Associations between dietary protein intake and liver cancer risk: findings from two population-based cohorts in China Authors: Qiu-Ming Shen, Zhuo-Ying Li, Dan-Ni Yang, Jia-Yi Tuo, Da-Ke Liu, Jie Fang, Yu-Ting Tan, Hong-Lan Li, Yong-Bing Xiang Published: 2025 ABSTRACT AbstractBackgroundThe association between dietary protein intake and liver cancer risk has rarely been reported. The aim of this study was to explore whether the intake of dietary protein from various sources is differentially associated with liver cancer risk in the Chinese population.MethodsThis large-scale prospective cohort study collected diet information by validated food frequency questionnaires at baseline in two population-based cohorts in Shanghai, China. Dietary protein intake was calculated using the China Food Composition Table. Cox proportional hazards regression models were utilized to estimate the hazard ratio and 95% confidence interval.ResultsA total of 60,915 men and 72,655 women were included in the analysis. By the end of 2020, 594 men and 335 women had been identified as incident cases of liver cancer during median follow-up time of 15.28 years and 20.01 years, respectively. In both men and women, total protein intake and major dietary protein sources showed null associations with the risk of liver cancer in categorical analysis based on quartiles, continuous analysis with increments per standard deviation, and trend analysis.ConclusionsThere was no evidence of a prospective association between total protein intake or any of the major dietary protein sources explored and the risk of liver cancer. Further validation through other large-scale cohort studies is needed in the future.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12889-025-25685-7.Keywords: Liver cancer, Diet, Protein, Nutrition, Cohort study Introduction China was among the countries with the highest burden of liver cancer. In 2022, there were 367,657 new cases, accounting for 42.5% of the global total [1]. The age-standardized incidence rate in China was estimated to be 15.0/100,000, substantially exceeding the world average level [2]. Diet and nutrition are important and modifiable influencing factors in cancer prevention. According to the Third Expert Report from World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) Continuous Update Project, dietary factors including alcoholic drinks and aflatoxin-contaminated foods are associated with an increased risk of liver cancer, whereas coffee consumption shows the opposite. Fish intake has also been identified as a potential dietary factor that may reduce the risk of liver cancer [3]. Compared with specific foods, the role of nutrients in liver cancer risk has less extensively investigated. Protein, as an essential macronutrient required at all stages of life, undergoes critical metabolism in the liver [4]. Although previous epidemiological studies have examined the association between dietary protein intake and the risk of cancers such as esophageal cancer [5], colorectal cancer [6], prostate cancer [7] and ovarian cancer [8], its association with liver cancer risk remains relatively underexplored. In laboratory animals, the effects of protein intake on the risk of liver cancer were diverse, depending on different dietary sources. For example, total protein intake was observed to exacerbate chemically induced preneoplastic hepatic lesions [9], while soy protein may lead to a diminishment of tumor promotion and related hepatic injury and inflammation [10]. In humans, evidence linking dietary protein intake to liver cancer risk is scarce, and the association remains poorly defined due to limited data. Meanwhile, existing observational studies on dietary protein and cancer risk have predominantly focused on total protein intake [11, 12], with some examining the effects of animal or plant protein [13, 14]. Evidence without further considering protein from specific sources may be simplistic and insufficient. Therefore, a thorough investigation regarding the associations between major dietary protein sources and liver cancer risk is warranted. In this study, we hypothesized that the intake of dietary protein from different sources would be differentially associated with the risk of liver cancer. Utilizing data from two population-based prospective cohorts in Shanghai, China, we investigated the associations between major dietary protein sources with liver cancer risk among Chinese males and females. The findings are expected to enhance our understanding of the dietary factors involved in the etiology of primary liver cancer. Methods Study population This study was based on two ongoing, population-based prospective cohorts, the Shanghai Men’s Health Study (SMHS) and Shanghai Women’s Health Study (SWHS), which recruited 61,469 men (January 2002 to June 2006) aged 40–74 years and 74,940 women (December 1996 to May 2000) aged 40–70 years in urban Shanghai. Details about the study design had been described previously [15, 16]. All participants have completed a structured questionnaire to obtain baseline information about demographic characteristics, lifestyle factors, dietary habits and medical history. The information about anthropometric measurements was also collected at baseline. The cohort protocol has been approved by the Institutional Review Boards of the Shanghai Cancer Institute and Vanderbilt University. Informed consent was obtained from all subjects involved in the study. The current study has been approved by the Renji Hospital Ethics Committee of Shanghai Jiao Tong University School of Medicine (KY2019-197). In this study, we excluded participants who met the following criteria: (1) cancer at baseline; (2) loss to follow-up shortly after the baseline survey; (3) diagnosis of cancer in situ during follow-up; (4) death from cancer with no cancer type or diagnosis date; (5) a cancer diagnosis that could not be confirmed; (6) reported extreme values for total calorie intake (men: <800 or >4,000 kcal/day, women: <500 or >3,500 kcal/day) [17]. Finally, a total of 60,915 men and 72,665 women were included (Fig. 1). Diagram of participant selection in a prospective cohort study of 60,915 men and 72,655 women Baseline data collection We used a structured questionnaire to collect information about age at interview, education level, income level, menopausal status (women only), smoking status, alcohol use, family history of liver cancer and medical history of type 2 diabetes, chronic hepatitis and cholelithiasis. Data on cirrhosis was collected in males only, as this variable was not included in the questionnaire administered to the female cohort. Smoking was defined as ever smoking at least one cigarette per day for more than six months continuously. Alcohol use was defined as ever consuming alcohol at least three times per week for more than six months continuously. Anthropometric measurements, including weight and height, were also collected during the baseline survey. Body mass index (BMI) was calculated as weight (in kilograms) divided by the square of height (in meters) and was categorized into four groups as underweight (< 18.5 kg/m2), normal weight (18.5 ~ 23.9 kg/m2), overweight (24.0 ~ 27.9 kg/m2), and obesity (≥ 28 kg/m2) by the Chinese classification [18]. Semi-quantitative physical activity questionnaire (PAQ) was used to evaluate participants’ physical activity in adulthood. We calculated the total physical activity by multiplying the MET value for each type of activity by the hours spent on that activity per week, and then summing the MET∙h/week values across all types of physical activity. Our previous study demonstrated that PAQ is a reproducible and valid tool for assessing physical activity [19, 20]. The correlation coefficients for the test-retest reliability of total energy expenditure derived from the PAQ were 0.68 and 0.70 in males and females, respectively. Additionally, the correlation coefficients for total energy expenditure based on the baseline PAQ, retest PAQ, and the average of the 7-day PAQ, ranged from 0.62 to 0.73 in males and 0.62 to 0.80 and in females. Semi-quantitative food-frequency questionnaires (FFQ) were used to assess usual food intake at baseline. FFQs used for the SMHS and SWHS included 81 and 71 food items, covering approximately 88.8% and 90.0% of foods commonly consumed in urban Shanghai at enrolment, respectively. Each participant was asked how frequently they consumed the food or food groups over the preceding 12 month, followed by a question about the average amount of consumption in liangs (50 g/liang) per unit of time. The FFQ can reliably and accurately measure usual intake of major nutrients and food groups, with the correlation coefficients for test-retest of major food groups ranging from 0.39 to 0.64 in men and 0.37 to 0.66 in women, and coefficients for the retest FFQ and the average of 24-hour dietary recalls ranging from 0.35 to 0.72 in males and 0.41 to 0.66 in women [21, 22]. In men, plasma samples collected at enrollment were assessed for HBsAg levels using the Architect HBsAg QT assay (Abbott Diagnostics). Further details regarding the quantification of HBsAg levels can be found in our previous publication [23]. According to the manufacturer’s guideline, an HBsAg titer below 0.05 IU/mL was defined as seronegativity. Measurement of dietary protein intake We adopted the China Food Composition Table [24] to calculate the amount of protein in each food. The amount of protein in a certain type of food (N) was formulated as. where F is the dietary intake of a certain type of food for one day (grams), P is the amount of edible parts per 100 g of food; A is the amount of protein per 100 g of edible parts. Due to minimal intake, protein from processed meat and salted fish were excluded from analysis. Major dietary protein sources investigated in this study included protein from red meat, protein from poultry, protein from fish and shrimp, protein from egg, protein from dairy, protein from legumes and products, protein from nuts and protein from cereal and their products. Food items used to classify major protein sources were shown in Table S1. We analyzed absolute protein intake because our primary hypothesis concerned the health implications of the actual consumption level, which is directly relevant for public health guidance. Liver cancer ascertainment The primary outcome of this study was incident liver cancer. Details about the follow-up of two cohorts can be found in previous articles [15, 16]. In brief, since the baseline survey, participants have been followed up every 3–4 years through in-person interviews to update their health status. If there was a new case of liver cancer during the follow-up, the staff will retrieve the medical records from the respective hospitals. Then the clinician and pathologist will review the diagnosis evidence according to the pathological examination, imaging examination and medical history to confirm the diagnosis. The records have also been annually linked with databases of the Shanghai Cancer Registry and Shanghai Resident Registry. Cases of primary liver cancer were identified using the International Classification of Diseases, 9th version (ICD-9) diagnosis code 155. Statistical analyses To describe the distribution of baseline characteristics, medians with interquartile ranges were calculated for continuous variables and counts with proportions were presented for categorical variables. Comparisons according to different levels of quartiles of total protein intake were performed using the Kruskal-Wallis test for continuous variables and χ2 test for categorical variables. Comparisons of baseline characteristics among liver cancer cases and non-cases were performed using the Mann-Whitney U test for continuous variables and χ2 test for categorical variables. Given low proportion of missing data, missing values were imputed using medians for continuous variables and modes for categorical variables. The distributions of dietary intakes of protein were described as median with interquartile ranges and were compared using the Mann-Whitney U test between liver cancer cases and non-cases. Hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations of total and specific-source protein consumption with liver cancer risk were estimated using Cox proportional hazards regression models. The follow-up year between the date of baseline FFQ and the diagnosis of liver cancer or censoring (death, loss to follow up or the last follow-up date on 31 December 2020) was used as a time scale in all Cox regression models. Proportional hazards assumption for dietary protein intake was tested using Schoenfeld residuals, and no evidence of violation of these assumptions has been detected. HRs (95%CIs) of both quartile categories and per standard deviation (SD) increment of total and sub-group protein intakes were estimated, except for protein from dairy. We performed tests for linear trend by entering the median value of each category as a continuous variable in the models. Protein from dairy was categorized into three groups based on the distribution of data. One group representing participants with no habitual intake of dairy products (protein intake from dairy = 0), and the remaining two groups representing participants consuming dairy products, classified by the median of protein intake from dairy. The multivariable Cox proportional hazards regression model was considered for confounding adjustment of age at interview (continuous), education level (elementary school and lower/middle school/high school/high profession, college or higher), income level (low/low to middle/middle to high/high), BMI (underweight/normal weight/overweight/obesity), menopausal status (females only) (yes/no), smoking status (yes/no), alcohol use (yes/no), physical activity (MET-hours/week, quartile), total energy intake (kcal/day, quartile), family history of liver cancer (yes/no) and medical history of type 2 diabetes (yes/no), chronic hepatitis (yes/no), cholelithiasis (yes/no) and cirrhosis (males only) (yes/no). Potential effect modification by BMI (< 24 or ≥ 24 kg/m2), menopausal status (yes/no) (only in females) and medical history of chronic hepatitis (yes/no) were investigated by adding the interaction terms to the Cox models. Significant interactions were further explored through stratified analyses, presenting the results for each subgroup accordingly. To test the robustness of our main results, sensitivity analysis has been performed by (1) excluding participants with less than 2 years of follow-up to minimize potential bias from reverse causation; (2) replacing binary variables for smoking status and alcohol use with continuous variables of the intensity of alcohol consumption (ethanol intake per day) and the cumulative amount of cigarette smoking (pack-year) for a more nuanced association; (3) substituting self-reported history of chronic hepatitis with HBsAg test results to enhance measurement accuracy; (4) adjusting protein intake for total energy intake by nutrient density method and standardized to 2000 kcal [25]. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC, USA). All p-values were two-sided with statistical significance set at P < 0.05. Results Baseline characteristics A total of 60,915 male and 72,665 female participants were included in the analysis. During 930,580.50 and 1,514,170.16 person-years of follow-up (15.28 years and 20.01 years on average), 594 men and 335 women were identified as incident cases of liver cancer in our two cohorts. As shown in Table 1, there existed significant differences in age at interview, education level, income level, BMI, physical activity, total energy intake, smoking, alcohol use, medical history of type 2 diabetes, cholelithiasis and chronic hepatitis across the quartiles of total protein intake in men. In addition to the aforementioned characteristics, menopausal status also differed between these groups in women (Table 2). The results of participant characteristics between liver cancer cases and non-cases can be found in Table S2. Table 1Baseline characteristics of male participants according to quartiles of total protein intakeCharacteristicsTotal protein intake (g)P valuebQ1≤ 55.0Q255.1–66.1Q366.2–79.0Q4≥ 79.1Age (years), median (IQR)55.0 (19.6)53.4 (16.4)52.7 (14.7)51.9 (12.8) < 0.0001 Education, N(%) < 0.0001 elementary school or lower1704 (11.2)939 (6.2)749 (4.9)608 (4.0) middle school5234 (34.4)5068 (33.3)4958 (32.6)4880 (32.1) high school5250 (34.5)5587 (36.7)5705 (37.5)5992 (39.4) high profession, college or higher3041 (20.0)3635 (23.9)3817 (25.1)3748 (24.6)Income level, N(%)a < 0.0001 low2093 (13.7)1826 (12.0)1765 (11.6)1930 (12.7) low to middle6925 (45.5)6426 (42.2)6416 (42.1)6233 (40.9) middle to high5000 (32.8)5517 (36.2)5492 (36.1)5376 (35.3) high1211 (8.0)1460 (9.6)1556 (10.2)1689 (11.1)BMI (kg/m2), median (IQR)23.3 (4.2)23.6 (4.0)23.7 (3.9)24.0 (3.9) < 0.0001 Physical activity (MET/week), median (IQR)50.1 (43.5)52.7 (42.6)55.3 (43.1)58.5 (46.5) < 0.0001 Total energy intake (kcal/day), median (IQR)1422.9 (296.7)1740.7 (293.1)1997.6 (343.9)2408.1 (508.4) < 0.0001 Smoking, N(%) < 0.0001 ever10,811 (71.0)10,460 (68.7)10,455 (68.7)10,659 (70.0) never4418 (29.0)4769 (31.3)4774 (31.4)4569 (30.0)Alcohol use, N(%) < 0.0001 ever5476 (36.0)4896 (32.2)4839 (31.8)5259 (34.5) never9753 (64.0)10,333 (67.9)10,390 (68.2)9969 (65.5)Medical history of type 2 diabetes, N(%)1092 (7.2)963 (6.3)884 (5.8)872 (5.7) < 0.0001 cholelithiasis, N(%)1281 (8.4)1208 (7.9)1091 (7.2)999 (6.6) < 0.0001 chronic hepatitis, N(%)426 (2.8)342 (2.3)388 (2.6)373 (2.5) 0.0203 cirrhosis, N(%)95 (0.6)94 (0.6)67 (0.4)93 (0.6)0.0968Family history of liver cancer, N(%)545 (3.6)527 (3.5)580 (3.8)533 (3.5)0.3617a. Low: male personal income < 500 RMB per month; low to middle: [500, 1000) RMB per month; middle to high: [1000, 2000) RMB per month; high: ≥2000 RMB per monthb. P values were obtained by Kruskal-Wallis test or χ2 testIQR interquartile range Baseline characteristics of male participants according to quartiles of total protein intake a. Low: male personal income < 500 RMB per month; low to middle: [500, 1000) RMB per month; middle to high: [1000, 2000) RMB per month; high: ≥2000 RMB per month b. P values were obtained by Kruskal-Wallis test or χ2 test IQR interquartile range Table 2Baseline characteristics of female participants according to quartiles of total protein intakeCharacteristicsTotal protein intake (g)P valuebQ1≤ 47.7Q247.8–57.6Q357.7–69.2Q4≥ 69.3Age (years), median (IQR)52.9 (19.2)50.4 (16.7)49.5 (14.6)49.2 (13.8) < 0.0001 Education, N(%) < 0.0001 elementary school or lower5886 (32.4)3945 (21.7)3046 (16.8)2626 (14.5) middle school6289 (34.6)6704 (36.9)6900 (38.0)7170 (39.5) high school4259 (23.4)5090 (28.0)5367 (29.5)5578 (30.7) high profession, college or higher1733 (9.5)2427 (13.4)2853 (15.7)2792 (15.4)Income level, N(%)a < 0.0001 low3594 (19.8)2815 (15.5)2581 (14.2)2703 (14.9) low to middle7150 (39.4)6924 (38.1)6833 (37.6)6889 (37.9) middle to high4660 (25.7)5203 (28.6)5363 (29.5)5195 (28.6) high2763 (15.2)3224 (17.8)3389 (18.7)3379 (18.6)BMI (kg/m2), median (IQR)23.7 (4.6)23.6 (4.5)23.7 (4.4)23.9 (4.4) < 0.0001 Physical activity (MET/week), median (IQR)97.5 (55.7)98.9 (56.1)100.1 (57.2)104.3 (58.1) < 0.0001 Total energy intake (kcal/day), median (IQR)1278.4 (251.5)1534.8 (221.7)1744.6 (257.3)2078.4 (418.3) < 0.0001 Smoking, N(%) < 0.0001 ever740 (4.1)451 (2.5)377 (2.1)444 (2.4) never17,427 (95.9)17,715 (97.5)17,789 (97.9)17,722 (97.6)Alcohol use, N(%) < 0.0001 ever397 (2.2)380 (2.1)352 (1.9)501 (2.8) never17,770 (97.8)17,786 (97.9)17,814 (98.1)17,665 (97.2)Menopausal status, N(%) < 0.0001 yes10,149 (55.9)8975 (49.4)8311 (45.8)8065 (44.4) no8018 (44.1)9191 (50.6)9855 (54.3)10,101 (55.6)Medical history of type 2 diabetes, N(%)815 (4.5)788 (4.3)685 (3.8)839 (4.6) 0.0004 cholelithiasis, N(%)2414 (13.3)2031 (11.2)1862 (10.3)1779 (9.8) < 0.0001 chronic hepatitis, N(%)447 (2.5)468 (2.6)466 (2.6)487 (2.7)0.6226Family history of liver cancer, N(%)571 (3.1)616 (3.4)607 (3.3)590 (3.3)0.5614a. Low: female family income < 10000 RMB per year; low to middle: [10000, 20000) RMB per year; middle to high: [20000, 30000) RMB per year; high: ≥30000 RMB per yearb. P values were obtained by Kruskal-Wallis test or χ2 testIQR interquartile range Baseline characteristics of female participants according to quartiles of total protein intake a. Low: female family income < 10000 RMB per year; low to middle: [10000, 20000) RMB per year; middle to high: [20000, 30000) RMB per year; high: ≥30000 RMB per year b. P values were obtained by Kruskal-Wallis test or χ2 test IQR interquartile range Distributions of dietary intake of protein Table 3 presented the distributions of dietary intakes of protein in liver cancer cases and non-cases. In men, no significant differences in intakes of total protein or major dietary protein sources were observed between the two groups, except for protein from dairy (P = 0.0142). In women, participants who had liver cancer had lower intakes of total protein (P = 0.0079) and protein from red meat (P = 0.0136), poultry (P = 0.0012), fish and shrimp (P < 0.0001) and egg (P = 0.0002). Table 3Medians (interquartile ranges) of baseline dietary protein intake of liver cancer cases and non-casesProtein (g)MalesFemalesNon-casesLiver cancer casesP valueaNon-casesLiver cancer casesP valueaTotal protein66.24 (24.07)65.45 (24.82)0.452457.72 (21.42)56.35 (19.85)0.0079Major protein sources Red meat7.64 (6.76)7.49 (6.94)0.36886.07 (5.68)5.52 (5.34)0.0136 Poultry1.27 (2.00)1.26 (2.02)0.08341.27 (2.01)0.90 (1.78)0.0012 Fish and shrimp4.24 (5.09)4.04 (5.03)0.58444.17 (4.98)3.57 (3.78)< 0.0001 Egg2.42 (3.50)2.80 (3.50)0.67242.80 (3.50)2.10 (4.09)0.0002 Dairy1.66 (5.80)2.49 (5.80)0.01420.80 (5.80)0.53 (5.80)0.5963 Legumes and their products8.70 (7.55)8.46 (7.96)0.82077.06 (6.92)7.46 (7.62)0.8841 Nuts0.08 (0.12)0.08 (0.12)0.66920.04 (0.06)0.04 (0.08)0.2159 Cereal and their products28.43 (11.12)28.05 (11.26)0.197421.94 (6.46)23.31 (7.14)0.1126a. P values were obtained by Mann-Whitney U test Medians (interquartile ranges) of baseline dietary protein intake of liver cancer cases and non-cases a. P values were obtained by Mann-Whitney U test Dietary protein intakes and liver cancer risk After adjustment for potential confounders, intakes of total protein and protein from major dietary protein sources were not associated with the risk of liver cancer in men (Table 4) and women (Table 5), except for protein from dairy in men [HRhigher than median consumer vs. non−consumer(95%CI) = 1.234 (1.019, 1.493)]. The majority of these findings remained unaltered in the sensitivity analyses when participants with less than 2 years of follow-up were excluded (Table S3), or when binary variables for smoking status and alcohol use were replaced with more detailed measurement (Table S4), or when self-reported history of chronic hepatitis were substituted with HBsAg test results (Table S5) or when protein intake were adjusted for energy intake by nutrient density method (Table S6). In men, the significant association between protein from dairy and liver cancer risk became null in three of the four sensitivity analyses, which indicated that the association may not be robust enough and could in fact be null. Table 4HRs and 95%CIs for risk of liver cancer according to dietary protein intake in menaProtein (g)Q1Q2Q3Q4P trendPer SD incrementNo. of casesHR (95% CI)No. of casesHR (95% CI)No. of casesHR (95% CI)No. of casesHR (95% CI)HR (95% CI)bTotal protein1591 (ref.)1460.983 (0.749, 1.290)1380.912 (0.659, 1.263)1511.024 (0.701, 1.494)0.86120.980 (0.858, 1.119)Major protein sources Red meat1631 (ref.)1380.937 (0.745, 1.180)1461.005 (0.796, 1.269)1471.054 (0.823, 1.350)0.55001.007 (0.922, 1.101) Poultry1561 (ref.)1411.012 (0.805, 1.273)1630.955 (0.764, 1.194)1340.945 (0.744, 1.202)0.57540.944 (0.862, 1.034) Fish and shrimp1471 (ref.)1591.181 (0.941, 1.482)1411.068 (0.843, 1.354)1471.170 (0.917, 1.491)0.37561.003 (0.920, 1.093) Egg1841 (ref.)1040.952 (0.747, 1.214)850.881 (0.678, 1.143)2211.087 (0.885, 1.334)0.33111.036 (0.959, 1.121) Dairyc2121 (ref.)1280.965 (0.773, 1.205)2541.234 (1.019, 1.493)-- Legumes and products1491 (ref.)1591.094 (0.873, 1.370)1320.917 (0.722, 1.165)1540.968 (0.763, 1.228)0.51840.964 (0.886, 1.050) Nuts1441 (ref.)2190.945 (0.764, 1.169)941.161 (0.893, 1.509)1370.879 (0.693, 1.115)0.44720.998 (0.920, 1.083) Cereal and products1651 (ref.)1390.853 (0.668, 1.089)1460.961 (0.728, 1.270)1440.893 (0.640, 1.246)0.64480.950 (0.841, 1.073)a. The models were adjusted for age at interview, education level, income level, BMI, smoking status, alcohol use, physical activity, total energy intake, family history of liver cancer and medical history of type 2 diabetes, chronic hepatitis, cholelithiasis and cirrhosisb. Risk estimates were for the associations between per standard deviation increase of dietary protein intake and liver cancerc. Protein from dairy was categorized into 3 groups because of the distribution of data. Participants who did not habitually consume dairy products (protein from dairy equals to 0) were used as the reference group. For participants consuming dairy products, they were classified into two groups based on the median of protein intake from dairyHR hazard ratios, CI confidence interval, Q quartile, SD standard deviation HRs and 95%CIs for risk of liver cancer according to dietary protein intake in mena a. The models were adjusted for age at interview, education level, income level, BMI, smoking status, alcohol use, physical activity, total energy intake, family history of liver cancer and medical history of type 2 diabetes, chronic hepatitis, cholelithiasis and cirrhosis b. Risk estimates were for the associations between per standard deviation increase of dietary protein intake and liver cancer c. Protein from dairy was categorized into 3 groups because of the distribution of data. Participants who did not habitually consume dairy products (protein from dairy equals to 0) were used as the reference group. For participants consuming dairy products, they were classified into two groups based on the median of protein intake from dairy HR hazard ratios, CI confidence interval, Q quartile, SD standard deviation Table 5HRs and 95%CIs for risk of liver cancer according to dietary protein intake in womenaProtein (g)Q1Q2Q3Q4P trendPer SD incrementNo. of casesHR (95% CI)No. of casesHR (95% CI)No. of casesHR (95% CI)No. of casesHR (95% CI)HR (95% CI)bTotal protein961 (ref.)841.031 (0.724, 1.470)881.183 (0.776, 1.804)671.019 (0.611, 1.700)0.93380.939 (0.778, 1.133)Major protein sources Red meat981 (ref.)870.999 (0.746, 1.339)871.090 (0.806, 1.473)630.882 (0.624, 1.245)0.55300.996 (0.878, 1.130) Poultry1091 (ref.)880.943 (0.710, 1.253)700.777 (0.570, 1.059)680.904 (0.655, 1.248)0.47311.023 (0.909, 1.152) Fish and shrimp1031 (ref.)951.108 (0.833, 1.473)831.063 (0.785, 1.438)540.753 (0.530, 1.070)0.07390.929 (0.813, 1.061) Egg1431 (ref.)730.762 (0.571, 1.017)210.938 (0.589, 1.495)980.875 (0.664, 1.153)0.47770.989 (0.879, 1.113) Dairyc1561 (ref.)881.132 (0.865, 1.480)911.152 (0.875, 1.517)-- Legumes and products891 (ref.)700.799 (0.582, 1.097)870.946 (0.699, 1.281)890.928 (0.677, 1.272)0.96630.931 (0.824, 1.051) Nuts831 (ref.)400.986 (0.675, 1.440)1181.110 (0.836, 1.473)941.139 (0.841, 1.542)0.41581.031 (0.941, 1.128) Cereal and products791 (ref.)710.863 (0.621, 1.201)961.154 (0.828, 1.607)891.068 (0.724, 1.575)0.50401.053 (0.912, 1.216)a. The models were adjusted for age at interview, education level, income level, BMI, menopausal status, smoking status, alcohol use, physical activity, total energy intake, family history of liver cancer and medical history of type 2 diabetes, chronic hepatitis and cholelithiasisb. Risk estimates were for the associations between per standard deviation increase of dietary protein intake and liver cancerc. Protein from dairy was categorized into 3 groups because of the distribution of data. Participants who did not habitually consume dairy products (protein from dairy equals to 0) were used as the reference group. For participants consuming dairy products, they were classified into two groups based on the median of protein intake from dairyHR hazard ratios, CI confidence interval, Q quartile, SD standard deviation HRs and 95%CIs for risk of liver cancer according to dietary protein intake in womena a. The models were adjusted for age at interview, education level, income level, BMI, menopausal status, smoking status, alcohol use, physical activity, total energy intake, family history of liver cancer and medical history of type 2 diabetes, chronic hepatitis and cholelithiasis b. Risk estimates were for the associations between per standard deviation increase of dietary protein intake and liver cancer c. Protein from dairy was categorized into 3 groups because of the distribution of data. Participants who did not habitually consume dairy products (protein from dairy equals to 0) were used as the reference group. For participants consuming dairy products, they were classified into two groups based on the median of protein intake from dairy HR hazard ratios, CI confidence interval, Q quartile, SD standard deviation Effect modification Table S7 showed the interactions between BMI, menopausal status and medical history of chronic hepatitis and dietary protein intake on the association with liver cancer risk. In men, the association between protein from red meat and liver cancer risk may differ by the status of baseline medical history of chronic hepatitis (Pinteraction = 0.0168). In women, the association between total protein intake and liver cancer risk may vary depending on menopausal status (Pinteraction = 0.0488). Stratification analysis revealed no significant association between dietary protein from red meat and liver cancer risk in male participants with [HR (95%CI) = 1.106 (0.942,1.299)] and without chronic hepatitis [HR (95%CI) = 0.972 (0.876,1.079)]. In addition, no significant association was observed between total protein intake and liver cancer risk in female participants, both premenopausal [HR (95%CI) = 1.185 (0.859,1.634)] and postmenopausal [HR (95%CI) = 0.855 (0.682,1.073)] (Figure S1). Discussion In two large population-based prospective studies conducted in Shanghai, China, our hypothesis that the intake of dietary protein from various sources is differentially associated with the risk of liver cancer in the Chinese population was not supported. The data provided no evidence to indicate total protein intake or major dietary protein sources were associated with liver cancer risk in either men or women. Our investigation extended beyond total protein to examine specific dietary sources, driven by the hypothesis that different sources of protein may be associated with liver cancer risk through distinct, and potentially opposing, biological pathways. This hypothesis is grounded in the distinct properties of protein from different dietary sources. At the metabolic level, proteins from different sources exhibit different amino acid profiles and digestion kinetics. For instance, animal proteins are generally highly digestible and provide a balanced profile of essential amino acids [27]. This can lead to a more potent stimulation of insulin-like growth factor 1 (IGF-1) and hepatic amino acid sensing pathways (e.g., the mTOR signaling pathway), which are considered promoters of cell proliferation and potential drivers of carcinogenesis [28, 29]. In contrast, plant proteins are incomplete, digested more slowly, and may result in more moderate metabolic and hormonal responses [30]. Beyond direct host metabolism, the interaction between dietary protein and the gut microbiota represents another critical point. Gut microbiota compositions are differentially modulated by protein sources [31]. Even within the category of animal proteins, microbiota composition can be differentially impacted by the differences between meat, fish, dairy and soy protein diets, illustrating the uniqueness of each protein source [32, 33]. Furthermore, the systemic metabolic fate of ingested proteins also differs by sources, where animal proteins can be converted by the gut microbiota to in higher amounts of detrimental metabolites, including ammonia, amines, and hydrogen sulfide that contribute to negative host health consequences. Plant proteins, on the other hand, are less associated with lower relative production of these metabolic byproducts [34]. Results from previous meta-analyses have suggested that total dietary protein intake had no significant associations with risk for several cancers, including esophageal [5], colorectal [6], prostate [7, 26] and ovarian cancer [8]. Analyses by protein type further revealed no clear associations for either animal protein or vegetable protein with the risk for these cancers [5–8, 26]. Our study also indicated no significant associations between total protein intake, major dietary protein sources, and liver cancer risk, which were consistent with the results for other cancer sites. Regarding specific protein sources, the evidence remains nuanced. For instance, soy protein was found to be associated with a 35% lower risk for breast cancer [35], yet no association was found with overall cancer risk [36] or colorectal cancer risk [37] in recent meta-analysis and pooled analysis. A meta-analysis of dairy protein reported a significant positive association with prostate cancer risk, though the lower bound of 95%CI for the relative risk was close to 1 [26]. Substituting plant protein for red meat protein, rather than white meat or other animal proteins, appeared to reduce colorectal cancer risk, with the reduction in risk being specifically attributed to the substitution of protein from bread, cereal, and pasta for red meat protein [38]. Evidence for the association between specific sources of protein and liver cancer risk is limited. One prospective study examining the association between intake of protein from dairy and hepatocellular carcinoma found that per 10% increase of protein from dairy was associated with a 9% higher risk of hepatocellular carcinoma [39]. However, the result of our study did not find a similar robust association. It should also be noted that factors other than protein itself, such as contamination of dairy products by aflatoxin in animal feeds [40], which is known to be a major risk factor for liver cancer, may also contribute to an increased risk of liver cancer [3]. Future studies with high quality data from large cohorts are needed to confirm the findings. The current study has several methodological considerations regarding dietary assessment that warrant discussion. The food items collected were primarily based on foods commonly consumed in urban Shanghai at the time of enrollment. For example, regarding nuts specifically, our selection was limited to peanuts, which was a common kind of nut available to our study population in the late 20th to early 21 st century. This reliance on a single variety may constrain the interpretation of the association for protein from nut and liver cancer risk. In addition, some major protein sources were not classified further. For instance, “Dairy” was not stratified by fat content (e.g., high-fat vs. low-fat dairy products), and “Cereal and products” lacked information on whole-grain foods. Aggregating data across these categories could potentially obscure true associations. These factors may also contribute to the lack of association between total or major dietary protein intake and liver cancer risk. Our study is the first to examine the associations between dietary protein intake of major sources and liver cancer risk among Chinese men and women. It is a prospective design with large sample size and a long-term follow-up. Besides, we extended the sources of dietary protein from total protein to more specific sources of protein, so that more detailed evidence could be provided. It is worth noting that some potential limitations also existed in our study. First, all diet and lifestyle information were collected at baseline, and their changes during follow-up were not taken into consideration. This approach may not accurately capture long-term exposure patterns and is susceptible to measurement error, which may potentially lead to an attenuation of the observed associations and limit our ability to detect true relationships. Future longitudinal studies with repeated measurements are necessitated to assess the association between dynamic changes of dietary protein sources and liver cancer risk. Second, protein from processed meat and salted fish were excluded from the analyses because only the frequencies that participants consumed them were collected in the questionnaires. Since most participants consumed these foods in low frequencies (never or less than 5 times/year), their contributions to protein intake were small, and their impact on the outcome was likewise negligible. Third, although we have included various covariates in the regression models, residual confounding due to unmeasured confounding cannot be fully eliminated. In particular, certain covariates were not available for both sexes. For instance, medical history of cirrhosis was only collected in men, and its absence may result in residual confounding that differs by sex. Fourth, we did not carry out the analysis for histological subtypes because the information for the subtype data for liver cancer cases was incomplete. Future studies with more comprehensive histological data could explore these associations in greater detail. Finally, study participants were recruited from an urban population in China, which may limit the generalizability of the findings to other geographic or demographic settings. Further research would be needed to assess the applicability of these findings in other populations. Conclusions Our results showed that there were no associations between dietary intakes of total or major protein sources and liver cancer risk in Chinese men and women. Further research that investigating the potential biological mechanisms or large-scale population-based cohort studies with repeated assessments of nutrition is needed to validate these findings. Supplementary Information Supplementary Material 1. Acknowledgements We thank all participants and staff from the Shanghai Men and Women’s Health Studies for their contribution to this research. Abbreviations Body mass index Confidence interval Food-frequency questionnaires Hazard ratio Physical activity questionnaire Shanghai Men’s health study Shanghai Women’s health study Authors’ contributions Y-BX designed research and obtained funding; Q-MS, Z-YL, D-NY, J-YT, D-KL, J-F, Y-TT, H-LL and Y-BX conducted the study, Q-MS and Y-BX analyzed the data and interpreted the results; Q-MS and Y-BX wrote the first draft; All authors reviewed and approved the final version of the paper; Y-BX has primary responsibility for final content. Funding This work was supported by the National Key Project of Research and Development Program of China [2021YFC2500404], and the subcontracts from grant of the US National Institutes of Health [UM1 CA173640, UM1 CA182910] for the parent cohorts. All funders had no role in the design, analysis or writing of this article. Data availability The data will be available on request pending approval by the scientific committee of the relevant institutes. Researchers interested in accessing the raw data should contact the corresponding author by E-mail ([ybxiang@shsci.org](mail to: ybxiang@shsci.org)) for more details about the data availability process, including submission requirements and approval procedures. Declarations Ethics approval and consent to participate This study was carried out in accordance with Declaration of Helsinki. The cohort protocol has been approved by the Institutional Review Boards of the Shanghai Cancer Institute and Vanderbilt University. Informed consent was obtained from all subjects involved in the study. The current study has been approved by the Renji Hospital Ethics Committee of Shanghai Jiao Tong University School of Medicine (KY2019-197). Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The data will be available on request pending approval by the scientific committee of the relevant institutes. Researchers interested in accessing the raw data should contact the corresponding author by E-mail ([ybxiang@shsci.org](mail to: ybxiang@shsci.org)) for more details about the data availability process, including submission requirements and approval procedures.