Diet and Risk of Myeloproliferative Neoplasms in Older Individuals from the NIH-AARP Cohort

Philadelphia chromosome–negative myeloproliferative neoplasms (MPN), a group of hematological malignancies characterized by overproliferation of myeloid cells, include three major subtypes: polycythemia vera (PV), essential thrombocythemia (ET), and primary myelofibrosis (PMF; ref. 1). On the basis of data from the Surveillance, Epidemiology and End Results program in the United States, the median age of diagnosis is 65, 67, and 69 years for PV, ET, and PMF, respectively (2), and the 5-year survival is approximately 78%, 79%, and 40% (3). Despite the exciting discoveries of JAK2, calreticulin, and myeloproliferative leukemia virus oncogene mutations in the pathogenesis of MPN (4), the etiology of MPN remains understudied. Two factors could have contributed to the paucity of research in this area: first, MPN are rare—the incidence of the three subtypes combined is approximately 2.8 per 100,000 per year in the United States, which makes it difficult to accumulate a large number of cases in etiological studies; second, MPN were not classified as malignancies by the World Health Organization until 2000 (5), and as such, were not required to be reported to cancer registries until 2001, hindering population-based case ascertainment in earlier efforts.

Existing epidemiologic studies of MPN have evaluated the potential etiological roles of environmental, lifestyle, and familial factors (6–14). None of them has assessed the possible role of diet, although dietary factors are considered to account for approximately 30% of cancer incidence in Western countries (15). For example, processed meat has been classified as “carcinogenic to humans” (Group 1) by the International Agency for Research on Cancer (16). On the other hand, there is uncertainty about the consumption of fruits and vegetables. As a few large prospective studies did not suggest a protective effect of fruit and vegetable consumption on cancer risk (17–19), the World Cancer Research Fund and the American Institute for Cancer Research concluded that a protective role for fruits, vegetables, and their constituents against cancer is “limited-suggestive” (2).

In this analysis, we conducted the first evaluation of diet factors and MPN risk in the NIH-AARP Diet and Health Study, a large prospective cohort of older individuals with extensive follow-up. Given the lack of studies on this topic and a concern about multiple comparison, we set out to focus our analysis on three macronutrients (carbohydrate, fat, and protein) and three food groups (meat, vegetables, and fruits). An earlier study using NIH-AARP data observed an association between total sugar intake and risk of leukemia (20), so we also planned to examine sugar intake in relation to MPN risk.

Design of the NIH-AARP cohort has been described previously (21). Briefly, during 1995 to 1996, a self-administered questionnaire on diet and health was mailed to approximately 3.5 million AARP members ages 50 to 71 years in six states (California, Florida, Louisiana, New Jersey, North Carolina, and Pennsylvania) and two metropolitan areas (Atlanta, Georgia, and Detroit, Michigan). Of the AARP members contacted, 566,398 satisfactorily completed and returned the questionnaire and were considered participants of the cohort at the baseline. The study was approved by the Special Studies Institutional Review Board of the NCI, and all participants gave written informed consent by virtue of completing and returning the questionnaire.

For this analysis, we excluded participants whose questionnaires were filled by a proxy responder (n = 15,760), who had prevalent cancer (n = 51,260), poor health (n = 8,365), end-stage renal disease (n = 769), cancer diagnoses reported only from death certificate (n = 4,117), zero person-years of follow-up (n = 48), missing body mass index (BMI; n = 11,390), total energy intake (n = 4,270), or BMI outside of three times the interquartile range (IQR) above the 75th or below the 25th percentile of the corresponding log-transformed value of the study population (n = 3,885). In addition, we excluded 3,485 participants whose BMI were lower than 18.5 kg/m², out of concern that their dietary intake might be distinct from the rest of the cohort. These exclusion criteria were consistent with those employed by another MPN analysis using the NIH-AARP data (14).

In the NIH-AARP cohort, incident cancer cases were identified through probabilistic linkage with population-based cancer registries in the eight original states and three additional states where participants most frequently moved during follow-up (Arizona, Nevada, and Texas). Cancer registry linkage with eight states identified approximately 90% of all cancer incidences in this cohort (3). MPN incidence during follow-up was ascertained from baseline through December 31, 2011. Because MPN were not required to be reported to cancer registries until 2001, the reporting of MPN prior to 2001 was sporadic and infrequent. Vital status was assessed through December 31, 2011, using the National Death Index. Incident MPN cases were identified using histologic codes from the International Classification of Diseases for Oncology, Third Edition: 9950 for PV, 9961 for PMF, 9962 for ET, 9960 for not otherwise specified MPN, and 9975 for unclassifiable MPN (5).

At baseline, participants were asked to report their usual dietary intake of foods and beverages over the past 12 months in both frequency of intake and portion size, using a 124-item food-frequency questionnaire developed and validated by the NCI. For this study, MyPyramid Equivalents Database values were used. We initially focused on macronutrients (i.e., carbohydrate, fat, and protein) and food groups that have been linked to cancer risk in previous studies (meat, vegetables, and fruits; refs. 2, 16). Given the positive finding regarding fruit intake, we further assessed groups of fruits and sugar intake from various sources.

The self-administered questionnaire also elicited information on a variety of demographic, medical, and lifestyle factors, including age at baseline, race/ethnicity, marital status, family history of cancer, personal history of multiple health conditions (e.g., heart disease, emphysema, diabetes, stroke), physical activity, and smoking history.

In this study, person-years of follow-up were calculated from the date the questionnaire was returned through the date of MPN diagnosis or the date of censoring, which included relocation out of the registry area, loss to follow-up, death, or the end of follow-up (December 31, 2011), whichever occurred earliest. Because MPN were not required to be reported to cancer registries until 2001, we also calculated person-years of follow-up starting from January 1, 2001.

All dietary factors were categorized into tertiles (low: first tertile; medium: second tertile; and high: third tertile) based on their distributions in the overall study population. For all variables, missing values were categorized as a separate group. Multivariable Cox proportional hazards regression models were utilized to estimate hazard ratios (HR) and 95% confidence intervals (CI), adjusting for age (<55, 55–59, 60–64, 65–69, and ≥70 years), sex, race (white, non-white, unknown), education (<12, 12–15, and ≥16 years, unknown), marital status (married, not married, unknown), self-reported chronic diseases (heart disease, emphysema, diabetes, and stroke as four binary variables), BMI (18.5–24.9, 25–29.9, ≥30 kg/m²), and total calories. Tests for linear trend were performed by ordering the ordinal exposure categories from the lowest to the highest levels and including the variable as a 1 degree-of-freedom linear term in the Cox regression models. All analyses were conducted for MPN overall and two specific types of MPN: PV (n = 190) and ET (n = 146). We chose not to carry out separate analyses for PMF due to the small number of patients identified (n = 67), or MPN that is unclassifiable or not otherwise specified (n = 87) due to the heterogeneity of this group.

As multiple comparison is a concern in any study with a large number of factors to be evaluated, we decided to focus on three macronutrients (carbohydrate, fat, and protein) and three food groups (meat, vegetables, and fruits). Because the initial analysis revealed that fruit consumption was associated with MPN risk and an earlier NIH-AARP analysis observed an association between total sugar intake and leukemia risk (20), we further examined different groups of fruits and sugar intake from various sources.

Since an earlier study linked coffee consumption and smoking to MPN risk (14), we also considered adjusting for these factors in our multivariable models. However, the adjustment of these factors had no appreciable impact on the HR estimates for any of the dietary factors, hence we decided not to include them as potential confounders.

All analyses described above were conducted with person-years of follow-up calculated from two different time points: (i) the date the baseline questionnaire was returned (i.e., 1995–1996) and (ii) the date when MPN were required to be reported to cancer registries (i.e., January 1, 2001). As MPN reporting prior to 2001 was sporadic and infrequent, we had more confidence in results from the analyses for which follow-up was considered to have begun in 2001 (Table 1; Figs. 1 and 2; Supplementary Fig. S1). However, the results from the analyses with baseline being the beginning of follow-up were very similar (Supplementary Table S1).

All statistical analyses were two-sided with a type I error of 0.05 for statistical significance and were performed using SAS Version 9.4 (SAS Inc.).

To help provide a context for the interpretation of our findings, we estimated statistical power for three comparison groups with equal sizes (i.e., three tertiles for dietary factors), assuming a R² of 0.1, 0.15, or 0.2 for the regression of one variable on the remaining variables in the model. If we focus on the follow-up from 2001 on, a statistical power of 0.8 or above is expected for HR ≥1.2 for the analysis of MPN as a group, HR ≥1.35 for the analysis of PV, and HR ≥1.45 for the analysis of ET.

A total of 463,049 participants were included in the final cohort. After a median follow-up of 15.5 (IQR = 14.4–15.6) years, 490 developed MPN (190 PV, 146 ET, 67 PMF, and 87 unclassifiable/not otherwise specified MPN). A majority of the MPN patients were male (n = 321, 65.5%), white (n = 444, 90.6%), and married (n = 344, 70.2%), whereas 220 (44.9%) ages ≥65 years at baseline.

The analysis with 2001 being the beginning of follow-up included 432,940 participants and 472 MPN patients (182 PV, 143 ET, 67 PMF, and 80 unclassifiable/not otherwise specified MPN).

Multivariable Cox proportional hazards regression analysis suggested that the risk of MPN overall, PV, or ET was not influenced by the intake of protein, fat, meat, red meat or processed meat (Table 1). Neither total vegetable and fruit intake nor total vegetable intake exhibited an association with the risk of MPN overall, PV, or ET. However, participants whose carbohydrate intake was in the third (i.e., highest) tertile had a higher risk of PV compared with those with carbohydrate intake in the first (i.e., lowest) tertile (HR = 1.78; 95% CI, 1.04–3.05). Compared with individuals who had low fruit consumption (≤1.20 cups/day), those with high levels of fruit consumption (≥2.27 cups/day) had a significantly increased risk of MPN (HR = 1.32; 95% CI, 1.04–1.67), and the magnitude of association was larger for PV (HR = 2.00; 95% CI, 1.35–2.95); a positive trend was also observed for both MPN overall (P_trend = 0.02) and PV (P_trend < 0.01). A closer examination of specific groups of fruits revealed a statistically significant, increased risk of MPN in cohort participants who had high levels of citrus, melon, and berry intake (HR = 1.28; 95% CI, 1.01–1.61), or medium (HR = 1.26; 95% CI, 1.00–1.59) or high (HR = 1.30; 95% CI, 1.03–1.65) consumptions of other fruits, as well as an increased risk of PV in cohort participants who had medium (HR = 1.51; 95% CI, 1.03–2.23) or high (HR = 1.80; 95% CI, 1.22–2.66) levels of citrus, melon, and berry intake, or medium (HR = 1.68; 95% CI, 1.14–2.47) or high (HR = 1.74; 95% CI, 1.17–2.58) levels of fruit juice consumption, compared with those with low consumption (i.e., whose intake was in the lowest tertile). On the other hand, we did not observe any evidence of association between total fruit intake or the consumption of any specific groups of fruits and the risk of ET (Table 1).

Total sugar intake did not appear to be a significant predictor for the risk of MPN overall, neither did sugar from natural sources, natural beverage sources, nonnatural sources, nonnatural beverage sources, or fructose (Supplementary Fig. S1).

As for PV, compared with individuals in the lowest tertile of exposure, those who had high level of total sugar intake (HR = 1.77; 95% CI, 1.12–2.79), medium (HR = 1.65; 95% CI, 1.12–2.44), or high (HR = 1.77; 95% CI, 1.16–2.71) level of sugar intake from natural sources, high level of sugar intake from natural beverage sources (HR = 1.57; 95% CI, 1.08–2.29), medium level of sugar intake from nonnatural beverage sources (HR = 1.76; 95% CI, 1.21–2.54), and medium (HR = 1.53; 95% CI, 1.04–2.27) or high (HR = 1.84; 95% CI, 1.21–2.79) level of fructose intake had an increased risk of PV. In addition, there was a significant trend for total sugar, sugar from natural sources, sugar from natural beverage sources, and fructose (Fig. 1).

We did not observe any association between the intake of total sugar or different sources of sugar and the risk of ET (Fig. 2).

In this large prospective cohort study of older individuals, we found that a higher level of sugar consumption was associated with an increased risk of MPN, and this association appeared to be driven by PV. PV patients also had a higher intake of total carbohydrate, sugar, and fructose. None of these factors were associated with the risk of ET, another common type of MPN. The other dietary factors that we assessed, including the intake of fat, protein, meat, and vegetables, did not influence the risk of MPN overall, PV or ET.

To the best of our knowledge, this study is the first to evaluate diet and the risk of MPN, so we are unable to compare the results directly with similar studies. However, fruits and vegetables are among the most widely studied dietary factors in relation to carcinogenesis. Although the consumption of fruits and vegetables has been inversely associated with the risk of multiple types of cancer in case–control studies (22–25), there was only a weak or null association in cohort studies of colorectal cancer (26, 27), breast cancer (28), lung cancer (29, 30), and cancer overall (18, 30–32). Earlier analyses of diet and the risk of other myeloid malignancies including myelodysplastic syndromes (33) and acute myeloid leukemia (34) in the NIH-AARP cohort did not implicate an etiologic role of fruit or vegetable consumption.

In our analysis, when we combined fruits and vegetables as a single group, there did not appear to be an association with MPN risk; the positive association we identified was specific to a higher level of fruit consumption. It is important to note that fruit consumption includes fresh fruits, dried fruits, and fruit juices. Most fruits are rich sources of sugar, particularly simple sugar, such as glucose and fructose (35). Fruit juices can contain more sugar per serving than fresh fruits. For example, the glycemic load values per serving are 6.2 for raw oranges and 13.4 for orange juice (36). When we examined specific groups of fruits, citrus, melon, and berry intake as well as fruit juices were significantly associated with an increased risk of PV, whereas the consumption of other fruits as a group did not appear to impact PV risk. A 2019 publication from a French cohort reported that the consumption of 100% fruit juices was positively associated with the risk of overall cancer (37), whereas another 2019 publication from a cohort in the United States linked the consumption of sugary beverages, including fruit juices, to increased all-cause mortality (38). However, our findings should not be interpreted in any way to discourage fruit consumption, especially when considering the rarity of PV and the lack of validation from other studies.

Our finding with regard to fructose intake was not surprising, given that fructose intake was highly correlated with total fruit consumption, with a correlation coefficient of 0.69 (P < 0.001) among all participants and 0.79 (P < 0.001) among the 490 participants who developed MPN in our study. Previous studies on the intake of sugar or fructose and cancer risk have reported inconsistent findings (39–43). In the NIH-AARP cohort, Tasevskaet and colleagues observed an increased trend of leukemia with total sugar (P_trend = 0.03) and added sugar (P = 0.02) among women, and women whose intake of added sugar was in the highest quintile experienced a 60% higher risk of leukemia than women in the lowest quintile (HR = 1.60; 95% CI, 1.03–2.48; ref. 20). In the same study, a higher fructose intake was associated with increased risks of bladder cancer in women and pleural cancer in men, but a reduced risk of oral cancer in men, ovarian cancer in women and lung cancer in both sexes (20). In another NIH-AARP study that focused on pancreatic cancer, a higher intake of both free fructose (Q5 vs. Q1; RR = 1.29; 95% CI, 1.04–1.59) and glucose (RR = 1.35; 95% CI, 1.1–1.67) increased disease risk (44). As sugar intake has increased dramatically in the United States over the last few decades (45–47) additional studies on their relation to cancer risk and potential mechanisms are warranted.

Multiple mechanisms may account for the role of fructose in the pathogenesis of MPN. First, fructose can produce advanced glycation end-products, which may be involved in the development and progression of cancer (48, 49). Second, higher fructose intake can lead to insulin resistance and hyperinsulinemia (20, 48), which can result in carcinogenesis through a highly active insulin-like growth factor pathway (50). The hematopoietic cells in MPN display a response to cytokines and growth factors such as erythropoietin, IL3, and insulin-like growth factor 1 (IGF1), especially PV (51, 52). Two IGF binding proteins (IGFBP-1 and IGFBP-3) are significantly elevated in MPN, and the IGF1 receptor signal pathway is more active in PV than in ET or PMF (53). Staerk and colleagues reported that the expression of JAK2V617F mutation rendered Ba/F3 cells hypersensitive to IGF1 stimulation (54). Recent study using transgenic mouse models of MPN showed that induction of JAK2V617F and JAK2 exon 12 mutations in hematopoietic cells leads to profound metabolic alternations including strongly increased metabolic demands, increase in glycolysis and vital dependence on glucose (55). Because the JAK2 mutations are present in approximately 95% of PV patients (4, 56), this could be the molecular mechanism connecting IGF1/IGF1 receptor signaling to PV. Given that the JAK2 mutation is less prevalent in ET (4), this may also help explain why higher fructose intake was not associated with ET risk in our cohort. Of course, the current study aimed to assess the possible role of diet in the etiology of MPN from an epidemiologic, and not physiologic or molecular perspective. Mechanistic evaluations would be important but are beyond the scope of our research.

This study has several strengths. It is based on a very large prospective cohort with extended follow-up that identified a relatively large number of cases, despite the rarity of MPN. The collection of dietary data predated MPN diagnosis, so our analysis was not prone to recall bias which might plague a case–control study. Because of the availability of data on sociodemographic and health-related factors in the NIH-AARP cohort, we were able to adjust for a number of covariates in our multivariable models, reducing the likelihood of confounding. Given the lack of existing studies on diet and MPN risk, our analysis was the first of its kind and filled an important knowledge gap.

On the other hand, we must highlight a few limitations that might inform the interpretation of our findings. First, although the number of cases was impressive given the low incidence of MPN, the sample size constrained some of the analyses, especially analyses for MPN subtypes. To preserve statistical power to the extent feasible, we made decisions a priori to focus only on a limited number of dietary factors and to categorize them into tertiles (instead of quintiles as used in some previous NIH-AARP analyses; refs. 20, 44). Second, the information on diet was obtained at baseline and was reflective of the participant's usual diet in the prior 12 months. Some changes in dietary intake over time was possible, leading to potential misclassification of exposure. Nevertheless, we are not concerned about dietary changes due to MPN incidence (i.e., reverse causality), given the lag time between baseline (1995–1996) and the first year that MPN was required to be reported to cancer registries (2001). The fact that MPN did not become reportable until 2001 also constitutes an important limitation. Of the 490 patients with MPN identified, 472 were diagnosed in or after 2001, and 18 were diagnosed between baseline and the end of 2000. We do not know why some cancer registries recorded the diagnosis of those 18 patients with MPN when MPN were not required to be reported. We addressed this by calculating person-years of follow up from two different time points (i.e., 1995–1996 vs. 2001) and found the two sets of results to be almost identical. Furthermore, we adjusted for a variety of potential confounders in the analysis, but residual confounding by unknown or unmeasured risk factors might have persisted. For example, we were unable to evaluate a possible role of pesticides or other chemicals used to grow vegetables and fruits. A Pennsylvania-based study of a PV cluster had raised concerns about exposure to hazardous chemicals and other environmental risk factors (57), but we did not have access to the residential histories of NIH-AARP participants and could not explore potential clustering in MPN incidence. Furthermore, sugar intake from different sources were correlated with each other and as such, it was difficult to single out any one source as being more relevant to the etiology of PV. Finally, NIH-AARP participants were 50 to 71 years of age at baseline, so most of the MPN cases we identified were ≥55 years at diagnosis. Although we do not consider this a major weakness given MPN's typical late onset in life (the median age of diagnosis ranges 65 to 69 years for PV, ET, and PMF; ref. 2), it impacted the generalizability of our results to the full age range of MPN patients.

In conclusion, in this prospective cohort study, participants who had a higher intake of sugar and fructose exhibited an increased risk of PV in older individuals. This finding needs to be replicated in other epidemiologic studies, preferably prospective cohorts, with adequate statistical power. Although PV is rare, recent findings from other studies regarding the possible impact of sugar intake on overall cancer risk and all-cause mortality underline the importance of potential mechanistic evaluations against the backdrop of rapidly increasing sugar consumption globally.

N.A. Podoltsev reports grants from The Frederick A. DeLuca Foundation during the conduct of the study, as well as personal fees from Pfizer, Agios Pharmaceuticals, Blueprint Medicines, Incyte, Novartis, Celgene, Bristol-Myers Squib, CTI BioPharma, and PharmaEssentia and other from Boehringer Ingelheim (principal investigator, clinical research reimbursement to institution), Astellas Pharma (principal investigator, clinical research reimbursement to institution), Daiichi Sankyo (principal investigator, clinical research reimbursement to institution), Sunesis Pharmaceuticals (principal investigator, clinical research reimbursement to institution), Jazz Pharmaceuticals (principal investigator, clinical research reimbursement to institution), Pfizer (principal investigator, clinical research reimbursement to institution), Astex Pharmaceuticals (principal investigator, clinical research reimbursement to institution), CTI BioPharma (principal investigator, clinical research reimbursement to institution), Celgene (principal investigator, clinical research reimbursement to institution), Genentech (principal investigator, clinical research reimbursement to institution), AI Therapeutics (principal investigator, clinical research reimbursement to institution), Samus Therapeutics (principal investigator, clinical research reimbursement to institution), Arog Pharmaceuticals (principal investigator, clinical research reimbursement to institution), and Kartos Therapeutics (principal investigator, clinical research reimbursement to institution) outside the submitted work. R. Wang reports grants from The Frederick A. Deluca Foundation during the conduct of the study, as well as grants from Celgene/Bristol-Myers Squibb outside the submitted work. A.M. Zeidan reports other from Celgene (consultancy/honoraria/research funding), AbbVie (consultancy/honoraria/research funding), Pfizer (consultancy/honoraria/research funding), Boehringer Ingelheim (consultancy/honoraria/research funding), Trovagene (consultancy/honoraria/research funding), Incyte (consultancy/honoraria/research funding), Takeda (consultancy/honoraria/research funding), Novartis (consultancy/honoraria/research funding), Astex Pharmaceuticals (research funding), MedImmune/AstraZeneca (research funding), Aprea (research funding), ADC Therapeutics (research funding), Otsuka (consultancy/honoraria), Jazz Pharmaceuticals (consultancy/honoraria), Agios (consultancy/honoraria), Acceleron (consultancy/honoraria), Astellas (consultancy/honoraria), Daiichi Sankyo (consultancy/honoraria), Cardinal Health (consultancy/honoraria), Taiho (consultancy/honoraria), Seattle Genetics (consultancy/honoraria), BeyondSpring (consultancy/honoraria), Ionis (consultancy/honoraria), and Epizyme (consultancy/honoraria) outside the submitted work. R.A. Mesa reports grants from Incyte, Cti, AbbVie, and Celgene and personal fees from Novartis, Sierra Oncology, Genentech, and Protagonist outside the submitted work. X. Ma reports grants from The Frederick A. Deluca Foundation during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

N.A. Podoltsev: Funding acquisition, methodology, writing–original draft, writing–review and editing. X. Wang: Formal analysis, methodology, writing–original draft, writing–review and editing. R. Wang: Conceptualization, formal analysis, writing–original draft, writing–review and editing. J.N. Hofmann: Methodology, writing–original draft, writing–review and editing. L.M. Liao: Methodology, writing–original draft, writing–review and editing. A.M. Zeidan: Methodology, writing–original draft, writing–review and editing. R.A. Mesa: Methodology, writing–original draft, writing–review and editing. X. Ma: Conceptualization, resources, formal analysis, methodology, writing–original draft, project administration, writing–review and editing.

N.A. Podoltsev and X. Ma received research funding (institutional) from the Frederick A. Deluca Foundation for this work. This research was supported in part by the Intramural Research Program of the NIH, NCI. Cancer incidence data from the Atlanta metropolitan area were collected by the Georgia Center for Cancer Statistics, Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, Georgia. Cancer incidence data from California were collected by the California Cancer Registry, California Department of Public Health's Cancer Surveillance, and Research Branch, Sacramento, California. Cancer incidence data from the Detroit metropolitan area were collected by the Michigan Cancer Surveillance Program, Community Health Administration, Lansing, Michigan. The Florida cancer incidence data used in this report were collected by the Florida Cancer Data System (FCDS; Miami, Florida) under contract with the Florida Department of Health (FDOH; Tallahassee, Florida). The views expressed herein are solely those of the authors and do not necessarily reflect those of the FCDS or FDOH. Cancer incidence data from Louisiana were collected by the Louisiana Tumor Registry, Louisiana State University Health Sciences Center School of Public Health, New Orleans, Louisiana. Cancer incidence data from New Jersey were collected by the New Jersey State Cancer Registry, the Rutgers Cancer Institute of New Jersey, New Brunswick, New Jersey. Cancer incidence data from North Carolina were collected by the North Carolina Central Cancer Registry, Raleigh, North Carolina. Cancer incidence data from Pennsylvania were supplied by the Division of Health Statistics and Research, Pennsylvania Department of Health, Harrisburg, Pennsylvania. The Pennsylvania Department of Health specifically disclaims responsibility for any analyses, interpretations, or conclusions. Cancer incidence data from Arizona were collected by the Arizona Cancer Registry, Division of Public Health Services, Arizona Department of Health Services, Phoenix, Arizona. Cancer incidence data from Texas were collected by the Texas Cancer Registry, Cancer Epidemiology and Surveillance Branch, Texas Department of State Health Services, Austin, Texas. Cancer incidence data from Nevada were collected by the Nevada Central Cancer Registry, Division of Public and Behavioral Health, State of Nevada Department of Health and Human Services, Carson City, Nevada. The authors are indebted to the participants in the NIH-AARP Diet and Health Study for their outstanding cooperation. They also thank Sigurd Hermansen and Kerry Grace Morrissey from Westat for study outcomes ascertainment and management and Leslie Carroll at Information Management Services for data support and analysis.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.