apparently because of the large variation in the bread types consumed in Finland. Serum enterolactone concentration has been analyzed in the Kuopio Breast Cancer Study, and the results will be presented later. The only human study published thus far showed that high urinary excretion of enterolactone and equol predicts a lower risk of breast cancer compared to that of women with low excretion levels (Ingram et al. 1997).
High consumption of fruit and vegetables rich in fiber and vitamins has been found to decrease the risk of all cancers. The association, however, has been more consistent for cancers of the lung and stomach than for hormone-related cancers (Steinmetz and Potter 1996). A cohort study in the United Kingdom found that after a 17-year follow-up the all-cause mortality of 11,000 vegetarians and health-conscious subjects was only one half that of the general population (Key et al. 1996). However, no decrease was found in breast cancer mortality. Most vitamin studies have focused on the possible protective effects of beta-carotene. Many of them, however, have been too small to detect any differences (see Kohlmeier and Mendez 1997). In this study, neither fruit nor vegetables were associated with the risk of breast cancer, but the increased intake of beta-carotene decreased the risk of postmenopausal breast cancer (Table 8). The associations between vitamins and breast cancer did not change when the dietary supplements were added to the nutrient intake. It may be that the protective effects of supplements exist only in a situation where the antioxidant intake is low, or consumption has been continued for a long time (Blot et al. 1993). In this study, it was impossible to separate subjects who had recently begun to use supplements from long-term users. It is also possible that beta-carotene may act as a marker for other potential factors, such as other carotenoids, flavonoids, phenols, or plant sterols.
The validity of nutrient intakes measured by FFQ was compared between the Nurses' Health Study (Willett et al. 1985), the ATBC Study (Pietinen et al. 1988), and the Kuopio Breast Cancer Study (Table 9). It seems that the results of the present Validation Study (II) were comparable to the validation study which included 173 female nurses (Willett et al. 1985), whereas the 168 Finnish men had slightly better correlation coefficients, especially for saturated fatty acids, fiber, and vitamin E (Pietinen et al. 1988). Women are more health-conscious than men, and thus tend to misreport their diet by responding in a manner better matching social norms and beliefs (Hebert et al. 1997).
The Validation Study (II), however, showed that it may be impossible to detect associations between rye products, total fat, saturated fatty acids, trans-fatty acids, vitamin C, and risk of breast cancer.
Table 9. Validity of the food frequency questionnaire against the diet records in the Nurses' Health Study (Willett et al. 1985), in the ATBC Study (Pietinen et al. 1988), and in the Kuopio Breast Cancer Study. Nutrient intakes adjusted for total energy intake.
|Nutrient||Nurses' Health Study||ATBC study||Kuopio Breast Cancer Study|
|Saturated fatty acids||0.49||0.62||0.49|
|Monounsaturated fatty acids||-||0.38||0.36|
|Polyunsaturated fatty acids||0.42||0.69||0.63|
|n-3 fatty acids||-||-||0.42|
|n-6 fatty acids||-||-||0.61|
1 Without supplements
Ecological and animal studies have found a strong relationship between low selenium intake and cancers of the breast and colon (Schrauzer et al. 1977, Willett et al. 1991). In case-control and cohort studies, selenium deficiency is more related to cancers of the digestive tract and lung than to breast cancer (Willett et al. 1991).
In this study, selenium status was analyzed in toenails, which are considered as appropriate material for assessing selenium intake over 6 to 12 months (Longnecker et al. 1993). Selenium was not associated with premenopausal breast cancer. In postmenopausal women, the relative risk of breast cancer was below unity in all three upper quintiles, but the findings were not statistically significant. This result was similar to that of a Finnish cohort study carried out between 1968 and 1972. In that study, the relative risk for the highest four quintiles of selenium intake was 0.52 compared to the lowest, but it was not statistically significant (Knekt et al. 1990). Low toenail selenium concentration was not associated with risk of breast cancer in a large cohort study in the United States (Hunter et al. 1990), in the Netherlands (van den Brandt et al. 1994), nor in a multicenter case-control study (van't Veer et al. 1996). Furthermore, the multicenter study found no interaction between toenail selenium and other antioxidants in adipose tissue (beta-carotene and alpha-tocopherol). Our results showed that the intake of retinol, beta-carotene, vitamin A, and vitamin C, used as covariates in the model, did not change the association between toenail selenium and breast cancer.
The most important question in the Selenium Study (V) is whether the selenium supplementation through fertilizers has changed the ranking of subjects by their selenium intake. It was possible to examine the question in the ATBC Study including 29,000 men (Hartman et al. 1998). No notable difference in ranking (r=0.83) was observed when selenium intake was calculated from the same food consumption data using a database based on pre-supplementation values, and then using that based on post-supplementation values. Another important question is the time frame between the recruitment of the case and the matched population control that was meant to be limited to one to two months. The interval was longer for a few subjects, but never more than one year. The time of recruitment was controlled for in the analyses.
There is a consensus based on epidemiological studies that more than two to three alcoholic drinks per day increase risk of breast cancer (Howe et al. 1991b, Smith-Warner et al. 1998). In the present study, the current alcohol consumption was not related to the risk of premenopausal or postmenopausal breast cancer. The alcohol consumption was, however, quite low, because the premenopausal women consumed, on average, only two drinks and postmenopausal women about one drink per week, and because 30% and over 50% of them were categorized as abstainers, respectively. The proportion of abstainers was quite high, but did not differ from the results of a preanalysis of lifetime alcohol consumption by birth cohorts (35- to 76-year-old subjects in the United States) (Russell et al. 1997). In that study, about 40% of subjects were abstainers in all birth cohorts through the decades. It should also be noted that variation in social norms of alcohol consumption affects differently cohorts born at different times (Lemmens 1998). Postmenopausal women in Finland belong to a female generation that has never consumed much alcohol. In the 1982 Dietary Survey of Finnish adults, the 55- to 64-year-old women consumed only 3 g alcohol per week, whereas the 25- to 34-year-old women consumed 17 g per week (Kleemola et al. 1994); the values were 10 g and 25 g per week in the 1992 Survey, respectively (National Public Health Institute 1998). Furthermore, it is possible that older women report their drinking habits more inconsistently than do younger women, which tendency was also observed in this study.
Consumption of coffee and alcohol seem to be negatively related to each other (Le Marchand et al. 1989, La Vecchia et al. 1992, Männistö et al. 1996). Therefore, it has been suggested that coffee and alcohol may be competitors in the diet. Coffee may affect estrogen metabolism by increasing sex hormone-binding globulin, and thus decrease the risk of breast cancer (Ferrini and Barrett-Connor 1996). In a cohort study in Norway, including 14,593 premenopausal women, high coffee consumption decreased the risk of breast cancer in lean women (BMI<24) but increased the risk in others (Vatten et al. 1990b). In Finland, coffee consumption is one of the highest in the world, on average over 400 g per day for women (National Public Health Institute 1998). This study showed a decreased risk of breast cancer among postmenopausal women drinking over 480 g coffee a day.
A challenge for breast cancer studies is to define the most sensitive period in a woman's lifetime in the development of breast cancer. Food frequency questionnaires typically evaluate dietary habits over the preceding 12 months, which may be totally irrelevant in examining a disease with a long latency period. As a biological explanation, it has even been speculated that intrauterine factors may advance mutations in the mammary gland (Michels et al. 1996). The most powerful biological explanation, however, is related to the hypothesis that the period between menarche and the first full-term pregnancy is particularly important because of changes in the breast tissue (Moolgavkar et al. 1980, MacMahon 1993, Colditz 1995).
It has been suggested that lifetime alcohol consumption may predict the alcohol exposure more accurately than would current alcohol consumption (Longnecker et al. 1995). The lifetime alcohol consumption questionnaire (AQ) was used in the Kuopio Breast Cancer Study to obtain information on alcohol consumption during different times (age of first use, before the age of 30, and cumulative lifetime consumption). In contrast to a study in the United States (Longnecker et al. 1995), no evidence on an association between past alcohol consumption and risk of breast cancer was found in this Finnish population.
The methods for assessing lifetime alcohol consumption have been reasonably good in reliability, between 0.67 and 0.90 (Lemmens et al. 1997). Although the validity for absolute amounts has been poor, the relative ranking has been reasonably stable, as shown by the validity for current alcohol consumption (between r=0.63 to 0.73) (Feunekes et al. 1999). It is, however, possible that all methods consistently measure alcohol consumption poorly, for example, with similar under-reporting. One study has examined the reporting of the initiation of alcohol and tobacco use among school children. When the questionnaire was repeated three times in 1989, 1991, and 1994, it was found that the older the students became, the older they reported the age when they started to drink alcohol (Engels et al. 1997).
In the Validation Study (II), the reliability of the current alcohol consumption was 0.64, and the validity between the first FFQ and the 14-day diet record was 0.80. Almost all subjects (95%) were categorized into the same or adjacent alcohol quintile by means of the FFQ and the diet records. The validity of the current alcohol consumption was also assessed by comparison of the first FFQ and the AQ, with the correlation was found to be higher for premenopausal women (r=0.80) than for postmenopausal women (r=0.40). In all, 64% of premenopausal and 70% of postmenopausal women were classified into the same category by the FFQ and the AQ. In general, the interview-based AQ yielded higher alcohol consumption figures than did the self-administered FFQ. The complexity of the lifetime alcohol consumption questionnaire may be lessened if it is administered by trained interviewers (Lemmens 1998).
Current alcohol consumption may influence the reporting of past alcohol consumption (Swanson et al. 1997), especially if the subject's drinking habits have changed many times (Lemmens 1998). In this study, the correlation between current alcohol consumption and lifetime alcohol consumption was high (r=0.74).
The variation in diet is often too narrow in studies within one country compared to the worldwide range. Thus, it is possible that some potential risk or protective factors for a disease cannot be recognized. Multicenter studies and pooling projects have been initiated to combine the advantages of individual and population-based studies. One example is the European Prospective Investigation on Cancer and Nutrition (EPIC), a very large cohort study involving almost half a million participants from nine Western European countries. The pilot phase concerning the accuracy of dietary assessment methods in each study center has been reported (Kaaks et al. 1997). The second example is the International Collaborative Study of Breast and Colorectal Cancer, in which the Kuopio Breast Cancer Study is one of the nine study centers (Boyle 1990).
Pooling of large cohort studies is a new approach to assess associations between diet and cancers. The Harvard Pooling Project includes nine large cohort studies, and it has produced, for example, a pooled analysis on the association between alcohol consumption and risk of breast cancer (Smith-Warner et al. 1998). New analyses concerning foods are in progress.
Some on-going intervention studies in humans are also of interest, for example, the Women's Health Initiative. That randomized trial including 48,800 women aged 50 to 79 was begun to determine whether a low fat diet decreases occurrence of breast cancer (Self et al. 1988). The last subjects of the trial were recruited during the summer of 1998.
The methods of nutritional epidemiology need to be developed by improvement in study hypotheses, study designs, and contribution of statistical methods, such as new models of measurement error (Carroll et al. 1995). The challenge is to understand and estimate better the problems in dietary assessment methods, for example, variation between and within individuals. Study designs may be improved by use of more than one method to assess the same exposure. In addition, it would be beneficial to find more biochemical indicators to estimate nutritional status. This study used, two different kinds of methods to assess alcohol consumption, two control groups to examine reporting bias under the threat of disease, and toenail selenium as a biochemical indicators of selenium status.
More information is needed on the effects of the different stages of disease on diet and diet reporting, such as presymptomatic illness, the threat of disease, and diagnosed breast tumor (changes in metabolism caused by disease or treatment and intentional changes in dietary habits in an attempt to improve prognosis). It is also unknown how much of the changes in reporting are truthful and how much reflect merely the wish for better dietary habits. The role should also be clarified for different kinds of fatty acids and phytoestrogens, for weight gain and for factors related to abdominal obesity, as well as for the interaction between genes and environmental factors in the etiology of breast cancer. Moreover, better knowledge is needed of the most sensitive periods in a woman's lifetime in the development of the breast.
This dissertation was part of the Kuopio Breast Cancer Study, a case-control study carried out in eastern Finland. The main objective of the diet study was to examine the possible role of body-size indicators, dietary factors and alcohol consumption in the development of breast cancer. Special attention was also paid to the study design and dietary assessment methods.
Weight and body mass index were not associated with risk of breast cancer. However, waist-to-hip ratio, which indicates body fat distribution more accurately, was a strong risk factor. Those premenopausal women categorized into the highest quintile of waist-to-hip ratio had a 4.6-fold and postmenopausal women a 2.6-fold risk of breast cancer compared to women in the lowest category. Further, in postmenopausal women, high body fat percent was associated with an increased risk of breast cancer.
The quality of dietary data was assessed by comparing the food frequency questionnaire (FFQ) to the 14-day diet record and by assessing recall bias under the threat of disease. The validity and reproducibility of the questionnaire were comparable to those in previous studies. This study showed that recall bias may distort the interpretation of results between dietary factors and disease in case-control studies. Those premenopausal women who were under the threat of breast cancer tended to misreport their consumption of milk products and tea. Recall bias was also related to the intake of fat and vitamins. In postmenopausal women, recall bias seemed to be associated with the reporting of milk products. It has been suggested that the recall bias of case-control studies may explain the discrepancies in results between case-control and cohort studies. The interpretation of results can be improved by using different types of methods to assess the same dietary factors.
Dietary factors did not play an important role in the occurrence of breast cancer in this study. However, a diet rich in milk increased, whereas high consumption of poultry decreased the risk of premenopausal breast cancer. Consumption of oil was negatively, and that of cream positively associated with risk of breast cancer in postmenopausal women. Suggestive evidence was found for butter, milk, and coffee. Among nutrients, high intakes of n-3 and n-6 polyunsaturated fatty acids and vitamin E were related to decreased risk of premenopausal breast cancer, while beta-carotene decreased the risk in postmenopausal women. This study supports the evidence that the quality of fat may be more important than total fat intake in the development of breast cancer. It is possible that dietary factors have independent effects, but they can also be merely indicators of a healthy lifestyle.
Current and past alcohol consumption were not associated with premenopausal or postmenopausal breast cancer among women with generally low alcohol consumption. When the validity of current alcohol consumption was assessed by comparing the food frequency questionnaire to the lifetime alcohol consumption questionnaire, it was found that older women were somewhat more inaccurate in their reporting than were younger women.
No association was found between toenail selenium concentration and risk of breast cancer, and this result was not changed by adjusting for intake of beta-carotene, retinol, vitamin E, and vitamin C.
Conclusion. Epidemiological studies furnish information on healthy dietary habits. The aim of this study was to find means for decreasing risk of breast cancer. These include maintenance of normal body size throughout one's lifetime and avoidance of high abdominal fat and body fat percent. High intake of fatty acids such as n-3 and n-6 polyunsaturated fatty acids that are generally considered healthy, certain vitamins, and the consumption of oil were also related to decreased risk of breast cancer. On the other hand, women do not need to be worried about one to three alcoholic drinks per week, although level of alcohol consumption was too low to exclude increased risk with high regular consumption. These findings are in line with general recommendations to lower the overall cancer incidence.