The Kuopio Breast Cancer Study is a multi-disciplinary cooperative project conducted by different departments of the University of Kuopio, Kuopio University Hospital, and the Department of Nutrition of the National Public Health Institute in Helsinki. The subjects of the project included all women who were referred to Kuopio University Hospital (North-Savo Health Care District) for breast examination between April 1990 and December 1995. The catchment area of the hospital is presented in Figure 2.
Figure 2. Catchment area of Kuopio University Hospital.
The project started as a case-control study, but all breast cancer cases could be followed up because they make their annual visits to Kuopio University Hospital. The long-term aim of this project is to decrease the incidence of and mortality from breast cancer in eastern Finland. The focus is on risk and protective factors for breast cancer, pathogenesis, diagnosis at an early stage of disease, estimation of prognosis, and improvement in treatment.
The Kuopio Breast Cancer Study follows the protocol of the International Collaborative Study of Breast and Colorectal Cancer coordinated by the European Institute of Oncology in Milan (the collaborative study was initiated as a SEARCH program in the International Agency for Research on Cancer). The collaborative study is based on the assumption that breast cancer and colorectal cancer may have common risk factors, particularly in the diet, such as alcohol, fat, cholesterol, and vitamin A (Boyle 1990). Study centers of the breast cancer study are situated in Canada, Finland, Greece, Ireland, Italy, Russia, Slovakia, Spain, and Switzerland. Each of the centers must collect at least 400 breast cancer cases (Boyle 1989). The multicenter study expands the variation in breast cancer incidence and in exposure variables compared to that of a single case-control study.
The diet study of the Kuopio Breast Cancer Study concentrates on associations between dietary factors, body-size indicators, and risk of breast cancer. In this dissertation, five of these studies are presented: Body-size Study (I), Validation Study (II), Dietary Study (III), Alcohol Study (IV) and Selenium Study (V). All studies are based on the original case-control study design.
Recruitment protocol of the Kuopio Breast Cancer Study
The recruitment protocol of subjects for the study in Kuopio University Hospital is shown in Figure 3. All those 1,919 women who had any suspected breast disease and who lived in the catchment area of the hospital during the study period from 1990 to 1995 were referred for further examination. All women were diagnosed and treated in Kuopio University Hospital. Originally, the subjects were referred to the hospital by a physician because of a suspected breast lump or breast symptom. Of these women, 516 (27%) were finally diagnosed with breast cancer, 717 (37%) had benign breast disease, and 686 (36%) were diagnosed as healthy (in terms of their breasts) after examination (later referred to as referral controls). Women with benign breast disease were excluded from Studies I-V. The clinical examination and interview of the subjects were carried out before their diagnosis was confirmed. This means that all subjects, both those with cancer and the referral controls, were similarly interviewed, and the diagnosis or treatment of cancer had no influence on the results.
Figure 3. Recruitment protocol of the Kuopio Breast Cancer Study.
All cases (n=516, age range from 23 to 91) had newly diagnosed, histologically confirmed breast cancer. Subjects who had a previous history of cancer in the past five years were ineligible. Subjects were asked to participate in the Kuopio Breast Cancer Study at their first visit to the hospital. Only twelve of the women who were later diagnosed with breast cancer refused to participate. The diet study included cases between 25 and 75 years of age according to the protocol of the International Collaborative Study of Breast and Colorectal Cancer (one case was under 25, and 76 cases were over 75). The subjects should also have had the ability to participate in an interview of sufficient quality, and thus the age of 75 was set as the upper limit. The diet study began in October 1990, and thus the first 56 breast cancer cases (recruited between April 1990 and September 1990) did not furnish any information on diet. The data of the diet study thus included 383 breast cancer cases aged between 25 and 75.
Referral controls (n=686) were women who were referred to the hospital because of a suspected breast lump or breast symptom, but who were subsequently diagnosed as healthy (as to their breasts). After the exclusions (53 referral controls under 25 or over 75, and 85 women recruited during April 1990 and September 1990 and for whom no information on diet was available), the data of the diet study included 548 referral controls 25 to75 years old.
Completeness of patient catchment
Our recruitment protocol missed 51 breast cancer cases and 5 referral controls within the hospital. These all were private patients who did not enter the hospital by the standard procedure. These missed subjects were recognized when clinical and pathology records of the hospital were compared to the data of the Kuopio Breast Cancer Study. Furthermore, eleven breast cancer cases and four referral controls were missed during the nurses' one-month strike in 1995.
The completeness of this data set was also tested by comparing the confirmed breast cancer cases (516+51+11=578) with the Finnish Cancer Registry. The comparison showed that 96% (578/604) of breast cancer cases who lived in the catchment area of Kuopio University Hospital were referred to the hospital; 26 breast cancer cases had been treated elsewhere.
In conclusion, the study material represents well the breast cancer cases in the catchment area of Kuopio University Hospital, since only 15% of eligible cases were missed (516 vs. 604).
One population control was selected for each breast cancer case after the cancer diagnosis had been confirmed. Controls should reflect the exposure distribution of the population that generates those cases (Ahlbom and Norell 1990). Therefore, a randomly selected group of subjects was drawn from the Finnish National Population Register covering the same catchment area (Figure 3). The population controls were individually matched with the breast cancer cases by the area of residence (urban/rural) and age (within ±5 years). The proportion of case-control pairs who lived in the same municipality was 87%, and the remaining pairs lived as near as possible, to keep the urban-urban or rural-rural matching.
In all, 663 population controls were invited to the hospital to be interviewed in parallel with the cases. Although the matching ratio was decided upon as 1:1, there were more population controls than breast cancer cases because one to four controls were selected for each case in the Validation Study (II) to ensure its completion on schedule. The participation rate of the population controls was 72% (64% during the Validation Study, 76% otherwise). The interval between the interview of a case and that of the corresponding population control was one to two months.
In Study I, the associations between body-size indicators, i.e., anthropometric measurements, and risk of breast cancer were examined according to the menopausal status of the women and estrogen-receptor status of the tumors. The subjects were 25 to 75-year-old women (339 breast cancer cases and 420 population controls) who were recruited to the Kuopio Breast Cancer Study during the first four years, from April 1990 to December 1994 (Figure 4). Seven cases and three population controls were excluded because of severe illness or inability to cooperate. Four breast cancer cases refused to participate in the study. The final data included 328 breast cancer cases and 417 population controls.
The body-size indicators with a value missing for less than 10 subjects were height, weight, body mass index, waist-to-hip ratio, or weight loss at age 22 to 44. Indicators with 10 to 20 missing values were body fat percent, fat weight, and lean body weight. Weight gain had a large number of missing values, especially in postmenopausal women (133). Estrogen-receptor status was not available for the tumors of 32 premenopausal and 52 postmenopausal women.
The mean age of the breast cancer cases was 54 years (SD 11), and 62% of them lived in towns. The corresponding values for the population controls were 53 years (SD 11) and 65%.
Figure 4. Recruitment time axis and final number of subjects in Studies I-V.
The first 250 population controls of the diet study, recruited between October 1990 and February 1991, were invited to the Validation Study (II), in which the reproducibility and validity of the FFQ designed for the diet study were investigated (Figure 4).
The first food frequency questionnaire (FFQ1) was sent to the subjects along with the invitation letter to attend the examination and interview. The purpose of the study and the importance of obtaining healthy population controls were explained in the letter. The controls were not paid for their participation, but a free health examination with laboratory analyses and anthropometric measurements was offered. Of the 250 eligible subjects, 198 visited the hospital and filled in the FFQ1 (79%), of these 188 agreed to continue, and 167 returned the first 7-day diet record (Figure 5). The second food frequency questionnaire (FFQ2) was sent to the subjects after an interval of three months (n=167); 160 of them returned it as well as the repeated 7-day diet record two weeks later. In analyses, the two 7-day diet records were combined into a 14-day diet record.
Eight women were excluded because of pregnancy, breast-feeding, fasting (more than two days), or incompletely filled in forms. As a result, the final Validation Study (II) included 152 population controls. Three women had only 13 diet-record days and two only 12 days because of fasting, illness, or unknown reasons. The aim was to collect over 100 subjects, which is a reasonable size for validation studies (Willett and Lenart 1998) .
Figure 5. Design of the Validation Study (II).
Study II was divided into two parts: the FFQ1 was compared with the FFQ2 in the reproducibility part, and both the FFQ1 and the FFQ2 with the 14-day diet record in the validation part. The interval of three months between the questionnaires was chosen because a few months' period between the original and the repeated questionnaire is recommended if the questionnaire concerns the intake over the past year. If the interval is too short, subjects may also remember their previous answers, which increases reproducibility artificially. On the other hand, with a long interval, reproducibility reflects not only the repeatability of the questionnaire but also the true changes in diet (Willett and Lenart 1998). The seasonal variation in the Finnish diet seems also to be covered, because the data were collected during 11 months (from October to August).
The mean age of the 152 subjects was 51 years (SD 9) and more than half of them (64%) lived in an urban area. The mean age of the original 250 population controls was 50 years (SD 9) with 62% of them living in an urban area.
The main aim of Study III was to examine associations between dietary factors (foods and nutrients such as fat, fiber, and vitamins) and risk of breast cancer. Furthermore, recall bias due to worry caused by the threat of disease was also examined. Because all subjects were diagnosed after the interview, it was possible to assess whether the self-reporting of diet between two control groups, the population controls and the referral controls, varied.
Study III included all breast cancer cases (n=383), population controls (n=663), and referral controls (n=548) aged 25 to 75 who were invited to the Kuopio Breast Cancer Study after October 1990 when the diet study began (Table 1, Figure 4). Exclusion criteria for participation in the Dietary Study (III) were as follows: refusal, pregnancy, or lactation during the Validation Study (II), inability to cooperate, other severe disease, or an unacceptable food frequency questionnaire (Table 1). The FFQ was not accepted if the energy intake per day was under 800 kcal, if more than 10 food items were skipped, if more than five food items had "unreasonably" high frequencies or if only one or two frequency categories were used repeatedly. These criteria had been modified from those reported by Block et al. (1990). Half of the population controls (n=90) who refused to participate were those who did not want to participate in the Validation Study (II) at the beginning of the project. Finally, the data of the Dietary Study included 310 breast cancer cases, 454 population controls, and 506 referral controls (81%, 68% and 92%, respectively, of enrolled 25- to 75-year-old women).
Table 1. Participation (n) and reasons for exclusions in Study III.
|Enrolled (25-75 year-old women)||383||663||548|
|Not willing to participate||12||188||2|
| Pregnancy or lactation during the
Validation Study (II)
|Controls who developed breast cancer||-||2||6|
|Unable to cooperate or too ill||25||8||9|
|Food frequency questionnaire missing or unacceptable||36||6||25|
|Final number of subjects in the Dietary
The association between lifetime alcohol consumption (current consumption, consumption at age of first use, cumulative consumption before age 30, and cumulative lifetime alcohol consumption) and risk of breast cancer was examined in Study IV. Two different kinds of measurements for alcohol consumption were used: the self-administered FFQ and an interview-based Lifetime Alcohol Consumption Questionnaire (AQ) designed for this study. The validity of the current alcohol consumption rates was assessed by comparing the results of both methods.
Study IV was based on the breast cancer cases (n=310) and the population controls (n=454) of the Dietary Study (III) (Figure 4). However, nine breast cancer cases and eleven population controls were excluded because of missing AQs. Thus, the final data included 301 breast cancer cases and 443 population controls.
The Selenium Study (V) was designed to assess the association between toenail selenium concentration and risk of breast cancer.
Study V was based on the breast cancer cases (n=310) and the population controls (n=454) of the Dietary Study (III) (Figure 4). Some exclusions, however, were made (21 breast cancer cases and 21 population controls) because of missing or insufficient samples, or unreasonably high selenium concentration (>3SD), which usually originates from selenium-containing anti-dandruff shampoo. The final data of Study V included 289 breast cancer cases and 433 population controls.
The main characteristics of the breast cancer cases, population controls and referral controls of the Dietary Study (III) are summarized in Table 2. The distributions by age, area of residence, and education were similar between the cases and the population controls both in premenopausal and in postmenopausal women. The first-degree relatives (mother and sisters) of the breast cancer cases had more diagnosed breast tumors than did relatives of the population controls (13% vs. 5% for premenopausal and 9% vs. 5% for postmenopausal women). The referral controls were the youngest both among premenopausal and among postmenopausal women, had the highest education among the premenopausal women and had the highest percentage of breast cancer in their family among the postmenopausal women.
The Kuopio Breast Cancer Study was approved by the Joint Ethics Committee of the University of Kuopio and Kuopio University Hospital. Participation was based on written consent by which subjects gave permission to use their medical records, and to monitor them for any future health conditions.
Table 2. Selected characteristics of the subjects in the Dietary Study (III).
|Number of subjects||119||178||324|
|More education than comprehensive school (%)||
|Family history of breast cancer (%)||
|Number of subjects||191||276||182|
|More education than comprehensive school (%)||
|Family history of breast cancer (%)||
The semi-quantitative food frequency questionnaire (FFQ) was developed for the Kuopio Breast Cancer Study from the longer questionnaire of the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (the ATBC Study) in Finland (Pietinen et al. 1988). The FFQ was designed to assess the subject's entire diet during the year before diagnosis. It included separate 110 food items, mixed dishes, and alcoholic beverages. Frequency of use was described by nine categories from "never" to "six or more times a day". The usual portion size was predefined in the FFQ, for example, a slice of bread, a banana, or a cup of coffee. Subjects were also asked to assess the frequency and quantity of dietary supplements used, and they were allowed to report additional foods consumed frequently but not listed in the FFQ.
For the subjects who reported using butter in cooking and usually ate homemade food (29%), butter was assumed to be the cooking fat in all recipes. For the other subjects, the cooking fat was assumed to consist of butter, margarine, and oil in equal proportions. This assumption was made, because a large proportion of Finnish employees have their lunch (the daily main meal) outside the home and are unaware of the types or amounts of fats used in cooking. The type of salad dressing usually chosen (oil and vinegar, mayonnaise, juices, or low-caloric) was also included individually in the analyses. Further, a limit for "unreasonably" high frequency was determined for 60% of food items or mixed dishes in order to exclude impossibly high consumption. In the analyses, unreasonably high frequencies were converted to one above the defined maximum so that the ranking between the subjects was still maintained. For example, "yogurt six or more times a day" was not accepted and was converted to "twice a day" (the highest reasonable frequency for yogurt was determined as "once a day"). However, less than 1% of frequencies were converted during the analyses and these conversions concerned 15% of subjects.
The FFQ was sent to the subjects along with a letter inviting them to a breast examination. The questionnaire was completed at home, and returned to the study nurse during the interview visit.
The subjects of the Validation Study (II) were asked to report all foods and beverages they consumed for two specified 7-day periods. The amounts of foods were estimated by a picture booklet including 126 color photographs of foods in varying portion sizes (Haapa et al. 1985) or by common household measurements whichever was more appropriate. Dietary supplements were not included in the diet record.
The lifetime alcohol consumption questionnaire (AQ) was completed together with the study nurse during the interview visit. The subjects were allowed freely to define any different drinking periods (age at start and end of drinking). Alcohol drinkers were especially asked to recall all special events in their life that were likely to be associated with changes in alcohol consumption, for example, leaving the parental home, pregnancy, divorce, and unemployment. The maximum number of drinking periods reported (when the subject used to drink alcohol in a constant pattern) was between two and five depending on alcoholic beverage. The subjects were asked to describe the number of predefined portions and the frequency of use for all alcoholic beverages consumed (beer, long-drink, wine, fortified wine and spirits) during each interval.
The following variables were calculated from the AQ: current consumption of alcohol (ethanol g/week), age at first use, alcohol consumption during the first year alcohol was consumed (g/week), total cumulative consumption before age of 30 (cumulative sum of all ethanol) and total lifetime alcohol consumption (cumulative sum). In this dissertation, the terms "alcohol consumption" and "alcohol intake" refer to intake as absolute alcohol.
All food frequency questionnaires and diet records were entered into the mainframe computer at the National Public Health Institute by students of nutrition science. The final approval was given by the nutritionist who used the exclusion criteria mentioned in sections 5.2.4.
Dietary data were converted into individual daily food consumption and nutrient intakes by the software and food composition database (FINELI) of the National Public Health Institute. This database, which is also the official Finnish National Food Composition Database, includes about 1,600 of the most important food items and mixed dishes in the Finnish diet, and over 200 dietary factors (Ovaskainen et al. 1996). The supplement database includes 778 dietary supplements, of which 75% had a known nutritional content (Kaartinen et al. 1997).
The food groups were selected on the basis of importance as macronutrient and micronutrient sources in the Finnish diet, or because of relation to reported risk of breast cancer. Thus, 17 of 32 food groups designed for the ATBC Study were chosen for this study. These groups are rye products, vegetables, fruit and berries, butter, soft margarine, oil, total milk, low-fat milk, sour milk, cream, cheese, beef and pork, poultry, fish, coffee, tea, and sugar. The formation of food groups has been explained in detail in Study III. For example, the group "Milk" consisted of all milk consumed in the diet independently of whether it was recorded as a milk item or derived from mixed dishes, e.g., from porridge made with milk.
Of the nutrients, the intakes were calculated of fat (total fat and fatty acids), fiber (total fiber and water-insoluble noncellulosic polysaccharide), vitamins (retinol, beta-carotene, vitamins E and C, with supplements), and alcohol. These nutrients were selected because earlier studies have suggested that they may be related to the development of breast cancer.
An envelope with instructions to cut clippings from the toenails was mailed to the subjects with the invitation letter to the hospital. They were asked to return the sample to the study nurse.
The toenails were analyzed for selenium at the National Public Health Institute. The nail samples were cleaned with 1% sodium dodecylsulfate for one hour. After one minute of ultrasonication, the samples were rinsed thoroughly with demineralized water and dried. The selenium concentration of the samples was analyzed by acid-digestion fluorimetry by use of 2,3-diaminonaphthalene to produce the piaselenole (Alfthan 1984). A standard reference material (BCR, Bovine Liver 185, Community Bureau of References, Brussels) was analyzed in each series (n=23), the mean ± SD value being 0.47 ± 0.01 mg/kg compared with the certified value of 0.47 ± 0.01 mg/kg. All analyses were done by the same laboratory personnel who were blinded for the case-control status of the subjects.
During the personal interview visit, which lasted approximately one hour, the study nurse filled in a questionnaire on socioeconomic background, medical history, family history of breast cancer, reproductive factors, physical activity, smoking, and current alcohol consumption. The same study nurse interviewed the subjects during the whole study, with the exception of the Validation Study (II), in which the information was collected by a student of nutrition science.
Height and weight were measured, and body mass index was calculated as weight (kg)/[height (m)]2. Waist-to-hip ratio was measured by two measures of waist and hip circumferences. Waist circumference was measured midway between the lower rib margin and the iliac crest, and the hip circumference at the widest circumference over the greater trochanters. Body composition was measured by near-infrared interactance (FUTREX 5000; Futrex, Gaithersburg, MD, USA). The device transmits near-infrared light into the biceps of the dominant arm at a wavelength that allows fat to absorb the light and lean mass to reflect it back. The light absorption is measured to determine percent body fat, and from that were calculated the values of fat weight (kg) and lean weight (kg). The weakness of this method is that it analyzes only the amount of fat at the point of measurement (biceps of the arms). Near-infrared interactance and skinfold thickness are both useful field techniques, since they are inexpensive, safe, simple, and rapid. Near-infrared interactance was chosen in order to eliminate observer errors.
Because near-infrared interactance measures adiposity indirectly, the degree of error deserves extensive attention. No such data, especially from the perspective of epidemiological studies, exist. However, the errors related to indirect measurements of obesity imply that its health effects may be underestimated in epidemiological studies (Willett 1998).
The samples for estrogen-receptor analysis were rapidly frozen (within 15 min of operation) in liquid nitrogen. Tumor blocks were cut into 8-µm slices which were mounted on glass slides coated with the tissue adhesive provided in the ER-ICA kit. The estrogen receptors were assayed using an ER-ICA kit (Abbot, North Chicago, IL, USA) at the laboratory of the Department of Pathology of Kuopio University Hospital. The laboratory and the assay is submitted to an inter-laboratory quality control program (Labquality, Helsinki, Finland). The immune-staining was performed according to the manufacturer's instructions. This method utilizes a monoclonal antibody and the peroxidase-anti-peroxidase (PAP) technique for visualization of estrogen receptors in a frozen section. A positive control slide provided by Abbott (North Chicago, IL, USA) was used along with a negative control slide (frozen section of the sample without antibody).
Tumors with strong staining (ER) were coded as strong expression (ER++) and completely ER-negative tumors as ER-. Tumors with weak or moderate cytoplasmic expression of ER were coded as ER+.
One population control (or 1-4 in Study II) was selected for each breast cancer case by individual matching for age (±5 years) and area of residence (urban/rural) in order to have similar distribution of these variables in both groups. However, because the individual matching was quite permissive and was based exclusively on age and area of residence, group matching was used in the analyses.
Analysis of variance (likelihood ratio test) was carried out to compare the distributions of selected variables between the cases and the population controls (and the referral controls in Study III). The accepted level of type I error was p<0.05. The data were analyzed with the SAS statistical software package of the National Public Health Institute computing system (SAS Institute Inc. 1989).
Associations between exposures and risk of breast cancer were evaluated by logistic regression using odds ratios (OR) and corresponding 95% confidence intervals (95% CI) by quintiles. Body-size indicators, consumption of foods and alcohol, intakes of nutrients, and toenail selenium concentration were the explanatory variables in the analyses. The matching variables, age and area of residence (urban/rural), were included in all models (Ahlbom and Norell 1990). Further adjustment was made for known risk factors for breast cancer including age at menarche, age at first full-term pregnancy, use of oral contraceptives, use of postmenopausal estrogen replacement therapy, family history of breast cancer, history of benign breast disease, education, current alcohol intake (except for Study IV), smoking, physical activity, body mass index, and waist-to-hip ratio (except for Study II). In addition, Study V also included a factor for the year of recruitment, because the selenium level annually changed in Finland as a result of selenium supplementation through fertilizers, and factors for intakes of retinol, beta-carotene, vitamin E, and vitamin C. Further, when past alcohol consumption was examined, a factor for the duration between age at interview and the age at first alcohol use was added to the models, whereas total alcohol consumption before age 30 was adjusted for the duration between age at interview and age 30.
Before calculating odds ratios, the dietary factors were log-transformed to reduce skewness and improve normality as required by most statistical methods. The formula log(x+1) was used because all subjects did not consume each food item, and the non-transformed value would then have been zero. We also employed the residual method by Willett (1998) to make the dietary factors independent of total energy intake. The summary of log-transformation, energy adjustment, and factors included in the multivariate models is presented in Table 3.
The study design to examine recall bias using two control groups (the population controls and the referral controls) in the Dietary Study (III) has been reported in detail in Study III. In this dissertation, the terms "recall bias" and "reporting bias" are used synonymously.
The data were analyzed separately for premenopausal and postmenopausal women. Women who were over 50 and used postmenopausal estrogen replacement therapy were classified as postmenopausal; otherwise the self-reported menopausal status was used. Tumors were also classified according to estrogen-receptor (ER) status in the Body-size Study (I).
Certain additional statistical methods were employed in the Validation Study (II). Intraclass correlation coefficients were calculated to measure agreement of the food consumption and nutrient intakes based on the original FFQ1 and the repeated FFQ2. The intraclass correlation coefficient is defined as the ratio of between-person variation to total variation. Thus, high intraclass correlation implies low within-person variation. The Pearson correlation coefficient was used to assess correlation between the food consumption (or nutrient intake) in the FFQ1 and in the 14-day diet record. Within-person variation in FFQ measurements induces attenuation in the correlation coefficient. This was corrected using the formula:
rc = ro * (1+s2intra/s2 inter)1/2
where rc denotes the corrected correlation, ro the observed correlation, s2intra intraindividual variance, and s2inter interindividual variance for the FFQ measurements. Because epidemiological data are often analyzed categorically, for example as quartiles, or quintiles, the degree of misclassification needs to be measured. The proportion of subjects correctly categorized in the same or adjacent quintile of food consumption and nutrient intakes was calculated based on the FFQ1 and the 14-day diet record.
The Pearson correlation coefficients and the degree of misclassification were also determined when the consumption of alcohol was assessed on the basis of the FFQ1 and the AQ in Study IV. Furthermore, the validity between toenail selenium concentration and dietary selenium intake was evaluated by the Pearson correlation coefficient.
Table 3. Summary of log-transformation (Log), energy adjustment by residual method and the covariates in multivariate models (known risk factors for breast cancer and other factors) used in studies I-V.
|Exposure||Log||Residual method||KRF2||Other factors|
|Current alcohol consumption||(Yes)1||(Yes)1||Yes||
Body mass index
|Past alcohol consumption||No||No||Yes||
Body mass index
Time between age at interview and age at first alcohol use
Time between age at interview and age 30
|Toenail selenium||Yes||No||Yes||Body mass index
Year of recruitment
Intake of antioxidants
1 Since log-transformation and energy-adjustment did not affect the results for current alcohol consumption, only the results without these adjustments are presented.
2 Known risk factors.