Ultra-Processed Products and Risk of Liver Cancer: A Prospective Cohort Study

Abstract

Background & Aims

Several studies have shown positive associations between ultra-processed foods and drinks and cancer risk. However, evidence remains limited for liver cancer. We aimed to evaluate the associations between ultra-processed foods and drinks and liver cancer risk.

Methods

We included 73,119 participants (22,431 Whites, 47,837 Blacks, 2851 other race) from the Southern Community Cohort Study. Ultra-processed products were defined based on the Nova classification using data from a validated food frequency questionnaire and calculated as percentage of daily foods by weight. Incident liver cancer and vital status were ascertained via linkages to state cancer registries and the National Death Index as of December 31, 2019.

Results

With a median of 13.9 year’s follow-up, we documented 453 incident liver cancer cases. Participants with higher intake of ultra-processed foods had an elevated risk of liver cancer (hazard ratios [HR] _{Tertile 3 vs. tertile 1} 1.69, 95% confidence intervals [CI]: 1.28-2.22; P _trend<0.001). The subclasses of ultra-processed foods, such as ultra-processed grains and fried potatoes (HR _{T3 vs. T1} 1.29, 95% CI: 1.01-1.65; P _trend=0.03), processed protein foods (HR _{T3 vs. T1} 1.49, 95% CI: 1.14-1.94; P _trend=0.007) and mixed dishes (HR _{T3 vs. T1} 1.39, 95% CI: 1.09-1.77; P _trend=0.01), were positively associated with liver cancer risk. No significant association was found for ultra-processed drinks (HR _{T3 vs. T1} 0.85, 95% CI: 0.67-1.07; P _trend=0.16).

Discussion

In a prospective cohort with predominantly low-income Southern US adults, we found certain ultra-processed foods were associated with a higher risk of liver cancer. Further studies are needed to confirm our findings.

Introduction

Liver cancer is the sixth leading cause of cancer incidence and the third leading cause of cancer death worldwide with a 5-year relative survival rate of <20% [1]. Established risk factors for liver cancer include chronic hepatitis B virus or hepatitis C virus infections, excessive alcohol intake, tobacco smoking, and metabolic disorders, including obesity and diabetes; however, more than 35% of US liver cancer cases are not attributable to these known risk factors [2]. The World Cancer Research Fund International/American Institute for Cancer Research concluded in their most recent report that the established dietary risk factors for liver cancer only include aflatoxin-contaminated foods and alcohol consumption. While evidence also suggests that coffee consumption is protective, the evidence for other dietary factors is limited.

Ultra-processed products (UPP), including ultra-processed foods and drinks, are foods undergoing a series of industrial processes and generally using colorants, color stabilizers, flavorings, sweeteners, and other additives [3]. The consumption of UPP worldwide has been increasing [4, 5] and contributes over 60% of total energy intake in the US [6]. Evidence continues to accumulate on positive associations of UPP intake with health outcomes [7, 8], including obesity [9, 10], type 2 diabetes [11, 12], metabolic syndrome [13–15], cardiovascular diseases [16, 17], and mortality [16, 18–20]. However, epidemiological evidence linking UPP intake and specific cancers, especially for liver cancer, is limited and inconsistent [21, 22]. Additionally, none of these studies evaluated subclasses of UPPs and liver cancer risk [21, 22].

Due to the biological impact of UPP on several pathways related to carcinogenesis in liver, such as inflammation, oxidative stress, insulin resistance, and microbiome dysbiosis, we hypothesized that higher consumption of UPP might influence liver cancer risk [23]. Prior studies have shown that food choices are related to race/ethnicity and income, with African Americans and low-income adults tending to consume energy-dense, nutrient-poor foods more frequently than their White and high-income counterparts [24]. Therefore, we tested our hypothesis using data from the Southern Community Cohort Study (SCCS) with participants primarily from low-income communities.

Methods

Study design and participants

The SCCS is a prospective cohort study conducted in 12 southern US states with 84,735 participants aged 40 to 79 years recruited during 2002-2009 [25, 26]. Details about the cohort enrollment were described elsewhere [25]. The SCCS received approval from the Institutional Review Boards of Vanderbilt University Medical Center and Meharry Medical College. All participants provided written informed consent. After excluding participants with invalid data on the demographics or participants who had incomplete food frequency questionnaires (FFQ) (>10 food items left blank) or reported implausible energy intakes (<600 or >8000 kcal/day) (n=7309), diagnosed liver cancer at baseline (n=24), loss of follow-up (n=3), or missing data in covariates (n=4280), we included 73,119 participants in current analyses (Supplementary Figure 1).

Dietary assessment

The dietary information was collected at enrollment using a validated semiquantitative FFQ, which included 89 food items covering the main sources of energy and nutrient intakes for African Americans and non-African Americans in the South. A validation study conducted among 275 SCCS participants indicated that the correlation between the FFQ and 24-hour dietary recall after adjusting total energy intake varied from 0.82 to 0.96 for macronutrients and from 0.73 to 0.95 for micronutrients [27]. The UPPs were defined based on the Nova classification [3] and calculated from the baseline FFQ. In brief, all food items from the FFQ were grouped into four mutually exclusive Nova groups: unprocessed/minimally processed foods, processed culinary ingredients, processed foods, and ultra-processed foods. The category of UPPs included formulations that normally undergo industrial processing like hydrolysis, or hydrogenation, extrusion, moulding, and pre-frying, which included 44 food items in the current study, categorizing into six subgroups based on What We Eat in America Food categories [28]: ultra-processed grains and fried potatoes, processed protein foods, condiments/fats/oils, sweets and snacks, mixed dishes, and non-coffee beverages (Supplementary Table 1). We assigned 100% weight for 32 food items and 30%, 50%, and 70% for 12 food items, considering the chance that those foods/food groups were likely to be UPP in the SCCS population [29]. We used percentage of food weight (in grams) from the UPPs as the exposure variable in the main analyses.

Liver cancer assessment

The main outcome was overall incident liver cancer (International Classification of Diseases, 10^th Revision: C22.0, C22.4, 22.7, 22.8, 22.9) obtained through linkage of the cohort with the 12 SCCS state cancer registries and the National Death Index up to December 31, 2019. We further applied hepatocellular carcinoma (HCC, C22.0) as our secondary outcome. Information on vital status was obtained by linkages to the National Death Index and Social Security Administration mortality files.

Covariate assessment

Information on demographic factors, lifestyle, and medical history was collected at baseline by in-person interview or mail using a structured questionnaire. The a priori covariates list included age (in years, continuous), sex (male, female), race (Whites, Blacks, others), annual household income (<$15,000, $15,000-<$25,000, ≥$25,000), education level (≤11 years, high school or GED, college or above), smoking status (never, former, current), drinking status (never drinker, drinker), physical activity (hours/week, continuous), aspirin use (yes, no), statin use (yes, no), family history of liver cancer (yes, no), self-reported hepatitis B or C infection (yes, no), self-reported diabetes (yes, no), total energy intake (in kcals/day, continuous) and body mass index (BMI, <25, 25-<30, ≥30 kg/m²) [30].

Statistical analyses

Age-adjusted baseline characteristics were presented as mean (standard deviation) for continuous variables and percentage for categorical variables. The follow-up time was calculated from the date of return of the baseline questionnaire to the date of liver cancer diagnosis, the date of death, or the last date of follow-up, whichever came first. We performed Cox proportional hazards regression models with follow-up years as underlying time scale, using tertiles of UPP intake and tested for linear trends using median values of each tertile as continuous variables. We adjusted for covariates a priori selected to reduce potential confounding (see above) [31]. Proportional hazards assumption was assessed by incorporating the product of UPP intake and follow-up time into the multivariable-adjusted model, with no violation observed.

Several sensitivity analyses were performed to test the robustness of our results. First, we conducted lag analyses excluding cases that occurred in the first 2 or 5 years of follow-up to address possible reverse causation due to undiagnosed liver cancer or pre-existing liver diseases. Second, we additionally adjusted coffee intake, dietary quality characterized by Healthy Eating Index-2010, and total food weight. Third, we used the absolute intake (in grams) of the UPPs consumption. Fourth, we evaluated the associations between UPPs intake and HCC risk, the major type of liver cancer. Fifth, we further conducted additional analyses of the components of UPPs. Finally, we tested multiplicative interaction terms using Wald test to evaluate whether UPPs-liver cancer associations differ by a priori major known liver cancer risk factors, such as sex, race, income, obesity, and diabetes.

All statistical analyses were performed using SAS software (version 9.4; SAS Institute Inc., Cary, NC, US). The statistical significance was evaluated using a 2-sided test at 0.05.

Results

Study population characteristics

With a median of 13.9 years of follow-up, we documented 453 incident liver cancer cases among 73,119 participants (men, 29,374; women, 43,745), among which 30.7% were White. People who consumed more UPPs were younger and less likely to be Whites, had more energy intake, a lower dietary quality, lower education, and income level, and they were more likely to be current smokers and drink less coffee (Table 1).

Table 1.

Baseline characteristics of participants from the Southern Community Cohort Study (N=73,119)

Characteristics	Percentage of food weight from ultra-processed products
	Tertile 1	Tertile 2	Tertile 3
N	24,373	24,373	24,373
Age at baseline (year),* mean (SD)	52.3 (8.9)	52.2 (8.7)	51.9 (8.4)
Physical activity (hours/week), mean (SD)	6.5 (4.6)	6.6 (4.8)	6.3 (4.8)
Total energy intake (kcals/day), mean (SD)	2157.2 (1170.2)	2757.2 (1526.6)	2732.2 (1502.6)
Healthy Eating Index-2010, mean (SD)	64.4 (11.7)	57.8 (10.0)	51 (10.0)
Total food weight (grams/day), mean (SD)	2396.4 (1140.8)	2504.6 (1333.5)	2477.6 (1213.4)
Body mass index (kg/m2), mean (SD)	30.1 (7.3)	30.3 (7.5)	30.7 (7.7)
Female, %	66.0	56.4	57.4
White, %	40.7	25.7	25.5
College or higher, %	47.2	36.6	32.1
Annual household income ≥$25,000, %	30.7	21.6	18.6
Current smoker, %	36.2	42.3	44.3
Alcohol drinker, %	52.3	53.3	53.1
Aspirin user, %	13.0	13.5	12.2
Statin user, %	16.5	15.7	15.0
Family history of liver cancer, %	2.1	1.9	1.7
Self-reported hepatitis B virus infection, %	1.3	1.3	1.4
Self-reported hepatitis C virus infection, %	2.8	3.6	3.6
Self-reported diabetes, %	20.6	22.2	21.7
Coffee intake ≥2 times/day, %	28.6	13.3	6.1

Abbreviations: MET, metabolic equivalent of task; SD, standard deviation.

^*Variable is not age-adjusted.

Associations between UPPs intake and liver cancer risk

A higher percentage of UPPs (including both food and drinks) was not associated with risk of liver cancer after controlling for covariates (HR _{Tertile 3 vs. Tertile 1}=0.94, 95% CI= 0.73-1.21; P _trend=0.52) (Table 2). When we analyzed the ultra-processed foods and drinks separately, a significantly positive association of ultra-processed foods with liver cancer was found in the multivariable-adjusted model (HR _{T3 vs. T1}=1.69, 95% CI=1.28-2.22; P _trend<0.001). A per 10% of foods from ultra-processed foods increment was associated with 20% higher risk of liver cancer (HR=1.20, 95% CI=1.09-1.32). We further adjusted the ultra-processed food for ultra-processed drinks (non-coffee beverage), and this did not materially change the HR estimates. The ultra-processed drink (non-coffee beverage) was not significantly associated with liver cancer risk (HR _{T3 vs. T1}=0.85, 95% CI=0.67-1.07; P _trend=0.16).

Table 2.

Ultra-processed products and risk of liver cancer in the Southern Community Cohort Study (N=73,119)

Percentage of food weight from ultra-processed products (%)	Percentage of food weight from ultra-processed products, Hazard Ratio (95% Confidence Interval)			P trend	Hazard Ratio 10% of gram (95% Confidence Interval)
	Tertile 1	Tertile 2	Tertile 3
Ultra-processed products (foods and drinks)
Median (interquartile), %	21.9 (9.9)	37.8 (7.2)	55.4 (14.3)
No. of cases	127	174	152
Age-adjusted model	1 (referent)	1.47 (1.17-1.85)	1.34 (1.05-1.71)	0.02	1.07 (1.01-1.13)
Multivariable model	1 (referent)	1.08 (0.85-1.37)	0.94 (0.73-1.21)	0.52	1.00 (0.94-1.06)
Ultra-processed foods
Median (interquartile), %	12.5 (4.8)	20.3 (4.0)	30.7 (8.5)
No. of cases	84	160	209
Age-adjusted model	1 (referent)	1.97 (1.51-2.56)	2.78 (2.16-3.60)	<0.001	1.45 (1.34-1.58)
Multivariable model	1 (referent)	1.56 (1.19-2.05)	1.69 (1.28-2.22)	<0.001	1.20 (1.09-1.32)
Mutually adjusted for ultra-processed drinks	1 (referent)	1.55 (1.18-2.04)	1.66 (1.24-2.20)	0.002	1.18 (1.06-1.30)
Ultra-processed drinks
Median (interquartile), %	1.3 (3.2)	12.9 (6.7)	33.3 (18.0)
No. of cases	157	166	130
Age-adjusted model	1 (referent)	1.08 (0.87-1.34)	0.87 (0.68-1.09)	0.17	0.92 (0.87-0.98)
Multivariable model	1 (referent)	0.95 (0.76-1.19)	0.85 (0.67-1.07)	0.16	0.93 (0.87-0.99)
Mutually adjusted for ultra-processed foods	1 (referent)	0.95 (0.76-1.19)	0.93 (0.73-1.18)	0.57	0.96 (0.90-1.03)

Multivariable model adjusted for age (continuous), sex (female, male), race (Whites, non-Whites), annual household income (<15k, 15k-<25k, ≥25k), education level (≤11 years, high school or GED, college or above), smoking status (never, former, current), drinking status (never drinker, drinker), physical activity (continuous), aspirin use (yes, no), statin use (yes, no), family history of liver cancer (yes, no), self-reported hepatitis B infection (yes, no), self-reported hepatitis C infection (yes, no), self-reported diabetes (yes, no), total energy intake (continuous), and body mass index (<25, 25-<30, ≥30 kg/m²).

Associations between specific UPPs groups and liver cancer risk

At baseline, the non-coffee beverages accounted for 47.2% of all UPPs, followed by ultra-processed grains and fried potatoes (15.3%), processed protein foods (14.5%), and mixed dishes (10.5%) (Figure 1). Results for specific food groups were generally similar to the main findings (Table 3). Ultra-processed grains and fried potatoes, processed protein foods (bacon, sausage, lunch meat, fried beef, chicken nuggets, et al), and mixed dishes (pizza, soups, hamburgers, et al) were positively associated with liver cancer risk (Ultra-processed grains and fried potatoes: HR _{T3 vs. T1}=1.29, 95% CI=1.01-1.65; P _trend=0.03; Processed protein foods: HR _{T3 vs. T1}=1.49, 95% CI=1.14-1.94; P _trend=0.007; Mixed dishes: HR _{T3 vs. T1}=1.39, 95% CI=1.09-1.77; P _trend=0.01). We did not find significant associations of sweets/snacks and condiments/fats/oils with liver cancer, which may be due to the narrow ranges of intake levels.

Figure 1.

Percent contributions of subclasses of ultra-processed products to total ultra-processed products based on grams per day (N=73,119).

Ultra-processed grains and fried potatoes: bran or high fiber cereals, other cold cereals, grits, oatmeal/cream of wheat/other hot cereals, white bread/rolls/dinner rolls/buns/bagets, dark or whole grain breads, corn bread/corn muffins/corn tortillas/hush puppies, fried potatoes, biscuits; Processed protein foods: bacon or breakfast sausage, fried beef, other ground beef, beef in mixed dishes roast beef/steak/beef barbeque, fried chicken or chicken nuggets, baked/broiled/boiled chicken or turkey, chicken in mixed dishes, seafood, bologna/salami/other lunch meats, hot dogs sausage, meat substitutes; Condiments, fats, and oils: margarine, regular salad dressing or mayonnaise, low fat or reduced fat salad dressing or mayonnaise, gravy added to potatoes/meat/ biscuits, cream or whipped cream; Sweets and snacks: ice cream, frozen yogurt/ice milk/sherbet, cookies, cake, baked or fried pies or cobblers, doughnuts/sweet rolls/pastry/danish/muffins/croissants, chocolate candy or candy bars, potato chips/corn chips/fried pork skins/cheese curls, crackers or pretzels, popcorn; Mixed dishes: pizza, soups or chowders, hamburgers/cheeseburgers/sloppy joes, macaroni and cheese, peanut butter; Non-coffee beverages: Kool-Aid, punch, Tang, lemonade, or sunny delight, Coke, Sprite, diet Coke, or other diet drinks.

Table 3.

Subclass of ultra-processed products and risk of liver cancer in the Southern Community Cohort Study (N=73,119)

Percentage of food weight from subgroups of ultra-processed products (%)*	Percentage of food weight from ultra-processed products, Hazard Ratio (95% Confidence Interval)			P trend	Hazard Ratio 10% of gram (95% Confidence Interval)
Ultra-processed foods	Tertile 1	Tertile 2	Tertile 3
Ultra-processed grains and fried potatoes	1 (referent)	1.10 (0.85-1.41)	1.29 (1.01-1.65)	0.03	1.47 (1.21-1.79)
Processed protein foods	1 (referent)	1.31 (1.01-1.71)	1.49 (1.14-1.94)	0.007	1.59 (1.14-2.23)
Condiments, fats, and oils	1 (referent)	1.11 (0.88-1.39)	1.07 (0.85-1.35)	0.64	1.51 (0.26-8.66)
Sweets and snacks	1 (referent)	0.94 (0.74-1.19)	1.09 (0.86-1.37)	0.35	1.05 (0.84-1.30)
Mixed dishes	1 (referent)	1.23 (0.96-1.58)	1.39 (1.09-1.77)	0.01	1.35 (1.07-1.71)
Ultra-processed drinks	Non-consumer	<Median	≥Median
Sweetened fruit drinks	1 (referent)	1.04 (0.80-1.34)	1.11 (0.86-1.45)	0.39	1.02 (0.91-1.16)
Regular soda	1 (referent)	0.99 (0.75-1.31)	0.78 (0.58-1.05)	0.02	0.90 (0.82-0.98)
Diet soda	1 (referent)	1.09 (0.89-1.32)	0.83 (0.51-1.35)	0.36	0.93 (0.82-1.07)

Model adjusted for age (continuous), sex (female, male), race (Whites, non-Whites), annual household income (<15k, 15k-<25k, ≥25k), education level (≤11 years, high school or GED, college or above), smoking status (never, former, current), drinking status (never drinker, drinker), physical activity (continuous), aspirin use (yes, no), statin use (yes, no), family history of liver cancer (yes, no), self-reported hepatitis B infection (yes, no), self-reported hepatitis C infection (yes, no), self-reported diabetes (yes, no), total energy intake (continuous), and body mass index (<25, 25-<30, ≥30 kg/m²).

^*Ultra-processed grains and fried potatoes: bran or high fiber cereals, other cold cereals, grits, oatmeal/cream of wheat/other hot cereals, white bread/rolls/dinner rolls/buns/bagets, dark or whole grain breads, corn bread/corn muffins/corn tortillas/hush puppies, fried potatoes, biscuits; Processed protein foods: bacon or breakfast sausage, fried beef, other ground beef, beef in mixed dishes roast beef/steak/beef barbeque, fried chicken or chicken nuggets, baked/broiled/boiled chicken or turkey, chicken in mixed dishes, seafood, bologna/salami/other lunch meats, hot dogs sausage, meat substitutes; Condiments, fats, and oils: margarine, regular salad dressing or mayonnaise, low fat or reduced fat salad dressing or mayonnaise, gravy added to potatoes/meat/ biscuits, cream or whipped cream; Sweets and snacks: ice cream, frozen yogurt/ice milk/sherbet, cookies, cake, baked or fried pies or cobblers, doughnuts/sweet rolls/pastry/danish/muffins/croissants, chocolate candy or candy bars, potato chips/corn chips/fried pork skins/cheese curls, crackers or pretzels, popcorn; Mixed dishes: pizza, soups or chowders, hamburgers/cheeseburgers/sloppy joes, macaroni and cheese, peanut butter; Sweetened fruit drinks: Kool-Aid, punch, Tang, lemonade, or sunny delight; Regular soda: Coke, Sprite; Diet soda: diet Coke, or other diet drinks.

Sensitivity and subgroup analyses

Similar results were observed when we excluded the liver cancer cases that occurred in the first 2 or 5 years (Supplementary Table 2). Further adjustment for coffee intake, dietary quality, or food weight did not materially change our findings (Supplementary Table 3). Using absolute UPPs in grams as the main exposure showed similar results (Supplementary Table 4). Results were similar for HCC (Supplementary Table 5). The results were generally consistent between sex, race, income, obesity, and diabetes groups (Supplementary Table 6) with all P _interaction ≥0.10.

Discussion

Based on a large prospective cohort study of ~73,000 predominantly low-income US adults with approximate 14 years’ follow-up, greater consumption of ultra-processed foods was associated with a higher risk of liver cancer, especially for ultra-processed grains and fried potatoes, processed protein foods, and mixed dishes. Participants in the upper tertile of ultra-processed foods intake had a 69% higher risk of liver cancer compared with those in the lower tertile. Our findings were generally similar regardless of subgroups, including sex, race, income, obesity, and diabetes.

Evidence is accumulating on associations between ultra-processed food products and cancer risk, though studies on liver cancer are still relatively limited. Recently, one study using the UK Biobank explored the associations of ultra-processed food consumption with incidence and mortality for 34 site-specific cancers and reported that ultra-processed food intake was not associated with incident liver cancer (n=157; HR _{per 10% increment}=1.03, 95% CI=0.89-1.20) [21]. Another cohort study using data from the European Prospective Investigation into Cancer and Nutrition study evaluated the associations between total ultra-processed foods and risk of cancer at 25 anatomical sites, including HCC [22]. Replacing 10% of ultra-processed food with an equal amount of minimally processed foods was associated with lower risk of HCC (n=215; HR=0.84, 95% CI=0.70-1.00), though the association was nonsignificant after Bonferroni correction. These studies have a relatively small sample size of liver cancer and did not further explore the associations between subclasses of ultra-processed foods and liver cancer risk. Our research presented novel findings indicating that the positive associations between ultra-processed foods and liver cancer in the current study population are mainly driven by processed grains and fried potatoes, processed protein foods, and mixed dishes. However, we cannot rule out that the insignificant associations of other two subclasses, sweets/snacks and condiments/fats/oils with liver cancer, which may be due to the narrow ranges of intake levels.

Previous studies combined ultra-processed foods and drinks together [21, 22]. In the current study, we did not find significant associations between total ultra-processed products and liver cancer risk. When using the absolute weight as a scale to measure the ultra-processed products intake, the drinks account for almost half of the total food weight from the UPPs (47.2%), which is much higher compared with previous studies (15%-30%) [12, 32]. Therefore, we provided separate analyses to better evaluate the effects of solid and liquid UPPs due to their difference in digestion and absorption [33]. We found that ultra-processed foods rather than drinks were associated with higher risk of liver cancer. The potential reason is largely unknown but could be due to the high energy density of solid foods. Notably, our previous study conducted among nearly 100,000 US postmenopausal women with a median of 21 years’ follow up, found that higher intake of sugar-sweetened beverages, the major type of ultra-processed drinks, was associated with elevated risk of liver cancer while nonsignificant positive association for artificially sweetened beverages [34]. In the present study involving both Black and White participants from predominantly low-income communities, we did not find a statistically positive association between ultra-processed drinks (including sugar-sweetened beverages and artificially sweetened beverages) and liver cancer risk. The discrepancy remains poorly understood, though it may be due to differences in study populations, potential interactions with other variables, duration of follow-up, and confounding structures, such as prevalent diabetes. Further studies in diverse populations are needed to examine the associations.

The potential link between ultra-processed food intake and liver cancer is biologically explainable. First, several studies showed that ultra-processed foods usually have a lower nutritional quality compared with unprocessed foods, with lower fiber and vitamins and higher content of saturated fat, energy density, added sugar, and salt [6, 35]. Previous studies have shown that diets with higher quality scores and low inflammatory potential were associated with lower risk of liver cancer [36, 37]. Second, potential carcinogenic components like certain additives (emulsifiers, sodium nitrate, and artificial sweeteners) or food contaminants like trans-fat or acrylamide have been found in UPPs [38]. For example, a French study, the NutriNet-Santé study, found that artificial sweeteners, especially aspartame and acesulfame-K, were positively associated with a higher risk of overall, breast, and obesity-related cancers [39]. Third, considering that ultra-processed foods generally have a high glycemic response and energy density, the ultra-processed foods-liver cancer risk may be mediated by overweight and obesity due to close relationships between them [40].

Our study has several strengths. Previous studies evaluating associations between total ultra-processed foods and cancer risk are conducted primarily in middle to high-income populations. The current study involved participants primarily from low-income communities, thereby broadening the scope of evidence. Ascertainment of liver cancer in this cohort was relatively complete via linkage with cancer registries in each of the 12 recruitment states and with the National Death Index. Additionally, we provided analyses for ultra-processed foods and drinks separately and delved deeper into the evaluation of the ultra-processed foods subclasses.

It’s important to acknowledge several limitations when interpreting our findings. We cannot rule out the misclassification of Nova UPP groups using data from FFQ since FFQ is not a perfect tool to capture the degree of food processing, especially with relatively limited food items. Therefore, further research is needed to confirm our findings. Additionally, since people with a better dietary quality by limiting UPP intake also had other health-related behaviors, we cannot completely rule out residual confounding such as undiagnosed diabetes and its medications, though we carefully controlled for potential confounders. Due to the observational design of our study, causal inference cannot be made.

Conclusion

In a prospective study of predominantly low-income Black and White Americans living in the southern US, we found that consumption of ultra-processed foods was associated with a higher risk of liver cancer. Further studies are needed to confirm our findings.

What is already known on this topic?

1. Evidence is accumulating on the detrimental effects of ultra-processed foods and drinks on medical conditions such as obesity and diabetes.

2. The association of ultra-processed foods and drinks with liver cancer remains poorly understood.

What this study adds?

1. Higher ultra-processed foods were positively associated with liver cancer risk in a cohort of predominantly low-income Southern US adults.

2. The positive associations were mainly driven by ultra-processed grains and fried potatoes, processed protein foods, and mixed dishes.

3. The positive associations were generally consistent between Black and White participants.

Abbreviations:

BMI

body mass index

CI

confidence interval

FFQ

food frequency questionnaire

HCC

hepatocellular carcinoma

HR

hazard ratio

SCCS

the Southern Community Cohort Study

UPP

ultra-processed products

Data Availability:

Data described in the manuscript, code book, and analytic code will be made available upon request pending application to and approval by the Southern Community Cohort Study.