Individuals aged 50 to 71 years were recruited from six U.S. states (California, Florida, Louisiana, New Jersey, North Carolina, and Pennsylvania) and two metropolitan areas (Atlanta, Georgia, and Detroit, Michigan) to form a large prospective cohort, the NIH-AARP Diet and Health Study. Questionnaires on demographic and lifestyle characteristics, including dietary habits, were mailed to 3.5 million members of AARP in 1995, described in detail elsewhere.13
The NIH-AARP Diet and Health Study was approved by the Special Studies Institutional Review Board of the U.S. National Cancer Institute. Completion of the baseline questionnaire was considered to imply informed consent.
A 124-item food frequency questionnaire (FFQ) (http://riskfactor.cancer.gov/DHQ/forms/files/shared/dhq1.2002.sample.pdf
) was completed at baseline. The FFQ collected information on the usual consumption of foods and drinks and portion sizes over the last twelve months. The validity of the FFQ was estimated using two 24-hour recalls,14
and the estimated energy adjusted correlations ranged from 0.36 to 0.76 for various nutrients, and attenuation factors ranged from 0.24 to 0.68. Red meat intake was calculated using the frequency of consumption and portion size information of all types of beef and pork and included bacon, beef, cold cuts, ham, hamburger, hot dogs, liver, pork, sausage, steak and meats in foods such as pizza, chili, lasagna, and stew. White meat included chicken, turkey, and fish and included poultry coldcuts, chicken mixtures, canned tuna, as well as low-fat sausages and low-fat hot dogs made from poultry. Processed meat included bacon, red meat sausage, poultry sausage, luncheon meats (red and white meat), cold cuts (red and white meat), ham, regular hotdogs and low-fat hotdogs made from poultry. The components constituting red or white and processed meats can overlap as both can include meats such as bacon, sausage, and ham, while processed meat can also included smoked turkey and chicken. However, these meat groups are not used in the same models thus they are not duplicated in any one analysis.
In order to investigate whether the overall composition of meat intake was associated with mortality, we created three diet types: high; medium; and low risk meat diet. To form these diet variables, red and white meat consumption was energy adjusted and split into two groups using the median values as a cutpoints. Individuals with red meat consumption in the upper half and white meat consumption in the lower half got a score of 1 (high risk meat diet), those with both red and white meat consumption in the same half got a score of 2 (medium risk meat diet), those with red meat consumption in the lower half and white meat consumption in the upper half got a score of 3 (low risk meat diet).
Cohort follow-up and case ascertainment
Cohort members were followed-up from the date the baseline questionnaire was returned (beginning 1995) through December 31, 2005 by annual linkage of the cohort to the National Change of Address database maintained by the U.S. Postal Service and through processing of undeliverable mail, other address change update services, and directly from cohort members' notifications. For matching purposes, we have virtually complete data on first and last name, address history, gender, and date of birth. Follow-up for vital status is performed by annual linkage of the cohort to the Social Security Administration Death Master File in the U.S. Verification of vital status and cause of death information is provided by follow-up searches of the National Death Index (NDI) Plus with the current follow-up for mortality covered until 2005.
Cause-Specific Case Ascertainment
Cancer (ICD9: 140-239; ICD10: C00-C44, C45.0, C45.1, C45.7, C45.9, C48-C97, D12-D48) - mortality included deaths due to cancers of the oral cavity and pharynx, digestive tract, respiratory tract, soft tissue (including heart), skin (excluding basal and squamous cell carcinoma), female genital system and breast, male genital system, urinary tract, endocrine system, lymphoma, leukemia, and other miscellaneous cancers.
Cardiovascular disease (CVD) (ICD9: 390-398, 401-404, 410-438, 440-448; ICD10: I00-I09, I10-I13, I20-I51, I60-I78) - mortality was from a combination of diseases of the heart, hypertension without heart disease, cerebrovascular diseases, atherosclerosis, aortic aneurysm and dissection, and other diseases of the arteries, arterioles, and capillaries.
Mortality from injuries and sudden deaths (ICD9: 800-978; ICD10: U01-U03, V01-Y09, Y35, Y85-Y86, Y87.0, Y87.1 Y89.0) - included accidents, adverse effects, suicide, self-inflicted injury, homicide, and legal intervention.
All others deaths included mortality from tuberculosis, human immunodeficiency virus, other infectious and parasitic diseases, septicemia, diabetes mellitus, Alzheimer's, stomach and duodenal ulcers, pneumonia and influenza, chronic obstructive pulmonary disease and allied conditions, chronic liver disease and cirrhosis, nephritis, nephrotic syndrome and nephrosis; congenital anomalies; certain conditions originating in the perinatal period, ill-defined conditions, and unknown causes of death.
Total mortality is a combination of all of the above mentioned causes of deaths.
A total of 617,119 persons returned the baseline questionnaire; of these, we excluded individuals who moved out of the eight study areas before returning the baseline questionnaire (n = 321), requested to be withdrawn from the study (n = 829), died before study entry (n = 261), had duplicate records (n = 179), indicated that they were not the intended respondent and did not complete the questionnaire (n = 13,442), provided no information on gender (n = 6), did not answer substantial portions of the questionnaire or had greater than 10 recording errors (n = 35,679). After these exclusions, we further removed individuals whose questionnaire was filled in by someone else on their behalf (n = 15,760). We excluded 4,849 subjects reporting extreme daily total energy intake defined as more than two inter-quartile ranges above the 75th percentile or below the 25th percentile and 140 people who had zero person years of follow up. After all exclusions, our analytic cohort consisted of 322,263 men and 223,390 women.
We estimated hazard ratios (HRs) and 95% confidence intervals (CIs) with time since entry into the study as the underlying time metric using Cox proportional hazards regression. Quintile cut-points were based on the entire cohort and multivariate adjusted HRs are reported using the lowest quintile as the referent category. The violation of the proportional hazard assumption was investigated by testing an interaction between a time dependent binary covariate, which indicated if follow-up was in the first 5 years or in the second 5 years, and the quintile terms for meat consumption. Dietary variables were energy adjusted using the nutrient density method and meat variables in each model added up to total meat (addition model). For example, one model contained both red and white meat while the processed meat model also contained a non-processed meat variable.
In order to address confounding we used forward stepwise variable selection to include covariates to develop the fully adjusted model. Smoking was the largest confounder of the association between meat intake and mortality. Physical activity and education were also important covariates, but not to the same degree as smoking. The final model included: age (continuous); education (less than 8 years or unknown, 8 to 11 years, 12 years (high school), some college, college graduate); marital status (married: yes/no); family history of cancer (yes/no) (cancer mortality only); race (non-Hispanic white, non-Hispanic black, Hispanic/Asian/Pacific Islander/American Indian/Alaskan native or unknown); body mass index (18.5-<25, 25-<30, 30-<35, ≥35 kg/m2); 31-level smoking history using smoking status (never, former, current), time since quitting for former smokers, and smoking dose; frequency of vigorous physical activity (never/rarely, 1-3 times/month, 1-2 times/week, 3-4 times/week, 5 or more times/week); total energy intake (continuous); alcohol intake (none, 0-<5, 5-<15, 15-<30, ≥30 g/day); vitamin supplement user (one or more supplement per month); fruit consumption (0 -< 0.7, 0.7-< 1.2, 1.2-< 1.7, 1.7-< 2.5, ≥2.5 servings/1000 kcal); vegetable consumption (0-<1.3, 1.3-< 1.8, 1.8-< 2.2, 2.2-<3.0, ≥3.0 servings/1000 kcal); and menopausal hormone therapy among women in the multivariate models.
In sub-analyses, we investigated the relation between meat intake and mortality by smoking status. We used median values of each quintile to test for linear trend with two-sided P-values. We also calculated population attributable risks (PAR) as an estimate of the percent of mortality that could be prevented if individuals adopted intake levels of participants within the first quintile. This was computed as one minus the ratio consisting of the sum of the estimated hazards (derived from the Cox proportional hazard models) of each member of the cohort divided by the sum of the estimated hazards where meat exposure was assigned to the lowest or highest quintile, depending on which quintile was the ideal level of meat consumption. The PAR was multiplied by 100 to convert them to a percentage. All statistical analyses were carried out using Statistical Analytic Systems (SAS) software (SAS Institute Inc, Cary, NC).