Association between dietary nutrient composition and the incidence of cirrhosis or liver cancer in the united states population

Authors

  • George N. Ioannou,

    Corresponding author
    1. Division of Gastroenterology, Department of Medicine, Veterans Affairs Puget Sound Health Care System, Seattle, WA
    2. Research Enhancement Award Program, Veterans Affairs Puget Sound Health Care System, Seattle, WA
    3. Division of Gastroenterology, Department of Medicine, University of Washington, Seattle, WA
    • Veterans Affairs Puget Sound Health Care System, Gastroenterology, S-111-Gastro, 1660 S. Columbian Way, Seattle, WA 98108
    Search for more papers by this author
    • fax: (206) 764-2232

  • Olivia B. Morrow,

    1. Research Enhancement Award Program, Veterans Affairs Puget Sound Health Care System, Seattle, WA
    Search for more papers by this author
  • Marah L. Connole,

    1. Research Enhancement Award Program, Veterans Affairs Puget Sound Health Care System, Seattle, WA
    Search for more papers by this author
  • Sum P. Lee

    1. Division of Gastroenterology, Department of Medicine, Veterans Affairs Puget Sound Health Care System, Seattle, WA
    2. Research Enhancement Award Program, Veterans Affairs Puget Sound Health Care System, Seattle, WA
    3. Division of Gastroenterology, Department of Medicine, University of Washington, Seattle, WA
    Search for more papers by this author

  • Potential conflict of interest: Nothing to report.

Abstract

Little is known about the impact of dietary factors on the progression of liver disease. Our aim was to determine whether dietary intake was associated with the risk of cirrhosis-related or liver cancer–related death or hospitalization in the U.S. population. Participants included 9221 persons aged 25-74 years without evidence of cirrhosis at entry into the study or during the first 5 years of follow-up, who were subsequently followed for a mean of 13.3 years as part of the first National Health and Nutrition Examination Survey. Dietary intake was ascertained at baseline using a 24-hour dietary recall questionnaire. During follow-up, 123 of 9221 participants had a diagnosis of cirrhosis (n = 118) or liver cancer (n = 5) in hospitalization records or death certificates, including 36 who were diagnosed only on the basis of death certificates. Participants who reported a diet high in protein were at a higher risk of hospitalization or death due to cirrhosis or liver cancer (P = 0.001), whereas those who reported a diet high in carbohydrates were at a lower risk (P = 0.003), after adjusting for potential confounders (daily consumption of protein, carbohydrate, fat, tea or coffee, and alcohol, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and subscapular-to-triceps skinfold ratio). Although total fat consumption was not significantly associated with the risk of cirrhosis or liver cancer, cholesterol consumption was associated with higher risk (P = 0.007), whereas serum cholesterol level was not associated with risk of cirrhosis or liver cancer. Conclusion: Diet may be an important and potentially modifiable determinant of liver disease. (HEPATOLOGY 2009.)

Dietary factors are likely to be important determinants of the development of hepatic steatosis and its progression to steatohepatitis for the following reasons: (1) Dietary factors are important and probably causative risk factors for obesity, insulin resistance, and diabetes, which are the most important, known risk factors for hepatic steatosis. (2) Dietary lipid composition influences both the quantity and composition of lipids that are delivered to the liver and incorporated into hepatocyte lipid droplets. This is important because recent data derived primarily from animal models suggest that specific lipid molecules, such as saturated fatty acids, may promote lipotoxicity,1, 2 whereas others, such as n-3 polyunsaturated fatty acids (PUFAs), may actually have beneficial effects.3, 4 Therefore, it is possible that the quantity and composition of dietary lipid can either promote or protect against the development or progression of hepatic steatosis. (3) In rabbits5, 6 and mice7 a high cholesterol diet has been shown to induce profound steatosis, inflammation, and centrilobular fibrosis.

Hepatic steatosis is the defining feature of nonalcoholic fatty liver disease (NAFLD) but it is also a common and probably pathogenetic feature of hepatitis C virus (HCV) infection and alcoholic liver disease. Although HCV infection and alcohol consumption can cause hepatic steatosis by themselves, obesity-related steatosis also occurs in the setting of HCV infection and alcoholic liver disease.8 Therefore, if dietary composition affects the development or progression of hepatic steatosis, it is likely to play a part in the natural history of the three most important liver conditions in the U.S.: NAFLD, HCV infection, and alcoholic liver disease.

Dietary nutrients may also cause hepatic injury through pathways that do not involve the development or progression of hepatic steatosis. Carbohydrates, proteins, and lipids are all extensively metabolized in the liver and it is conceivable that they may influence the progression of chronic liver disease, either positively or negatively. In hepatitis B virus (HBV) transgenic mice, a diet low in animal protein was associated with decreased liver injury and decreased incidence of hepatocellular carcinoma.9, 10 In the presence of oxidative stress, dietary cholesterol may be oxidized in the liver to oxysterols, which can induce cell damage and malignant transformation and regulate signal transduction pathways.

Our aim was to investigate whether dietary nutrient composition was associated with the subsequent development of cirrhosis or liver cancer in a representative sample of the U.S. population.

Abbreviations

BMI, body mass index; HBV, hepatitis B virus; HCV, hepatitis C virus; ICD-9, International Classification of Diseases, 9th Revision; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis; NHANES, National Health and Nutrition Examination Survey; NHEFS, NHANES I Epidemiologic Follow-up Study; PUFA, polyunsaturated fatty acid.

Patients and Methods

Study Design.

Data were derived from the first National Health and Nutrition Examination Survey (NHANES I) performed by the National Center for Health Statistics. NHANES I is a cross-sectional study of a nationwide probability sample from the civilian, noninstitutionalized population of the coterminous United States conducted between 1971-1975.11 The survey included 14,407 participants aged 25-74 years who completed extensive dietary questionnaires and underwent physical examinations and laboratory investigations. The NHANES I Epidemiologic Follow-up Study (NHEFS)12 sought to locate these 14,407 individuals in the years 1982-1984, 1986, 1987, and 1992 and collected data on specific health conditions that they developed in the intervening period through personal interviews, hospitalization records, and death certificates. We merged together NHANES I and NHEFS to form a nationally representative cohort of 14,407 persons with ≈20 years of follow-up. We used the NHANES I-NHEFS cohort to determine whether dietary composition at baseline is associated with the subsequent incidence of death or hospitalization related to liver cancer or cirrhosis.

Study Population.

Among 14,407 NHANES I participants aged 25-74 years, 13,861 were successfully traced in at least one of the four follow-up occasions (that is 1982-1984, 1986, 1987, or 1992). In order to limit our analysis to incident cases of cirrhosis and liver cancer, we attempted to exclude participants who suffered from these conditions at the time of entry into the study. Specifically, we excluded participants who reported at baseline ever being told by a physician that they had jaundice (n = 886), hepatitis (n = 47), or a malignant tumor (n = 47), who had hepatomegaly or splenomegaly at baseline examination (n = 237), or whose level of serum albumin was less than 3 g/dL (n = 10). Serum bilirubin levels and platelet counts, which may be abnormal in advanced cirrhosis, were available only in a small minority of participants and therefore could not be used to identify participants with possible cirrhosis. Because cirrhosis and liver cancer may be present for a long time before they are clinically diagnosed, we also excluded participants who were diagnosed with these conditions within the first 5 years of follow-up, or who had less than 5 years of follow-up (n = 687). We excluded 46 participants with missing values in any of the following potential confounding variables included in our models: age, gender, race, alcohol consumption, body mass index (BMI), subscapular-to-triceps skinfold ratio, geographical location, and educational attainment. The 24-hour dietary recall questionnaire, as well as questions on diabetes and coffee consumption, were administered only to the subsample of NHANES participants recruited at locations 1-65. This subsample was designed so that it would still constitute a random sample of the U.S. population. An additional 2,680 NHANES I participants did not complete a 24-hour dietary recall questionnaire and did not answer questions on diabetes or coffee consumption because they were not recruited at locations 1-65, leaving 9221 participants in the current analyses.

Ascertainment of Baseline Dietary Intake.

A dietary interview was conducted by professional staff at entry into the study in 1971-1975, including a 24-hour dietary recall questionnaire. The 24-hour recall provides information on specific food items and their quantities ingested for all regular meals and between-meal foods or snacks consumed from midnight to midnight on the day preceding the interview. A Nutrient Composition Data Bank was then used to calculate the grams of protein, carbohydrate, and fat consumed during the 24-hour period. In addition, the 24-hour intake of the following specific fats was calculated: cholesterol, saturated fatty acids, oleic acid (a monounsaturated fatty acid), and linoleic acid (an n-6 PUFA). The estimated dietary intake of other lipids of interest such as n-3 PUFAs or n-6 PUFAs other than linoleic acid was not provided by the NHANES I investigators.

Hospitalizations and Deaths Due to Liver Cirrhosis or Liver Cancer During Follow-up.

Deaths and hospitalizations due to liver cirrhosis or liver cancer that occurred during follow-up were ascertained from hospitalization records and death certificates. Specially trained NHEFS personnel abstracted the diagnoses in these documents into International Classification of Diseases, 9th Revision (ICD-9) diagnosis codes. We used the ICD-9 code 155.0 for liver cancer and the following ICD-9 codes for cirrhosis: alcoholic cirrhosis 571.2; cirrhosis without mention of alcohol 571.5; pigmentary cirrhosis 275.0; esophageal varices 456.0-456.2; hepatic coma 572.2; portal hypertension 572.3; and hepatorenal syndrome 572.4. Esophageal varices, hepatic coma, portal hypertension, and hepatorenal syndrome have been included in the diagnosis of liver cirrhosis because the overwhelming majority of these conditions in the U.S. are the result of liver cirrhosis. If acute necrosis of the liver (ICD-9 code 570.0) was diagnosed together with hepatic coma or hepatorenal syndrome, then the person was considered not to have cirrhosis. Other complications of cirrhosis such as ascites or peritonitis have not been included as evidence of cirrhosis because they are commonly caused by other conditions. The date of the first hospital admission for each condition was used as the date of incidence. For subjects who had a death certificate recording one of these conditions but did not have a hospitalization for any of them, the date of death was used as the date of incidence. Henceforth we use the term “incidence of cirrhosis or liver cancer” for brevity, but what we have ascertained is the incidence of hospitalizations and deaths related to cirrhosis or liver cancer.

Ascertainment of Potential Confounding Variables.

We considered the following baseline characteristics, which may be associated with both dietary intake and development of cirrhosis or liver cancer as potential confounders: age; BMI; subscapular to triceps skinfold ratio (a measure of central as opposed to peripheral subcutaneous fat); race, categorized as white (n = 7644) and nonwhite (n = 1638, of whom 1545 were black and only 93 of other race, too small a number for additional racial categories); gender; alcohol consumption over the previous 12 months, categorized as none, >0 to <1 drink/day, 1 to <2 drinks/day, and ≥2 drinks/day; coffee or tea consumption categorized into <1, 1-2, and >2 cups/day; educational attainment, categorized as completion of high school or not; self-reported diabetes mellitus; and geographical area of residence in the U.S., categorized into Northeast, Midwest, South, and West.

Viral hepatitis B and C testing was not available in 1971-1974 when the NHANES I participants were recruited. We wanted to ensure that viral hepatitis C, an important cause of cirrhosis in the U.S., was not associated with the dietary consumption of carbohydrates, proteins, and fats (including cholesterol, saturated fatty acids, oleic acid, and linoleic acid) in order to exclude the possibility that viral hepatitis was an important source of unmeasured confounding in our NHANES I cohort. To do this, we used data from NHANES III, a cross-sectional study conducted between 1988-1994 that included measurements of viral hepatitis serologies.

Statistical Analysis.

The Cox proportional-hazards model13 was used to determine the hazard ratio comparing persons within different quartiles of a dietary nutrient intake with respect to the risk for cirrhosis or liver cancer with or without adjusting for the potential confounders listed above. For each macronutrient we present results from two statistical models, corresponding to “percent energy” and “absolute energy.”14 In “percent energy” models each macronutrient is modeled as the percent of total energy that it provides calculated as the number of calories from each macronutrient divided by the total number of calories consumed (for alcohol, models will use categorized drinks per day); total energy is included as a linear covariate, with or without adjusting for the potential confounders listed above. “Percent energy” models can be interpreted as the effect of substituting energy from each specific macronutrient for other macronutrients, thus keeping the total dietary energy intake constant. In “absolute energy” models, each macronutrient is modeled as the absolute energy that it provides, adjusted for the energy provided by each of the other macronutrients, with or without adjusting for potential confounders. “Absolute energy” models can be interpreted as increasing energy from a specific macronutrient while keeping the energy from each other macronutrient constant. All dietary components were categorized into quartiles and modeled either as dummy variables (where each quartile is compared to the lowest quartile) or as a single variable to give a “test of trends” for increasing quartile. The date 5 years from ascertainment of dietary intake was used as time zero for the Cox proportional-hazards models because any cases occurring within the first 5 years were excluded. We performed sensitivity analyses in which we varied the number of years following entry into the cohort that we excluded from analysis from 0 to 6 years.

We performed additional analyses limited to persons (n = 7032) who were hospitalized at least once or who died during follow-up, such that all had hospitalization records or death certificates available to look for the diagnosis of cirrhosis or liver cancer. We also determined whether dietary composition was independently associated with the number of hospitalizations during follow-up.

Due to the importance of obesity and alcohol consumption in liver disease, especially as causes of hepatic steatosis, we performed subgroup analyses stratifying persons into obese (BMI >25 kg/m2) or nonobese, and into persons with excessive alcohol consumption or not. Alcohol consumption was considered excessive in women who consumed ≥1, and in men who consumed ≥2 alcoholic drinks/day.15, 16

Using the NHANES III cross-sectional data (1988-1994), we investigated whether there was an association between quartiles of dietary carbohydrate, protein, or fat intake and the presence of HCV infection, with or without adjusting for the confounders listed above by logistic regression. HCV infection was defined by the presence of HCV RNA.

Results

During an average follow-up time of 13.3 years, 123 out of 9,221 participants had a new diagnosis of cirrhosis (n = 118) or liver cancer (n = 5) in hospitalization records or death certificates, including 36 who were diagnosed only on the basis of death certificates (Table 1). Patients who developed cirrhosis or liver cancer were older, more obese with more central fat distribution, had lower educational attainment and higher alcohol consumption, and were more likely to be male, diabetic, and nonwhite.

Table 1. Baseline Characteristics of NHANES I Participants Presented According to the Subsequent Occurrence of Hospitalization or Death Related to Cirrhosis or Liver Cancer
Baseline CharacteristicsIncident Cirrhosis-Related or Liver Cancer-Related Hospitalization or Death
Yes (N = 123)No (N = 9098)
Age (years), mean (SD)53.5 (13.7)48.7 (15.7)
BMI (kg/m2), mean (SD)26.9 (5.4)25.7 (5.2)
Subscapular to triceps skinfold ratio, mean (SD)1.28 (0.6)1.07 (0.5)
Nonwhite race, %2418
Male, %5838
Diabetes, %4.13.4
Alcohol consumption (drinks per day), %  
 None3444
 >0 to <13945
 1 to <2127
 ≥2155
Coffee or tea consumption (cups per day), %  
 <11419
 1-24547
 24135
High school graduate, %3752
Geographical region, %  
 Northeast2022
 Midwest1824
 South3427
 West2827

The total number of calories and the total quantity of fat consumed were not associated with the incidence of cirrhosis or liver cancer in univariate or multivariate analyses (Table 2). Increasing carbohydrate consumption was associated with significantly reduced incidence of cirrhosis or liver cancer in unadjusted and adjusted analyses in both “percent” and “absolute” energy models (adjusted hazard ratio of 0.42, 95% confidence interval [CI] 0.2-0.8, comparing the top to the bottom quartile). In contrast, protein consumption was associated with a significantly increased risk of cirrhosis or liver cancer after adjusting for potential confounders in absolute energy models, but not in percent energy models.

Table 2. Association Between Dietary Composition and the Incidence of Cirrhosis and Liver Cancer
Dietary IntakeNumber of Subjects N = 9221Person-yearsDeaths or Hospitalizations Related to Cirrhosis or Liver Cancer N = 123Deaths or Hospitalizations Related to Cirrhosis or Liver Cancer per 100,000 Person-yearsUnadjusted Hazard Ratio (95% CI)Adjusted* Hazard Ratio (95% CI)P-value of Test for Trends in the Adjusted* Hazard Ratio
  • Data from the first National Health and Nutrition Examination Survey.

  • *

    In “Absolute Energy” models, each macronutrient is modeled as the absolute energy that it provides, and adjusted for the energy provided by each of the other macronutrients, as well as daily alcohol consumption, coffee or tea consumption, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and subscapular-to-triceps skinfold ratio. In “Percent Energy” models each macronutrient is modeled as the percent of total energy that it provides (calculated as the number of calories from each macronutrient divided by the total number of calories consumed), and adjusted for total dietary energy, as well as daily alcohol consumption, coffee or tea consumption, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and subscapular-to-triceps skinfold ratio.

Dietary calories per day       
 0-1167229529,0263211011† 
 1168-1589230530,28527890.81 (0.5-1.4)0.83 (0.5-1.4)† 
 1590-2146231230,92529940.88 (0.5-1.5)0.80 (0.5-1.4)† 
 ≥2147230932,100351090.99 (0.6-1.6)0.83 (0.5-1.5)†0.7
Protein (g/day)   Absolute Energy Model
 0-45230929,337268911 
 46-63230430,11320660.75 (0.4-1.4)1.0 (0.5-1.9) 
 64-88230430,775371201.31 (0.8-2.2)2.2 (1.2-4.0) 
 ≥89230432,111401251.31 (0.7-2.3)2.4 (1.2-4.9)0.001
Protein (% of total energy intake)   Percent Energy Model
 0-13.1230330,375309911 
 13.2-15.8231230,75618580.59 (0.3-1.1)0.63 (0.3-1.1) 
 15.9-19.3230030,538331081.09 (0.7-1.8)1.12 (0.7-1.9) 
 19.3230630,667421371.39 (0.9-2.2)1.37 (0.8-2.2)0.07
Carbohydrate (g/day)   Absolute Energy Model
 0-126230829,3624114011 
 127-176230530,276311020.74 (0.5-1.2)0.72 (0.4-1.2) 
 177-239230331,12126840.59 (0.4-0.99)0.52 (0.3-0.9) 
 ≥240230531,57725790.52 (0.3-0.9)0.42 (0.2-0.8)0.003
Carbohydrate (% of total energy intake)   Percent Energy Model
 0-37.4230230,7155016311 
 37.5-45.0230631,075321040.63 (0.4-0.9)0.76 (0.5-1.2) 
 45.1-52.2229930,56824790.48 (0.3-0.8)0.57 (0.3-0.9) 
 52.2231329,96217570.35 (0.2-0.6)0.42 (0.2-0.8)0.002
Fat (g/day)   Absolute Energy Model
 0-42229029,0103512111 
 43-62231030,32526860.72 (0.4-1.2)0.75 (0.4-1.3) 
 63-89230930,95126840.73 (0.4-1.2)0.64 (0.3-1.2) 
 ≥ 90231232,050361120.94 (0.6-1.5)0.72 (0.4-1.4)0.3
Fat (% of total energy intake)   Percent Energy Model
 0-30.6228629,5783311211 
 30.7-36.4232030,81825810.73 (0.4-1.2)0.89 (0.5-1.5) 
 36.5-42.2229931,35531990.89 (0.5-1.4)1.08 (0.6-1.8) 
 >42.2231630,585341111.0 (0.6-1.6)1.14 (0.7-1.9)0.5

Although total fat consumption was not significantly associated with cirrhosis or liver cancer, we identified associations with specific lipid components (Table 3). In particular, higher consumption of cholesterol was associated with a higher risk of cirrhosis or liver cancer in both unadjusted and adjusted analyses. Persons in the top quartile of cholesterol consumption were more than twice as likely to develop cirrhosis or liver cancer compared to persons in the bottom quartile of cholesterol consumption. Consumption of saturated fatty acids, oleic acid, or linoleic acid was not associated with the risk of cirrhosis or liver cancer. In contrast, consumption of fatty acids other than the ones quantified by the NHANES I investigators was associated with a reduced risk of cirrhosis or liver cancer in total energy models (but not in percent energy models). Given that saturated fatty acids, oleic acid (the most common dietary monounsaturated fatty acid), linoleic acid (an n-6 PUFA), and cholesterol were quantified by NHANES I investigators, the “other fats” consist mostly of n-3 and n-6 PUFAs (other than linoleic acid).

Table 3. Association Between Dietary Lipid Consumption and the Incidence of Cirrhosis and Liver Cancer
Dietary IntakeNumber of Subjects N = 9221Person-yearsDeaths or Hospitalizations Related to Cirrhosis or Liver Cancer N = 123Deaths or Hospitalizations Related to Cirrhosis or Liver Cancer per 100,000 Person-yearsUnadjusted Hazard Ratio (95% CI)Adjusted* Hazard Ratio (95% CI)P-value of Test for Trends in the Adjusted* Hazard Ratio
  • Data from the first National Health and Nutrition Examination Survey.

  • *

    In “Absolute Energy” models, each macronutrient is modeled as the absolute energy that it provides, and adjusted for the energy provided by each of the other macronutrients, as well as daily alcohol consumption, coffee or tea consumption, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and subscapular-to-triceps skinfold ratio. In “Percent Energy” models each macronutrient is modeled as the percent of total energy that it provides (calculated as the number of calories from each macronutrient divided by the total number of calories consumed), and adjusted for total dietary energy, as well as daily alcohol consumption, coffee or tea consumption, gender, race, age, educational attainment, US geographical region, diabetes, body mass index, and subscapular-to-triceps skinfold ratio.

  • Only absolute energy models are presented for cholesterol because it represents a negligible percentage of ingested dietary energy.

  • Calculated as the daily consumption of total fat minus the daily consumption of saturated fat, oleic acid, linoleic acid and cholesterol.

Saturated fatty acids (g/day)   Absolute Energy Model
 0-14229328,9683010411 
 15-22230630,51030980.98 (0.6-1.6)1.61 (0.8-3.2) 
 23-33231030,79724780.76 (0.4-1.3)1.25 (0.5-3.2) 
 ≥34231232,061391221.2 (0.7-1.9)1.98 (0.7-5.9)0.3
Saturated fatty acids (% of total energy intake)   Percent Energy Model
 0-10230529,5673110511 
 10.1-12.9230030,82225810.77 (0.5-1.3)0.90 (0.5-1.5) 
 13.0-15.9230831,043321030.98 (0.6-1.6)1.12 (0.7-1.9) 
 >15.9230830,904351131.08 (0.7-1.8)1.24 (0.7-2.1)0.3
Oleic acid (g/day)   Absolute Energy Model
 0-16229929,3513511911 
 17-24230930,29422730.61 (0.4-1.04)0.47 (0.2-0.9) 
 25-35230730,71229940.79 (0.5-1.3)0.52 (0.2-1.3) 
 ≥36230631,979371161.0 (0.6-1.6)0.43 (0.1-1.4)0.2
Oleic acid (% of total energy intake)   Percent Energy Model
 0-11.1230529,9613612011 
 11.2-13.9230630,93724780.65 (0.4-1.1)0.72 (0.4-1.2) 
 14-16.7230231,03629930.78 (0.5-1.3)0.90 (0.5-1.5) 
 >16.7230830,402341120.93 (0.6-1.5)0.92 (0.6-1.5)0.9
Linoleic acid (g/day)   Absolute Energy Model
 0-3228528,8953311411 
 4-6232030,54529950.83 (0.5-1.4)0.98 (0.6-1.7) 
 7-10230931,09226840.76 (0.5-1.3)0.86 (0.5-1.6) 
 ≥11230731,804351100.96 (0.6-1.6)1.20 (0.6-2.2)0.7
Linoleic acid (% of total energy intake)   Percent Energy Model
 0-2.1230529,8173612111 
 2.2-3.4230430,68727880.73 (0.4-1.2)0.74 (0.4-1.2) 
 3.5-5.2230230,66125820.67 (0.4-1.1)0.66 (0.4-1.1) 
 >5.2231031,171351120.93 (0.6-1.5)1.09 (0.7-1.7)0.9
Cholesterol (mg/day)   Absolute Energy Model
 0-156228929,782206711 
 157-294230531,06327871.3 (0.8-2.4)1.52 (0.8-2.9) 
 295-510231230,94329941.4 (0.8-2.5)1.66 (0.9-3.1) 
 ≥511231530,548471542.3 (1.4-3.9)2.45 (1.3-4.7)0.007
All other dietary fat (g/day)   Absolute Energy Model
 0-4226829,2263512011 
 5-8228630,22827890.75 (0.5-1.2)0.70 (0.4-1.2) 
 9-13236931,481361140.95 (0.6-1.5)0.78 (0.4-1.4) 
 ≥13229831,40125800.66 (0.4-1.1)0.46 (0.2-0.9)0.05
All other dietary fat (% of total energy intake)   Percent Energy Model
 0-3.4229730,5744113411 
 3.5-4.7232131,04625810.60 (0.4-0.9)0.63 (0.4-1.0) 
 4.8-6.2229430,548341110.83 (0.5-1.3)0.85 (0.5-1.4) 
 >6.2230930,16823760.57 (0.3-0.9)0.63 (0.4-1.0)0.1

Exploratory subgroup analyses by BMI and alcohol consumption are shown in Table 4. The most significant associations that we observed in the entire population (that is, with cholesterol, carbohydrate, and protein intake) were only observed among persons with elevated BMI (≥25 kg/m2) and not among persons with normal BMI (<25 kg/m2). In contrast, the associations between dietary nutrients and cirrhosis or liver cancer were similar among persons with or without excessive alcohol consumption.

Table 4. Association Between Dietary Composition and the Incidence of Cirrhosis and Liver Cancer Presented According to Body Mass Index and Alcohol Consumption Category
Number of SubjectsBMI < 25BMI ≥ 25Alcohol Consumption ≥1 Drink Per Day for Women or ≥2 Drinks Per Day for Men
4,6014,620No 8,110Yes 1,111
Deaths or Hospitalizations Related to Cirrhosis or Liver Cancer47768538
 HR*P ValueHR*P ValueHR*P ValueHR*P Value
  • Data from the first National Health and Nutrition Examination Survey.

  • *

    HR is the adjusted Hazard Ratio for each increasing quartile of dietary intake (equivalent to a test of trends across quartiles). In “Absolute Energy” models, each macronutrient is modeled as the absolute energy that it provides, and adjusted for the energy provided by each of the other macronutrients, as well as daily alcohol consumption, coffee or tea consumption, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and subscapular-to-triceps skinfold ratio. In “Percent Energy” models each macronutrient is modeled as the percent of total energy that it provides (calculated as the number of calories from each macronutrient divided by the total number of calories consumed), and adjusted for total dietary energy, as well as daily alcohol consumption, coffee or tea consumption, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and subscapular-to-triceps skinfold ratio.

Protein        
Absolute energy1.030.91.66<0.0011.310.051.610.027
Percent energy0.940.71.340.081.220.061.080.6
Carbohydrate        
Absolute energy0.950.80.63<0.0010.750.020.700.04
Percent energy0.680.0130.790.0350.770.010.700.07
Total fat        
Absolute energy1.200.40.760.0481.00.80.740.1
Percent energy1.150.31.010.91.10.20.920.6
Saturated fat        
Absolute energy1.500.21.110.71.190.41.320.4
Percent energy1.370.0260.980.81.190.080.910.6
Oleic acid Absolute energy1.130.70.640.080.880.60.620.2
Percent energy1.110.40.940.51.040.70.920.6
Linoleic acid        
Absolute energy0.991.01.120.61.110.40.890.5
Percent energy0.980.91.030.81.110.30.810.2
Cholesterol        
Absolute energy1.040.81.460.0021.330.0171.330.1
Other dietary fat        
Absolute energy0.690.030.900.40.860.20.770.2
Percent energy0.810.10.950.60.920.40.830.2

In the analyses presented in Tables 2 we excluded the first 5 years of follow-up, reasoning that cirrhosis or liver cancer that was present and undiagnosed at entry into the cohort would have led to hospitalization or death within 5 years. We also varied the number of years from entry into the cohort that were excluded from analysis from 0 to 6 and found in each analysis that protein and cholesterol consumption were associated with higher risk of cirrhosis or liver cancer, whereas carbohydrate and “other fat” consumption were significantly associated with lower risk of cirrhosis or liver cancer. When we repeated our analyses using only the 7032 participants who were hospitalized at least once or who died during follow-up, we found near-identical results. Dietary content of protein, carbohydrate, total fat, saturated fatty acid, oleic acid, linoleic acid, cholesterol, or “other fat” was not independently associated with the number of hospitalizations during follow-up.

Because dietary cholesterol consumption was so strongly associated with the incidence of cirrhosis or liver cancer, we evaluated whether serum cholesterol levels were also associated with cirrhosis or liver cancer, but we found no such association in either univariate or multivariate analyses. Therefore, ingested, dietary cholesterol was associated with cirrhosis or liver cancer, but hepatically synthesized cholesterol, which is the main determinant of serum cholesterol, was not associated with cirrhosis or liver cancer.

Finally, we used NHANES III data collected between 1988-1994 (Table 5) to investigate the association between dietary composition and the presence of HCV infection (which was not ascertained in NHANES I). There was no association between any measure of dietary composition and HCV infection after adjusting for the same confounders that we adjusted for in our NHANES I analyses. Therefore, it is extremely unlikely that the significant associations that we present in Tables 2 and 3 would have changed if we could have adjusted for HCV infection. Also, the lack of association between HCV infection and dietary composition strongly suggests that the presence of underlying liver disease does not cause a change in dietary intake and, instead, makes it more plausible that differences in dietary intake of proteins, carbohydrates, cholesterol, and perhaps other lipid components contribute to the development of cirrhosis or liver cancer.

Table 5. Association Between Dietary Composition and HCV Infection
Dietary IntakeNumber of Subjects N = 13,629Number (and %) with HCV N = 268Odds Ratio (95% CI)Adjusted* Odds Ratio (95% CI)Adjusted Odds Ratio (95% CI)
  • Cross-sectional data from the third National Health and Nutrition Examination Survey, 1988-1994.

  • *

    Adjusted for alcohol consumption, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and waist circumference.

  • Adjusted for daily consumption of protein, carbohydrate, or fat (for these variables) or for saturated fatty acids, oleic acid, linoleic acid, and cholesterol (for the lipid components), as well as alcohol consumption, gender, race, age, educational attainment, U.S. geographical region, diabetes, body mass index, and waist circumference.

Protein (g/day)     
 0-49338561 (1.8)111
 50-70343464 (1.9)1.0 (0.7-1.5)1.0 (0.7-1.4)0.9 (0.6-1.4)
 71-97339149 (1.5)0.8 (0.5-1.2)0.6 (0.4-0.9)0.6 (0.4-0.9)
 ≥98341994 (2.8)1.5 (1.1-2.1)0.9 (0.7-1.3)0.8 (0.5-1.3)
Carbohydrate (g/day)     
 0-166340955 (1.6)111
 167-231340452 (1.5)0.9 (0.6-1.4)0.9 (0.6-1.3)1.0 (0.7-1.5)
 232-313340772 (2.1)1.3 (0.9-1.9)1.1 (0.8-1.6)1.3 (0.9-1.9)
 ≥314340989 (2.6)1.6 (1.2-2.3)1.2 (0.8-1.7)1.4 (0.9-2.1)
Fat (g/day)     
 0-45340455 (1.6)111
 46-69341763 (1.8)1.1 (0.8-1.6)1.0 (0.7-1.4)1.1 (0.7-1.6)
 70-102340665 (1.9)1.2 (0.8-1.7)0.9 (0.6-1.4)1.0 (0.6-1.6)
 ≥103340285 (2.5)1.6 (1.1-2.2)1.0 (0.7-1.4)1.0 (0.6-1.6)
Saturated fatty acids (g/day)     
 0-14343455 (1.6)111
 15-23338056 (1.7)1.0 (0.7-1.5)0.9 (0.6-1.3)1.1 (0.7-1.8)
 24-35341569 (2.0)1.3 (0.9-1.8)1.0 (0.7-1.5)1.4 (0.8-2.5)
 ≥36340088 (2.6)1.6 (1.2-2.3)1.1 (0.8-1.6)1.6 (0.8-3.1)
Oleic acid (g/day)     
 0-16338355 (1.6)111
 17-24344261 (1.8)1.1 (0.8-1.6)0.9 (0.6-1.4)0.8 (0.5-1.4)
 25-37338066 (2.0)1.2 (0.8-1.7)0.9 (0.7-1.4)0.7 (0.4-1.4)
 ≥38342486 (2.5)1.6 (1.1-2.2)1.0 (0.7-1.4)0.6 (0.3-1.3)
Linoleic acid (g/day)     
 0-6333553 (1.6)111
 7-11350568 (1.9)1.2 (0.9-1.8)1.1 (0.8-1.6)1.2 (0.8-1.8)
 12-18338669 (2.0)1.3 (0.9-1.8)1.1 (0.7-1.6)1.2 (0.7-1.9)
 ≥19340378 (2.3)1.5 (1.0-2.1)1.0 (0.7-1.5)1.1 (0.7-1.8)
Cholesterol (mg/day)     
 0-125338258 (1.7)111
 126-217340551 (1.5)0.9 (0.6-1.3)0.8 (0.5-1.1)0.7 (0.5-1.1)
 218-388343156 (1.6)1.0 (0.7-1.4)0.7 (0.5-1.1)0.7 (0.4-1.1)
 ≥3893411103 (3.0)1.8 (1.3-2.5)1.1 (0.8-1.6)1.1 (0.7-1.7)

Discussion

Our results show for the first time that dietary nutrient composition is a strong predictor of hospitalization or death due to cirrhosis or liver cancer in the U.S. population. In particular, we identified that protein and cholesterol consumption were associated with elevated risk, whereas consumption of carbohydrates was associated with reduced risk of hospitalization or death related to cirrhosis or liver cancer. We also observed a weaker association between consumption of PUFAs and reduced risk of cirrhosis or liver cancer, although PUFA consumption was not ascertained well in NHANES I.

The strong association between cholesterol intake and cirrhosis or liver cancer is potentially our study's most important finding. In rabbits5, 6 and mice7 a high cholesterol diet has been shown to induce profound steatosis, inflammation, and centrilobular fibrosis. It was postulated that the hepatic steatosis, which was predominantly microvesicular, was the result of a cholesterol-induced reduction in mitochondrial beta oxidation of fatty acids.5 Some dietary cholesterol is already oxidized to oxysterols and it is also possible for dietary cholesterol to be converted in the liver to oxysterols, which promote cytotoxic and carcinogenic effects. Hepatic oxidative stress is a prominent feature of nonalcoholic steatohepatitis (NASH), HCV infection, and alcoholic liver disease, and could lead to generation of oxysterols from dietary cholesterol. Despite the profound importance of cholesterol in nonhepatic diseases such as atherosclerosis, dietary cholesterol and oxysterols have received little attention as potential lipotoxic molecules in the human liver. We are not aware of any other human studies linking cholesterol intake to human liver disease. Our finding that dietary cholesterol but not serum cholesterol was associated with cirrhosis or liver cancer could have profound implications if confirmed by other studies. For example, it would suggest that drugs blocking intestinal cholesterol absorption could have more beneficial effects to the liver than drugs blocking hepatic cholesterol synthesis, which lower serum cholesterol levels by up-regulation of hepatic low-density lipoprotein receptors.

We did not have strong a priori hypotheses as to the direction of associations between protein or carbohydrate consumption and cirrhosis or liver cancer. Thus, our findings of a positive association of protein consumption and a negative association of carbohydrate consumption with hospitalization or death related to cirrhosis or liver cancer should be considered with caution. High protein intake in humans, in particular, animal protein, has been associated with hepatocellular carcinoma in an ecological study from rural China.17 In HBV transgenic mice, a diet low in animal protein was associated with decreased liver injury and decreased incidence of hepatocellular carcinoma.9, 10 (We did not have information on whether ingested protein was of animal or plant origin.) Twenty-five patients with biopsy-proven NASH had diets richer in protein (as well as saturated fat and cholesterol and poorer in PUFAs and fiber) than 25 age-, BMI-, and gender-matched healthy controls.18 We are not aware of any studies investigating the effects of carbohydrate intake on the progression of liver disease in humans. Contrary to our findings, a recent study found that patients with NAFLD consumed 2-3 times more fructose than controls, which may be linked to the development of steatosis because the hepatic metabolism of fructose favors de novo lipogenesis.19 Our data did not allow us to determine the specific fructose intake; we speculate that fructose intake was much lower in the period of our study (1971-1992) than it is now due to the dramatic increase in the consumption of soft drinks containing high fructose corn syrup.

Although hepatic steatosis is defined by excessive deposition of lipid within hepatocytes, little is known about the exact nature of these lipids in humans. Evidence is emerging from cell culture and animal model experiments that specific lipid molecules, such as saturated fatty acids, may exert lipotoxicity,1, 2 whereas others, such as n-3 PUFAs, may actually have beneficial effects.3, 4 In human studies, hepatic levels of n-3 PUFAs were shown to progressively decrease from control patients to patients with NAFLD to NASH.20, 21 In a randomized controlled trial, n-3 PUFA supplementation reduced ultrasonographic steatosis, alanine aminotransferase levels, and insulin resistance in patients with suspected NAFLD.22 We did not find any association between dietary intake of saturated fatty acids or the monounsaturated fatty acid oleic acid and cirrhosis or liver cancer. Although we did not have direct estimates of dietary PUFAs, especially n-3 PUFAs, we approximated PUFA intake by subtracting the intake of saturated fat, oleic acid, linoleic acid, and cholesterol from total fat intake. Using this approximation, we identified that persons in the top quartile of PUFA intake were half as likely to develop cirrhosis or liver cancer than persons in the bottom quartile in absolute energy models, but not in percent energy models; however, the test of trends across quartiles of “other fat” intake was just short of statistical significance (P = 0.05) even in the absolute energy models. Future studies should investigate these associations with more specific ascertainment of n-3 and n-6 PUFA intake.

Subgroup analyses showed that the significant associations of protein, carbohydrate, and cholesterol intake with cirrhosis or liver cancer that we described in the entire study population, were limited to overweight or obese persons (BMI ≥25 kg/m2) (Table 4). No such associations were observed in normal-weight persons. In contrast, alcohol consumption did not appear to modify substantially the associations between dietary nutrients and cirrhosis or liver cancer. This suggests that high protein and cholesterol and low carbohydrate intake are more likely to mediate their hepatic effects through obesity-related fatty liver disease.

Our study has some limitations, including the absence of data on HCV infection, which is a major cause of cirrhosis and liver cancer in the U.S. However, we have shown that in the more recent NHANES III study, HCV infection was not associated with any of the dietary components that we investigated; hence, HCV infection is unlikely to be an important source of uncontrolled confounding. It is possible that our results are subject to misclassification because a 24-hour dietary recall may not reflect long-term dietary intake accurately. However, such inaccuracies in the assessment of dietary intake using a 24-hour dietary recall are expected to be random and to have no relation to the study's outcome, that is, the development of cirrhosis or liver cancer. It is well described that such random, nondifferential misclassification of exposure does not lead to spurious associations, but, rather, tends to reduce real associations toward the null.23 Therefore, we believe that the associations that we describe between cholesterol, protein, and carbohydrate intake and the development of cirrhosis or liver cancer are likely to be true and the magnitude of these associations is likely to be even greater than what we report. Multiple studies have used the NHANES 24-hour dietary recall data to study the associations between baseline dietary factors and the subsequent development of disease or mortality as we did.24–26 The quality of the NHANES 24-hour dietary recalls is considered very good because the National Center for Health Statistics employed specially trained, experienced interviewers who performed more than 20,000 24-hour dietary recalls as part of NHANES I.

Although we simultaneously adjusted for all the dietary components that we investigated, it is possible that these dietary components may be surrogate markers for the intake of other known or unknown nutrients that could have important effects on the liver, such as dietary antioxidants. We could not determine from the available 24-hour dietary recall data whether the protein intake that was associated with increased risk of cirrhosis-related death or hospitalization was animal protein or plant protein. The associations that we describe could occur if an underlying chronic liver disease were to influence taste, smell, and food preference before the development of cirrhosis. However, in the NHANES III study there was no association between the presence of a chronic liver disease (HCV infection) and dietary intake. Our study measures hospitalizations and deaths due to cirrhosis or liver cancer, not the actual incidence. Thus, patients with undiagnosed cirrhosis, or diagnosed cirrhosis that did not lead to hospitalization or death, were not captured.

At the same time, our study has some important strengths. Dietary intake was ascertained prospectively in a large, nationally representative sample of the U.S. population with 13 years of prospective follow-up (after excluding from analysis the first 5 years following entry into the study). Hospitalizations or deaths due to cirrhosis were ascertained prospectively by specially trained personnel without any knowledge of baseline dietary intake. Many determinants of liver disease progression are currently unknown, as evidenced by the fact that we cannot predict accurately which patients with any of the major liver diseases (HCV, HBV, NAFLD, and alcoholic liver disease) will progress to cirrhosis and which ones will have a relatively benign course. Our study raises the possibility that dietary factors may be important, modifiable, and hitherto unrecognized determinants of liver disease progression.

Ancillary