NEJM a écrit :
Study Population
The NIH–AARP Diet and Health Study has been described previously.32 Between 1995 and 1996, a total of 617,119 AARP members, 50 to 71 years of age, returned a comprehensive questionnaire assessing diet and lifestyle. Participants resided in six states (California, Florida, Louisiana, New Jersey, North Carolina, and Pennsylvania) and two metropolitan areas (Atlanta and Detroit). Of the respondents, 566,401 completed the questionnaire satisfactorily. Completion of the self-administered questionnaire was considered to imply informed consent to participate in the study.
We excluded from these analyses 15,760 persons whose questionnaires were completed by a spouse or other surrogate correspondent, as well as 51,234 persons with cancer, 65,044 with heart disease, 10,459 who had had a previous stroke, 2082 who did not provide information on coffee use, 15,820 who did not provide information on cigarette smoking, 3731 with an extremely low or high caloric consumption (two times as high as the 75th percentile of caloric intake or two times as low as the 25th percentile of caloric intake), and 11 who died before their completed questionnaire was received. The resulting analytic cohort included 229,119 men and 173,141 women. The NIH–AARP Diet and Health Study was approved by the Special Studies Institutional Review Board of the National Cancer Institute.
Assessment of Exposure
Participants completed a baseline questionnaire that assessed demographic and lifestyle characteristics and 124 dietary items, as previously described.32 Consumption of fruits, vegetables, red meat, white meat, and saturated fat were adjusted for total energy intake with the use of the nutrient-density approach (i.e., measured per 1000 kcal per day for food groups and as a percentage of total energy for saturated fat).
Coffee consumption was assessed according to 10 frequency categories, ranging from 0 to 6 or more cups per day. In addition, 96.5% of coffee drinkers provided information on whether they drank caffeinated or decaffeinated coffee more than half the time, and we used this information to categorize coffee drinkers.
In a subgroup of 1953 study participants who also completed a 24-hour dietary-recall questionnaire on 2 nonconsecutive days,33 the Spearman coefficient for the correlation between coffee consumption as assessed with this questionnaire and coffee consumption as assessed with the baseline food-frequency questionnaire was 0.80. The respective Spearman correlation coefficients for caffeinated and decaffeinated coffee were 0.64 and 0.48, respectively. Among participants who completed the 24-hour dietary-recall questionnaire, 79% drank ground coffee, 19% drank instant coffee, 1% drank espresso coffee, and 1% did not specify the type of coffee they consumed.
Cohort Follow-up
Participants were followed from baseline (1995–1996) until the date of death or December 31, 2008, whichever came first, by means of linkage to the National Change of Address database maintained by the U.S. Postal Service, specific change-of-address requests from participants, and updated addresses returned during other mailings. Vital status was assessed by periodic linkage of the cohort to the Social Security Administration Death Master File, linkage with cancer registries, questionnaire responses, and responses to other mailings.
Causes of Death
Specific causes of death were obtained through follow-up linkage to the National Death Index Plus, maintained by the National Center for Health Statistics. We used the International Classification of Diseases, Ninth Revision (ICD-9), and International Classification of Diseases, 10th Revision (ICD-10) codes to classify the underlying cause of death (obtained from death certificates) as follows: cancer (ICD-9, 140–239; ICD-10, C00–C97 and D00–D48), heart disease (ICD-9, 390–398, 401–404, 410–429, and 440–448; ICD-10, I00–I13, I20–I51, and I70–I78), respiratory disease (e.g., pneumonia, influenza, chronic obstructive pulmonary disease, and associated conditions) (ICD-9, 480–487 and 490–496; ICD-10, J10–J18 and J40–J47), stroke (ICD-9, 430–438; ICD-10, I60–I69), injuries and accidents (e.g., accident, suicide, and homicide) (ICD-9, 800–978; ICD-10, V01–X59, Y85–Y86, U03, X60–X84, Y87.0, U01–U02, X85–Y09, Y35, Y87.1, and Y89.0), diabetes (ICD-9, 250; ICD-10, E10–E14), infections (e.g., tuberculosis, septicemia, and other infectious and parasitic diseases) (ICD-9, 001–139; ICD-10, A00–B99), and all other causes.
State data on the incidence of cancer were obtained from the Arizona Cancer Registry, the Georgia Center for Cancer Statistics, the California Cancer Registry, the Michigan Cancer Surveillance Program, the Florida Cancer Data System, the Louisiana Tumor Registry, the New Jersey State Cancer Registry, the North Carolina Central Cancer Registry, the Pennsylvania Cancer Registry, and the Texas Cancer Registry.
Statistical Analysis
Coffee consumption was tabulated according to a number of dietary and lifestyle factors. Hazard ratios and 95% confidence intervals for mortality associated with coffee consumption were estimated with the use of Cox proportional-hazards regression models, with person-years as the underlying time metric; results calculated with age as the underlying time metric were similar. We tested the proportional-hazards assumption by modeling the interaction of follow-up time with coffee consumption and observed no significant deviations. Analyses were conducted with the use of SAS software, version 9.1. Statistical tests were two-sided, and P values of less than 0.05 were considered to indicate statistical significance.
We present risk estimates separately for men and women. Multivariate models were adjusted for the following baseline factors: age; body-mass index (BMI); race or ethnic group; level of education; alcohol consumption; the number of cigarettes smoked per day, use or nonuse of pipes or cigars, and time of smoking cessation (<1 year, 1 to <5 years, 5 to <10 years, or ?10 years before baseline); health status; presence or absence of diabetes; marital status; level of physical activity; total energy intake; consumption of fruits, vegetables, red meat, white meat, and saturated fat; and use of any vitamin supplement (yes vs. no). In addition, risk estimates for death from cancer were adjusted for history of cancer (other than nonmelanoma skin cancer) in a first-degree relative (yes vs. no). For women, status with respect to postmenopausal hormone therapy was also included in multivariate models. Less than 5% of the cohort lacked any single covariate; for each covariate, we included an indicator for missing data in the regression models, if necessary. In a sensitivity analysis, we adjusted for propensity scores34 that reflected associations of coffee consumption with the other variables in the multivariate-adjusted models. Results obtained with the use of propensity-score adjustment were very similar to those from multivariate-adjusted models (Table 1 in the Supplementary Appendix, available with the full text of this article at NEJM.org).
Hazard ratios for death associated with categories of coffee consumption (<1, 1, 2 or 3, 4 or 5, and ?6 cups per day), as compared with no coffee consumption, were estimated from a single model. Tests of linear trend across categories of coffee consumption were performed by assigning participants the midpoint of their coffee-consumption category and entering this new variable into a separate Cox proportional-hazards regression model.
In secondary analyses, we determined risk estimates for categories of consumption of caffeinated and decaffeinated coffee and examined associations among prespecified baseline subgroups based on the following: follow-up time; age; cigarette-smoking status; presence or absence of diabetes; BMI; alcohol consumption; self-reported health; high or low consumption of red meat, white meat, fruits, and vegetables; use or nonuse of any vitamin supplement; and, in women, use or nonuse of postmenopausal hormone therapy. For these analyses, we combined the categories of 4 or 5 cups of coffee per day and 6 or more cups per day to preserve numbers in the top category of consumption. P values for interactions were computed by means of likelihood-ratio tests comparing Cox proportional-hazards models with and those without cross-product terms for each level of the baseline stratifying variable, with coffee consumption as an ordinal variable. For total mortality, we performed 12 interaction tests for men and 13 interaction tests for women. We also performed interaction tests for smoking status with eight different outcomes for both men and women. Several differences (P<0.05) would be expected by chance alone.
