Introduction

Prostate cancer is the most commonly diagnosed malignancy and the second leading cause of cancer death in American black and white men [1]. However, black men have a 70% higher incidence and a more than twofold higher mortality rate compared with white men [2], and they are diagnosed at younger ages and with more aggressive disease [3]. The cause (s) of these black–white differences in risk remain unclear [3], but are likely multifactorial [4], including a combination of environmental exposures (e.g., dietary and nutrient intake [5]), delays in disease detection, differential genetic susceptibility (i.e., chromosome 8q24), or tumor biology (i.e., DNA methylation) [6], and socioeconomic factors [4]. Examination of race-specific modifiable factors potentially related to prostate cancer risk may therefore provide insights into this racial disparity, and opportunities for risk reduction. To date, our knowledge concerning possible dietary, nutrient, and health-related contributors to prostate cancer risk is based on research in predominantly non-Hispanic white populations. Small sample sizes and number of cases for black men and other racial/ethnic minorities have limited prospective race-specific investigations [7]. As such, most evaluation of these factors in black men have been in case–control [8,9,10,11,12,13,14,15] or retrospective cohort studies [16]. There remains a paucity of data prospectively examining the directionality and magnitude of race-specific associations between diet and health-related risk factors and prostate cancer among black and white men, particularly within individual cohorts.

The present study examines the race-specific relationship between diet and nutrient intakes, and health-related factors in relation to prostate cancer risk in the National Institutes of Health (NIH)-AARP Diet and Health Study. Beyond highlighting black–white differences in prostate cancer risk, this analysis considers whether previously identified risk factor associations are consistent in black and white men, and whether they explain some of the excess risk observed in black men.

Methods

Study population

The NIH-AARP (formerly the American Association of Retired Persons) Study is a large cohort of adults aged 50–69 years who were enrolled between 1995 and 1996. As previously described, [17] the cohort includes individuals residing in six US states (California, Florida, Louisiana, New Jersey, North Carolina, and Pennsylvania), and two metropolitan areas (Atlanta, Georgia and Detroit, Michigan). [17] A baseline questionnaire including a detailed 124-item food frequency instrument and other baseline characteristics was completed satisfactorily by 567,169 respondents [17]. A supplementary Risk Factor Questionnaire (RFQ) was completed by a subset of these individuals (approximately 339,000) [17], providing information on screening with prostate-specific antigen (PSA) and digital rectal examination (DRE) within the 3 years prior to baseline (responses: no; yes, once; yes, more than once; and don’t know).

From the 566,398 respondents with sufficient dietary data on the baseline questionnaire, we excluded: all women (n = 225,467), men who had proxy questionnaires (n = 15,760), a prior history of cancer (n = 27,289), self-reported poor health (n = 4958) end-stage renal disease (n = 485), cancers reported by autopsy or death certificate only (n = 2742), zero follow-up time (n = 21), total energy intake beyond twice the interquartile range of Box-Cox log-transformed intake (n = 2218), races other than non-Hispanic black or white (n = 13,976), and first incident cancer other than prostate cancer (n = 45,592). After exclusions, our analytic sample consisted of 227,890 non-Hispanic men: 221,032 white and 6858 black. A subset of these individuals (130,371 white (13,079 cases) and 3217 black (520 cases)) completed the RFQ (Supplementary Fig. 1). To maximize statistical power, our primary analyses used data from the baseline questionnaire.

The NIH-AARP Diet and Health Study was approved by the Special Studies Institutional Review Board of the U.S. National Cancer Institute. All participants provided written informed consent.

Code availability

All computer code used to generate results for this study can be accessed by contacting the corresponding author.

Cancer ascertainment

Cases were identified through linkage with state-based cancer registries as previously described [18]. First primary incident prostate cancer (International Classification of Diseases for Oncology, 3rd Edition code C619) and vital status, using the National Death Index, were ascertained through 31 December 2011 and included cases without disease staging information (809 white and 35 black). Advanced prostate cancers were defined as clinical stage T3–T4, N1, or M1 based on the American Joint Committee on Cancer’s Tumor-Node-Metastasis (i.e., TNM) classification system, or fatal disease, and all other prostate cancer cases are defined as non-advanced. Information on Gleason grade was unavailable.

Statistical analysis

Bivariate analyses were conducted using chi-square tests for categorical variables and t-tests for continuous variables. Person-years of follow-up were calculated from the date of return of the baseline questionnaire to the earliest of the following dates: prostate cancer diagnosis, moved out the registry area, death, or the end of the follow-up. Cox proportional hazards regression, with person-years of follow-up as the time metric, was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the risk of prostate cancer. The proportional hazard assumption was assessed and confirmed by modeling cross-product terms between each factor and time.

Using information assessed from the baseline questionnaire, our analysis included factors identified in step-wise selection (P ≤ 0.25) for the association with overall prostate cancer risk in all men. Selection models began with the following factors: height (cm); body mass index (BMI): normal weight (18.5−< 25 kg/m2), overweight (25−< 30 kg/m2), and obese ( ≥ 30 kg/m2)); alcohol consumption: never, < 1, 1–3, >3−< 6, and ≥6 drinks per day; smoking status: never/rarely, former, and current; self-reported history of diabetes: yes/no; physical activity (≥20 min causing increased breathing/heart rate/sweating): never/rarely, 1–3 times/month, or 1–2, 3–4, or ≥5 times/week. Average daily dietary intakes included red meat (g/day), pyramid servings for all sources of: poultry, fish, dairy, fruit, vegetables, and tomatoes, as well as overall vitamin D (µg), alpha-tocopherol (mg), and, beta-carotene (µg). Quintiles of overall intake were estimated based on the baseline distribution among all men. We also considered use (yes/no) of any individual vitamin or mineral supplements (vitamins A, C, E, beta-carotene, calcium, folic acid, iron, selenium and zinc), and/or multi-vitamins (e.g., therapeutic, stress-tab, or one-a-day) within the year prior to baseline. Estimates of supplemental vitamin D were only available from multivitamin sources.

Based on step-wise selection, the following factors were examined together in race-specific models: history of diabetes; height; BMI; alcohol consumption; smoking status; and average daily dietary intakes of: red meat (g); pyramid servings of tomato [one large tomato or eight ounces of tomato juice] and dairy [one cup (244 ml) milk or yogurt, 1.5 ounces (42.5 g) of natural cheese, or 2 ounces (56.0 g) of processed cheese]; and vitamin D (µg). All models adjusted for the following potential confounders as demonstrated by >10% changes in the parameter estimates: age (continuous plus a 3 knot spline term [19]); family history of prostate cancer (father, brother, or son); attained education (≤11 years, high school graduate and some college, college and post graduate); marital status (married/living as married, never married, separated, divorced, widowed, unknown); as well as quintiles of total energy intake (kcal/day). Indicator variables were used for missing covariate data; overall, values were missing in 3% of white men and 6% of black men. Confounding by screening with PSA and/or DRE was evaluated in the subset of men who completed the RFQ.

Interaction between each factor and race was examined by adding cross-product terms to the model. Trend tests were evaluated treating medians values for quintile-specific categories as continuous in the model and testing the statistical significance of the corresponding regression coefficient. Sensitivity analyses were conducted excluding the first 2 years of follow-up to examine whether associations differed after exclusion of prevalent cases. To evaluate how each factor influenced the black–white disparity in risk, we examined the change in the HR for race following addition of each factor to the model. Models began with an indicator variable for black vs. white race alone, followed by age and family history. Each diet, nutrient, and health-related factor was then added to the model based on the order identified from forward step-wise selection results.

A high degree of correlation (Pearson r ≥ 0.70) between certain variables (e.g., vitamin D and dairy), made it challenging to disentangle individual associations. As such, these variables were not included in models together. Models including dietary intakes adjusted for total energy intake by adding it as a covariate. We used the P-value for equal or unequal variances where appropriate for all groups being statistically compared.

All statistical tests used a two-sided Type I error of 0.05 for statistical significance, and all analyses were carried out using the Statistical Analysis System version 9.3 (SAS Institute, Inc., Cary, NC).

Results

During a median of 15.5 years of follow-up, 28,845 incident prostate cancers were identified in white men (3898 advanced), and 1417 in black men (209 advanced). The baseline distribution of each factor by race and case status is summarized in Table 1. Among white men, compared with non-cases, cases were older, reported more screening, and were more likely to be college educated, married, and drink ≥ 6 drinks/day, but were less likely to be obese or current smokers. Among black men, height was one of the few factors that differed between cases and non-cases (cases were slightly shorter on average), and was the only factor that did not differ between black and white controls other than family history of prostate cancer. For both racial groups, cases were more likely than non-cases to have a history of diabetes (Table 1).

Table 1 Distribution of baseline characteristics, mean (SD) or percent
Full size table

Statistically significant risk factor–prostate cancer associations for overall disease were mainly evident for white men, with many risk estimates similar in magnitude but nonsignificant for black men (Table 2). This includes inverse associations with BMI, current and former smokers, and tomato consumption, and positive associations with alcohol and red meat consumption (Pinteraction > 0.05 for all factors) (Table 2). History of diabetes was the only factor statistically significant and similar in magnitude for white men (HR = 0.72, 95% CI: 0.69–0.76) and black men (HR = 0.77, 95% CI: 0.65–0.90, Pinteraction = 0.66). Whereas, the association with height qualitatively differed between white men (HR = 1.03, 95% CI:1.01–1.04) and black men (HR = 0.91, 95% CI:0.85–097, Pinteraction = 0.003) (Table 2).

Table 2 Race-specific hazard ratio (HR) and 95% confidence interval (CI) for the association with overall prostate cancer
Full size table

Race-specific estimates were similar for advanced and non-advanced disease in both racial groups; exceptions include the statistically significant associations with height and history of diabetes among black men that were only evident for non-advanced disease (Table 3). Interaction between race and height was evident for both advanced (Pinteraction = 0.04) and non-advanced (Pinteraction = 0.01) disease (Table 3).

Table 3 Race-specific hazard ratio (HR) and 95% confidence interval (CI) for the association with prostate cancer by disease type
Full size table

In sensitivity analyses restricted to men with at least 2 years of follow-up, race-specific estimates were largely unchanged, including the interaction between race and height for overall (Pinteraction = 0.002) and non-advanced (Pinteraction = 0.01), but not advanced disease (Pinteraction = 0.24) (Supplementary Table 1). The positive trend in the association with frequency of alcohol consumption was, however, evident in both white (HR≥6 drinks/day vs. never = 1.35, 95% CI:1.12–1.63, Ptrend = 0.01) and black men (HR≥6 drinks/day vs. never = 2.54, 95% CI:1.22–5.32, Ptrend = 0.03; N = 11 cases with ≥ 6 drinks/day) (Supplementary Table 1).

Controlling for prostate cancer screening yielded similar race-specific estimates, however, without evidence of interaction between race and height (Supplementary Table 2). Additionally, previously apparent black–white differences in the association with vitamin D intake became statistically significant for non-advanced disease: (HRQ5 vs. Q1white men = 1.14, 95% CI:1.06–1.23, Ptrend = 0.0001; black men HR = 0.73, 95% CI:0.49–1.07, Ptrend = 0.18; Pinteraction = 0.02). Tests for interactions between race and supplemental or supplemental plus dietary vitamin D intake did not reach statistical significance (data not shown), nor was there evidence of interaction with any of the other evaluated factors (Supplementary Table 2). Among men with at least 2 years of follow-up and controlling for prostate cancer screening, the racial difference in the association with height was only evident for non-advanced disease (Pinteraction = 0.05), and no longer statistically significant for vitamin D intake and non-advanced disease (Pinteraction = 0.06) (data not shown).

In cumulatively adjusted models, adjusting for age, family history of prostate cancer, and each of the factors we examined increased HR for black men compared with white men by 32.9% for overall prostate cancer, and 12.4% for advanced disease, relative to models with race alone. This includes <10% individual changes in the HR associated with adjustment for factors associated with risk in black men (i.e., diabetes, height, and dietary vitamin D) (Fig. 1, HRs and 95% CI are presented in Supplementary Table 3).

Fig. 1
figure 1

The cumulative change in the hazard ratio for the association between black race and risk of overall and advanced prostate cancer after adjustment. The initial model had an indicator variable for black vs. white race alone, followed by adjustment for age, and then family history of prostate cancer. Each diet and health-related factor was subsequently added to the model based on the order identified using forward selection; starting with diabetes and ending with either dairy or dietary vitamin D. Due to high correlation (correlation coefficient ≥0.70), models with dairy and dietary vitamin D are mutually exclusive. The total cumulative change is the percentage change in the hazard ratio between the race alone vs. the final cumulative model (ending with dairy)

Full size image

Discussion

In this prospective cohort, investigation of race-specific associations between dietary, nutrient, and health-related factors and prostate cancer risk, a history of diabetes was the only factor that was both significantly associated with risk, and of a similar inverse direction and magnitude, in black and white men. We also found evidence of racial variation in the associations with attained height and dietary vitamin D intake. Adjustment for all of the investigated factors, including those associated with risk in both racial groups (i.e., diabetes), substantially increased, rather than decreased, the black–white difference in risk.

The latter observation from our study is consistent with findings from prospective analyses of black and white men in the Multiethnic Cohort (MEC) Study [20] and the Health Professionals Follow-up Study [21]. In both cohorts, adjustment for several hypothesized dietary and lifestyle factors increased, rather than  decreased, the relative risk for race/ethnicity [20, 21]. This suggests many of the identified prostate cancer risk factors do not adequately explain risk in black men. Evaluation of whether risk associations are consistent and applicable to a broader at-risk population is important within the context of racial/ethnic disparities, as the assumption of risk factor homogeneity may mask or prevent the discovery of important racial differences that underlie some of the persistent risk differences.

The protective association with a history of diabetes we observed, with a weaker association in black men, has been observed in some [20, 22], but not all studies [23]. In meta-analyses, the overall prostate cancer risk estimate for diabetes is protective, particularly with increasing duration of diabetes [23]. Suspected mechanisms for this association, particularly with type II diabetes, include inhibitory effects of hypoinsulinemia on bioavailable insulin growth factor-I (IGF-I), and alterations in circulating androgens and leptin [24].

The aforementioned MEC study, is similar to the NIH-AARP cohort with respect to average years of follow-up (13.9 vs. 12.7, respectively), number of black prostate cancer cases (1486 vs. 1417, respectively) and some of the examined factors [20]. The MEC study found no evidence of racial variation for any of the evaluated factors, however, which included BMI, smoking status, history of diabetes, alcohol consumption, and height [20]. In both studies, observed associations with BMI, alcohol consumption, and smoking status in white men were not significant in black men. However, our analysis identified racial variation in the prostate cancer risk association with height and dietary vitamin D. Attained height, an indicator of early life nutrition, IGF-I concentrations [25], and heredity, has been associated with increased prostate cancer risk in a dose-response manner [26]. Prior studies evaluating the impact of height by race have found modest increases in overall prostate cancer risk in white men [27], similar to previous reports for white men in the NIH-AARP cohort [28], with either no association [10, 13], or a suggestive protective relationship in black men [29,30,31], whereas others found no racial difference in the positive [32, 33] or null association [20]. Racial differences in the association with prostate cancer risk, however, may be explained by racial variation in the IGF system [34] and its influence on height [35].

Although dietary vitamin D was not associated with risk in black men in our study, for non-advanced disease we found evidence of racial variation in the association, with a positive association in white men and suggestively inverse relation in black men. However, adjusting for dietary vitamin D did not attenuate the relative risk associated with black race. Our findings for white men are consistent with the current literature indicating a positive association with dietary [36] and circulating [37] vitamin D and prostate cancer risk [36]. Our finding of a suggestive inverse association in black men is consistent with both preclinical studies showing a protective role of vitamin D in prostate cancer carcinogenesis [38] and findings from our prospective analysis of serum 25-hydroxyvitamin D and prostate cancer risk in black men [39]. No association has previously been found between vitamin D intake and prostate cancer risk in the few observational studies evaluating this association in black men [40, 41]. In addition to having lower solar ultraviolet B radiation production of 25-hydroxyvitamin D due to greater skin pigmentation [42], black populations have lower intake of vitamin D relative to white populations [43]. Thus, race may be a proxy for vitamin D insufficiency, and lower circulating vitamin D may contribute to black–white differences in prostate cancer risk. A possible mechanism underlying this difference may be related to black–white differences in vitamin D-mediated immune response and inflammation gene expression in the prostate [44, 45].

The present analysis of a large-scale prospective cohort included more than 10 years of average follow-up and information on multiple potential confounders, including prostate cancer screening practices. However, even with a considerable number of black cases, our analysis was limited in power that may have impacted our ability to detect associations among black men, particularly for advanced disease, and to identify heterogeneity in the associations by race. Additionally, measurement error in the questionnaire data, including the  food frequency questionnaire, may have influenced our diet and nutrient risk associations. As such, our results should be interpreted cautiously given the potential for chance findings. Further research is needed to reconcile whether certain risk factor–prostate cancer associations in black men are truly null or missed because of limited statistical power.

Similar to prior studies with race-specific estimates of risk, we found that few of the evaluated dietary, nutrient, and health-related factors were associated with prostate cancer risk in black men. Additionally, adjustment for these factors—which primarily explain risk in non-Hispanic white men—widen the black–white difference in risk. This overall lack of association in black men is in part due to their relatively smaller sample size in these studies, limiting the ability to detect risk associations in this group. Multiple inter-related risk factors, including as-yet determined factors associated with black race, likely contribute to  the risk difference. The current challenge of identifying factors that meaningfully contribute to this well-known racial risk disparity underscores the need for large-scale prospective studies of racial/ethnic minority populations.

Disclaimer

The opinions and conclusions expressed in this article are solely the views of the authors and do not necessarily reflect those of the Food and Drug Administration.