NHANES is one of the National Center for Health Statistics’ primary initiatives (NCHS) (21). The program conducts a two-year survey cycle of a nationally representative sample of Americans who are not institutionalized (22). The NCHS Research Ethics Committee approved all procedures, and each subject provided informed consent (23). The cross-sectional study was deemed exempt from ethical review by the Academic Review Board due to the deidentified publicly available data used.

Data from four NHANES cycles (2003-2010) were analyzed, involving 41,156 participants ( Figure 1 ). We excluded females (n=20,785) and males under 40 (n=13,231). Further exclusion criteria were: (1) Missing data about dietary (n=639); (2) missing PSA measurements (n=746); (3) missing education status (n=4); (4) unknown BM (n=89); (5) missing smoking status (n=2); (6) missing hypertension information (n=2). After the screening, 514 men with PSA ≥4 and the remainder with PSA<4 were included.

NHANES nutritional assessment includes a 24-hour dietary recall interview, conducted over two non-consecutive days to gather participants’ food intake data. Initial interviews were held at mobile examination centers (MECs) (24), and the subsequent interview occurred three to ten days later via phone (25). To minimize bias and improve reliability, the mean of the two measures was used. CDAI was calculated using a modified version of an existing model (26). Each micronutrient (vitamins E, A, C, selenium, zinc, and total carotenoids) was standardized by dividing by the standard deviation after subtracting the mean, which was then summed to derive the CDAI composite score. The formula is as follows:

PSA testing was available to male participants aged ≥40 years who had no history of prostate cancer, prostate infection, inflammation, cystoscopy, rectal exam, or recent prostate biopsy. PCa risk was evaluated using the tPSA and f/t PSA ratios. Individuals with tPSA greater than 10 ng/mL, or tPSA between 4 and 10 ng/mL and an f/t PSA ratio of 25% or lower were categorized as high risk for PCa; others were considered low risk.

Based on previous research findings, the following possible confounding variables were chosen for this study that may influence the risk of PCa: race, age, education, PIR, alcohol use, BMI, diabetes, hypertension, and total cholesterol levels (3, 27). Smoking is a known risk factor for various cancers, including prostate cancer, primarily through mechanisms such as increasing oxidative stress, DNA damage, and inflammation pathways. In the context of prostate cancer, smoking has been associated with more aggressive tumor types and poorer clinical prognosis (28). Regular physical activity has been shown to reduce the risk of prostate cancer, likely through mechanisms such as reducing systemic inflammation, improving immune function, and regulating hormone levels (e.g., testosterone). Physical activity also affects oxidative stress levels and overall health, which may, in turn, influence the role of dietary antioxidants (29). Therefore, considering these biological mechanisms, smoking and physical activity were included as covariates in the analysis to control for their potential confounding effects on the relationship between CDAI and prostate cancer.

We then grouped these covariates. Participants were categorized into two groups based on age (65 years):<65 years and ≥65 years. Educational attainment was divided into three categories: high school, higher education (above high school education for high education level), and lower education (below high school education for low education level). PIR< 2 was considered low, and PIR ≥ 2 was considered high. BMI was categorized into three groups: group 1 (<25), group 2 (≥25 and<30), and group 3 (>30). Participants were classified as smokers if they answered “yes” to smoking (SMQ020). In terms of alcohol consumption, the question about drinking (ALQ130) is used to distinguish between drinkers and non-drinkers. People who drink <12 drinks in a year are categorized as non-drinkers, while those who drink ≥12 drinks are categorized as drinkers.

Individuals with an average of three systolic blood pressure readings of ≥140 mmHg and diastolic blood pressure readings of ≥90 mmHg, those on antihypertensive medication, and those who responded “yes” to the question “Have you been told you have hypertension?” were considered patients with hypertension. Participants were considered to have diabetes if they answered “yes” to having diabetes, used glucose-lowering drugs or insulin, or had glycosylated hemoglobin (≥6.5%) and fasting blood glucose (≥126 mg/dl) values indicating diabetes. Total cholesterol levels were categorized as low (<240 mg/dl) or high (≥240 mg/dl).

During the processing phase, NHANES sample weights were applied to guarantee the study population’s national representation. Basic attributes were described after participants were classified into four CDAI categories. Weighted percentages (%) were used to compare categorical variables using a chi-square test. The results were displayed as mean (± SD) after weighted linear regression was used to compare continuous variables.

We first investigated the association between CDAI and prostate cancer risk in populations under various age groups due to the significant influence of age on prostate cancer risk (30, 31). We then explored the moderating effect of age on CDAI and prostate cancer risk. To ascertain whether a noteworthy trend or difference existed, the study performed univariate analyses and constructed three adjusted models to assess the connection between prostate cancer risk and CDAI. Model 1 was the baseline model, Model 2 added race, education, and PIR, and Model 3 further added BMI, diabetes, hypertension, smoking, alcohol consumption, cholesterol, and moderate and vigorous activity. Subsequently, the dose-response relationship was validated by GAM and threshold effect analysis was performed using RCS and smoothed curve fitting. CDAI and its components were separately regressed on prostate cancer risk to explore the relative effects. In addition, stratified associations between CDAI and UUI were explored by subgroup analyses, and interaction tests were performed.

For all statistical studies, R (version 4.4.0) was utilized. Statistical significance was established when the two-sided p-value was less than 0.05. Data were analyzed from November 1, 2023, to August 17, 2024.

The screening criteria led to the selection of 5658 eligible participants from NHANES 2003–2010 ( Figure 1 ). Among them, 377 were at high risk for prostate cancer and 5281 were not. The weighted estimates for the baseline characteristics of the study population are shown in Table 1 . Compared to those with lower levels of CDAI, those with higher levels of CDAI were more likely to be under 65 years of age, non-Hispanic, have an education level above high school, have a high PIR, be non-smokers, engage in vigorous physical activity, engage in moderate physical activity, not have hypertensive disease, and not have diabetes.

Higher CDAI values were associated with lower levels of total PSA and free PSA, while the free-to-total PSA ratio was unaffected. The number of people at high risk for prostate cancer decreased significantly with higher CDAI values. In addition, we performed population-weighted analyses based on prostate cancer risk ( Supplementary Table S1 ). The CDAI value for those at high risk for prostate cancer was -1.25 (-3.05, 0.82), which was significantly lower than the value of -0.58 (-2.49, 1.77) for those not at high risk (p=0.001).

The connection between prostate cancer and CDAI was initially explored by univariate analysis of total PSA and prostate cancer risk. High prostate cancer risk was positively associated with age ≥65 years, non-Hispanic black men, and hypertension. It was negatively associated with high school education or higher, BMI ≥25, and vigorous and moderate physical activity ( Supplementary Table S2 ). Both total PSA (Q3, p=0.002; Q4, p<0.001) and high prostate cancer risk (Q3, p=0.004; Q4, p=0.001) were negatively associated with Q3 and Q4 of the CDAI quartiles.

The distribution of CDAI in relation to total PSA and prostate cancer risk in different age groups was analyzed and visualized as violin plots ( Figure 2 ). In a moderated effects analysis with CDAI as the independent variable, hr-PCa as the dependent variable, and age as the moderator variable, the results indicated a significant moderating effect of age between CDAI and hr-PCa (B = -0.0097, P = 0.004). Visualized as an interaction plot showing the relationship between CDAI and hrPCa in different age groups ( Figure 3 ). The findings demonstrated that the impact of CDAI on hr-PCa was enhanced with age, especially at the old age group (70 years), where the negative effect of CDAI was most significant. While in the younger age group (49 years), the effect of CDAI was weaker or even tended to be insignificant.

Subgroup analysis was conducted by dividing participants into four groups based on age: 40 ≤ age< 50, 50 ≤ age< 60, 60 ≤ age< 70, and 70 ≤ age ( Supplementary Table S2.1 ). Overall Population: The intervention or exposure factor overall: reduced the risk of the outcome by approximately 7%, and the result was highly statistically significant. Age Subgroup: A significant effect was observed in the population aged 70 and above (OR: 0.94, P = 0.016), indicating that the intervention or exposure factor has a protective effect on the outcome in this group.

To further explore the association between CDAI and the high risk of prostate cancer, weighted logistic regression analyses were performed under three models ( Figure 4 ). CDAI was converted to quartiles (Q1-4; Q1 was used as a reference). In Model One, no variables were added. In Model Two, PIR, race, and education were taken into account. Model Three was built on Model Two with adjustments for BMI, smoking, alcohol consumption, hypertension, diabetes, total cholesterol, moderate activity, and vigorous activity. In analyses with total PSA, Q3 and Q4 were negatively correlated in all three models. In model 3, OR=0.74, p=0.022 for Q3 and OR=0.70, p=0.012 for Q4. In the analysis of the prostate cancer risk, Q3 and Q4 were also negatively correlated in all three models. In model 3, OR=0.72, p=0.036 for Q3 and OR=0.72, p=0.040 for Q4.

Under a fully adjusted model, the relationship between CDAI and high prostate cancer risk was investigated using smoothed curve fitting ( Supplementary Figure S1 ). The RCS results for the dose-response relationship are displayed ( Figure 5 ). Under the fully adjusted model, prostate cancer risk and total PSA were not nonlinearly correlated with CDAI (total PSA, p for nonlinear = 0.052, prostate cancer risk, p for nonlinear = 0.348). At low CDAI levels, the OR was greater than 1, suggesting a lower antioxidant index may be associated with higher health risk. The reference CDAI value at OR = 1 was -0.635, which can help guide men in avoiding prostate cancer. As the CDAI increased, the OR gradually decreased below 1, suggesting that a lower risk of prostate cancer is linked to a greater CDAI.

In addition, a threshold effect analysis between CDAI and prostate cancer risk was conducted ( Table 2 ). In the model that has been entirely modified, the inflexion point (k) was 6.311 for t-PSA and 6.699 for hr-PCa (both log-likelihood ratios< 0.001). The linear model more fully explained the association between CDAI and high prostate cancer risk [OR = 0.943, P< 0.0001 (t-PSA); OR = 0.949, P = 0.0021 (hr-PCa)]. Furthermore, when CDAI > 6.699, the PCa risk was decreased by 58% for every additional CDAI unit (OR = 0.419, P = 0.0033).

To explore the connection between CDAI, each antioxidant ingredient, and PCa risk, a Pearson’s correlation analysis was conducted ( Supplementary Figure S2 ). Z-scores were calculated for each of the six antioxidant components that make up the CDAI. There was a moderately high positive association between CDAI and the levels of each antioxidant, implying that CDAI better reflects overall antioxidant intake. Additionally, CDAI and prostate cancer risk were negatively correlated (r = -0.054).

Regression analyses of CDAI and its components were performed ( Supplementary Table S3 ) and visualized under a fully adjusted model ( Figure 6 ). CDAI, zinc, and selenium showed significant negative correlations with t-PSA and hr-PCa. Vitamin A and C had no significant effect on prostate cancer risk, suggesting their role might be limited. Vitamin E was significantly negatively associated with total PSA and insignificantly negatively associated with a high risk of prostate adenocarcinoma. Total carotenoids were insignificantly negatively correlated with total PSA and weakly negatively correlated with hr-PCa (p=0.043).

Stratified by age, race, PIR, education, hypertension, diabetes, BMI, smoking status, alcohol consumption, total cholesterol, and moderate- and vigorous-intensity exercise ( Table 3 ), CDAI was significantly negatively correlated with both tPSA and hrPCa. In examining the relationship between CDAI and hr-PCa, an OR of 0.95 with p = 0.002 suggests that a decreased risk of PCa is linked to a greater CDAI.

The negative association between hrPCa and CDAI was more pronounced among those aged >65, other Hispanic, non-Hispanic white, with less than a high school education, low PIR, BMI between 25 - 30, no vigorous physical activity, hypertensive, without diabetes, with low cholesterol, and who did not consume alcohol. However, the interaction effect in both tPSA and hrPCa did not reach significance. Therefore, although a strong negative correlation was observed in some subgroups (e.g., low PIR, low education level, etc.), overall, the protective effect of CDAI was broadly applicable to different male populations.