Meat consumption and risk of hepatobiliary cancers in the National Institutes of Health–AARP Diet and Health Study

Abstract

Background

We investigated the relationship between intakes of red, white, and processed meats with liver cancer—including hepatocellular carcinoma and intrahepatic cholangiocarcinoma, gallbladder cancer, and other biliary tract cancers.

Methods

The analytic cohort consisted of 480 347 US adults in the prospective National Institutes of Health–AARP Diet and Health Study who were cancer-free at baseline at ages 50-71 years. With a median follow-up of 19.68 years, we identified 1150 participants with incident liver cancer (219 intrahepatic cholangiocarcinomas and 931 hepatocellular carcinomas), 231 with incident gallbladder cancer, and 472 with other incident biliary tract cancers (272 extrahepatic cholangiocarcinomas). At baseline, a self-administered food frequency questionnaire assessed usual dietary intake. We used multivariable Cox proportional hazards models to estimate the associations of meat type with hepatobiliary cancers. We used substitution models with the “leave-one-out” approach as our primary analysis and addition models as a supplemental analysis.

Results

Replacing red meat with white meat was inversely associated with liver cancer (hazard ratio [HR]_{50 g/1000 kcal} = 0.62, 95% CI = 0.51 to 0.77), hepatocellular carcinoma (HR_{50 g/1000 kcal} = 0.63, 95% CI = 0.50 to 0.80), and intrahepatic cholangiocarcinoma (HR_{50 g/1000 kcal} = 0.56, 95% CI = 0.35 to 0.90). Because of the symmetry of substitution models, replacing white meat with red meat yielded hazard ratios equal to the reciprocal of these values, indicating increased risk for the same cancer sites. No associations were observed for meat intake and gallbladder or other biliary tract cancers.

Conclusions

Our study indicates replacing intake of red meat with white meat could lower risk of liver cancer by nearly 40%, whereas replacing white meat with red meat could increase the risk by more than 60%.

Introduction

In the United States, the incidence of liver cancer rose steadily from the 1990s to 2015, when it plateaued among males but continued to rise among females.¹ Projections for 2025 estimate over 42 000 new diagnoses will be made in the United States.² Although not among the top 10 cancers in incidence in the United States, liver cancer accounts for an outsized number of deaths—ranking as the sixth most common cause of cancer mortality in the United States, with a projected 30 000 deaths in 2025.² Cancers of the gallbladder and other sites of the biliary tract share some common risk factors with liver cancer but, at present, are understudied because of lower case numbers.

In 2015, an expert panel of the International Agency for Research on Cancer classified consumption of processed meat as a group 1 human carcinogen based on sufficient evidence of an association with colorectal cancer.³ Consumption of red meat was classified as a probable carcinogen in humans (group 2A) because of associations with colorectal, pancreatic, and prostate cancers.³ There are several hypotheses addressing the underlying carcinogenic mechanisms of these meats. The curing, preservation, and cooking processes can introduce cancer-promoting compounds such as heterocyclic aromatic amines, polycyclic aromatic hydrocarbons, nitrites, and N-nitrosamines to meats.^4-7 Because worldwide per capita meat consumption nearly doubled in the last 50 years, largely driven by economic development and the adoption of more Western-style diets, it is increasingly important to understand whether these foods are associated with less common cancers of the digestive tract.^8,9

The World Cancer Research Fund recently concluded that evidence is limited concerning the association of meat consumption with liver cancer and that little evidence exists concerning the association of meat consumption with gallbladder cancer.^10,11 However, there is some evidence suggesting fish—and white meat more broadly—is inversely associated with risk of liver cancer. A prospective cohort study in Japan reported a strong inverse relationship between consumption of fish high in n-3 polyunsaturated fatty acids and hepatocellular carcinoma, the most common histologic type of liver cancer.¹² The European Prospective Investigation into Cancer and Nutrition cohort similarly showed an inverse relationship between fish and hepatocellular carcinoma, although no notable association was observed between red or processed meat and hepatocellular carcinoma.¹³ Two previous studies of liver cancer risk in the National Institutes of Health [NIH]–AARP Diet and Health Study found increased risk with red meat intake, inverse associations with white meat intake, and no associations with processed meat intake.^14,15 With an additional 13 years of follow-up and nearly 3 times the number of participants with incident liver cancer as the prior investigations, we can further examine whether heterogeneity in the liver–meat relationship exists between different histologic types of liver cancer: hepatocellular carcinoma and intrahepatic cholangiocarcinoma.^14,15 Additionally, these previous studies focused on the independent associations of red and white meat, whereas we primarily investigated the effects of replacing one meat type with another. This is more likely to reflect real-life dietary changes, as dietary intake is compositional and therefore is of greater public health significance. To date, this is also among the largest prospective investigations into the relationship of meat intake and gallbladder and other biliary tract cancers.

Methods

Study population

The NIH-AARP Diet and Health Study is a large-scale prospective cohort study that has been described in detail previously.¹⁶ To summarize, in 1995, a self-administered baseline questionnaire was sent to 3.5 million members of AARP, ages 50-71 years, located in 6 states (California, Florida, Pennsylvania, New Jersey, North Carolina, and Louisiana) and 2 metropolitan areas (Detroit, Michigan, and Atlanta, Georgia). The baseline questionnaire collected information pertaining to demographics, diet, and other health behaviors. A total of 566 398 participants returned completed baseline questionnaires. A risk factor questionnaire, which asked about additional cancer risk factors such as past diet, medication use, and more detailed physical activity information, was subsequently returned by 333 000 of these participants. Of the participants who returned the baseline questionnaire, we excluded those who completed the baseline questionnaire via proxy (n = 15 760), had zero person-years (n = 52), had a self-reported or prevalent cancer prior to baseline (n = 51 120), self-reported end-stage renal disease (n = 997), or lacked a cancer registry report (n = 14 364). Additionally, participants with extreme caloric intakes (n = 3758), defined as 2 interquartile ranges above the 75th percentile or below the 25th percentile of the Box–Cox transformed caloric values, were excluded. The final analytic cohort consisted of 480 347 participants (285 634 male and 194 713 female) (Figure S1). For the gallbladder cancer analysis, 33 052 additional participants who had received surgery to remove their gallbladder were also excluded. The NIH-AARP Diet and Health Study was approved by the Special Studies institutional review board at the National Cancer Institute. All participants provided informed consent via their completion and subsequent return of the baseline questionnaire.

Outcome ascertainment

Participants were cancer-free at enrollment and were followed until the occurrence of their first primary cancer diagnosis, death, relocation outside the follow-up ascertainment area, or the end of follow-up (December 31, 2018). Participants with incident cancers of interest to this study were classified using histology and morphology codes from the International Classification of Diseases for Oncology, Third Edition, and Surveillance, Epidemiology, and End Results Program incidence site recode. Hepatocellular carcinoma was defined using site code C220 and histology codes 8170-8175. One case with site code C221 and histology code 8170 (hepatocellular carcinoma, not otherwise specified) was also classified as hepatocellular carcinoma. Intrahepatic cholangiocarcinoma was identified using site code C221 and histology codes 8000, 8010, 8012, 8140, 8160-8162, 8180, 8480, and 8481. In this investigation, liver cancer cases were defined as the sum of hepatocellular carcinoma and intrahepatic cholangiocarcinoma cases. Gallbladder cancer cases were identified by site code C230 and histology codes 8000, 8010, 8041, 8046, 8070, 8140, 8160, 8210, 8240, 8246, 8255, 8260, 8261, 8310, 8480, 8481, 8490, 8560, 8570, 8574, and 8980. Other biliary tract cancers were identified by site codes C240 (extrahepatic cholangiocarcinoma); C241 (cancer of the Ampulla of Vater); C248 (overlapping lesions of the biliary tract); C249 (cancer of the biliary tract, not otherwise specified); and histology codes 8000, 8010, 8140, 8160, 8162, 8210, 8240, 8246, 8260, 8261, 8263, 8480, 8481, 8490, 8500, and 8980.

Exposure assessment

Data on meat intake were collected via the self-administered 124-item food frequency questionnaire, which was administered to all participants at baseline and gathered information on the frequency of consumption and portion size of meat products and other foods over the previous 12 months.¹⁶ The frequency was collected across 10 predefined categories ranging from never to 2 or more times per day for foods and never to 6 or more times per day for beverages. Three categories were provided for portion sizes, with the unit and amount depending on the food or beverage. Nutrient content and portion sizes were estimated using the US Department of Agriculture’s 1994-1996 Continuing Survey of Food Intakes by Individuals.¹⁷ After processing, data for all meats were reported in grams per day. Red meat consisted of processed red meat (ham, bacon, hotdogs, cold cuts, sausages, and the meat components of pizza) and unprocessed red meat (steak, beef roast, hamburgers, pork, liver, and the meat components of beef stew, meatloaf, lasagna, chili, and tomato sauces with meat). White meat consisted of processed white meat (poultry cold cuts, low-fat hotdogs, and low-fat sausages) and unprocessed white meat (chicken, turkey, fish, and ground poultry). White meat was also looked at in terms of fish and poultry (chicken, poultry cold cuts, ground poultry, turkey, and low-fat hotdogs and sausages). Total processed meat was the sum of red and white processed meats. The nutrient density method was used to adjust all meat variables, standardizing them for total energy intake and resulting in units of grams per 1000 kcal of energy per day. Red meat intake was validated using two 24-hour dietary recalls within a year of the baseline questionnaire, showing correlations of 0.62 for men and 0.70 for women between the food frequency questionnaire and 24-hour recall.¹⁶

Statistical analysis

We used Cox proportional hazards regression models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for fixed 50 g per 1000 kcal increases in total red meat, unprocessed red meat, processed red meat, total white meat, unprocessed white meat, processed white meat, fish, poultry, and total processed meat in relation to liver, gallbladder, and other biliary tract cancers. This dose corresponds to the roughly 75th percentile of white and red meat intakes in our cohort. We used substitution and addition models to investigate the effects of these different meats.¹⁸ Substitution models using the “leave-one-out” approach comprised the main analysis.¹⁹ With this, one meat type was substituted for another meat type, which was left out of the model, while a total meat intake variable was included. For example, red meat was included in a model with total meat. This showed the effect of substituting red meat for white meat, which was left out of the model. By keeping total meat intake constant, every fixed unit increase for red meat was met by an equal decrease in white meat. Because substitution models are symmetric, the hazard ratio for substituting red meat for white meat is the reciprocal of the hazard ratio for substituting white meat for red meat. Although the direction of substitution changes, the magnitude of the effect remains the same on the log scale. For the total processed meat analysis, total unprocessed meat was left out of the model while adjusting for total meat. For the meat subcomponent analyses, both components (eg, unprocessed red meat and processed red meat) making up the greater category (eg, red meat) were included in the same model along with total meat, while the other greater category (eg, white meat) was left out. Addition models included the separate types of meat (eg, red and white meat) but did not adjust for total meat intake. This displayed the independent effect of increasing each meat by keeping the consumption of the corresponding second meat type fixed but, therefore, allowing an overall increase in total meat intake. Person-years of follow-up was used as the underlying time metric, and all associated tests were 2-sided. We tested the proportional hazards assumption for liver cancer (P = .43), gallbladder cancer (P = .32), and other biliary tract cancers (P = .92) by including an interaction term between person-years and meat intake and found no violations. We also used joint Cox proportional hazards models to test for heterogeneity between the subtypes of liver cancer.²⁰

Cox models were adjusted for the following confounders: sex, entry age, family history of cancer, aspirin usage in the previous 12 months, race and ethnicity, educational status, diabetes status, smoking status and dose, body mass index (BMI), physical activity, alcohol intake, coffee intake frequency, fruit intake, vegetable intake, Healthy Eating Index (HEI) score, and total energy. As described above, substitution models were additionally adjusted for total meat consumption, while addition models were jointly adjusted for either red or white meat intake, or processed or unprocessed meat intake. Sex-specific medians were imputed for missing values of BMI. One category included Asian, Pacific Islander, American Indian, and Alaskan Native participants because of small numbers in each individual group. For categorical variables, missing values were represented by a separate category in the models. All variables had less than 5.2% missing values except for aspirin usage (39.3%), which was collected in a subset of participants via the risk factor questionnaire.

In supplemental analyses, we performed a lag analysis, estimating hazard ratios during 5-year follow-up periods to evaluate potential reverse causation. We also ran addition models with quintiles of meat intake to mitigate potential effects of outliers and facilitate comparisons with previous investigations.^14,15 Because of low consumption of processed white meat, a category was created for nonconsumers, and quartiles of consumers comprised the remaining 4 categories. We stratified by diabetes status and level of BMI (<25, 25 to <30, >30 kg/m²) to further examine the associations with meat intake, as these are known risk factors for liver cancer and gallbladder cancer;^10,11,21 participants with missing values for a stratifying variable were excluded from the analysis. We also ran our main analysis with a 25 g per 1000 kcal/day dose for each meat, as this corresponds to approximately the median intake of red and white meat and therefore may more closely represent the meat consumption patterns of participants.

All analyses were conducted using SAS version 9.4 (SAS Institute) and R Statistical Software (v4.4.1; R Core Team 2024). All statistical tests for significance were 2-sided, and P values less than .05 were deemed statistically significant. Strengthening the Reporting of Observational Studies in Epidemiology – Nutritional Epidemiology guidelines were used for reporting results and preparing the manuscript.²²

Results

A total of 480 347 participants were included in the analytic cohort at baseline. There were 1150 incident liver cancers (219 intrahepatic cholangiocarcinomas and 931 hepatocellular carcinomas), 231 gallbladder cancers, and 472 other biliary tract cancers (272 extrahepatic cholangiocarcinomas). There were 7 910 937 person-years of follow-up, with a median of 19.68 years. Mean intakes of processed, total red, and total white meat were 11.36, 34.54, 33.55 g per 1000 kcal/day, respectively (Table S1). Red unprocessed meat accounted for 75% of total red meat consumed, while poultry accounted for 67% of total white meat consumed by participants. Those in the lowest quintile of red meat consumption correspondingly ate the highest amount of white meat (Table 1). Consumption of processed meat appeared to increase with higher consumption of red meat, because of the most processed meat in this cohort being derived from red meat sources such as bacon, ham, hot dogs, and cold cuts. Participants in the highest quintile of red meat intake were more often male, non-Hispanic White, current smokers, had diabetes, consumed coffee more than 3 times per day, had lower HEI scores, had higher BMI and daily energy intake, and were less likely to have a college degree and exercise regularly than the lowest quintile.

Table 1.

National Institutes of Health–AARP Diet and Health Study sample characteristics by quintile of red meat intake (n = 480 347)

Characteristic	Red meat intake quintile, g/1000 kcal
	Quintile 1	Quintile 3	Quintile 5
Male, No. (%)	44 260 (46.1)	56 851 (59.2)	70 679 (73.6)
Age, mean (SD), y	62.33 (5.4)	62.14 (5.3)	61.39 (5.4)
Total meat, mean (SD), grams per 1000 kcal	46.37 (33.0)	63.93 (22.9)	99.82 (27.0)
Red meat, mean (SD), grams per 1000 kcal	9.27 (4.5)	31.41 (3.0)	67.62 (16.8)
White meat, mean (SD), grams per 1000 kcal	37.10 (32.8)	32.52 (22.7)	32.21 (21.3)
Processed meat, mean (SD), grams per 1000 kcal	5.01 (6.6)	10.56 (7.2)	19.66 (13.9)
Family history of cancer, No. (%)	46 317 (48.2)	47 218 (49.2)	45 884 (47.8)
Race, No. (%)
Asian, Pacific Islander, or American Indian or Alaskan Native	2602 (2.7)	1395 (1.5)	1153 (1.2)
Hispanic	2340 (2.4)	1697 (1.8)	1830 (1.9)
Non-Hispanic Black	5869 (6.1)	3405 (3.5)	2356 (2.5)
Non-Hispanic White	83 546 (87.0)	88 462 (92.1)	89 521 (93.2)
Other or missing	1712 (1.8)	1110 (1.1)	1209 (1.2)
Smoking status and dose, No. (%)
Never smoked	39 222 (40.9)	33 869 (35.3)	28 627 (29.8)
Quit, smoked ≤20 cigarettes per day	28 737 (29.9)	27 085 (28.2)	23 504 (24.5)
Quit, smoked >20 cigarettes per day	17 471 (18.2)	20 135 (21.0)	24 440 (25.4)
Current, smokes ≤ 20 cigarettes per day	5021 (5.2)	7711 (8.0)	9074 (9.4)
Current, smokes >20 cigarettes per day	1662 (1.7)	3778 (3.9)	6724 (7.0)
Missing	3956 (4.1)	3491 (3.6)	3700 (3.9)
Physical activity, No (%)
Never	3883 (4.0)	4025 (4.2)	5168 (5.4)
Rarely	9974 (10.4)	13 020 (13.6)	15 896 (16.6)
1-3 times per month	9749 (10.2)	13 356 (13.9)	15 277 (15.9)
1-2 times per week	16 968 (17.7)	21 548 (22.3)	22 473 (23.4)
3-4 times per week	28 791 (30.0)	25 921 (27.0)	21 818 (22.7)
≥5 times per week	25 379 (26.3)	17 276 (18.0)	14 404 (14.9)
Missing	1325 (1.4)	923 (1.0)	1033 (1.1)
College or postgraduate education, No. (%)	42 653 (44.4)	36 509 (38.0)	33 386 (34.8)
Body mass index, mean (SD), kg/m2	25.76 (4.7)	27.13 (4.9)	28.30 (5.3)
Energy intake, mean (SD), kcal/day	1700.22 (770.3)	1829 (788.6)	1992.51 (860.7)
Alcohol intake, mean (SD), grams per 1000 kcal	5.64 (12.5)	6.24 (11.1)	5.03 (8.4)
Diabetes, No. (%)	5945 (6.2)	7944 (8.3)	12 858 (13.4)
Used aspirin in last 12 months, No. (%)	42 385 (44.1)	43 298 (45.1)	41 848 (43.6)
Missing	35 388 (36.8)	37 506 (39.0)	40 371 (42.0)
Coffee intake, No. (%)
<1 cup per day	33 457 (34.8)	24 158 (25.1)	23 320 (24.3)
1 cup per day	17 012 (17.7)	15 927 (16.6)	14 314 (14.9)
2-3 cups per day	34 441 (35.9)	40 681 (42.4)	39 871 (41.5)
>3 cups per day	11 159 (11.6)	15 303 (15.9)	18 564 (19.3)
Vegetable intake, mean (SD), cups per day	2.1 (1.5)	1.9 (1.2)	2.0 (1.2)
Fruit intake, mean (SD), cups per day	2.8 (2.2)	2.0 (1.5)	1.5 (1.2)
Health Eating Index score, mean (SD)	72.3 (9.0)	68.1 (8.9)	62.0 (8.8)

In the substitution models, replacing white meat with red meat intake was associated with a statistically significant increase in liver cancer risk (HR_{50 g/1000 kcal} = 1.60, 95% CI = 1.31 to 1.97). This association held for hepatocellular carcinoma (HR_{50 g/1000 kcal} = 1.58, 95% CI = 1.25 to 1.98) and intrahepatic cholangiocarcinoma (HR_{50 g/1000 kcal} = 1.78, 95% CI = 1.11 to 2.84) (Figure 1). Overall, associations with liver cancer were stronger when replacing white meat with unprocessed red meat compared with processed red meat, although it should again be noted that most (74%) red meat intake was from unprocessed red meat. Addition models were also used, giving the independent effects of each type of meat on cancer risk. This analysis showed statistically significant increased risk of liver cancer overall for total red meat (HR_{50 g/1000 kcal} = 1.15, 95% CI = 1.00 to 1.32) and unprocessed red meat (HR_{50 g/1000 kcal} = 1.26, 95% CI = 1.07 to 1.49), as well as increased risk of hepatocellular carcinoma for unprocessed red meat (HR_{50 g/1000 kcal} = 1.23, 95% CI = 1.03 to 1.48) (Table S2). The quintile analysis using the addition models yielded similar results for liver cancer and its histologies across red meat types as the continuous analysis (Table S3). There were statistically significant trends across quintiles of total red meat intake for overall liver cancer and gallbladder cancer and quintiles of unprocessed red meat intake for overall liver cancer, hepatocellular carcinoma, and gallbladder cancer.

Figure 1.

Adjusted associations^a of liver, gallbladder, and other biliary tract cancer incidence by types of red meat intake^b in substitution models in the National Institutes of Health–AARP Diet and Health Study.^aAdjusted for sex, entry age (years, continuous), family history of cancer (yes or no), usage of aspirin (yes or no), race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian, Pacific Islander, or American Indian or Alaskan Native), education (11 years or less, 12 years or completed high school, post high school, some college, college and postgraduate, or unknown), smoking status and dose (never smoked, quit and smoked ≤20 cigarettes per day, quit and smoked >20 cigarettes per day, currently smoking ≤20 cigarettes per day, currently smoking >20 cigarettes per day), diabetes status (yes or no), body mass index (kg/m², continuous), and intakes of total meat (grams per 1000 kcal per day, continuous), alcohol (grams of alcohol from alcoholic drinks per 1000 kcal/day, continuous, total energy (kilocalories per day, continuous), coffee intake frequency (<1 time per day, 1 to <2 time per day, 2 to <3 times per day, >3 times per day), fruit and vegetable intake (cups per day), physical activity (never, rarely, 1-3 times per month, 1-2 times per week, 3-4 times per week, ≥5 times per week, or unknown), and Healthy Eating Index score. ^bPer 50 g per 1000 kcal of intake, which corresponds to approximately 75th percentile of total red meat intake. Abbreviations: CI = confidence interval; HR = hazard ratio.

Replacing white meat with red meat intake did not increase risk of gallbladder cancer (HR_{50 g/1000 kcal} = 1.29, 95% CI = 0.87 to 1.93) or other biliary tract cancers (HR_{50 g/1000 kcal} = 0.94, 95% CI = 0.71 to 1.26) in substitution models (Figure 1). The independent effects of all types of red meat were null for other biliary tract cancers, but the risk of gallbladder cancer increased with total red meat (HR_{50 g/1000 kcal} = 1.48, 95% CI = 1.08 to 2.01) and unprocessed red meat (HR_{50 g/1000 kcal} = 1.52, 95% CI = 1.05 to 2.18) (Table S2). In the quintile analysis with independent effects of meat intake, there was a statistically significant increased risk of gallbladder cancer with total red meat (HR_Q5vsQ1 = 1.83, 95% CI = 1.14 to 2.94; P_trend = .025) and unprocessed red meat (HR_Q5vsQ1 = 1.90, 95% CI = 1.14 to 3.17; P_trend = .020) (Table S3).

Conversely, replacing red meat intake with white meat resulted in lower risk of liver cancer overall (HR_{50 g/1000 kcal} = 0.62, 95% CI = 0.51 to 0.77); this association was largely driven by replacing red meat with poultry (HR_{50 g/1000 kcal} = 0.58, 95% CI = 0.46 to 0.73) (Figure 2). This statistically significant inverse association was observed for substituting red meat for total white meat for hepatocellular carcinoma (HR_{50 g/1000 kcal} = 0.63, 95% CI = 0.50 to 0.80) and intrahepatic cholangiocarcinoma (HR_{50 g/1000 kcal} = 0.56, 95% CI = 0.35 to 0.90), as well as for poultry for hepatocellular carcinoma (HR_{50 g/1000 kcal} = 0.59, 95% CI = 0.45 to 0.76) and intrahepatic cholangiocarcinoma (HR_{50 g/1000 kcal} = 0.52, 95% CI = 0.30 to 0.88). Substituting red meat for fish showed no statistically significant association with any cancer site in this study. We observed strong inverse associations for the replacement of red meat with processed white meat and hepatocellular carcinoma, although this could be a spurious result because of the low intake of this meat type in our cohort (Table S1). The independent associations of white meat types were strong in the liver overall for total white meat (HR_{50 g/1000 kcal} = 0.71, 95% CI = 0.62 to 0.83) and poultry (HR_{50 g/1000 kcal} = 0.66, 95% CI = 0.55 to 0.79) and hepatocellular carcinoma for total white meat (HR_{50 g/1000 kcal} = 0.71, 95% CI = 0.60 to 0.83) and poultry (HR_{50 g/1000 kcal} = 0.65, 95% CI = 0.53 to 0.80) (Table S2). The quintile analysis with independent effects showed nearly identical lowered risks in the highest quintile for white meat and its components for all cancer sites, with statistically significant trends across quintiles of total white meat, processed white meat, unprocessed white meat, and poultry for liver cancer overall and for hepatocellular carcinoma (Table S3).

Figure 2.

Adjusted associations^a of liver, gallbladder, and other biliary tract cancer incidence by types of white meat intake^b in substitution models in the National Institutes of Health–AARP Diet and Health Study.^aAdjusted for sex, entry age (years, continuous), family history of cancer (yes or no), usage of aspirin (yes or no), race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian, Pacific Islander, or American Indian or Alaskan Native), education (11 years or less, 12 years or completed high school, post high school, some college, college and postgraduate, or unknown), smoking status and dose (never smoked, quit and smoked ≤20 cigarettes per day, quit and smoked >20 cigarettes per day, currently smoking ≤20 cigarettes per day, currently smoking >20 cigarettes per day), diabetes status (yes or no), body mass index (kg/m², continuous), and intakes of total meat (grams per 1000 kcal/day, continuous), alcohol (grams of alcohol from alcoholic drinks per 1000 kcal/day, continuous), total energy (kilocalories per day, continuous), and coffee intake frequency (<1 time per day, 1 to <2 time per day, 2 to <3 times per day, >3 times per day), fruit and vegetable intake (cups per day), physical activity (never, rarely, 1-3 times per month, 1-2 times per week, 3-4 times per week, ≥5 times per week, or unknown), and Healthy Eating Index score. ^bPer 50 g per 1000 kcal of intake, which corresponds to approximately 75th percentile of total white meat intake. Abbreviations: CI = confidence interval; HR = hazard ratio.

There were no statistically significant associations for gallbladder cancer or other biliary tract cancers with the replacement of red meat with total white meat or any of its subcomponents (Figure 1). The independent effects of white meat were null for gallbladder and other biliary tract cancers across white meat types, although there was increased risk of extrahepatic cholangiocarcinoma from unprocessed white meat in females (HR_{50 g/1000 kcal} = 1.43, 95% CI = 1.03 to 1.98) (Table S2). Only poultry intake was independently associated with extrahepatic cholangiocarcinoma in the quintile analysis of white meat types (HR_{Q5 v Q1} = 1.50, 95% CI = 1.01 to 2.23; P_trend = .067) (Table S3).

There were no observed statistically significant associations for the replacement of unprocessed meat with processed meat, comprised of both red and white processed meats, and any cancer site (Figure S2). There was a minor inverse association for overall liver cancer (HR_{50 g/1000 kcal} = 0.73, 95% CI = 0.54 to 0.99) and hepatocellular carcinoma (HR_{50 g/1000 kcal} = 0.69, 95% CI = 0.49 to 0.96) with processed meat independently (Tables S1 and S2).

Further analyses stratified by sex showed statistically significant increased risk for intrahepatic cholangiocarcinoma from the replacement of white meat with total red meat (HR_{50 g/1000 kcal} = 2.15, 95% CI = 1.16 to 4.01) and unprocessed red meat (HR_{50 g/1000 kcal} = 2.55, 95% CI = 1.32 to 4.95) among males only (Table S4). There were also statistically significant inverse associations among males for intrahepatic cholangiocarcinoma and the replacement of red meat with total white meat (HR_{50 g/1000 kcal} = 0.46, 95% CI = 0.25 to 0.86), unprocessed white meat (HR_{50 g/1000 kcal} = 0.43, 95% CI = 0.22 to 0.83), and poultry (HR_{50 g/1000 kcal} = 0.37, 95% CI = 0.18 to 0.77). Upon testing the interactions between the meat variables and sex, the results were only statistically significant for intrahepatic cholangiocarcinoma with total white meat (P_interaction = .04), unprocessed white meat (P_interaction = .03), and poultry (P_interaction = .03). The lag analysis displayed little difference in risk of liver, gallbladder, or other biliary tract cancers with any of the meat types over 5-year follow-up periods (Table S5). Tests for heterogeneity among liver cancer histologies showed no statistically significant differences in risk with any meat type (red: P =  .62; white: P =  .62; processed: P =  .92). There were few differences in meat intake and cancer associations by self-reported diabetes status, although some cancer sites lacked sufficient power among participants with diabetes because of low case numbers (Table S6). Stratification by BMI showed generally consistent increased risk associated with the substitution of white meat for red meat intake and lowered risk associated with replacing red meat with total white meat and poultry across all categories of BMI for liver cancer, hepatocellular carcinoma, and intrahepatic cholangiocarcinoma (Table 2). Testing across these categories showed there was only a statistically significant difference in risk associated with processed red meat and hepatocellular carcinoma (P =  .03); all other combinations of meat type and cancer outcomes showed no statistically significant differences by BMI strata, indicating no evidence of effect modification. Results for 25 g per1000 kcal per day doses of meat intake using the main substitution models were expectedly weaker compared with the 50 g per 1000 kcal per day dose, but the same statistically significant associations persisted (Table S7).

Table 2.

Adjusted associations^a of liver, gallbladder, and other biliary tract cancers with types of meat intake^b across strata of body mass index (kg/m²) in substitution models in the National Institutes of Health–AARP Diet and Health Study

		Meat type, HR (95% CI)
Site	Body mass index category, kg/m2	Red	Red unprocessed	Red processed	White	White unprocessed	White processed	Fish	Poultry	Processed
Liver	<25	2.11 (1.35 to 3.31)	2.10 (1.28 to 3.44)	2.16 (0.98 to 4.73)	0.47 (0.30 to 0.74)	0.51 (0.32 to 0.80)	0.21 (0.04 to 1.06)	0.56 (0.28 to 1.10)	0.44 (0.26 to 0.74)	1.17 (0.58 to 2.34)
	25 to <30	1.65 (1.20 to 2.26)	1.88 (1.33 to 2.65)	1.07 (0.59 to 1.93)	0.61 (0.44 to 0.83)	0.64 (0.46 to 0.89)	0.32 (0.11 to 0.96)	0.90 (0.56 to 1.45)	0.51 (0.35 to 0.74)	0.73 (0.44 to 1.21)
	≥30	1.25 (0.89 to 1.78)	1.41(0.97 to 2.06)	0.85 (0.44 to 1.63)	0.80 (0.56 to 1.13)	0.83 (0.58 to 1.18)	0.54 (0.17 to 1.71)	0.81 (0.44 to 1.47)	0.79 (0.54 to 1.16)	0.72 (0.41 to 1.26)
Hepatocellular carcinoma	<25	2.42 (1.44 to 4.05)	2.43 (1.38 to 4.26)	2.39 (0.99 to 5.73)	0.41 (0.25 to 0.69)	0.44 (0.26 to 0.76)	0.18 (0.03 to 1.18)	0.53 (0.25 to 1.15)	0.37 (0.20 to 0.67)	1.19 (0.55 to 2.58)
	25 to <30	1.52 (1.07 to 2.15)	1.67 (1.14 to 2.45)	1.12 (0.60 to 2.10)	0.66 (0.47 to 0.93)	0.71 (0.49 to 1.01)	0.30 (0.09 to 1.03)	0.94 (0.56 to 1.60)	0.57 (0.38 to 0.85)	0.76 (0.44 to 1.32)
	≥30	1.25 (0.85 to 1.85)	1.47 (0.96 to 2.24)	0.74 (0.35 to 1.54)	0.80 (0.54 to 1.18)	0.86 (0.57 to 1.28)	0.35 (0.09 to 1.42)	0.76 (0.39 to 1.51)	0.81 (0.53 to 1.24)	0.57 (0.30 to 1.09)
Intrahepatic cholangiocarcinoma	<25	1.42 (0.57 to 3.55)	1.36 (0.48 to 3.88)	1.60 (0.27 to 9.57)	0.71 (0.28 to 1.77)	0.74 (0.29 to 1.88)	0.34 (0.01 to 8.93)	0.65 (0.15 to 2.75)	0.73 (0.27 to 1.99)	1.07 (0.22 to 5.16)
	25 to <30	2.43 (1.12 to 5.29)	3.26 (1.45 to 7.31)	0.63 (0.11 to 3.62)	0.41 (0.19 to 0.90)	0.41 (0.18 to 0.91)	0.45 (0.04 to 5.15)	0.71 (0.23 to 2.19)	0.32 (0.13 to 0.79)	0.55 (0.14 to 2.19)
	≥30	1.35 (0.62 to 2.92)	1.26 (0.53 to 2.96)	1.69 (0.44 to 6.53)	0.74 (0.34 to 1.61)	0.68 (0.30 to 1.53)	1.63 (0.28 to 9.58)	0.91 (0.27 to 3.06)	0.69 (0.29 to 1.63)	1.70 (0.57 to 5.07)
Gallbladder	<25	1.30 (0.58 to 2.91)	1.75 (0.77 to 3.98)	0.30 (0.03 to 2.63)	0.77 (0.34 to 1.72)	0.85 (0.38 to 1.90)	0.07 (0.00 to 3.76)	0.91 (0.26 to 3.14)	0.73 (0.30 to 1.75)	0.22 (0.04 to 1.38)
	25 to <30	0.97 (0.50 to 1.90)	0.92 (0.42 to 1.97)	1.18 (0.31 to 4.48)	1.03 (0.53 to 2.01)	1.02 (0.51 to 2.03)	1.19 (0.18 to 8.05)	1.11 (0.38 to 3.26)	1.01 (0.49 to 2.07)	1.19 (0.39 to 3.62)
	≥30	1.53 (0.76 to 3.08)	1.44(0.67 to 3.13)	1.86 (0.55 to 6.36)	0.65 (0.32 to 1.31)	0.55 (0.26 to 1.15)	2.53 (0.72 to 8.87)	0.83 (0.28 to 2.41)	0.60 (0.28 to 1.30)	2.25 (0.89 to 5.66)
Other biliary tract	<25	0.93 (0.55 to 1.58)	0.71 (0.37 to 1.34)	1.86 (0.77 to 4.49)	1.07 (0.63 to 1.83)	1.11 (0.65 to 1.90)	0.62 (0.10 to 3.76)	1.24 (0.60 to 2.58)	1.01 (0.56 to 1.81)	1.42 (0.61 to 3.33)
	25 to <30	1.05 (0.67 to 1.64)	1.07 (0.64 to 1.78)	0.99 (0.42 to 2.35)	0.95 (0.61 to 1.50)	1.02 (0.65 to 1.62)	0.37 (0.07 to 1.88)	0.87 (0.41 to 1.86)	0.98 (0.60 to 1.59)	0.76 (0.35 to 1.64)
	≥30	0.83 (0.46 to 1.48)	0.72 (0.37 to 1.40)	1.27 (0.45 to 3.58)	1.21 (0.68 to 2.16)	1.21 (0.67 to 2.20)	1.14 (0.20 to 6.63)	0.80 (0.28 to 2.27)	1.33 (0.73 to 2.40)	1.26 (0.50 to 3.15)
Extrahepatic cholangiocarcinoma	<25	0.50 (0.23 to 1.10)	0.54 (0.21 to 1.35)	0.41 (0.06 to 2.72)	1.98 (0.91 to 4.31)	2.01 (0.92 to 4.40)	1.54 (0.16 to 14.52)	2.24 (0.84 to 6.02)	1.87 (0.81 to 4.32)	0.53 (0.12 to 2.40)
	25 to <30	1.01 (0.58 to 1.76)	0.84 (0.44 to 1.60)	1.70 (0.63 to 4.62)	0.99 (0.57 to 1.73)	0.97 (0.55 to 1.73)	1.27 (0.26 to 6.17)	0.60 (0.21 to 1.68)	1.12 (0.63 to 2.00)	1.61 (0.68 to 3.81)
	≥30	0.79 (0.37 to 1.66)	0.80 (0.35 to 1.85)	0.75 (0.17 to 3.38)	1.27 (0.60 to 2.67)	1.31 (0.61 to 2.79)	0.84 (0.07 to 9.56)	1.02 (0.28 to 3.66)	1.34 (0.62 to 2.91)	0.77 (0.21 to 2.81)

Abbreviations: CI = confidence interval; HR = hazard ratio.

^aAdjusted for sex, entry age (years, continuous), family history of cancer (yes or no), usage of aspirin (yes or no), race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian, Pacific Islander, or American Indian or Alaskan Native), education (11 years or less, 12 years or completed high school, post high school, some college, college and postgraduate, or unknown), smoking status and dose (never smoked, quit and smoked ≤20 cigarettes per day, quit and smoked >20 cigarettes per day, currently smoking ≤20 cigarettes per day, currently smoking >20 cigarettes per day), diabetes status (yes or no), intakes of total meat (grams per 1000 kcal per day, continuous) and alcohol (grams of alcohol from alcoholic drinks per 1000 kcal per day, continuous), total energy (kilocalories per day, continuous), coffee intake frequency (<1 time per day, 1 to <2 time per day, 2 to <3 times per day, >3 times per day), fruit and vegetable intake (cups per day), physical activity (never, rarely, 1-3 times per month, 1-2 times per week, 3-4 times per week, 5 or more times per week, or unknown), and Healthy Eating Index score.

^bPer 50 g per 1000 kcal of intake.

Discussion

With over 1100 participants with incident liver cancer and now more than 20 years of follow-up, this study has more than twice the case numbers and follow-up time as prior investigations into the associations of meat intakes with liver cancer in the NIH-AARP Diet and Health cohort study. To our knowledge, this study is among the largest prospective investigations into the relationship between meat intake and gallbladder and other biliary tract cancers and strengthens the current evidence base for liver cancer and its 2 most common histologies. We observed that, when replacing white meat, every increase of 50 g per 1000 kcal per day in total red meat intake was associated with a corresponding increase in risk for liver cancer overall (60%), hepatocellular carcinoma (58%), and intrahepatic cholangiocarcinoma (78%). This was largely because of unprocessed red meat, which was associated with elevated risks ranging from 75% to 94% for these cancers (Figure 1). Replacing red meat with white meat had a strong inverse relationship with risk of liver cancer (38% reduction), hepatocellular carcinoma (37%), and intrahepatic cholangiocarcinoma (44%). Results for fish were null for all cancers studied, while there was an inverse association with poultry consumption and liver cancer (42%), hepatocellular carcinoma (41%), and intrahepatic cholangiocarcinoma (48%). The main substitution model results for the other cancer sites were statistically nonsignificant for all meat types.

In totality, there is mixed epidemiological evidence on whether red meat is associated with an increased risk of liver cancer.^23-25 A recent meta-analysis showed a slight increase, but statistically nonsignificant, in relative risk for hepatocellular carcinoma associated with red meat.²⁶ Our results suggest that unprocessed red meat is associated with a higher risk of liver cancer compared with processed red meat. Red meat has high levels of bioavailable heme iron and is correspondingly the most common dietary source of heme iron.^27,28 This is notable because research shows that heme iron can catalyze oxidative reactions that damage DNA.²⁹ In the digestive tract, heme also stimulates the endogenous production of N-nitroso compounds, which are known to be carcinogenic.³⁰ The liver specifically serves as an important reservoir for iron in the body.³¹ Additional evidence shows that excess iron in the liver contributes to a myriad of diseases including cancer through oxidative damage, potentially indicating the importance of total iron given the routes of uptake in the liver.^32,33 Our results could further be explained in part by unprocessed red meats in this study containing higher levels of heme iron, with these levels additionally increasing with the doneness of the meat after cooking.²⁸ Unprocessed red meats, such as beef in hamburgers and steaks, are more likely to be cooked to higher levels of doneness than processed red meats, which can introduce more carcinogenic compounds to the meat product. The fat content could also play a role. Red meats are a common dietary source of saturated fatty acids, which can be pro-inflammatory and accumulate in the liver—contributing to fatty liver disease—and have been associated with hepatocellular carcinoma in another observational study.^15,34 Stratifying by BMI showed generally consistent results across groups, although there were statistically significant differences in risk for hepatocellular carcinoma in relation to processed red meat, with the risks being highest within the lowest BMI category. Studies suggest that energy and protein intake are sometimes underreported in interviews and with food frequency questionnaires among individuals with higher BMIs, which could explain our observed difference in risk profiles.^35,36

Our results are in line with the findings from the aforementioned meta-analysis that showed white meat intake was associated with a statistically significant decrease in risk of hepatocellular carcinoma.²⁶ White meat has lower heme iron and saturated fat content compared with red meat, potentially explaining the protective substitution effect we observed.²⁸ However, our results indicate that this reduced risk from white meat for liver cancer overall, hepatocellular carcinoma, and intrahepatic cholangiocarcinoma also exists independently of substituting red meat for white meat intake—signaling a potential protective mechanistic effect from white meat. More studies are needed to further verify this independent inverse association. Poultry was the main type of white meat in the study, comprising 67% of total white meat intake. Consistent with the current findings, a case-control study looking at components of the Chinese HEI and the HEI 2015 found a statistically significant reduction in risk of primary liver cancer associated with poultry intake.³⁷ Fewer prospective studies separate poultry from total white meat, although among those that did, there were mixed results with liver cancer.^13,38 Higher poultry consumption, however, has been shown to be associated with healthier lifestyles and diets, so there could be residual confounding affecting our results,³⁹ although we attempted to mitigate this by adjusting for dietary quality and physical activity. Notably, fish intake in this study consisted mostly of canned tuna, which commonly contains elevated levels of mercury and other heavy metals. Exposure to these elements may promote carcinogenesis and other toxic effects, potentially explaining the null associations we observed compared with the inverse associations for hepatocellular carcinoma seen in other studies.^26,40,41 Dietary data in the European Prospective Investigation into Cancer and Nutrition cohort included an array of fish products including fresh, canned, salted, and smoked fish, among others. This wider variety of fish types enabled a more robust analysis into its relationship with hepatocellular carcinoma in which they found a statistically significant inverse association.¹³ Fish are high in n-3 polyunsaturated fatty acids, which are known to reduce inflammation in the body, potentially contributing to reduced cancer risks seen with fish in the literature.⁴²

There is little evidence in the literature pertaining to biliary tract cancers and meat intake. A case-control study in India observed a statistically nonsignificant increased risk with red meat and gallbladder cancer but had small case numbers and did not report the actual intake ranges of red meat.⁴³ In 2020, the first prospective study of biliary tract cancers and meat reported no associations with fish and poultry but did find an inverse association with red meat among men.⁴⁴ Our findings show that the independent effects of total red and unprocessed red meats increased gallbladder cancer risk, while other biliary tract cancers had null associations across all types of meat intake in this study. The associations for extrahepatic cholangiocarcinoma differed noticeably from intrahepatic cholangiocarcinoma. A potential explanation could involve the embryological origins of the 2 segments of the biliary tract. Early in development, the hepatic diverticulum consists of 2 buds. The cranial bud gives rise to hepatocytes and intrahepatic cholangiocytes, both of which differentiate from hepatoblasts, while the caudal bud goes on to form the gallbladder and extrahepatic cholangiocytes.^45-47 It is possible that these differing embryological origins could contribute to the observed differences of extrahepatic cholangiocarcinoma and intrahepatic cholangiocarcinoma while also providing a basis for the similarity in results between intrahepatic cholangiocarcinoma and hepatocellular carcinoma.

Our study had several notable strengths in addition to its large size. Because of its prospective design, dietary information for all participants was collected before case ascertainment therefore preventing differential recall bias by case status. With 1150 participants with incident liver cancer, we were able to achieve higher statistical power in this analysis than previous cohort studies to date and to look at hepatocellular carcinoma and intrahepatic cholangiocarcinoma independently. The lag analysis we performed by looking at 5-year segments of follow-up time confirmed that reverse causation was unlikely to explain the observed associations. Our study also had several limitations. Approximately 90% of the participants were of European ancestry, which may limit the generalizability of our findings to the population at large. Even in this large study, we had relatively small numbers of participants with gallbladder and other biliary tract cancers, potentially limiting our ability to detect possible effects with meat intake. Additionally, the dietary data were only collected at one point in time, so the analysis does not reflect any variations in participants’ diets over time and how these variations may affect cancer risk. We also lacked data on chronic hepatitis B or C virus infections, which are known risk factors for hepatocellular carcinoma. This could mean our results overestimate the risk of hepatocellular carcinoma and therefore overall liver cancer associated with meat consumption, although to what degree is unknown. Finally, the observational nature of this study prevents us from establishing true causality between meat intake and these cancers. More work is needed to elucidate any potential mechanisms of carcinogenesis.

To summarize, our findings suggest that red meat intake is associated with an increased risk of liver cancer overall, as well as with hepatocellular carcinoma and intrahepatic cholangiocarcinoma. White meat was inversely associated with risk of liver cancer in both substitution and addition models, indicating that the inverse association may derive from components of white meat and not just indirectly through the consumption of less red meat. Results for gallbladder and other biliary tract cancers are null. Our results are of particular importance because meat consumption is ultimately a modifiable risk factor, so dietary guidelines could be adjusted to minimize the population-wide risk and public health burden of these cancers.

Author contributions

David Wahl (Formal analysis, Writing—original draft, Writing—review & editing), Erikka Loftfield (Formal analysis, Supervision, Writing—review & editing), Sémi Zouiouich (Formal analysis), Linda M. Liao (Conceptualization, Data curation, Formal analysis, Writing—review & editing), Hyokyoung G. Hong (Formal analysis, Writing—review & editing), Katherine A. McGlynn (Conceptualization, Formal analysis, Supervision, Writing—review & editing), and Rashmi Sinha (Conceptualization, Formal analysis, Supervision, Writing—review & editing).

Supplementary material

Supplementary material is available at JNCI Cancer Spectrum online.

Funding

This work was supported by the Intramural Research Program of the National Cancer Institute at the National Institutes of Health.

Conflicts of interest

The authors declare no potential conflicts of interest.

Data availability

Access to the data from the NIH-AARP Diet and Health Study can be requested online at https://dietandhealth.cancer.gov/.