Dietary Acid Load and the Risk of Pancreatic Cancer: A Prospective Cohort Study

Pancreatic cancer is a highly lethal malignancy, with a mortality-incidence ratio of nearly 1 (1). In the United States, it is the fourth leading cause of cancer-related death, with estimated deaths of 47,050 in 2020 (2). Hence, it is of paramount importance to identify modifiable risk factors for improving primary prevention of this cancer. One of such candidate factors is dietary intake, as it has been implicated to play a critical role in the etiology of pancreatic cancer (3).

Dietary intake can significantly affect the acid-base balance of human body through supplying acid or base precursors (4, 5). Observational studies found that dietary acid load, a measure reflecting the intake of acidogenic foods, was positively associated with risks of metabolic syndrome (6) and type 2 diabetes (7), two well-established cancer risk factors (8, 9). Also, experimental studies found that increasing the acidity of tumor microenvironment would promote the invasion or metastasis of cancer cells (10–12). These findings together suggest a potentially detrimental role of increasing dietary acid load in cancer development (13). Indeed, a large-scale prospective study showed that higher dietary acid load was associated with a higher risk of breast cancer (14). However, to our knowledge, whether such a positive association is also observed for pancreatic cancer has not been determined previously.

A Western dietary pattern is characterized by the richness of acidogenic foods (e.g., meat and fish), resulting in chronic metabolic acidosis (15). Thus, clarifying the potential association between dietary acid load and the risk of pancreatic cancer will have important public health implications. Hence, we performed a prospective study of 95,708 American adults to examine the hypothesis that higher dietary acid load is associated with an increased risk of pancreatic cancer.

We reported this study in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology statement (16).

Our study population was identified from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, a multicenter clinical trial that was designed to examine whether screening for prostate, lung, colorectal, and ovarian cancers could decrease mortality due to these cancers. Study design of the PLCO Cancer Screening Trial has been reported elsewhere (17). Briefly, individuals aged 55 to 74 years were invited to participate in this trial in ten U.S. screening centers (Minneapolis, Denver, Salt Lake City, Marshfield, Detroit, Birmingham, St Louis, Pittsburgh, Washington, and Honolulu) from November 1993 to September 2001. After exclusions (see Fig. 1 for the relevant exclusion criteria), a total of 154,887 eligible individuals (76,678 men and 78,209 women) were randomized into screening group and control group in equal proportions.

In this study, the following individuals were further excluded: (i) individuals with an invalid diet history questionnaire [DHQ; the invalid DHQ refers to the absence of DHQ completion date, death date before DHQ completion date, ≥8 missing DHQ items, or the presence of extreme energy intakes (top 1% or bottom 1%); n = 41,440]; (ii) individuals failing to complete a baseline questionnaire (n = 8,518); (iii) individuals diagnosed with any cancer before the completion of DHQ or baseline questionnaire (n = 9,095); (iv) individuals diagnosed with primary adenocarcinoma of endocrine pancreas (n = 17); (v) individuals receiving a diagnosis of pancreatic cancer following the diagnosis of other cancers during the trial (n = 39); and (vi) individuals with outcome events observed between randomization and DHQ completion (n = 70). Finally, a total of 95,708 individuals were included in this study (Fig. 1). Written informed consents were obtained from all participants. The PLCO Cancer Screening Trial was approved by the Institutional Review Boards of the NCI and each screening center.

We used potential renal acid load (PRAL; ref. 18) and net endogenous acid production (NEAP; ref. 4) to assess dietary acid load of each participant. These two scores were calculated by the following formulas:

PRAL score:

• PRAL (mEq/d) = 0.4888 × protein intake (g/d) + 0.0366 × phosphorus (mg/d) – 0.0205 × potassium (mg/d) – 0.0125 × calcium (mg/d) – 0.0263 × magnesium (mg/d)

NEAP score:

• NEAP (mEq/d) = [54.5 × protein intake (g/d) ÷ potassium intake (mEq/d)] – 10.2

For PRAL score, a negative value indicates an alkaline-forming potential, while a positive value indicates an acid-forming potential; for NEAP score, greater values indicate greater acid-forming potentials. Serval studies had confirmed the validity of PRAL and NEAP scores estimated from the above formulas and suggested that these scores yielded appropriate estimates to predict urine pH (4, 18–20). Notably, dietary nutrient intakes used for the calculation of PRAL and NEAP scores were adjusted for energy intake from diet using residual method (21).

The information of nutrient intakes was collected via the above-mentioned DHQ, a self-administered 137-item food frequency questionnaire, which is developed by the NCI for evaluating serving size and frequency of food consumption of each participant over the past 12 months. Of note, the DHQ was administered across both screening and control groups of the PLCO Cancer Screening Trial approximately three years after trial entry. The validity of the DHQ had been confirmed in the Eating at America's Table Study (22). Daily gram amounts of each food were approximated by multiplying serving size by food frequency; daily nutrient intake was approximated on the basis of national dietary data [USDA's 1994–96 Continuing Survey of Food Intakes by Individuals (23) or the Nutrition Data Systems for Research (24)].

In the PLCO Cancer Screening Trial, incident pancreatic cancer was ascertained mainly via a mailed annual study update form that required each participant to report whether they were diagnosed with any cancer, the date of cancer diagnosis, as well as the type and the site of cancer. For participants who did not return the form, they were contacted repeatedly via telephone or e-mail. Pancreatic cancer reported on the annual study update form was further confirmed by carefully checking the relevant medical records. Family reports and death certificates were used as additional sources for the ascertainment of pancreatic cancer. Of note, in this study, pancreatic cancer referred to primary adenocarcinoma of the exocrine pancreas only (ICD-O-2 code: C25.0-C25.3 and C25.7-C25.9).

Sex, race, educational level, body height and weight, aspirin use, history of diabetes, history of hepatitis or cirrhosis, history of emphysema, family history of pancreatic cancer, and smoking status were assessed through a self-administrated baseline questionnaire. Body mass index (BMI) was calculated as body weight (kg) divided by height squared (m²). Physical activity level, which was defined as total time of moderate-to-vigorous activity per week in this study, was assessed through a self-administrated supplemental questionnaire. Healthy Eating Index 2015, a measure reflecting an individual's diet quality, with higher values indicating higher diet quality, was calculated using the method described in the literature (25). Remaining covariates, including age at DHQ completion, single or multivitamin supplement use, energy intake from diet, alcohol intake, food consumption, and nutrient intake, were assessed through the aforementioned DHQ.

Continuous variables were expressed as mean (SD), and categorical variables were expressed as numbers (percentage). To maximize statistical power and minimize selection bias, we used multiple imputation with chained equations to impute 24,148 missing values (25.2% of sample size) of the variable “physical activity level” under an assumption that these values were missing at random, with the number of imputations set at 25 (26). All variables involved in statistical analyses were included in the imputation model.

HRs and 95% confidence intervals (CI) were estimated for the association between PRAL and NEAP scores and the risk of pancreatic cancer using Cox proportional hazards regression, with person-year as time metric. In this study, person-year was calculated from DHQ completion to the date of pancreatic cancer diagnosis, the end of follow-up (i.e., December 31, 2009), the date of death, or loss to follow-up, whichever came first. No evidence indicating the violation of proportional hazards assumption was found using Schoenfeld residuals. PRAL and NEAP scores of the entire study population were split into sex-specific quartiles, as the distribution of these scores possibly differ by sex. Results for each of sex-specific quartiles were combined for overall results. The first quartile was used as the reference group in all Cox regression analyses. To examine a linear trend across quartiles of PRAL or NEAP score, the median of each quartile was first assigned to each participant in the quartile, and then treated as a continuous variable in Cox regression model for testing its statistical significance. Covariate inclusion for multivariable Cox regression was determined according to our knowledge of existing literature and the change-in-estimate strategy (27). Specifically, model 1 (the basic model) was adjusted for age and sex; model 2 (the model that included known pancreatic cancer risk factors) was further adjusted for smoking status, history of diabetes, alcohol intake, BMI, and family history of pancreatic cancer; model 3 additionally included factors that were confounders (energy-adjusted intakes of dietary fiber and carbohydrate) and energy intake from diet.

We used restricted cubic spline functions (28) with three knots placed at the 10th, 50th, and 90th percentiles to visually inspect the shape of the association of PRAL and NEAP scores with the risk of pancreatic cancer. For comparison with the results from Cox regression, the reference level was set at the median of first quartile (i.e., -26.8 and 21.3 mEq/day for PRAL and NEAP, respectively). To decrease the potential influence of extreme values on the results, individuals in the top 2.5% or bottom 2.5% of PRAL or NEAP score were excluded from the dose–response analysis. A P for nonlinearity was calculated by testing the null hypothesis that the regression coefficient of the second spline was equal to zero (28).

To identify the potential effect modifiers of the association of PRAL and NEAP scores with the risk of pancreatic cancer, we performed several predefined subgroup analyses after stratifying for age (≥ 65 vs. < 65 years), follow-up length (≥ 6 vs. < 6 years), BMI (≥ 25 vs. < 25 kg/m²), single or multivitamin supplement use (yes vs. no), smoking status (current or former vs. never), and physical activity level (≥ median vs. < median). A P_interaction was calculated by comparing models with and without interaction terms prior to performing the above subgroup analyses to avoid spurious subgroup differences.

We performed the following sensitivity analyses to examine the robustness of our results: (i) excluded individuals with extreme energy intake values, which refer to <800 or >4,000 kcal/day for men and <500 or >3,500 kcal/day for women (29); (ii) excluded pancreatic cancer cases diagnosed within the first two years of follow-up to examine the potential impacts of reverse causation; (iii) excluded individuals with history of diabetes, liver comorbidity, or emphysema, as these conditions could lead to acid–base disturbance (30); (iv) repeated the main analyses in individuals with complete data; and (v) additional adjusted for Healthy Eating Index-2015 to determine whether the observed associations were mediated by diet quality.

Statistical analyses were performed using STATA software (version 12.0, StataCorp). The results were regarded to be statistically significant when a two-tailed P value was less than 0.05.

The PLCO Screening Trial concept was approved by the Institutional Review Board of the NCI and each screening center. Written informed consent was obtained from all individuals.

In the whole study population, the mean value (SD) was -9.8 (16.4) for PRAP score and 33.8 (11.7) for NEAP score. Table 1 shows baseline characteristics of study population by sex-specific quartiles of PRAL score. Compared with participants in the lowest quartile of PARL score, those in the highest quartile were more likely to have a history of diabetes and be never smokers, had higher BMI and higher energy intake from diet, but had a lower physical activity level, lower Healthy Eating Index-2015, and lower alcohol consumption. In addition, participants in the highest versus the lowest quartiles of PRAL score had higher intakes of red meat, fish, total protein, total fat, calcium, and phosphorus but lower intakes of fruits, vegetables, coffee, carbohydrate, dietary fiber, folate, magnesium, and potassium.

During 848,534.0 person-years of follow-up, a total of 337 incident pancreatic cancer cases were observed, with the overall incidence rate of 3.97 cases per 10,000 person-years. The mean (SD) follow-up duration was 8.87 (1.90) years. In the fully adjusted model, the highest versus the lowest quartiles of PRAL score was found to be associated with an increased risk of pancreatic cancer (HR_{quartile 4 vs. 1}: 1.73; 95% CI, 1.21–2.48; P_trend = 0.001; Table 2) in the whole study population. A similar association was observed for NEAP score (fully adjusted HR_{quartile 4 vs. 1}: 1.64; 95% CI: 1.14–2.36; P_trend = 0.006). When conducting Cox regression analyses in men and women separately, we found that the positive association of PRAL score with the risk of pancreatic cancer did not change materially in men and in women, although the association in women did not reach statistical significance (Table 2). Similar results were obtained for NEAP score when Cox regression analyses were conducted in men and women separately.

Restricted cubic spline functions revealed that both PRAL (P_nonlinearity = 0.012) and NEAP (P_nonlinearity = 0.013) scores were positively related to the risk of pancreatic cancer in a nonlinear dose–response fashion (Fig. 2). Subgroup analyses revealed that the positive associations of PRAL (P_interaction = 0.018) and NEAP (P_interaction = 0.018) scores with the risk of pancreatic cancer were more pronounced in participants aged <65 years than in those ≥65 years. In addition, the positive associations with PRAL (P_interaction<0.001) and NEAP (P_interaction<0.001) scores were more pronounced among individuals with follow-up length of ≥6 years. The initial associations did not alter materially in a series of sensitivity analyses, including sensitivity analyses excluding pancreatic cancer cases observed within the first two years of follow-up (i.e., two-year lagged analysis; Table 4).

On the basis of prospective data of 95,708 American adults, we found that higher dietary acid load, as assessed by PRAL and NEAP scores, was associated with a higher risk of pancreatic cancer, even after controlling for the potential confounders; this positive association was more pronounced in participants aged <65 years than in those aged ≥65 years. Moreover, we found that the positive association of dietary acid load with the risk of pancreatic cancer was nonlinear, suggesting that with changes in dietary acid load, the risk of pancreatic cancer would change in a nonparallel fashion.

Previously, diet-induced acidosis has been suggested to be capable of promoting tumor initiation and progression through modulating molecular and cellular activities (13). However, only two epidemiologic studies had investigated the association of dietary acid load with the risk of developing cancer to date (14, 31). A prospective study of 27,096 male smokers found that the highest versus the lowest quintiles of dietary acid load score was not associated with the risk of bladder cancer (446 bladder cancer cases; fully adjusted HR_{quintile 5 vs. 1}: 1.15; 95% CI, 0.86–1.55; P_trend = 0.38; ref. 31). In contrast, a prospective study of 43,570 women found that compared with those in the lowest quartile of dietary acid load score, women in the highest quartile were found to be at an increased risk of breast cancer (1,614 breast cancer cases; fully adjusted HR_{quartile 4 vs. 1}: 1.21; 95% CI, 1.04–1.41; P_trend = 0.04; ref. 14). On the basis of prospective data from the PLCO Cancer Screening Trial, we found that higher dietary acid load scores were associated with a higher risk of pancreatic cancer. On the one hand, this finding can contribute to our understanding of the role of diet-induced acidosis in the etiology of pancreatic cancer. On the other hand, it indicates that encouraging consumption of more alkaline foods (e.g., fruit and vegetables) and decreasing consumption of acidogenic foods (e.g., meat and cheese) may represent an attractive approach to reduce the risk of pancreatic cancer, which is particularly relevant in the context that modern Western diets are gradually shifting toward more acidifying (32).

Intriguingly, our subgroup analysis observed that the positive association of PRAL and NEAP scores with the risk of pancreatic cancer was more pronounced in subjects aged <65 years than in those aged ≥65 years. In fact, Akter and colleagues also observed a stronger positive association of PRAL score with all-cause mortality in subjects aged <55 years than in those aged ≥55 years (P_interaction = 0.005; ref. 33). The exact reasons for these observations are unclear. One straightforward explanation is that dietary acid load has some interactions with age in biological pathways, resulting in that younger subjects are more susceptible to diet-induced acidosis than older subjects. An alternative explanation is that older subjects are more likely to use diuretics, given that chronic kidney disease is more common in older subjects than in younger subjects (34); thus, acids from diet can be quickly excreted by the kidneys among older subjects, which can weaken or even eliminate the potential influence of diet acid load on human body. Indeed, a cross-sectional study of 30,528 subjects found that the prevalence of acidosis (defined as bicarbonate <22 mEq/L) was decreased in older compared with younger subjects (8% vs. 12%), especially in subgroup with estimated glomerular filtration rate of <45 ml/min per 1.73 m² (15% vs. 54%; ref. 35). Unfortunately, we cannot confirm whether the latter explanation is the case in our study because of the unavailability of data on diuretic use in the PLCO Cancer Screening Trial. In addition, it should be reminded that we cannot rule out the possibility that age difference in the association of dietary acid load with the risk of pancreatic cancer we observed is a chance finding, despite that it may be explained by the abovementioned facts. In addition, our subgroup analyses found that the positive associations with PRAL and NEAP scores were more pronounced among individuals followed-up ≥ 6 years than those followed-up <6 years. This finding supports the notion that the positive association between dietary acid load scores and the risk of pancreatic cancer we observed is not due to reverse causation, which is actually consistent with the results from the two-year lagged analyses.

The positive association of dietary acid load with the risk of pancreatic cancer is biologically plausible. Diet-induced acidosis can lead to the increase in serum levels and the bioactivity of cortisol (36, 37), which has been found to promote insulin resistance (38). Meanwhile, low-grade metabolic acidosis resulting from long-term consumption of high protein is found to be associated with increased levels of serum insulin-like growth factor 1 (13). Of note, insulin resistance and insulin-like growth factor 1 have been demonstrated to promote the development and progression of pancreatic cancer (39, 40). In addition, metabolic acidosis is found to reduce circulating adiponectin levels through inhibiting the transcription of adiponectin gene (41); both experimental and epidemiologic studies have suggested the protective role of adiponectin against the occurrence of pancreatic cancer (42, 43). Furthermore, emerging evidence indicates that microenvironmental acidity can inhibit antitumor immunity and eventually promotes immune escape and cancer progression (44), which may also explain the positive association of dietary acid load with the risk of pancreatic cancer, at least in part.

The strengths of our study mainly include the prospective study design, the large sample size, the robustness of results in sensitivity analyses, and the long duration of follow-up. However, our study also has several limitations. First, we had adjusted for a wide range of possible confounders, but our results could be still subject to residual confounding because of unrecognized or unmeasured confounders. For example, it is well known that kidney function plays a critical role in maintaining the acid–base balance of human body; hence, the association of dietary acid load with the risk of pancreatic cancer could be confounded by this factor. However, we could not control this potential confounding effect as the data on kidney function were not collected in the PLCO Cancer Screening Trial. Nevertheless, the positive association of dietary acid load with the risk of pancreatic cancer is expected to be stronger in individuals with impaired kidney function than in those with normal kidney function. Likewise, PRAL and NEAP scores may be correlated to other exposures that are associated with the risk of pancreatic cancer, such as insulin resistance (45, 46). Unfortunately, we could not control the potential impacts of these exposures on the association between dietary acid load score and the risk of pancreatic cancer due to the unavailability of the relevant data. Second, our results were originated from a U.S. population ages 55 to 74 years with a relatively alkaline diet, and thus might not be generalized to other age groups or other populations with different dietary habits. Third, nutrient data used for the calculation of PRAL and NEAP scores were collected at baseline in the PLCO Cancer Screening Trial. The assessment of dietary intake at one time point may lead to nondifferential bias, as dietary behaviors could change over time. Nonetheless, a classic assumption in nutritional epidemiology is that an adult's dietary habits would not change dramatically during a short period. Moreover, it has been suggested that the method only using baseline diet generally yields a weaker association than that using the cumulative averages (47). Finally, in subgroup analyses, it was found that the positive association of PRAL and NEAP scores with the risk of pancreatic cancer could not be modified by follow-up length, BMI, single or multivitamin supplement use, smoking status, and physical activity level. However, it should be reminded that these findings were based on a small number of pancreatic cancer cases, resulting in that likelihood ratio test may have inability to detect the potential interactions between PRAL and NEAP scores and the above stratification factors due to the insufficient power. Therefore, our findings from subgroup analyses should be treated with caution and need to be confirmed by future studies.

In conclusion, dietary acid load, as reflected by PRAL and NEAP scores, is found to be positively associated with the risk of pancreatic cancer in this population. These findings suggest that reducing dietary acid load may be beneficial in decreasing the risk of pancreatic cancer. As contemporary Western diets are rich in acidogenic foods, our findings could have substantial public health implications. Future studies should validate our results in other populations and settings.

No disclosures were reported.

L.-W. Shi: Writing–original draft. Y.-L. Wu: Resources. J.-J. Hu: Writing–review and editing. P.-F. Yang: Writing–original draft. W.-P. Sun: Writing–review and editing. J. Gao: Software, formal analysis, methodology. K. Wang: Software, formal analysis, methodology. Y. Peng: Writing–review and editing. J.-J. Wu: Validation, methodology, writing–review and editing. G.-C. Zhong: Conceptualization, data curation, supervision, project administration, writing–review and editing.

The authors sincerely appreciate the NCI for access to NCI's data collected by the PLCO Screening Trial. The statements contained herein are solely those of the authors and do not represent or imply concurrence or endorsement by NCI. This study was supported by Medical Science and Technology Innovation Fund of Chongqing General Hospital (grant no. Y2020MSXM01).

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.