IJE Advance Access published online on April 4, 2008
International Journal of Epidemiology, doi:10.1093/ije/dyn060
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Use of baseline and updated information on alcohol intake on risk for breast cancer: importance of latency
1Centre for Alcohol Research, National Institute of Public Health, Øster Farimagsgade 5A, 2nd floor, DK-1399 Copenhagen K, Denmark.
2Department of Biostatistics, Institute of Public Health, University of Copenhagen, Øster Farimagsgade 5, PO Box 2099, DK-1014 Copenhagen K, Denmark.
3Institute of Cancer Epidemiology, Danish Cancer Society, Strandboulevarden 49, DK-2100 Copenhagen Ø, Denmark.
*Corresponding author. Lau Caspar Thygesen, Centre for Alcohol Research, National Institute of Public Health, Øster Farimagsgade 5A, 2nd floor, DK-1399 Copenhagen K, Denmark. E-mail: lct{at}niph.dk
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
Background Alcohol intake has been shown to be associated with an increased risk for breast cancer. In the analysis of longitudinal prospective cohort studies, however, the analysis of repeated measurements of alcohol intake might not be straightforward.
Methods In this analysis of the Copenhagen City Heart Study, in which alcohol intake was measured four times, 9318 Danish women with no previous diagnosis of cancer were followed for breast cancer for 27 years, from 1976 to 2002. During follow-up, breast cancer was diagnosed in 476 women.
Results The association between alcohol intake at first measurement (baseline alcohol intake) and breast cancer was positive and approximately linear. When alcohol intake was updated during follow-up, no association was observed between breast cancer and alcohol intake. It is suggested that this difference in results may be attributable to long latency time between alcohol intake and breast cancer occurrence, because a markedly increased risk was estimated on the basis of direct lagging of risk time.
Conclusions Our results support the hypothesis that baseline alcohol intake is more strongly associated with breast cancer risk than updated intake, and we suggest that this is due to the long latency between alcohol intake and breast cancer.
Keywords Alcohol drinking, cohort study/longitudinal study, breast neoplasm
Accepted 28 February 2008
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
The association between alcohol intake and breast cancer has been studied in numerous case-control and cohort studies.1–3 Recent, large pooled studies suggest an increased risk even at low intake.4,5 It has been estimated that 4% of new breast cancer cases in developed countries are attributable to alcohol intake;4 in Italy, where the average alcohol intake is high, the proportion was 10.7%.6 Alcohol intake therefore, ranks as an important modifiable risk factor for breast cancer. In most previous prospective studies, only one baseline measurement of alcohol intake was used, on the assumption that alcohol measured at one time will predict an individual's intake over a longer period. Changes in drinking habits and random measurement error may, however, result in underestimation of the importance of alcohol intake if only baseline information is used.7,8 In some prospective cohort studies, like the Copenhagen City Heart Study,9 repeated measurements of alcohol intake during follow-up are available, so that alcohol intake can be updated to obtain the most accurate information during follow-up.
In epidemiological studies, information on alcohol intake should optimally comprise the interval during which alcohol influences carcinogenesis. In all cancers, there is a long, highly variable lag between exposure and diagnosis.10 In studies of long-term lifestyle factors like alcohol intake, the time of exposure is not clear, because it is not known when exposure reaches the critical level at which the disease process starts.11 If alcohol intake influences early-stage carcinogenesis, the effect of time since exposure should be an increasing relative rate,12,13 and baseline measurements of alcohol might be more appropriate than updated information. In contrast, if alcohol intake influences carcinogenesis at a penultimate stage, updated information on alcohol intake might be more appropriate, and baseline measurements should be ignored. In prospective studies with repeated measures of alcohol intake, latency can be evaluated directly by lagging risk time.
We therefore conducted a prospective cohort study with 27 years of follow-up to determine the influence of both baseline and updated alcohol intake on the risk for breast cancer and also the modifying influence of latency.
|
Materials and methods |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
Study design and population
The Copenhagen City Heart Study is a large prospective cohort study initiated in 1976 in which 19 698 individuals from central Copenhagen (Østerbro and Nørrebro) were invited to a health examination and to fill in a questionnaire concerning health behaviour. The sample was randomly selected by age strata from a population of 90 000 inhabitants of this well-defined area. A detailed description of the study has been published previously.9 At the first examination, 14 223 persons participated (72%), and the cohort was re-examined in 1981–83, 1991–94 and 2001–04, supplemented with younger participants. At all three follow-ups, participants were invited to undergo a new health examination and to fill in similar questionnaires on health behaviour. We call these four examinations ‘waves’ 1–4. Table 1 shows the number of participating women in the four waves.
|
Exposure measurements
In wave 1, the participants were asked to describe their intake of alcoholic drinks in a multiple-choice format. Drinks were classified as beverages containing 12.8 g of alcohol in a 33-cl bottle of beer, 11.0 g in a glass of wine and 14.0 g in a drink of liquor. The choices were ‘hardly ever/never’, ‘monthly’, ‘weekly’ or ‘daily’. If daily alcohol intake was reported, the average number of beverages per week was recorded. Supplementary, non-daily alcohol intake was estimated as previously described14 and was added up to estimate the aggregate intake of alcohol in g/day. In waves 2–4, the average numbers of type-specific beverages consumed per week were reported, which were then added up to the aggregate intake in g/day.
Measurements of potential confounders
We included several factors as potential confounders in the statistical model. Smoking habit was included in five categories (‘never’, ‘ex-smoker’ and ‘daily smoker’ of 1–14, 15–24 or 
25 g/day). The categories of physical activity we used were ‘sedentary’, ‘light’, ‘moderate’ and ‘heavy’. Body mass index was categorized as <20, 20–24, 25–29, 30–34 or 35 kg/m2 and educational level as 0–7, 8–11 or 12–24 years of education. Use of hormonal therapy was included as a binary variable, and menopausal status and age were categorized into six categories (pre-menopausal, <37 years of age, 37–44, 45–49, 50–54 and 55 years). Parity was included in four categories (nulliparous, one or two children, three children and four or more children). The main criterion for inclusion of confounders was that they had been shown previously to be associated with breast cancer risk. The potential confounders might have changed during follow-up, but, in order to obtain meaningful estimates, baseline alcohol intake was adjusted for confounders measured at baseline and updated alcohol intake was adjusted for updated confounders.
Measurement of outcome
By means of the personal identification number, which has been assigned to all residents of Denmark since April 1968, linkage was made with the National Central Person Registry and the Danish Cancer Registry. The vital status of each member of the cohort was determined in the National Central Person Registry. Information on cancer occurrence until December 31, 2002 was obtained through linkage to the files of the Danish Cancer Registry, which has collected information on all individuals in Denmark with a diagnosis of cancer since 1943. The registration is based on notification forms that are completed by hospital departments and practising physicians whenever cancer is diagnosed or found at autopsy, and whenever there are changes in an initial diagnosis. The cases recorded manually are supplemented by unrecorded cases found by computerized linkage to death certificate files and to the National Registry of Patients.15 Medically qualified personnel supervise the entire process. Ambiguous or contradictory information leads to queries in 
10% of the notifications received. Comprehensive evaluation has shown that the Registry is 95–99% complete and valid.16–19
We classified the breast cancers according to codes 170.0–170.5 of the modified Danish version of the International Classification of Diseases, 7th revision. Between the first examination and the end of 2002, the cohort experienced 476 primary breast cancers, of which 460 were histologically confirmed (431 carcinomas and 29 adenocarcinomas).
Statistical analyses
A Cox regression analysis was performed with SAS/STAT software version 8.220 and R software version 2.1.121 to estimate the hazard ratio (HR) of breast cancer by alcohol intake, while taking potential confounding variables into account. Because age was a strong predictor of breast cancer, age was used as the time scale with delayed entry, thereby adjusting for the effect of age on the risk estimates.
In our first analysis, we included baseline alcohol intake by ignoring the repeated measures and using only the baseline measurement. In this analysis, breast cancer incidence during follow-up was related to alcohol intake at the time each woman first participated. In a second analysis, we evaluated the association between the most recently reported alcohol intake and breast cancer, representing the short-term risk. We treated each interval between waves for each woman as one observation (subject interval). This approach is equivalent to the algorithm that all subject intervals are pooled as if the information recorded at each interval is a new observation for evaluating breast cancer risk during that interval by alcohol intake status at the start of the interval. This procedure is described thoroughly by Cupples and colleagues.22 We updated alcohol intake on the exact date of examination, thereby assuming no latency between alcohol intake and breast cancer.
Latency was evaluated in detail by relating updated alcohol intake to breast cancer incidence after a delay equal to the latency. Each subject interval then consisted of updated alcohol intake and risk time a number of years later, equal to the assumed latency. We thereby assumed that breast cancer occurring during the latency could not plausibly be related to the updated alcohol exposure. Latencies of 4, 8, 12, 16 and 20 years were used to compare the HR for different latencies.
Alcohol intake was categorized into <1.71 g/day (one drink/week), 1.71–12, 13–24, 25–48 and >48 g/day. To study the shape of the curve of the relation between alcohol intake (baseline and updated) and breast cancer further, we examined the possible non-linear relation between alcohol intake and breast cancer by using penalized cubic spline regression with approximately four degrees of freedom.23,24 Penalized cubic splines are a graphical representation of dose–response relations with no a priori assumptions about the shape of the curve.
We tested whether alcohol intake is significantly associated with breast cancer risk by testing whether all regression coefficients of alcohol intake jointly equalled zero by a Wald chi-square test with four degrees of freedom. Additionally, we used a trend test in which alcohol intake was included as a continuous variable in the model (Wald chi-square test with one degree of freedom). We give two-sided P-values and confidence intervals throughout this article. We examined plots of log(time) by log[-log(survival probability)] for alcohol intake and confounders, which should yield parallel curves if the assumption of proportional hazards is satisfied. These plots did not indicate violation of that assumption.
We recruited 9318 women with no diagnosed malignant disease before study entry. Women participated from their first examination to the date of cancer diagnosis (breast or any other cancer other than non-melanoma skin cancer), date of death, emigration or disappearance, or end of follow-up (December 31, 2002), whichever came first.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
Characteristics of the study population
The number of participating women in the four waves of the Copenhagen City Heart Study decreased during follow-up (Table 1). Women with higher alcohol intake were more likely to smoke more in waves 1–3 and to have a higher educational level in all four waves (Table 1). Alcohol intake was highest among women with a body mass index of 20–24.9 kg/m2 and higher among pre-menopausal than post-menopausal women. Nulliparous women and women with few children had higher alcohol intakes than women with more children.
The median alcohol intake decreased between wave 1 and wave 2 and increased between wave 2 and wave 4 (Table 1), while the mean alcohol intake increased between wave 1 and wave 4 (data not shown). The four repeated measurements of alcohol intake were correlated, with correlation coefficients of 0.49–0.64 for adjacent waves; the coefficient decreased for waves further apart in time (data not shown). The correlation between alcohol intake at the first and the last wave was the lowest, 0.39. The percentages of women in the same category of alcohol intake at adjacent waves were 50–55%; the percentages of women whose category decreased during follow-up were 18–25% and the percentages whose category increased were 22–28%. The percentages of women whose intake increased by two categories in adjacent waves were 4–6%, and the percentages of women whose intake decreased by two categories in adjacent waves were 2–5%.
Baseline and updated alcohol intake in relation to breast cancer
A positive relationship between baseline alcohol intake and breast cancer risk, after adjustment for confounders measured at baseline, was observed (Table 2). Increased risks were observed for women with alcohol intakes of 13–24 g/day (multivariate HR = 1.36; 95% CI: 1.01–1.81) and >48 g/day (HR = 4.64; 1.67–12.9). The trend test showed a significant association with alcohol intake (P < 0.001); no association with updated alcohol intake was observed (Table 2). After multivariate adjustment, both the overall test (P = 0.66) and the trend test (P = 0.13) yielded high P-values. For both baseline and updated alcohol intake, the risk estimates for almost all categories were attenuated after multivariate adjustment.
|
Examination of the association between alcohol intake and breast cancer with a penalized cubic spline regression curve (Figure 1) showed that baseline alcohol intake had a linear effect on the risk for breast cancer, whereas updated alcohol intake showed a threshold effect, with no increased risk for breast cancer at alcohol intakes below 40 g/day. The HR for baseline alcohol intake included as a continuous variable (10 g/day) on the risk for breast cancer was 1.20 (95% CI, 1.09–1.31) after adjustment for baseline confounders.
|
When both baseline and updated alcohol intake were included in the same model, the positive association between baseline alcohol intake and breast cancer risk was maintained (P overall = 0.04), with a significantly increased risk at >48 g/day (HR = 5.54; 95% CI: 1.45–21.2). The trend test was significant (P = 0.005), supporting the hypothesis that baseline alcohol intake had an effect even after adjustment for updated alcohol intake. Conversely, after adjustment for baseline alcohol intake, updated alcohol intake showed no association with breast cancer risk. Simultaneous adjustment for both baseline and updated confounders only marginally changed the estimates associated with baseline and updated alcohol intake.
The influence of baseline alcohol intake was similar for pre- and post-menopausal women. For updated alcohol intake, some divergence was seen between pre- and post-menopausal women, although none of the risk estimates was statistically significant (data not shown). Body mass index did not modify the influence of baseline or updated alcohol intake, and further adjustment for height did not affect the reported results (data not shown).
Latency
The estimates of the effect of updated alcohol intake on the risk for breast cancer increased almost monotonically when longer latency was included (Table 3). For an alcohol intake of 12–23 g/day, the risk estimates increased markedly, from HR = 1.18 for a 0-year time window to HR = 2.48 for a 20-year window, while for alcohol intake above 48 g/day the estimates increased from HR = 1.05 for 0 years to HR = 10.5 for a 20-year window. This increase in risk estimates was similar for all four alcohol intake categories, although some of the confidence intervals were very wide because few cases were available for long latencies. The estimates for 0 years correspond to the risk estimates of updated alcohol intake in Table 2.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
In this large, long-term prospective cohort study, we found a strong positive dose–response relation between baseline alcohol intake and breast cancer; however, inclusion of the most recent alcohol intake markedly attenuated the association. Longer latency increased the risk estimates, supporting the probability that the difference in results with the two methods might be due to a long latency between alcohol intake and breast cancer.
Previous prospective studies of use of baseline and updated information on risk factors for cardiovascular disease showed that the strongest risk estimates were for updated information.7,25 This has been ascribed to increasing non-differential misclassification of baseline information on risk factors with longer follow-up.8 Willett and Stampfer26 proposed that the association between alcohol intake and breast cancer might be underestimated by at least 50% when based on a single assessment of alcohol intake. In two meta-analyses, of 10 and 13 cohort studies, respectively, the association between alcohol intake and breast cancer was less strong in the studies with the longest follow-up,1,2 which the authors ascribed to the possibility that the intake measure carries less information about intake during a more aetiologically relevant period2 and to the ageing of the population, as older women are generally at higher risk for breast cancer, which could lead to a smaller relative risk associated with alcohol intake.1
Our findings were not consistent with these results, as we found a pronounced increase of effect with longer latency.1,2 Our results imply that the latency between exposure to alcohol and breast cancer is at least 20 years (Table 3), because the risk estimates increased when longer latency was inferred. We were unable to determine whether the latency was even longer with 27 years of follow-up. Armenian and Lilienfeld27 summarized much of the early literature on latency for seven specific cancers (not including breast cancer) and concluded that the median latencies range from 6 to 7 years for leukaemias to 36 years for lung cancer. In studies on atomic bomb survivors, the exposure was instantaneous, so that the time of exposure can be set unambiguously and analysis of latency is straightforward12 (pp. 463–64). In a study of long-term lifestyle factors like alcohol intake, however, the time of exposure is not clear,11 and the effect of latency may be difficult to separate from the effect of cumulative exposure (product of duration and intensity). Although the effect of latency can be inferred, as in this study, information on cumulative exposure was not available in the Copenhagen City Heart Study, e.g. as information on the start of alcohol intake or periods of alcohol abstinence. Nevertheless, if the effect of cumulative exposure explains our findings on latency, we would expect the risk estimates with updated alcohol intake to be stronger than those with baseline information, because the repeated measurements of alcohol intake were correlated in this study. Furthermore, in a review of exposure time–response relationships in cancer epidemiology,10 Thomas stated that, if an exposure acts at the first stage of a multistage model of carcinogenesis, the effect of time since exposure should increase relative rates. Our results support the hypothesis that alcohol intake acts at an early stage of breast neoplasia.
Biological mechanisms
In further support of our interpretation of the results, biological evidence indicates a long latency between alcohol intake and breast cancer, as alcohol influences an early stage of carcinogenesis.12,28 One proposed mechanism is that alcohol increases the breast area occupied by mammographically dense tissue,29 which is associated with breast cancer.30 Another hypothesis is that cumulative lifetime exposure to oestrogens increases the risk for breast cancer,31 and studies of both pre- and post-menopausal women support the hypothesis that alcohol intake increases oestrogen concentrations.32,33
Alternatively, alcohol may contribute to breast cancer risk at a late stage of neoplasia. For instance, high alcohol intake is associated with increased risks for regional tumours34 and for breast cancer recurrence and death.35 Even if these associations are attributable to confounding, e.g. that women who drink alcohol attend screening services later than non-drinkers, the mechanism is unknown. In a review, Singletary and Gapstur13 concluded that the biological evidence for a longer latency between alcohol intake and breast cancer is stronger than that for a short latency.
Alcohol information
Self-reported information on alcohol intake might underestimate the true intake,36 although studies to compare self-administered questionnaires with intake assessed by detailed diet records showed correlations above 0.80.37–39 Systematic underreporting of consumption would result in an overestimate of the relative risk for breast cancer for a given level of alcohol consumption, while random misclassification among all women would have the opposite effect. Moreover, the shape of the dose–response curve might change if heavy drinkers are more likely to underreport intake than moderate drinkers. These types of measurement error are inevitable but counterweight each other; it is not possible to evaluate their overall effect on our results. Differential misclassification, whereby women with and without breast cancer misreport to different extents, was minimized in this study, because information on alcohol intake was collected prospectively.
There might have been a difference in the alcohol intake at baseline of women who participated only once and those who participated in more waves. The mean, median and interquartile ranges of baseline alcohol intake were almost identical for these two groups of women (data not shown), so that this does not explain the difference in results between baseline and updated alcohol intake.
The information on alcohol intake in wave 1 was different from that in the later waves, because it was partly conditional on covariates in wave 2, as described by Becker et al.14 To investigate if this influenced our results, we excluded observations from wave 1 and conducted the analyses shown in Figure 1. The results were almost completely identical (data not shown), indicating that the information on alcohol intake in wave 1 did not influence the results.
An alternative to using repeated measures of alcohol intake would be to average previous alcohol intake at different times during follow-up. As noted, updated alcohol intake was only most recent intake. Average alcohol intake could be interpreted as an intermediate approach between baseline and updated intake, because it combines the two. The risk estimates for average alcohol intake were between that for baseline and that for updated alcohol intake, except for an intake of 1.71–12 g/day, for which there was a slightly increased risk with an intake below 48 g/day and a significantly increased risk with an intake above 48 g/day (HR = 2.75; 95% CI: 1.00–7.51).
Perspectives
Previous studies have suggested that updated exposure information is more precise and therefore superior to baseline information.8 One conclusion of our results may be that this does not apply in cancer epidemiology, because of a pronounced effect of latency. If the latency is long, variations in alcohol intake during follow-up may equate to non-differential misclassification. The effect will be strong if the variation in exposure during follow-up is large and independent of previous exposure. Although our data showed dependency between repeated measures of alcohol intake, the correlation became weaker with time.
In this study, with relatively crude information about alcohol intake during follow-up, inferring longer time lags gave stronger risk estimates than updated information without time lags. Thus, when analysing exposure that varies over time and with long follow-up times in cancer epidemiology, it is important to take explicit account of latency.
Large meta-analyses and pooled analyses have shown a linearly increasing risk for breast cancer with alcohol intake;1,2,4,5 e.g. in a large pooled analysis of 322 647 women with 4335 cases of invasive breast cancer, the risk was 1.09 (95% CI: 1.04–1.13) for a 10-g/day increase in alcohol consumption.5 Our risk estimate for baseline alcohol intake was stronger, HR = 1.20 (95% CI: 1.09–1.31) for a 10-g/day increase; however, as the confidence interval is wide, the finding might not be in conflict with the previous result. In the six pooled studies,5 only one measurement of alcohol intake was used, corresponding to baseline alcohol intake in our study. We suggest that the results of the combined analyses1,2,4,5 are not biased towards zero because of random misclassification of alcohol intake during follow-up.
|
Conclusion |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
The results of this study support those of other prospective studies in which one measurement of alcohol was used, that the risk for breast cancer increases linearly with increasing intake of alcohol. When updated alcohol intake was included, no association was observed. A strongly increased risk was observed when risk time was lagged directly, supporting the hypothesis that the latency between alcohol intake and breast cancer is long. This conclusion is supported by biological and epidemiological studies of latency.
|
Acknowledgements |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
The National Institute of Public Health, Denmark and the Danish Cancer Society supported this study, and we would like to thank the Copenhagen City Heart Study for making data available.
Conflict of interest: None declared.
KEY MESSAGES
|
|
References |
|---|
|
TopAbstractIntroductionMaterials and methodsResultsDiscussionConclusionAcknowledgementsReferences |
|---|
1 Ellison RC, Zhang Y, McLennan CE, Rothman KJ. Exploring the relation of alcohol consumption to risk of breast cancer. Am J Epidemiol (2001) 154:740–47.
2 Longnecker MP. Alcohol beverage consumption in relation to risk of breast cancer: meta-analysis and review. Cancer Causes Control (1994) 5:73–82.[CrossRef][Web of Science][Medline]
3 Roth HD, Levy PS, Shi L, Post E. Alcoholic beverages and breast cancer: some observations on published case-control studies. J Clin Epidemiol (1994) 47:207–16.[CrossRef][Web of Science][Medline]
4 Collaborative Group on Hormonal Factors in Breast Cancer. Alcohol, tobacco and breast cancer – collaborative reanalysis of individual data from 53 epidemiological studies, including 58 515 women with breast cancer and 95 067 women without the disease. Br J Cancer. (2002) 87. 1234–45.
5 Smith-Warner SA, Spiegelman D, Yaun SS, et al. Alcohol and breast cancer in women. A pooled analysis of cohort studies. JAMA (1998) 279:535–40.
6 Mezzetti M, La Vecchia C, Decarli A, Boyle P, Talamini R, Franceschi S. Population attributable risk for breast cancer: diet, nutrition, and physical exercise. J Natl Cancer Inst (1998) 90:389–94.
7 Emberson JR, Shaper AG, Wannamethee SG, Morris RW, Whincup PH. Alcohol intake in middle age and risk of cardiovascular disease and mortality: accounting for intake variation over time. Am J Epidemiol (2005) 161:856–63.
8 MacMahon S, Peto R, Cutler J, et al. Blood pressure, stroke and coronary heart disease. Lancet (1990) 335:765–74.[CrossRef][Web of Science][Medline]
9 Appleyard M, Hansen AT, Schnohr P, Jensen G, Nyboe J. The Copenhagen City Heart Study. A book of tables with data from the first examination (1976–78) and a five year follow-up (1981–83). J Soc Med (1989) 170(Suppl 41):1–160.
10 Thomas DC. Statistical methods for analyzing effects of temporal patterns of exposure on cancer risk. Scand J Work Environ Health (1983) 9:353–66.[Web of Science][Medline]
11 Rothman KJ. Induction and latent periods. Am J Epidemiol (1981) 114:253–59.
12 Thomas DC. Models for exposure-time-response relationships with applications to cancer epidemiology. Annu Rev Public Health (1988) 9:451–82.[CrossRef][Web of Science][Medline]
13 Singletary KW, Gapstur SM. Alcohol and breast cancer. Review of epidemiologic and experimental evidence and potential mechanisms. JAMA (2001) 286:2143–51.
14 Becker U, Deis A, Sørensen TIA, et al. Alcohol intake in a population study. Assessment and characterization. Alcologia (1995) 7:35–42.
15 National Board of Health. In: The Activity in the Hospitals of Denmark 1979 [in Danish] (1981) Copenhagen: Statistics Denmark.
16 Jensen AR, Overgaard J, Storm HH. Validity of breast cancer in the Danish Cancer Registry. A study based on clinical records from one county in Denmark. Eur J Cancer Prev (2002) 11:359–64.[CrossRef][Web of Science][Medline]
17 Kjaergaard J, Clemmensen IH, Storm HH. Validity and completeness of registration of surgically treated malignant gynaecological diseases in the Danish National Hospital Registry. J Epidemiol Biostat (2001) 6:387–92.[CrossRef][Medline]
18 Storm HH. Completeness of cancer registration in Denmark 1943–1966 and efficacy of record linkage procedures. Int J Epidemiol (1988) 17:44–49.
19 Storm HH, Michelsen EV, Clemmensen IH, Pihl J. The Danish Cancer Registry – history, content, quality and use. Dan Med Bull (1997) 44:535–39.[Web of Science][Medline]
20 SAS Institute Inc. SAS/STAT Software. (1999) Cary, North Carolina: SAS Institute Inc.
21 R Development Core Team. R: A Language and Environment for Statistical Computing. (2005) (Accessed October 3, 2007). Vienna, Austria: R Foundation for Statistical Computing. Available at: http://www.R-project.orgISBN 3-900051-07-0.
22 Cupples LA, D’Agostino RB, Anderson K, Kannel WB. Comparison of baseline and repeated measure covariate techniques in the Framingham Heart Study. Stat Med (1988) 7:205–18.[Web of Science][Medline]
23 Greenland S. Dose-response and trend analysis in epidemiology: alternatives to categorical analysis. Epidemiology (1995) 6:356–65.[Web of Science][Medline]
24 Eisen EA, Agalliu I, Thurston SW, Coull BA, Checkoway H. Smoothing in occupational cohort studies: an illustration based on penalized splines. Occup Environ Med (2004) 61:854–60.
25 Emberson JR, Whincup PH, Morris RW, Walker M. Re-assessing the contribution of serum total cholesterol, blood pressure and cigarette smoking to the aetiology of coronary heart disease: impact of regression dilution bias. Eur Heart J (2003) 24:1719–26.
26 Willett WC, Stampfer MJ. Sobering data on alcohol and breast cancer. Epidemiology (1997) 8:225–27.[Web of Science][Medline]
27 Armenian HK, Lilienfeld AM. The distribution of incubation periods of neoplastic diseases. Am J Epidemiol (1974) 99:92–100.
28 Pöschl G, Seitz HK. Alcohol and cancer. Alcohol Alcohol (2004) 39:155–65.
29 Vachon CM, Kuni CC, Anderson K, Anderson VE, Sellers TA. Association of mammographically defined percent breast density with epidemiologic risk factors for breast cancer (United States). Cancer Causes Control (2000) 11:653–62.[CrossRef][Web of Science][Medline]
30 Boyd NF, Lockwood GA, Byng JW, Tritchler DL, Yaffe MJ. Mammographic densities and breast cancer risk. Cancer Epidemiol Biomarkers Prev (1998) 7:1133–44.
31 Bernstein L, Ross RK. Endogenous hormones and breast cancer risk. Epidemiol Rev (1993) 15:48–65.
32 Sarkola T, Makisalo H, Fukunaga T, Eriksson CJ. Acute effect of alcohol on estradiol, estrone, progesterone, prolactin, cortisol, and luteinizing hormone in premenopausal women. Alcohol Clin Exp Res (1999) 23:976–82.[CrossRef][Web of Science][Medline]
33 Purohit V. Moderate alcohol consumption and estrogen levels in postmenopausal women: a review. Alcohol Clin Exp Res (1998) 22:994–97.[CrossRef][Web of Science][Medline]
34 Vaeth PAC, Satariano WA. Alcohol consumption and breast cancer stage at diagnosis. Alcohol Clin Exp Res (1998) 22:928–34.[CrossRef][Web of Science][Medline]
35 Hebert JR, Hurley TG, Ma Y. The effect of dietary exposures on recurrence and mortality in early stage breast cancer. Breast Cancer Res Treat (1998) 51:17–28.[CrossRef][Web of Science][Medline]
36 Doll R, Peto R, Hall E, Wheatley K, Gray R. Mortality in relationship to consumption of alcohol: 13 years’ observation of male British doctors. BMJ (1994) 309:911–18.
37 Giovannucci E, Colditz G, Stampfer MJ, et al. The assessment of alcohol consumption by a simple self-administered questionnaire. Am J Epidemiol (1991) 133:810–17.
38 Goldbohm RA, van den Brandt PA, Brants HA, et al. Validation of a dietary questionnaire used in a large-scale prospective cohort study on diet and cancer. Eur J Clin Nutr (1994) 48:253–65.[Web of Science][Medline]
39 Gronbaek M, Heitmann BL. Validity of self-reported intakes of wine, beer and spirits in population studies. Eur J Clin Nutr (1996) 50:487–90.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||