Total Antioxidant Capacity and Pancreatic Cancer Incidence and Mortality in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial

Despite its relatively low incidence, pancreatic cancer is the third most common cause of cancer-related death in the United States, with a total of 53,670 incident cases and 43,090 cancer-related deaths in 2017 (1). Although the specific mechanisms underlying the initiation and progression of pancreatic cancer remain to be elucidated, oxidative stress caused by the overproduction of reactive oxygen species is regarded as a potential mechanism (2, 3). Antioxidant agents can decrease the level of reactive oxygen species in the human body; thus, it is expected that increasing antioxidant intake is beneficial in the prevention of pancreatic cancer.

Total antioxidant capacity (TAC) is defined as the total amount of moles of oxidants neutralized by one liter of food extracts (4). TAC reflects overall antioxidant potential from foods, beverages, and supplements, with the consideration of the potential synergistic interactions between them (4). Hence, investigating TAC instead of individual antioxidants in relation to cancer risk may provide more convincing evidence on the role of antioxidant intake in cancer prevention. Higher TAC has been associated with lower risks of gastric cancer (5), prostate cancer (6), breast cancer (7), and hepatocellular carcinoma (8). To our knowledge, only one hospital-based case–control study from Italy has investigated the association of TAC with pancreatic cancer incidence, with an inverse association observed (9). However, given different genetic backgrounds and dietary habits across populations, whether the observation from the European population can be generalizable to other populations needs to be further confirmed. More importantly, case–control studies, especially hospital-based ones, are susceptible to recall bias and selection bias, and cannot establish a temporal association. In addition, no study has investigated the association of TAC with pancreatic cancer mortality.

Clarifying the association of TAC with pancreatic cancer incidence and mortality can aid in tailoring diet-based interventions to reduce the global burden of this disease. Therefore, we used prospective data from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial to examine the hypothesis that higher TAC is associated with lower risks of pancreatic cancer incidence and mortality in the U.S. population.

The PLCO Screening Trial is a multicenter randomized controlled trial designed to investigate whether selected screening examinations can reduce mortality from prostate, lung, colorectal, and ovarian cancers. Design and implementation of this trial have been reported in detail elsewhere (10). Briefly, approximately 155,000 individuals aged 55 to 74 years were enrolled in 10 screening centers across the United States from November 1993 to September 2001. Individuals were excluded if they (i) were participating in another cancer prevention or screening trial, (ii) were receiving treatment for cancer (excluding squamous and basal cell skin cancer), (iii) received a screening examination for colorectal or prostate cancer recently, or (iv) had a history of PLCO cancers (10). Eligible individuals were randomized into the intervention group or the control group in equal proportions, with individuals in the intervention group receiving screening examinations and those in the control group receiving usual care.

In this study, following individuals were further excluded: (i) individuals diagnosed with any cancer before completing a diet history questionnaire (DHQ; including those diagnosed with any cancer before randomization; n = 11,882); (ii) individuals failing to complete a DHQ (n = 34,401); (iii) individuals with an invalid DHQ [the invalid DHQ refers to the presence of ≥8 missing frequency responses or extreme values of calorie intake (i.e., top 1% or bottom 1%)] (n = 4,841); (iv) individuals failing to return or complete the baseline questionnaire (n = 7,675); and (v) individuals with outcome events (incident pancreatic cancer or pancreatic cancer–related death) occurred between baseline questionnaire completion and DHQ completion (n = 70). After exclusions, a total of 96,018 individuals were included in our analysis. Importantly, for cancer incidence, all eligible individuals were followed up through December 31, 2009, whereas for cancer mortality, all eligible individuals were followed up through December 31, 2015. Of note, all patients dying from pancreatic cancer included in the current analysis had received a diagnosis of pancreatic cancer before December 31, 2009. The timeline and follow-up scheme of PLCO Screening Trial are shown in Fig. 1. The PLCO Screening Trial concept was approved by the Institutional Review Board of the National Cancer Institute and each screening center. Written informed consent was obtained from all individuals.

Baseline information, including sex, height, body weight, race, education degree, aspirin use, smoking status, diabetes, and family history of pancreatic cancer, was collected through a self-administered baseline questionnaire. Body mass index (BMI) was computed as body weight (kg) divided by height squared (m²). Remaining baseline information, including age at DHQ completion, energy intake from diet, multivitamin supplement use, and alcohol drinking status, was collected through a DHQ (version 1.0, National Cancer Institute, 2007). The DHQ was a self-administered food frequency questionnaire and was developed to assess the frequency and serving size of dietary intake and supplement use over the previous year. The performance of DHQ had been validated against four 24-hour dietary recalls in a nationally representative sample of 1,640 subjects; it was found that the DHQ had good performance in estimating dietary intake (11).

Nutrient data for TAC assessment were collected through the abovementioned DHQ. On the basis of national dietary data from Continuing Survey of Food Intakes by Individuals and Nutrition Data Systems for Research, daily nutrient intake of each participant was calculated using the DietCalc software. We employed the Antioxidant Food Table developed by Carlsen and colleagues (12) to estimate an individual's TAC intake from diet and/or supplements. This database provides the antioxidant content of more than 3,100 foods, beverages, and supplements, which was measured by a modified version of the ferric-reducing ability of plasma (FRAP) assay (13). The FRAP assay measures the ability of each sample to reduce ferric ion (Fe³⁺) to ferrous ion (Fe²⁺) and is a highly reproducible and commonly used assay for antioxidant measurements (4, 13). Each nutrient in the DHQ was assigned a FRAP value, which was expressed as antioxidant content in mmol per 100 g nutrient. When a FRAP value was not provided for a certain nutrient, after consultation with nutrition experts (Y.Q.L. Wu and Y. Zhao), the FRAP value of the nutrient with similar antioxidant profile was used. When several FRAP values were available for a certain nutrient, the mean of all available values was assigned to this nutrient. Daily nutrient intake of each participant was multiplied by the corresponding FRAP value, and then the resulting values of all nutrients were summed together to produce an overall FRAP score to reflect an individual's TAC intake. Higher FRAP scores represent higher intakes of antioxidants. Of note, in line with previous studies (14, 15), coffee consumption was not considered in TAC assessment, as the main contributors to antioxidant capacity of coffee in vitro are Maillard products (16) and their absorption and antioxidant capacity in vivo remain unclear (17). TAC used for all analyses was adjusted for energy intake using the residual method (18). Three exposure variables for TAC were created, namely total TAC (the antioxidant capacity was from diet and supplements), dietary TAC (the antioxidant capacity was from diet only), and supplemental TAC (the antioxidant capacity was from supplements only).

Incident pancreatic cancer cases were ascertained mainly through an annual study update form where subjects were asked to report if they were diagnosed with any cancer, the type of cancer, date of diagnosis, the location of diagnosis, and contact information of their healthcare providers. Death certificates and family reports were employed as additional sources for pancreatic cancer ascertainment. Reports of pancreatic cancer were followed up and relevant medical records were abstracted using a standardized form and were reviewed carefully for cancer ascertainment. In this analysis, pancreatic cancer was defined as primary adenocarcinoma of exocrine (ICD-O-2 code C25.0-C25.3, C25.7-C25.9) or endocrine pancreas (ICD-O-2 code C25.4). Vital status was confirmed mainly through the annual study update form and was adjudicated via linkage with the US National Death Index. Causes of death were obtained from death certificates.

Statistical analyses were performed using STATA software (version12.0, StataCorp). The statistical significance level was set at P < 0.05 under a two-tailed test. Continuous variables are expressed as mean (SD), and categorical variables are expressed as counts (percentage). The distribution of TAC was divided into quartiles. Baseline characteristics of study population were shown by quartiles of energy-adjusted total TAC. Between-group differences in baseline characteristics were compared using the ANOVA test for continuous variables and the χ² test for categorical variables.

We employed Cox proportional hazards regression to estimate hazard ratios (HR) and 95% confidence intervals (CI) of the association between TAC and pancreatic cancer incidence, with person-year as the time metric. Here, person-year was calculated from the date of DHQ completion to the date of pancreatic cancer diagnosis, loss to follow up, death, or the end of follow-up, whichever occurred first (Fig. 1). The proportional hazards assumption for Cox regression model was verified using the Schoenfeld residuals (19). We did not find evidence that TAC or any covariate violated the proportional hazards assumption (all P > 0.05). To reduce the potential impacts of competing risk bias on the association of TAC with pancreatic cancer mortality, we employed competing risk regression to estimate subdistribution HRs (SHR) and 95% CIs, with nonpancreatic cancer causes of death as competing events (20).

In above regression models, we employed the first quartile of TAC as the referent. Covariate selection was based on the change-in-estimate strategy (21) and a literature review. Specifically, model 1 was adjusted for age and sex; model 2 was adjusted for age, sex, race, BMI, educational level, aspirin use, smoking status, alcohol consumption, diabetes, family history of pancreatic cancer, and energy intake. For the association of dietary TAC with pancreatic cancer incidence and mortality, model 2 was further adjusted for supplemental TAC; for the association of supplemental TAC with pancreatic cancer incidence and mortality, model 2 was further adjusted for dietary TAC. To examine a linear trend across quartiles of TAC, we first assigned the median value of each quartile to each individual in the quartile and then regarded it as a continuous variable in regression models. Moreover, TAC was regarded as a continuous variable after a log₂ transformation, which represents a doubling of TAC intake.

We performed a series of subgroup analyses after stratifying for age (≥60 vs. <60 years), sex (male vs. female), BMI (≥25 vs. <25 kg/m²), multivitamin supplement use (yes vs. no), current smoking (yes vs. no), and alcohol consumption (≥30 vs. <30 g/day). The significance of the interaction between TAC and stratification factors was evaluated by a likelihood ratio test, which compares models with and without interaction terms. Notably, to increase statistical power, subgroup analyses were based on log₂ HR of the association between total TAC and pancreatic cancer incidence and mortality. To examine the robustness of our results, we performed following sensitivity analyses: excluding pancreatic endocrine tumors (only for pancreatic cancer incidence), excluding cases/deaths occurring within the first two years of follow-up, excluding subjects with extreme values of energy intake (extreme values of energy intake are defined as <800/>4,000 kcal/day for men and <500/>3,500 kcal/day for women; ref. 22), and using TAC without adjustment for energy intake. We employed restricted cubic spline models (23) with 4 knots at the 5th, 35th, 65th, and 90th percentiles to explore the potential nonlinear dose–response relationship of total and dietary TAC levels to pancreatic cancer incidence and mortality. Importantly, we excluded subjects with extreme TAC intake (i.e., <1th or >99th percentile) from the dose–response analysis to reduce the potential impacts of extreme values on the relevant results. The minimum TAC intake (total TAC: 1.24 mmol/day; dietary TAC: 1.52 mmol/day) was employed as the referent. We obtained a P_nonlinearity by testing the null hypothesis that the estimated values of both the second and third splines equal to zero (23). All dose–response analyses were based on the most fully adjusted risk estimates (i.e., those from model 2).

In the whole study population, total TAC (without adjustment for energy intake) ranged from 0.87 to 90.89 mmol/day, with an average level of 12.25 mmol/day. The main contributors to total TAC were multivitamin supplements (22.3%), fruits (20.5%), vegetables (12.7%), tea (10.0%), beverages (except tea) (7.8%), and cereals and breads (6.1%). Table 1 presents baseline characteristics of study population by quartiles of energy-adjusted total TAC. Compared with individuals with a low level of total TAC, those with higher levels generally had higher energy intakes and educational levels, were more likely to use aspirin and multivitamin supplements, and were less likely to be current smokers but were more likely to be current alcohol drinkers.

We identified a total of 393 pancreatic cancer cases during a mean follow-up of 8.86 ± 1.91 years (851,087.4 person-years), with an overall incidence rate of 4.62 cases per 10,000 person-years. Table 2 summarizes the results of univariable and multivariable Cox regression analyses on TAC and pancreatic cancer incidence. A significant inverse association between total TAC and the risk of developing pancreatic cancer was observed in univariable analysis, which persisted after adjustment for potential confounders (HR _{quartile 4 vs. quartile 1}, 0.53; 95% CI, 0.39–0.72; P_trend = 0.0002; log₂ HR, 0.81; 95% CI, 0.74–0.89; model 2). Similar results were obtained for the association of dietary TAC with pancreatic cancer incidence (fully adjusted HR _{quartile 4 vs. quartile 1}, 0.66; 95% CI, 0.48–0.90; P_trend = 0.0086; log₂ HR, 0.75; 95% CI, 0.66–0.84). However, both univariable and multivariable analyses did not detect a significant association between supplemental TAC and pancreatic cancer incidence.

Subgroup analyses did not find evidence of effect modification by age, sex, BMI, multivitamin supplement use, current smoking, and alcohol consumption for the association of total TAC with pancreatic cancer incidence (all P_interaction > 0.05; Supplementary Fig. S1). Sensitivity analyses showed that the aforementioned findings on TAC and pancreatic cancer incidence did not change materially after excluding patients with pancreatic endocrine tumors (n = 17), excluding cases observed within the first two years of follow-up (n = 61), excluding subjects with extreme values of energy intake (n = 2,686), and using TAC without adjustment for energy intake (n = 96,018; Supplementary Table S1). Restricted cubic spline model revealed that both total and dietary TAC levels were associated with pancreatic cancer incidence in a nonlinear inverse dose–response manner (all P_nonlinearity < 0.05; Fig. 2).

Over an average follow-up of 13.33 ± 3.43 years (1,280,193.9 person-years), we observed a total of 353 pancreatic cancer–related deaths, with an overall mortality rate of 2.76 deaths per 10,000 person-years. After the full adjustment for confounders (model 2), individuals in the highest quartile of total TAC were found to be at a reduced risk of death from pancreatic cancer (SHR _{quartile 4 vs. quartile 1}, 0.52; 95% CI, 0.38–0.72; P_trend = 0.0003; log₂ SHR, 0.81; 95% CI, 0.73–0.89) compared with those in the lowest quartile (Table 3). Likewise, individuals with higher levels of dietary TAC were found to have a lower risk of pancreatic cancer–related death (fully adjusted SHR _{quartile 4 vs. quartile 1}, 0.64; 95% CI 0.46–0.89; P_trend = 0.0112; log₂ SHR, 0.73; 95% CI, 0.65–0.83). However, supplemental TAC was found to be not associated with pancreatic cancer mortality in univariable or multivariable analyses.

The significant inverse association between total TAC and pancreatic cancer mortality could be not modified by predefined stratification factors (all P_interaction > 0.05; Supplementary Fig. S2). In sensitivity analyses, the initial inverse association of total and dietary TAC levels with pancreatic cancer mortality remained significant; the null association of supplemental TAC with pancreatic cancer mortality did not alter substantially (Supplementary Table S2). A nonlinear inverse dose–response association with pancreatic cancer mortality was identified for total and dietary TAC (all P_nonlinearity < 0.01; Fig. 3).

In this cohort of 96,024 American adults, we observed that higher total and dietary TAC levels, as reflected by higher FRAP scores, were associated with lower risks of pancreatic cancer incidence and mortality in a nonlinear inverse dose–response manner, even after adjustment for potential confounders; these observations remained in sensitivity analyses, indicating the robustness of our findings. Moreover, subgroup analyses showed that risk reductions of pancreatic cancer incidence and mortality did not differ significantly between predefined subgroups. No association was observed for supplemental TAC and pancreatic cancer incidence and mortality.

Antioxidants, including vitamin C and vitamin E, have been found to decrease oxidative DNA damage (24). Hence, increasing antioxidant intake is hypothesized to be associated with a reduced risk of developing pancreatic cancer. However, a meta-analysis of 12 prospective studies showed that there was no association of consumption of fruits and vegetables with pancreatic cancer incidence (25); a recent large-scale prospective study even revealed a positive association between vegetable intake and pancreatic cancer incidence in ever smokers (26). In addition, consumption of green tea and intakes of vitamin C and vitamin E were found to be not associated with pancreatic cancer incidence (27, 28). The abovementioned findings pose a great challenge to the hypothesis that antioxidants have beneficial effects on the risk of developing pancreatic cancer. However, it should be reminded that these studies fail to consider the potential synergic or additive effects among dietary components with antioxidative properties (29). Therefore, their results may not fully reflect the role of antioxidants in the prevention of pancreatic cancer. In this study, we employed TAC, a maker considering the potential synergistic interactions of antioxidant nutrients, to reflect an individual's overall antioxidant intake, and revealed a significant inverse association of total and dietary TAC levels with pancreatic cancer incidence, which is in line with the results of a case–control study (9). In fact, similar to our findings, adherence to Mediterranean diet, which is characterized by a high intake of dietary antioxidants, has been associated with a reduced risk of pancreatic cancer incidence (30). Overall, our findings support that a diet rich in antioxidants has beneficial effects on the risk of developing pancreatic cancer, which has important implications for public health and decision making.

To date, four prospective studies have investigated the association of TAC with overall cancer mortality in the general population, with one showing an inverse association (31) and remaining three showing a null association (32–34). On the basis of a total of 353 pancreatic cancer–related deaths, we observed a significant inverse association of dietary TAC with pancreatic cancer mortality, indicating that increasing consumption of foods rich in antioxidants is beneficial in reducing pancreatic cancer mortality in the general population. Interestingly, a case-series study of 814 patients with glioma revealed a null association between dietary TAC and overall survival (35). In addition, our study did not find a significant association between supplemental TAC and pancreatic cancer mortality. However, a prospective study of 2,264 women with breast cancer showed that frequent use of vitamin E and vitamin C was associated with a reduced risk of breast cancer recurrence (36). Taken together, whether our findings on TAC and mortality from the general population could extend to cancer survivors remains unknown, and the role of antioxidants in improving the prognosis of patients with cancer needs to be further studied.

Interestingly, in our study, multivariable Cox regression analyses showed that supplemental TAC was not associated with pancreatic cancer incidence and mortality; furthermore, our subgroup analyses did not find evidence of effect modification by multivitamin supplemental use. The exact reasons for these phenomena are unknown. One straightforward explanation is that the dose of supplemental multivitamins is insufficient to exert their protective effects. However, this seems to be not the case in our study, as supplemental TAC contributed to more than 20% of total TAC. Another biologically possible explanation is that antioxidant supplements have antagonistic interactions with dietary components (29), which abolish their protective effects. In addition, natural antioxidants from foods and beverages possibly have different biological activity and potency from the synthetic antioxidants used in supplements (37). Thus, the bioavailability of antioxidant supplements could decrease significantly in the human body, resulting in the loss of their protective effects. In fact, several meta-analyses of randomized controlled trials have consistently found that antioxidant supplements have no primary or secondary preventive effects against cancer (38–42). On the basis of these findings, individuals are encouraged to consume more antioxidant-rich foods or beverages instead of antioxidant supplements.

Cigarette smoke is a rich source of reactive nitrogen species and oxidants. It has been found that current smokers have lower circulating levels of antioxidants compared with never smokers (43, 44), possibly resulting from lower intakes of dietary antioxidants (45) and increased utilization of these compounds (46). Hence, it is possible that the association of TAC with health outcomes will be modified by smoking status. Indeed, a prospective study showed that the inverse association between dietary TAC and gastric cancer incidence was significant in current smokers, but not in never smokers (5). However, in this study, our subgroup analyses on total TAC and pancreatic cancer incidence and mortality did not find evidence of modification effect by smoking status. Likewise, a case–control study on dietary TAC and pancreatic cancer incidence also found that there was no statistically significant difference in incidence reduction between ever and never smokers (5). These findings appear to indicate that the protective effects of antioxidants on pancreatic cancer incidence and mortality are not associated with the concentrations of circulating antioxidants in the human body. However, it should be reminded that the failure to observe a significant interaction between TAC and smoking status in our study may be due to the insufficient power, as the proportion of current smokers was less than 10%, leading to a small number of outcomes events observed among current smokers (59 incident cases and 53 deaths). Therefore, more studies with large sample size are needed to determine whether the association of TAC with pancreatic cancer incidence and mortality could be modified by smoking status.

Interestingly, our multivariable analyses showed that the HRs for total and dietary TAC barely changed after adjustment for the potential confounders. Of note, such a phenomenon was also observed in several prospective studies on TAC and health outcomes (7, 14, 47). One possible explanation is the inappropriate selection of covariates. However, this seems to be not the case for our study, as we used the change-in-estimate strategy (21) and literature review to determine covariates, which have been widely used in epidemiology. Another more conceivable explanation is that selected covariates contributed so insubstantially to the observed associations; in other words, total and dietary TAC levels by themselves are strong independent predictors of pancreatic cancer incidence and mortality.

Our study has several limitations. First, antioxidant defense system in the human body consists of exogenous and endogenous antioxidants. However, the FRAP assay only considers the antioxidant activity of foods and supplements in vitro. Therefore, our results on TAC and pancreatic cancer incidence and mortality may not fully reflect the situation in vivo. Second, residual confounding is always a concern in observational studies, including ours. Despite the full adjustment for potential confounders available in the PLCO Screening Trial; however, we cannot exclude the possibility that our results were distorted by unrecognized or unmeasured confounders. Third, nutrient intake from diet and supplements was evaluated only at baseline. The use of nutrient intake at one time point possibly introduces the nondifferential misclassification of exposure, as an individual's dietary habit could change over time. Nevertheless, it has been found that the methods using baseline diet information only usually result in weaker associations than those using the cumulative averages (48). Finally, we used death certificates to identify the causes of death. However, the information from death certificates is not accurate in some circumstances. Hence, our findings may be subject to misclassification bias (49).

In conclusion, in the U.S. general population, dietary but not supplemental TAC level, as suggested by the FRAP score, is inversely associated with risks of pancreatic cancer incidence and mortality in a nonlinear dose–response pattern. These findings indicate that a diet rich in antioxidants could be beneficial in decreasing pancreatic cancer incidence and mortality. More studies are needed to determine whether our findings can be modified by smoking status, and to validate them in other populations.

No potential conflicts of interest were disclosed.

The statements contained herein are solely those of the authors and do not represent or imply concurrence or endorsement by NCI.

Conception and design: G.-C. Zhong, J.-Y. Pu, J.-P. Gong

Development of methodology: J.-Y. Pu, K. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.-C. Zhong

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.-Y. Pu, K. Wang

Writing, review, and/or revision of the manuscript: G.-C. Zhong, Y.-L. Wu, Z.-J. Yi, L. Wan, F.-B. Hao, Y. Zhao, J.-P. Gong

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G.-C. Zhong

Study supervision: J.-Y. Pu, J.-P. Gong

The authors thank Miss You-Qi-Le Wu (Department of Nutrition and Food Hygiene, School of Public Health and Management, Chongqing Medical University, Chongqing, China) for her thoughtful suggestions in total antioxidant capacity assessment. The authors thank the National Cancer Institute for access to NCI's data collected by the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial.

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.