ART ICLECoffee, Tea, and Caffeine Consumption and Incidence of Colon and Rectal Cancer

ART ICLECoffee, Tea, and Caffeine Consumption and Incidence of Colon and Rectal Cancer Karin B. Michels, Walter C. Willett, Charles S. Fuchs, Edward Giovannucci Background: Frequent coffee consumption has been associated with a reduced risk of colorectal cancer in a number of case control studies. Cohort studies have not revealed such an association but were limited in size. We explored the association between consumption of coffee and tea and the incidence of colorectal cancer in two large prospective cohorts of women and men. Methods: We used data from the Nurses Health Study (women) and the Health Professionals Followup Study (men). Consumption of coffee and tea and total caffeine intake were assessed and updated in 1980, 1984, 1986, 1990, and 1994 among women and in 1986, 1990, and 1994 among men. The incidence of cancer of the colon or rectum was ascertained through 1998. Hazard ratios were calculated using Cox proportional hazards models that adjusted for potential confounders. All tests of statistical significance were two-sided. Results: During almost 2 million person-years of, 1438 cases of colorectal cancer were observed. Consumption of caffeinated coffee or tea with caffeine or caffeine intake was not associated with the incidence of colon or rectal cancer in either cohort. For both cohorts combined, the covariate-adjusted hazard ratio for colorectal cancer associated with consumption of each additional cup of caffeinated coffee was 0.99 (95% confidence interval [CI] = 0.96 to 1.03). However, participants who regularly consumed two or more cups of decaffeinated coffee per day had a 52% (95% CI = 19% to 71%) lower incidence of rectal cancer than those who never consumed decaffeinated coffee (crude incidence rate of 12 cases of rectal cancer per 100 000 person-years of among participants consuming two or more cups of decaffeinated coffee per day and crude incidence rate of 19 cases of rectal cancer per 100 000 person-years of followup among participants who never consumed decaffeinated coffee). Conclusions: Consumption of caffeinated coffee, tea with caffeine, or caffeine was not associated with incidence of colon of rectal cancer, whereas regular consumption of decaffeinated coffee was associated with a reduced incidence of rectal cancer. [J Natl Cancer Inst 2005;97:282 92] Results of epidemiologic studies have not resolved whether coffee consumption is related to colorectal cancer risk. A recent report by the World Cancer Research Fund concluded that the available evidence was not sufficient to draw any firm conclusions about a decreased risk of colorectal cancer associated with coffee consumption (1 ). However, some researchers contend that a link between high consumption of coffee and a low incidence of colorectal cancer has been firmly established (2 ). A meta-analysis of coffee consumption and colorectal cancer risk revealed that conflicting results have been obtained in case control and cohort studies (3 ). The combined results of 12 case control studies suggested that coffee consumption is associated with a reduced risk of colorectal cancer (odds ratio for high versus low consumption = 0.72; 95% confidence interval [CI] = 0.61 to 0.84). By contrast, the combined results of five prospective cohort studies revealed no association between coffee consumption and colorectal cancer risk, but the number of cases in each study was small (relative risk = 0.97; 95% CI = 0.73 to 1.29). The constituents of coffee might have genotoxic, mutagenic, or antimutagenic properties, any of which could influence colorectal cancer risk. For example, caffeine has been reported to inhibit chemical carcinogenesis and UVB light- induced carcinogenesis in animal models (4 6 ). Conversely, caffeine has also been found to be mutagenic (7 9 ). Coffee consumption has also been speculated to decrease the risk for colorectal cancer because it increases large bowel motility in the rectosigmoid region, which might decrease contact between bowel contents and colon epithelia and thus decrease mucosal damage (10 ). Coffee may also prevent mucosal damage by reducing the excretion of bile acid and sterols into the bowel (11 ). Furthermore, caffeine has been reported to lower insulin sensitivity (12 ), and hyperinsulinemia has been hypothesized to increase risk of colon cancer (13 ). Another caffeinated beverage, black tea, has been suggested to have anticarcinogenic properties due to its flavonoids, which have antioxidative effects (14 ). Affiliation of authors: Obstetrics and Gynecology Epidemiology Center, Brigham and Women s Hospital, and Harvard Medical School (KBM); Departments of Epidemiology (KBM, WCW, EG) and Nutrition (WCW, EG), Harvard School of Public Health, Boston; Channing Laboratory, Department of Medicine, Brigham and Women s Hospital, and Harvard Medical School, Boston (KBM, WCW, CSF, EG); Department of Adult Oncology, Dana-Farber Cancer Institute, Boston, MA (CSF). Correspondence to: Karin B. Michels, ScD, Obstetrics and Gynecology Epidemiology Center, Brigham and Women s Hospital, 221 Longwood Ave., Boston, MA 02115 (e-mail: kmichels@rics.bwh.harvard.edu ). See Notes following References. DOI: 10.1093/jnci/dji039 Journal of the National Cancer Institute, Vol. 97, No. 4, Oxford University Press 2005, all rights reserved. 282 ARTICLES Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005

Given the widespread consumption of coffee worldwide and the high incidence of colorectal cancer in industrialized countries and in South America, where coffee consumption is particularly high, any substantial association between the two could have considerable public health implications. We examined the relations between the consumption of caffeinated and decaffeinated coffee, the consumption of tea with caffeine, and total caffeine intake and the incidences of colon and rectal cancers among participants in the Nurses Health Study (NHS) and the Health Professionals Follow-up Study (HPFS). To our knowledge, these are the largest prospective studies considering this issue. S UBJECTS AND METHODS Study Cohorts The NHS was initiated in 1976, when 121 700 female registered nurses aged 30 55 years completed a self-administered questionnaire that provided information on demographics, lifestyle, and medical history. The HPFS comprises 51 529 male health professionals, including dentists, veterinarians, pharmacists, optometrists, osteopaths, and podiatrists, who were 40 75 years of age at enrollment in 1986. Participants in both cohorts have been followed through mailed self-administered biennial questionnaires that have updated information on lifestyle factors and disease. The study populations for the present analyses consisted of all women who were free of cancer, Crohn s disease, and ulcerative colitis in 1980 and who had completed the 1980 food frequency questionnaire and reported having a total caloric intake of between 500 and 3500 calories per day (N = 87 794 women) and all men who were free of cancer in 1986 and who had completed the 1986 food frequency questionnaire and reported having a total caloric intake of between 800 and 4200 calories per day (N = 46 099 men). A greater proportion of HPFS participants than of NHS participants completed the food frequency questionnaire (100% versus 96%) because the food frequency questions were part of the enrollment questionnaire for HPFS, whereas diet was first assessed in the NHS 4 years after study initiation. This study and the NHS were approved by the Institutional Review Board (IRB) of the Brigham and Women s Hospital (Boston, MA); the HPFS was approved by the IRB of Harvard School of Public Health (Boston, MA). Ascertainment of Cases and Follow-Up On each biennial questionnaire, we asked participants in each cohort whether they had been diagnosed during the previous 2 years with cancer of the colon or rectum. Deaths were reported to us primarily through family members; we also used the National Death Index and the U.S. Postal Service to identify deaths among participants who did not respond to a mailed questionnaire. We estimate that more than 98% of deaths were ascertained (15 ). When a participant (or the next of kin, for decedents) reported a diagnosis of cancer, we sought permission to obtain the relevant medical records and pathology reports. Study physicians blinded to all questionnaire data reviewed the medical records and extracted information from them on the histologic type, anatomic location, and stage of the cancer. We included only invasive adenocarcinoma in this analysis; cases of carcinomas in situ were not considered. For this analysis, participants were followed through 1998 (with a cutoff date of June 1, 1998, for the NHS and January 31, 1998, for the HPFS). Follow-up rates for the cohorts were calculated as the proportion of the number of person-years actually followed out of the total possible number of person-years followed. The followup rates for the populations studied in this analysis were 98.5% for the NHS population and 97.0% for HPFS population. Dietary Assessment Dietary intake data were collected repeatedly from members of both cohorts with the use of a validated self-administered semiquantitative food frequency questionnaire (16,17 ). Among participants in the NHS, diet was assessed in 1980, 1984, 1986, 1990, and 1994; among participants in the HPFS, diet was assessed in 1986, 1990, and 1994. In the NHS, the food frequency questionnaire used for the 1980 dietary assessment consisted of 61 food items and included questions about consumption of coffee with caffeine (in cups), tea with caffeine (in cups), cola or other caffeinated sodas (in glasses), and chocolate (in 1-ounce servings). The 1984 food frequency questionnaire was expanded to include a question about the consumption of decaffeinated coffee (in cups); the unit for sodas was changed to glass, bottle, or can and the unit for chocolate was changed to bars or pieces. The questionnaires used in 1986, 1990, and 1994 were similar to the 1984 questionnaire. In the HPFS, the 1986, 1990, and 1994 food frequency questionnaires were similar to the expanded NHS questionnaires. The expanded questionnaires used in both cohorts provided nine mutually exclusive response possibilities to describe the participant s frequency of intake of a particular item: never or less than once per month; 1 3 times per month; 1 time per week; 2 4 times per week; 5 6 times per week; 1 time per day; 2 3 times per day; 4 5 times per day; 6 or more times per day. Participants were asked to report their average intake of one cup of the respective beverages or one serving of any food over the preceding year. We then converted the responses for individual beverages to an average daily intake for each participant. We used information obtained from U.S. Department of Agriculture food-composition sources (18 ) to calculate caffeine intake. The caffeine contents used for these calculations were 137 mg caffeine per cup of coffee, 47 mg caffeine per cup of tea, 46 mg caffeine per can or bottle of cola beverage, and 7 mg caffeine per serving of chocolate. The reproducibility and validity of the food frequency questionnaire for both women and men have been reported previously (16,17 ). The validity of the questionnaires with respect to individual food items has been documented by comparing the responses of 173 women from the NHS and 127 men from the HPFS to two food frequency questionnaires that were administered approximately 12 months apart and multiple 7-day diet records obtained during the 1-year interval (19,20 ). Correlation coefficients for the average consumption for coffee and tea as assessed by the food frequency questionnaires and the diet records, correcting for within-person weekly variation in diet, were 0.78 and 0.93, respectively, in the NHS (19 ) and 0.93 and 0.77, respectively, in the HPFS (20 ). Statistical Analysis Daily consumption of coffee, tea, and caffeine was calculated from the frequencies that were pre-specified on the food frequency questionnaire. Categories of frequency of coffee and tea consumption were created, and the lowest intake category was Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005 ARTICLES 283

Table 1. Age-standardized distribution of covariates during, by frequency of caffeinated and decaffeinated coffee and tea consumption and caffeine intake * Caffeinated coffee consumption (cups/day) Decaffeinated coffee consumption (cups/day) Covariate Never ½ 1 2 3 4 5 >5 Never ¼ ½ 1 1.9 2 Women Total no. person-years 240507 178813 216149 545379 218771 66403 327938 177942 151244 132475 119285 Age, y 52.4 55.7 55.3 54.9 53.5 51.7 55.2 56.6 57.6 57.4 56.8 Height, cm 164 164 164 164 164 164 164 164 164 164 164 Body mass index, kg/m 2 25.8 25.8 25.5 25.3 25.1 24.9 25.9 25.8 25.8 25.7 25.5 Family history of colorectal 12.9 13.0 13.1 13.2 12.8 12.8 13.6 13.5 13.3 13.9 13.8 cancer, % Prior sigmoidoscopy, % 40.6 41.4 39.6 37.7 33.8 29.7 36.3 41.1 42.7 43.9 44.7 Pack-years of smoking 35 y in 0.8 0.9 0.9 1.1 1.3 1.5 1.3 1.2 1.2 1.3 1.5 the past Alcohol consumption, g/day 3.8 4.5 5.4 7.0 7.0 6.3 6.3 5.6 5.1 5.3 6.0 Physical activity, h/week 3.2 3.2 3.2 3.1 3.0 3.0 2.9 3.1 3.1 3.1 3.1 Premenopausal, % 31.0 31.6 31.6 31.3 30.6 29.1 24.1 24.6 24.5 24.6 23.8 Current postmenopausal 20.8 25.1 22.8 22.2 19.3 14.3 23.5 29.5 30.3 29.2 27.4 hormone use, % Regular aspirin use ( 1 times/ 36.5 40.1 40.2 41.5 43.0 38.3 43.4 43.6 43.4 42.8 45.4 wk), % Vitamin supplement use, % 43.5 49.8 45.3 43.4 39.1 31.1 42.5 51.4 53.4 54.2 51.2 Total caloric intake, kcal/day 1668 1718 1686 1708 1718 1706 1726 1768 1758 1761 1769 Red meat consumption, 0.25 0.22 0.23 0.25 0.27 0.31 0.23 0.20 0.18 0.20 0.21 servings/day Caffeinated coffee consumption, 3.4 3.5 3.2 3.1 3.1 cups/day chosen as referent group. Caffeine intake was calculated from the responses to questions about consumption of coffee with caffeine, tea with caffeine, carbonated beverages with caffeine, and chocolate and was used in the analysis as an energy-adjusted residual. Residuals were obtained by regressing log-transformed caffeine on log total energy and determining the antilog of the resulting residual value (18 ). The caffeine residual, which retains the units of intake (in milligrams) when the antilog is used, was divided into quintiles on the basis of the distributions of caffeine intakes among women and among men. To represent the long-term consumption patterns for individual subjects as accurately as possible and to reduce random within-person variation in beverage consumption, we modeled the incidence of colorectal cancer in relation to the cumulative average intake of coffee, tea, and caffeine from all dietary questionnaires available up to the start of each 2-year interval (21 ). Thus, for the female participants we used dietary data from the 1980 questionnaire for analyses of colorectal cancers diagnosed from 1980 through 1984; the average of the dietary Men Total no. person-years 126850 104803 81524 133469 43313 10131 198099 94043 79083 56535 58775 Age, y 57.8 59.1 59.1 58.4 56.9 55.5 56.8 57.9 60.0 60.1 59.0 Height, cm 178 178 178 178 179 179 178 178 178 178 178 Body mass index, kg/m 2 25.5 25.7 25.7 25.9 26.2 26.1 25.8 25.6 25.7 25.8 26.0 Family history of colorectal 9.6 10.1 9.8 9.8 9.0 8.7 9.3 10.2 9.9 9.9 10.3 cancer, % Prior sigmoidoscopy, % 53.7 56.4 51.8 52.4 49.6 42.3 48.2 55.2 57.0 58.2 55.8 Pack-years of smoking before 4.0 4.7 5.5 6.4 8.2 9.6 5.1 4.9 5.2 6.0 7.8 age 30 y Alcohol consumption, g/day 7.1 9.5 11.1 13.3 13.7 14.7 10.5 10.2 10.1 10.8 12.8 Physical activity, METS/week 31.4 31.6 30.2 30.7 29.3 25.6 29.9 31.9 32.1 31.5 29.6 Regular aspirin use ( 2 times/ 34.2 38.0 37.4 38.7 37.1 35.8 34.0 37.0 39.4 40.0 39.9 week), % Vitamin supplement use, % 51.7 51.2 50.0 47.9 45.9 42.7 47.6 51.3 50.9 51.1 50.6 Total caloric intake, kcal/day 1894 1914 1939 1982 2052 2175 1951 1961 1925 1943 1957 Red meat consumption, 0.14 0.14 0.16 0.17 0.19 0.21 0.17 0.15 0.14 0.15 0.15 servings/day Caffeinated coffee consumption, 2.79 3.04 2.72 2.69 2.47 cups/day (Table continues) data from the 1980 and 1984 questionnaires was used for analyses of colorectal cancers diagnosed from 1984 through 1986; the average of the dietary data from the 1980, 1984, and 1986 questionnaires was used for analyses of colorectal cancers diagnosed from 1986 through 1990; and so forth. Similarly, for the male participants, dietary data from the 1986 questionnaire were used for analyses of colorectal cancers diagnosed from 1986 through 1990; the average of the dietary data from the 1986 and 1990 questionnaires was used for analyses of colorectal cancers diagnosed from 1990 through 1994; and the average of the dietary data from the 1986, 1990, and 1994 questionnaires was used for analyses of colorectal cancers diagnosed from 1994 through 1998. Consumption of decaffeinated coffee among women was first assessed in 1984. Therefore, all analyses pertaining to decaffeinated coffee intake in women used data collected in 1984 through 1998. We carried out separate analyses for colon and rectal cancers. Cancer incidence rates associated with each category of coffee intake were calculated by dividing the number of new cases of 284 ARTICLES Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005

Table 1 (continued). Tea consumption (cups/day) Caffeine intake (quintiles) Covariate Never ¼ ½ 1 2 1 2 3 4 5 Women Total no. person-years 297609 380689 348019 230854 207919 280355 295596 301955 301041 300857 Age, y 53.2 55.2 55.3 54.2 52.4 54.1 54.5 54.6 54.3 54.0 Height, cm 164 164 164 164 164 164 164 164 164 164 Body mass index, kg/m 2 25.2 25.5 25.6 25.5 25.4 25.7 25.6 25.4 25.4 25.2 Family history of colorectal 12.7 13.0 13.2 13.3 13.1 13.1 13.4 13.2 13.1 12.5 cancer, % Prior sigmoidoscopy, % 35.7 38.2 39.3 38.7 38.3 41.7 40.5 39.5 36.8 32.4 Pack-years of smoking 35 y 1.2 1.1 1.0 0.9 1.0 0.8 0.9 1.0 1.1 1.4 in the past Alcohol consumption, 6.7 6.3 5.6 5.4 4.9 4.1 5.3 6.8 6.7 6.4 g/day Physical activity, h/wk 3.0 3.2 3.2 3.2 3.1 3.2 3.2 3.2 3.1 3.0 Premenopausal, % 30.3 31.6 31.3 31.1 31.0 31.1 31.7 31.7 31.0 29.8 Current postmenopausal 17.6 23.5 24.0 21.6 19.9 22.3 22.6 22.5 21.3 19.9 hormone use, % Regular aspirin use 37.0 41.4 41.9 41.8 39.6 37.6 40.4 42.3 41.5 40.4 ( 1 times/wk), % Vitamin supplement use, % 38.1 45.0 46.7 44.4 40.9 47.7 46.3 44.9 41.3 37.1 Total caloric intake, kcal 1606 1691 1740 1740 1742 1701 1730 1749 1692 1625 Red meat consumption, 0.26 0.23 0.23 0.25 0.28 0.23 0.25 0.26 0.25 0.25 servings/day Men Total no. person-years 171396 145549 98404 47676 33646 99129 101852 101871 104148 104801 Age, y 57.9 58.3 59.4 58.5 57.5 59.3 58.6 58.9 58.1 57.2 Height, cm 178 178 178 178 178 178 178 178 178 178 Body mass index, kg/m 2 25.8 25.7 25.7 25.7 25.9 25.2 25.7 25.8 25.9 26.1 Family history of colorectal 9.1 10.0 10.1 10.1 9.8 9.7 9.9 9.9 9.7 9.5 cancer, % Prior sigmoidoscopy, %, 47.0 51.7 51.6 49.2 49.1 54.0 55.6 52.9 52.6 50.4 Pack-years of smoking 6.0 5.2 4.9 5.3 6.1 3.8 4.4 5.4 6.1 7.6 before age 30 y Alcohol consumption, 11.6 10.6 9.5 9.8 9.9 7.5 9.2 10.9 13.4 11.9 g/day Physical activity, METS/wk 29.2 32.3 31.6 30.5 29.2 33.1 31.5 30.7 30.5 28.0 Regular aspirin use 2 34.9 37.7 39.3 37.5 37.9 33.7 37.5 38.0 38.6 37.5 times/week, % Vitamin supplement use, % 48.8 50.4 50.1 49.9 48.9 52.2 51.4 49.8 49.5 45.4 Total caloric intake, kcal 1887 1956 1970 2001 2052 1922 1984 1990 2047 1784 Red meat consumption, 0.15 0.15 0.15 0.17 0.18 0.14 0.15 0.16 0.17 0.16 servings/day * Mean values given for covariates unless otherwise indicated. METS = metabolic equivalents; = not applicable. Mother, father, or sibling. Ever having had a prior sigmoidoscopy Includes nondrinkers. Mother or father. colon or rectal cancer by the number of person-years of followup. The number of person-years of for each participant was calculated from the month we receive the completed 1980 questionnaire (NHS) or the 1986 questionnaire (HPFS) to the date of diagnosis of colon or rectal cancer, the date of death, or the end of (June 1, 1998, for the NHS and January 31, 1998, for the HPFS), whichever occurred first. Participants who reported that they had been diagnosed with Crohn s disease, ulcerative colitis, or cancers other than nonmelanoma skin cancer were excluded at baseline, and was censored when these diseases were diagnosed during. We used a Cox proportional hazards model to calculate the relative hazard of developing colon or rectal cancer associated with coffee, tea, or caffeine intake ( 22 ). The proportional haz- ards model permits the simultaneous adjustment for multiple potential confounders, including time-dependent covariates. Proportionality of hazards was confirmed by visual examination of associations across intervals of time. Coffee, tea, and caffeine consumption were used as time-dependent variables and were cumulatively updated every 4 years, as described above. Regression models were adjusted for age (in months), family history of colorectal cancer (yes/no), history of sigmoidoscopy or colonoscopy (ever/never), height (in centimeters; continuous variable), body mass index (in kilograms per meter squared; continuous variable), moderate to vigorous physical activity (women: <1 hour/week, 1 1.9 hours/week, 2 3.9 hours/ week, 4 6.9 hours/week, 7 hours/week; men: quintiles of metabolic equivalents [working metabolic rate/resting metabolic Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005 ARTICLES 285

Table 2. Cumulative updated caffeinated coffee consumption and covariate-adjusted hazard ratio of colorectal, colon, and rectal cancers * HR for 1 additional cup Frequency of coffee consumption of coffee/day Cancer Never 0.5 cup/day 1 cup/day 2 3 cups/day 4 5 cups/day >5 cups/day Ptrend (95% CI) rate/week]), regular aspirin use (women: never or <1/week, 1 6/week, 7/week; men: <2/week, 2/week), pack-years of smoking (both sexes: <10 pack-years, 10 pack-years; among women: pack-years of smoking 35 years in the past; among men: pack-years of smoking before age 30 years), ever-use of vitamin supplement (use of multivitamins or vitamins A, C, or E: yes/no), alcohol consumption (none, <10 g/day, 10 19.9 g/ day, 20 29.9 g/day, 30 g/day), total caloric intake (kilocalories; continuous variable), red meat consumption (<1/week, 1/ week, 2 4/week, 5 6/week, 1/day), and (among women) menopausal status (premenopausal, postmenopausal, uncertain menopausal status) and postmenopausal hormone use (never, current, past). Covariates were assessed repeatedly and updated throughout the analysis. Total caloric intake was included in the covariate-adjusted model to control for confounding by total energy intake and to minimize extraneous variation due to general under- or overreporting of intake of food items on the food frequency questionnaire (23 ). Body mass index and height were included in the analytic model to capture the relation between body composition and size and colorectal cancer risk (24 ). Hazard ratios (HRs) for Colorectal cancer HR, women (95% CI) 1.00 (referent) 1.11 0.99 1.08 0.95 0.94.52 0.99 (0.86 to 1.43) (0.77 to 1.27) (0.88 to 1.34) (0.73 to 1.23) (0.63 to 1.40) (0.95 to 1.04) HR, men (95%CI) 1.00 (referent) 1.00 0.98 0.93 1.04 1.09.99 1.00 (0.77 to 1.28) (0.75 to 1.28) (0.73 to 1.19) (0.74 to 1.48) (0.55 to 2.17) (0.94 to 1.06) HR, pooled (95% CI) 1.00 (referent) 1.05 0.99 1.02 0.98 0.98.60 0.99 (0.88 to 1.26) (0.82 to 1.18) (0.87 to 1.19) (0.80 to 1.21) (0.69 to 1.38) (0.96 to 1.03) No. cases 259 243 228 497 164 40 1431 No. person-years of 367357 283616 297672 678849 262085 76534 1966113 Colon cancer HR, women (95% CI) 1.00 (referent) 1.03 0.92 1.02 0.85 0.85.26 0.98 (0.78 to 1.36) (0.70 to 1.20) (0.82 to 1.28) (0.64 to 1.12) (0.55 to 1.32) (0.94 to 1.03) HR, men (95% CI) 1.00 (referent) 1.09 1.04 1.01 0.91 1.39.88 0.99 (0.83 to 1.45) (0.77 to 1.41) (0.77 to 1.33) (0.59 to 1.36) (0.69 to 2.78) (0.92 to 1.06) HR, pooled (95% CI) 1.00 (referent) 1.06 0.97 1.02 0.86 0.98.31 0.98 (0.87 to 1.29) (0.79 to 1.19) (0.85 to 1.21) (0.68 to 1.09) (0.68 to 1.41) (0.95 to 1.02) No. of cases 214 200 185 413 123 35 1170 No. person-years of 367423 283665 297738 678969 262125 76542 1966462 Rectal cancer HR, women (95% CI) 1.00 (referent) 1.71 1.58 1.58 1.80.29 1.05 (0.88 to 3.34) (0.82 to 3.06) (0.87 to 2.88) (0.94 to 3.44) (0.95 to 1.17) HR, men (95% CI) 1.00 (referent) 0.63 0.78 0.65 1.33.81 1.02 (0.35 to 1.15) (0.43 to 1.44) (0.37 to 1.14) (0.69 to 2.56) (0.89 to 1.18) HR, pooled (95% CI) 1.00 (referent) 1.08 1.55.31 1.04 (0.69 to 1.69) (0.97 to 2.45) (0.96 to 1.13) No. of cases 45 43 43 84 45 260 No. person-years of 367647 283917 297968 679527 338867 1967926 * HR = hazard ratio; CI = confidence interval. The highest frequency of consumption category for rectal cancer is 4 cups/day. Wald test for trend; two-sided Adjusted for age, family history of colorectal cancer, history of sigmoidoscopy, height, body mass index, pack-years of smoking, physical activity, aspirin use, vitamin supplement intake, alcohol consumption, red meat consumption, total caloric intake, and, among women in addition for menopausal status, postmenopausal hormone use. Data were not combined because of statistically significant heterogeneity (at P <.05) in HR estimates from women and men. an increase in intake by one serving (e.g., one cup) per day were obtained by using daily consumption as a continuous variable. Because the two cohorts differed by sex, time, food frequency questionnaires, and covariates, we performed separate analyses for each cohort and then combined the results by using a fixed-effects model that weighted the two relative risk estimates by the inverse of the standard error (25 ). Tests of heterogeneity were used to evaluate whether associations differed between women and men; results are shown separately whenever statistically significant heterogeneity was observed. Analyses were stratified by current smoking status and by alcoholic beverage consumption (both of which were timedependent covariates). All tests of statistical significance were two-sided. R ESULTS From 1980 through 1994, participants in the NHS gradually decreased the amount of caffeinated coffee they consumed and the number of NHS participants who did not drink coffee also decreased. In 1980, 22.5% of the women did not drink coffee, 286 ARTICLES Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005

Table 3. Cumulative updated decaffeinated coffee consumption and covariate-adjusted hazard ratio of colorectal, colon, and rectal cancer * Frequency of decaffeinated coffee consumption HR (95% CI) for 1 additional cup Cancer Never ¼ cup/day ½ cup/day 1 1.9 cups/day 2 cups/day Ptrend of coffee/day Colorectal cancer HR, women (95% CI) 1.00 0.85 0.76 0.76 0.80.08 0.96 (referent) (0.68 to 1.07) (0.60 to 0.98) (0.58 to 0.98) (0.61 to 1.04) (0.89 to 1.04) HR, men (95% CI) 1.00 0.77 0.63 0.72 0.85.51 0.96 (referent) (0.60 to 1.00) (0.48 to 0.84) (0.54 to 0.96) (0.64 to 1.12) (0.88 to 1.05) HR, pooled (95% CI) 1.00 0.82 0.70 0.74 0.82.08 0.96 (referent) (0.69 to 0.97) (0.58 to 0.85) (0.61 to 0.90) (0.67 to 0.99) (0.91 to 1.02) No. of cases 463 212 167 146 150 1138 No. person-years of 526036 271985 230327 189009 178059 1395416 Colon cancer HR, women (95% CI) 1.00 1.00 0.85 0.84 0.90.20 0.98 (referent) (0.78 to 1.29) (0.64 to 1.13) (0.63 to 1.13) (0.67 to 1.21) (0.90 to 1.07) HR, men (95% CI) 1.00 0.79 0.66 0.71 0.94.84 1.00 (referent) (0.60 to 1.05) (0.48 to 0.89) (0.51 to 0.99) (0.70 to 1.28) (0.91 to 1.09) HR, pooled (95% CI) 1.00 0.90 0.76 0.78 0.92.43 0.99 (referent) (0.75 to 1.09) (0.62 to 0.93) (0.63 to 0.97) (0.74 to 1.14) (0.93 to 1.05) No. of cases 357 176 136 116 128 913 No. person-years of 526140 271972 230297 189072 178089 1395570 Rectal cancer HR, women (95% CI) 1.00 0.46 0.52 0.51 0.48.15 0.88 (referent) (0.27 to 0.79) (0.30 to 0.89) (0.28 to 0.90) (0.26 to 0.89) (0.72 to 1.07) HR, men (95% CI) 1.00 0.48 0.69 0.76 0.50.22 0.87 (referent) (0.21 to 1.13) (0.31 to 1.53) (0.34 to 1.70) (0.20 to 1.27) (0.67 to 1.14) HR, pooled (95% CI) 1.00 0.46 0.57 0.58 0.48.06 0.88 (referent) (0.29 to 0.73) (0.36 to 0.89) (0.36 to 0.93) (0.29 to 0.81) (0.75 to 1.03) No. of cases 105 36 31 30 22 224 No. person-years of 526655 272237 230558 189149 178286 1396885 * HR = hazard ratio; CI = confidence interval. Wald test for trend; two-sided. Adjusted for age, family history of colorectal cancer, history of sigmoidoscopy, height, body mass index, pack-years of smoking, physical activity, aspirin use, vitamin supplement intake, alcohol consumption, red meat consumption, total caloric intake, consumption of caffeinated coffee, and, among women in addition for menopausal status, postmenopausal hormone use. and 25.0% drank four or more cups of caffeinated coffee per day. In 1994, 13.2% of the women did not drink coffee, and 14.2% drank four or more cups per day. A similar, but weaker, trend toward decreased coffee consumption was observed among participants in the HPFS. In 1986, 29.9% of the men did not drink caffeinated coffee, and 10.8% drank four or more cups per day. In 1994, 22.3% of the men drank no coffee and 8.8% drank four or more cups daily. Similar patterns emerged for tea consumption in the two cohorts. Accordingly, mean caffeine intake decreased from 391 mg/day (standard deviation [SD] = 270 mg/day) in 1980 to 242 mg/day (SD = 207 mg/day) in 1994 among all women and from 227 mg/day (SD = 227 mg/day) in 1986 to 221 mg/day (SD = 221 mg/day) in 1994 among all men. The number of participants who reported drinking decaffeinated coffee increased over time. Among women, 49.3% drank decaffeinated coffee with any regularity in 1984 and 71.5% did so in 1994. Among men, 48.8% drank decaffeinated coffee with any regularity in 1986, whereas 62.5% did so in 1994. Coffee consumption varied greatly among the participants in each cohort. Although in both cohorts the most frequent category of caffeinated coffee consumption selected was two to three cups per day, considerable numbers of participants reported that they either consumed no coffee or more than five cups per day ( Table 1 ). Higher consumption of caffeinated coffee was associated with a lower mean body mass index among women and with a higher mean body mass index among men. In both cohorts, higher caffeinated coffee consumption was also associated with a lower frequency of sigmoidoscopy, a lower use of vitamin supplements, and higher frequencies of smoking, alcohol consumption, aspirin use, and red meat consumption. Higher consumption of decaffeinated coffee was associated with higher frequencies of sigmoidoscopy, vitamin supplement use, and smoking in both cohorts and with a higher frequency of aspirin use among men. In both cohorts, higher tea consumption was associated with lower alcohol consumption. Patterns for caffeine intake were similar to those for caffeinated coffee consumption ( Table 1 ). During 1 479 804 person-years of among women, we documented 731 cases of colon cancer and 155 cases of rectal cancer; during 511 801 person-years of among men, we documented 446 cases of colon cancer and 106 cases of rectal cancer. Total coffee consumption, which included caffeinated and decaffeinated coffee consumption, was not associated with the incidence of colorectal, colon, or rectal cancer. The covariate-adjusted hazard ratios for colorectal cancer for each additional cup of any coffee consumed were 0.99 (95% confidence interval [CI] = 0.95 to 1.04) among women and 0.98 (95% CI = 0.92 to 1.03) among men. Results of our analysis of the association between caffeinated coffee consumption and incidence of colorectal cancer are Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005 ARTICLES 287

Table 4. Cumulative updated tea consumption and covariate-adjusted hazard ratio of colorectal, colon, and rectal cancer * HR for 1 additional cup Frequency of tea consumption of tea/day Cancer Never ¼ cup/day ½ cup/day 1 cup/day 2 cups/day Ptrend (95% CI) Colorectal cancer HR, women (95% CI) 1.00 (referent) 0.84 0.83 0.95 0.96.43 1.02 (0.69 to 1.03) (0.68 to 1.02) (0.76 to 1.18) (0.76 to 1.22) (0.96 to 1.09) HR, men (95% CI) 1.00 (referent) 1.00 0.93 1.14 1.12.32 1.03 (0.80 to 1.25) (0.72 to 1.19) (0.84 to 1.53) (0.78 to 1.59) (0.92 to 1.14) HR, pooled (95% CI) 1.00 (referent) 0.91 0.87 1.01 1.01.23 1.02 (0.78 to 1.06) (0.74 to 1.02) (0.84 to 1.21) (0.83 to 1.22) (0.97 to 1.08) No. of cases 367 373 303 205 154 1402 No. person-years of 469005 526238 446423 278529 241565 1961760 Colon cancer HR, women (95% CI) 1.00 (referent) 0.92 0.90 1.04 1.08.18 1.04 (0.74 to 1.15) (0.71 to 1.13) (0.81 to 1.32) (0.83 to 1.39) (0.97 to 1.11) HR, men (95% CI) 1.00 (referent) 1.08 0.94 1.31 1.05.40 1.01 (0.84 to 1.38) (0.70 to 1.24) (0.95 to 1.80) (0.69 to 1.59) (0.90 to 1.15) HR, pooled (95% CI) 1.00 (referent) 0.99 0.91 1.13 1.07.11 1.03 (0.84 to 1.17) (0.76 to 1.09) (0.93 to 1.37) (0.86 to 1.33) (0.97 to 1.10) No. of cases 288 310 243 176 129 1146 No. person-years of 469096 526318 446534 278567 241591 1962106 Rectal cancer HR, women (95% CI) 1.00 (referent) 0.58 0.61 0.65 0.54.27 0.93 (0.37 to 0.91) (0.39 to 0.96) (0.39 to 1.09) (0.29 to 0.99) (0.79 to 1.11) HR, men (95% CI) 1.00 (referent) 0.73 0.93 0.49 1.34.59 1.08 (0.43 to 1.24) (0.54 to 1.60) (0.19 to 1.25) (0.67 to 2.68) (0.87 to 1.35) HR, pooled (95% CI) 1.00 (referent) 0.64 0.73 0.61.59 0.99 (0.46 to 0.90) (0.51 to 1.03) (0.39 to 0.96) (0.86 to 1.13) No. of cases 79 63 60 29 25 256 No. person-years of 469499 526692 446800 278799 241774 1963564 * HR = hazard ratio; CI = confidence interval. Wald test for trend; two-sided. Adjusted for age, family history of colorectal cancer, history of sigmoidoscopy, height, body mass index, pack-years of smoking, physical activity, aspirin use, vitamin supplement intake, alcohol consumption, red meat consumption, total caloric intake, and, among women in addition for menopausal status, postmenopausal hormone use. Data were not combined because of statistically significant heterogeneity (at P <.05) in HR estimates from women and men. presented in Table 2. Because the hazard ratios adjusted for age only did not differ appreciably from the hazard ratios adjusted for all covariates, we present only the latter in Table 2. We found no significant association between consumption of caffeinated coffee and the incidence of colorectal cancer ( Table 2 ). Participants who reported never drinking caffeinated coffee had a crude incidence rate of colorectal cancer of 71 cases per 100 000 personyears of while those who reported drinking more than five cups of caffeinated coffee per day had a crude incidence rate of colorectal cancer of 52 cases per 100 000 person-years of. Among women and men combined, the pooled hazard ratio for colorectal cancer for one additional cup of caffeinated coffee per day was 1.01 (95% CI = 0.97 to 1.04) when we adjusted for age only and 0.99 (95% CI = 0.96 to 1.03) when we adjusted for age, family history of colorectal cancer, history of sigmoidoscopy, height, body mass index, pack-years of smoking, physical activity, aspirin use, vitamin supplement intake, total caloric intake, alcohol consumption, red meat consumption, and, among women, menopausal status and postmenopausal hormone use. Compared with no consumption, consumption of four or more cups of coffee per day was associated with a slight although not statistically significant increase in rectal cancer (pooled HR = 1.55, 95% CI = 0.97 to 2.45). No statistically significant trend of higher rectal cancer incidence emerged with increasing coffee consumption (Ptrend =.31). When we only used coffee consumption reported in 1980 for women and 1986 for men and did not update consumption during, the results did not differ from the results presented for cumulatively updated coffee consumption (data not shown). Women and men who drank decaffeinated coffee had a lower incidence of colorectal cancer, particularly rectal cancer, than women and men who never drank decaffeinated coffee ( Table 3 ). Participants who reported never drinking decaffeinated coffee had a crude incidence rate of rectal cancer of 19 cases per 100 000 person-years of while those who reported drinking two or more cups of decaffeinated coffee per day had a crude incidence rate of rectal cancer of 12 cases per 100 000 person-years of. Participants who reported any regular consumption of decaffeinated coffee had approximately half the risk of rectal cancer than those who did not drink decaffeinated coffee. Women and men who drank two or more cups of decaffeinated coffee per day had a 52% (95% CI = 19% to 71%) lower risk of rectal cancer than women and men who did not drink decaffeinated coffee after adjusting for all covariates used in this analyses as well as for caffeinated coffee consumption; however, the relation between consumption frequency and rectal 288 ARTICLES Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005

Table 5. Cumulative updated caffeine consumption and covariate-adjusted hazard ratio of colorectal, colon, and rectal cancer * HR for Quintiles of caffeine intake additional 100 mg per caffeine 1 2 3 4 5 Ptrend per day (95% CI) Median intake, mg/day Women 72 181 290 396 602 Men 15 76 165 307 499 Colorectal cancer HR, women (95% CI) 1.00 1.05 0.99 1.02 1.02 0.95 1.00 (referent) (0.85 to 1.30) (0.80 to1.23) (0.82 to 1.27) (0.82 to1.27) (0.97 to 1.03) HR, men (95% CI) 1.00 0.84 0.88 0.96 0.88 0.90 1.00 (referent) (0.64 to 1.09) (0.68 to1.15) (0.74 to1.25) (0.67 to1.16) (0.96 to1.04) HR, pooled (95% CI) 1.00 0.96 0.95 1.00 0.96 0.99 1.00 (referent) (0.81 to1.14) (0.80 to1.12) (0.84 to1.18) (0.81 to 1.14) (0.98 to1.03) No. of cases 284 287 291 299 277 1438 No. person-years of 379483 397448 403825 405190 405658 1991604 Colon cancer HR, women (95% CI) 1.00 1.05 0.97 1.05 0.97 0.71 1.00 (referent) (0.83 to 1.32) (0.76 to 1.22) (0.83 to 1.33) (0.76 to 1.23) (0.96 to 1.03) HR, men (95% CI) 1.00 0.87 0.91 0.99 0.85 0.67 1.00 (referent) (0.64 to 1.17) (0.68 to 1.23) (0.74 to 1.32) (0.63 to 1.16) (0.96 to 1.05) HR, pooled (95% CI) 1.00 0.97 0.94 1.03 0.92 0.97 1.00 (referent) (0.81 to 1.17) (0.79 to 1.14) (0.85 to 1.23) (0.76 to 1.11) (0.97 to 1.02) No. of cases 233 237 238 252 217 1177 No. person-years of 379562 397519 403912 405243 405719 1991955 Rectal cancer HR, women(95% CI) 1.00 1.08 1.13 0.87 1.28 0.50 1.02 (referent) (0.65 to 1.81) (0.67 to 1.89) (0.50 to 1.51) (0.76 to 2.13) (0.95 to 1.10) HR, men(95% CI) 1.00 0.69 0.75 0.86 1.01 0.54 1.01 (referent) (0.37 to 1.30) (0.41 to 1.39) (0.47 to 1.57) (0.55 to 1.83) (0.92 to 1.11) HR, pooled (95% CI) 1.00 0.91 0.95 0.86 1.15 0.89 1.02 (referent) (0.61 to 1.35) (0.64 to 1.41) (0.57 to 1.30) (0.78 to 1.70) (0.96 to 1.08) No. of cases 51 50 53 47 60 261 No. person-years of 379808 397849 404221 405551 406014 1993443 * HR = hazard ratio; CI = confidence interval. Wald test for trend; two-sided. Adjusted for age, family history of colorectal cancer, history of sigmoidoscopy, height, body mass index, pack-years of smoking, physical activity, aspirin use, vitamin supplement intake, alcohol consumption, red meat consumption, total caloric intake, and, among women in addition for menopausal status, postmenopausal hormone use. cancer risk was not linear. Colon cancer incidence was lowest among participants who drank a daily average of half a cup of decaffeinated coffee. When we performed separate analyses for colon cancers at proximal and distal sites, we found that the reduction in colon cancer risk associated with consumption of decaffeinated coffee was restricted to men and women who had been diagnosed with proximal site colon cancers but that this risk reduction was not statistically significant (data not shown). When we chose participants who reported that they never drank caffeinated or decaffeinated coffee as the referent group, the results did not differ appreciably (data not shown). When we used information about decaffeinated coffee consumption from the first questionnaires only (the 1984 questionnaire for women and the1986 questionnaire for men), the results were similar to those obtained using updated information about decaffeinated coffee consumption (data not shown). Tea consumption was not associated with the incidence of either colon or rectal cancer ( Table 4 ). Similarly, there was no significant association between total caffeine intake and the incidence of either colon or rectal cancer ( Table 5 ). When we restricted our analyses to women and men who did not currently smoke cigarettes, the results were not materially different from the results observed among the entire study population ( Table 6 ). Similarly, results of analyses that were restricted to participants who did not smoke cigarettes and did not drink alcoholic beverages were not materially different from those obtained for the entire study population ( Table 6 ). D ISCUSSION In these two large prospective cohorts in which coffee and tea consumption was assessed repeatedly, consumption of caffeinated coffee and tea and overall caffeine intake were not associated with the incidence of either colon or rectal cancer. However, participants who reported any regular consumption of decaffeinated coffee had a statistically significantly lower incidence of rectal cancer in both cohorts than participants who reported that they did not drink decaffeinated coffee. To our knowledge, this is the largest study on coffee and tea consumption and the incidence of colon and rectal cancer. A meta- Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005 ARTICLES 289

Table 6. Cumulative updated consumption of caffeinated coffee, decaffeinated coffee, tea, and caffeine and covariate-adjusted hazard ratio of colorectal cancer among nonsmokers and among women and men who do not smoke and do not drink alcohol * Nonsmokers HR for 1 additional cup Frequency of coffee consumption of coffee/day Never 0.5 cup/day 1 cup/day 2 3 cups/day 4 5 cups/day >5 cups/day Ptrend (95% CI) HR, women (95% CI) 1.00 1.01 0.91 1.01 0.91 0.88.52 0.99 (referent) (0.79 to 1.30) (0.72 to 1.17) (0.82 to 1.24) (0.70 to 1.17) (0.57 to 1.36) (0.95 to 1.04) HR, men (95% CI) 1.00 1.04 1.25 0.91 1.12 1.10.96 1.00 (referent) (0.67 to 1.62) (0.83 to 1.88) (0.60 to 1.37) (0.64 to 1.96) (0.42 to 2.87) (0.91 to 1.10) HR, pooled (95% CI) 1.00 1.02 0.99 0.99 0.94 0.92.55 0.99 (referent) (0.82 to 1.27) (0.80 to 1.22) (0.82 to 1.19) (0.75 to 1.19) (0.62 to 1.36) (0.96 to 1.03) No. of cases 188 160 181 405 130 30 1094 No. person-years of 283 658 213 615 244 605 562 249 203 998 52 047 1 560172 HR for 1 Frequency of decaffeinated coffee consumption additional cup of coffee/day Never ¼ cup/day ½ cup/day 1 1.9 cups/day 2 cups/day Ptrend (95% CI) HR, women (95% CI) 1.00 0.87 0.77 0.75 0.78.04 0.95 (referent) (0.70 to 1.10) (0.60 to 0.99) (0.57 to 0.97) (0.59 to 1.04) (0.87 to 1.03) HR, men (95% CI) 1.00 0.59 0.66 0.46 0.92.77 0.98 (referent) (0.36 to 0.96) (0.41 to 1.06) (0.26 to 0.84) (0.61 to 1.40) (0.86 to 1.11) HR, pooled (95% CI) 1.00 0.81 0.74 0.69 0.82.13 0.96 (referent) (0.66 to 1.00) (0.60 to 0.93) (0.54 to 0.88) (0.65 to 1.04) (0.89 to 1.03) Total no. of cases 324 144 123 95 107 793 Total person-years of 375680 201829 176013 146038 133803 1033363 HR for 1 Frequency of tea consumption additional cup of tea/day Never ¼ cup/day ½ cup/day 1cup/day 2 cups/day Ptrend (95% CI) HR, women (95% CI) 1.00 0.79 0.82 1.01 1.00.09 1.05 (referent) (0.64 to 0.97) (0.66 to 1.00) (0.81 to 1.25) (0.79 to 1.26) (0.98 to 1.12) HR, men (95% CI) 1.00 0.92 1.11 1.41 0.92.89 0.95 (referent) (0.63 to 1.34) (0.73 to 1.67) (0.88 to 2.25) (0.44 to 1.52) (0.79 to 1.14) HR, pooled (95% CI) 1.00 0.82 0.87 1.07 0.97.10 1.04 (referent) (0.68 to 0.98) (0.72 to 1.04) (0.88 to 1.31) (0.78 to 1.21) (0.98 to 1.10) No. of cases 253 258 239 177 130 1057 No. person-years of 332 797 409 465 366 844 232 804 202 926 1 544 836 HR for an Quintiles of caffeine intake additional 100 mg caffeine/ 1 2 3 4 5 P trend day (95% CI) HR, women (95% CI) 1.00 1.03 1.00 1.03 1.02.91 1.00 (referent) (0.83 to 1.27) (0.81 to 1.24) (0.83 to 1.27) (0.82 to 1.28) (0.97 to 1.03) HR, men (95% CI) 1.00 0.68 0.93 0.80 0.87.96 0.99 (referent) (0.43 to 1.09) (0.61 to 1.42) (0.51 to 1.24) (0.56 to 1.34) (0.93 to 1.05) HR, pooled (95% CI) 1.00 0.96 0.99 0.98 0.99.99 1.00 (referent) (0.79 to 1.16) (0.82 to 1.20) (0.81 to 1.19) (0.81 to 1.20) (0.97 to 1.03) No. of cases 216 215 228 217 206 1082 No. person-years of 309 091 321 871 320 374 314 834 298 136 1 564 306 (Table continues) analysis of coffee consumption and the risk of colorectal cancer that used the combined data from all prospective studies included 931 cases of colorectal cancer (3 ) ; the present study comprised 1433 cases of colorectal cancer. Results of case control studies that have examined this association have suggested that coffee consumption is associated with a decreased risk of colorectal cancer, whereas results of prospective cohort studies have suggested that coffee consumption is not associated with a reduction in colorectal cancer incidence (3,26 ). Results from case control studies have to be evaluated with caution because these studies are subject to selection bias and to differential misclassification of coffee consumption because participants recall of coffee consumption from the time before their cancer diagnosis may be influenced by their current coffee consumption and by their disease status. In addition, most of the case control studies were conducted in Europe, where coffee preparation and brewing methods may be different from those used in the United States. Results from only one case control study that was carried out in the United States using hospital-based control subjects suggested that coffee consumption is inversely associated with the risk of colorectal cancer (27 ). 290 ARTICLES Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005

Table 6 (continued). Nonsmokers and non-drinkers of alcoholic beverages HR for an Quintiles of caffeine intake additional 100 mg caffeine/ 1 2 3 4 5 Ptrend day (95% CI) HR, women (95% CI) 1.00 0.99 1.08 1.05 0.93.72 0.98 (referent) (0.74 to 1.32) (0.81 to 1.24) (0.78 to 1.41) (0.69 to 1.26) (0.94 to 1.03) HR, men (95% CI) 1.00 0.74 1.28 1.01 1.05.55 1.01 (referent) (0.42 to 1.31) (0.78 to 2.11) (0.60 to 1.70) (0.62 to 1.78) (0.94 to 1.08) HR, pooled (95% CI) 1.00 0.93 1.13 1.04 0.96.97 0.99 (referent) (0.72 to 1.21) (0.88 to 1.45) (0.80 to 1.35) (0.74 to 1.25) (0.96 to 1.03) No. of cases 133 113 130 119 113 608 No. person-years of 184146 162545 146783 150710 159633 803817 * HR = hazard ratio; CI = confidence interval. Wald test for trend; two-sided. Adjusted for age, family history of colorectal cancer, history of sigmoidoscopy, height, body mass index, pack-years of smoking, physical activity, aspirin use, vitamin supplement intake, alcohol consumption, red meat consumption, total caloric intake, and, among women in addition for menopausal status, postmenopausal hormone use Additionally adjusted for consumption of caffeinated coffee. Our unanticipated finding of an inverse association between decaffeinated coffee consumption and the incidence of rectal cancer is consistent with a threshold effect rather than a dose response relation. Only one other study (28 ) has examined associations between decaffeinated coffee consumption and the risks of colon and rectal cancer, perhaps because dietary assessment instruments do not always differentiate between caffeinated and decaffeinated coffee. In that case control study, conducted in Italy, the odds ratios associated with decaffeinated coffee consumption were 0.92 (95% CI = 0.72 to 1.18) for colon cancer and 0.88 (95% CI = 0.65 to 1.20) for rectal cancer; however, in that study population only 4% of the case patients and control subjects consumed decaffeinated coffee (28 ). Our data suggest that individuals who regularly drink decaffeinated coffee are more health conscious in their behaviors than those who do not drink decaffeinated coffee, in that a higher percentage had undergone sigmoidoscopy screening for colorectal cancer and reported using vitamin supplements. Many individuals who drink decaffeinated coffee may be former caffeinated coffee drinkers who like the taste of coffee but cannot or do not want to drink caffeinated coffee. Thus, residual confounding by healthy lifestyle and chance should be considered as possible explanations for the observed inverse association between decaffeinated coffee consumption and the incidence of colorectal cancer, especially given that we observed no dose response relation. However, participants who reported that they regularly consumed decaffeinated coffee were not more physically active, did not drink less alcohol or smoke less, and did not consume less red meat than those who did not drink decaffeinated coffee. Furthermore, although tea drinkers had healthy lifestyle patterns similar to those of decaffeinated coffee drinkers, the incidence rates of colorectal cancer were not lower among those who frequently consumed tea than among those who did not. Few of the participants in our study reported very high consumption of either tea or decaffeinated coffee. The consistency of the inverse association between decaffeinated coffee consumption and rectal cancer incidence across both cohorts and the observation that the association between decaffeinated coffee consumption and the incidence of colon cancer was restricted to cancers in the proximal colon among both women and men lend credibility to our results. A number of biologic mechanisms could underlie the observed inverse association between decaffeinated coffee consumption and colorectal cancer. Given its constituents, intake of caffeinated coffee could either increase or decrease the incidence of colorectal cancer. Decaffeinated coffee, however, may not contain potentially harmful constituents, such as caffeine, or have the homocysteine-raising effect of coffee (29 ). It has been reported that both regular and decaffeinated coffee increase rectosigmoid motility (10 ). However, we previously found that, among the women in our cohort, the self-reported frequency of bowel movements was not associated with colon or rectal cancer risk (30 ). By contrast, in another study population constipation was associated with an increased risk of colon cancer (31 ). Among participants in the NHS, we found a U-shaped relation between coffee consumption and bowel movement: modest consumption was associated with a decreased risk of constipation, whereas frequent consumption was associated with increased risk of constipation, possibly because of the dehydrating effect of high coffee intake (32 ). Reports on the association between tea consumption and the risk of colorectal cancer have been inconsistent. Tea consumption was associated with an increased incidence of colon cancer but not rectal cancer in a cohort of Finnish men (33 ). In a Swedish case control study, tea consumption was associated with a reduced risk of rectal cancer for participants who drank two or more cups per day; no association between tea consumption and colon cancer risk was found (34 ). Frequent tea consumption was associated with a decreased incidence of colon cancer in the National Health and Nutrition Examination Survey I Epidemiologic Follow-up study (35 ). Other studies (28,36,37 ) have found no association between self-reported tea consumption and the risk of colorectal cancer. Data collected in any observational study are measured with error, and nondifferential measurement error has to be considered as a possible explanation for the lack of association we observed for coffee consumption and incidence of colorectal cancer. However, results from our validation study suggest that coffee consumption was reported with good accuracy. Moreover, our findings are largely consistent with previous cohort studies on this topic ( 3, 26 ). Furthermore, the repeated assessments Journal of the National Cancer Institute, Vol. 97, No. 4, February 16, 2005 ARTICLES 291