Serum Biomarkers of Iron Status and Risk of Hepatocellular Carcinoma Development in Patients with Nonalcoholic Fatty Liver Disease

The mortality rate of primary liver cancer in the United States has doubled since the mid-1980s (1). Worldwide, the death toll of primary liver cancer is projected to reach one million in 2030 with the current trend (2, 3). Hepatocellular carcinoma (HCC) is the major subtype of primary liver cancer (4), accounting for approximately 75% to 90% of total primary liver cancer cases (5). Incidence rates of HCC vary dramatically across different geographic regions: the highest rate in sub-Saharan Africa and Eastern Asia (>20.0 per 100,000 persons) and the lowest in North America and Northern Europe (<5.0 per 100,000 persons; ref. 6). Chronic infection with hepatitis B virus (HBV) and/or hepatitis C virus (HCV) accounts for approximately 50% or more of all HCC cases in the United States (7). Heavy alcohol use and cigarette smoking are also recognized as risk factors for HCC (8). In high-risk regions, exposure to dietary aflatoxin plays a significant role in the development of HCC (9).

Nonalcoholic fatty liver disease (NAFLD) and its associated conditions including obesity and diabetes has been emerged as major risk factors for HCC (10). NAFLD is considered as one of the most common chronic liver diseases in developed countries (11). Globally, it is estimated that the prevalence of NAFLD is 30% to 40% in men and 15% to 20% in women (12). NAFLD encompasses a wide spectrum of liver conditions from simple steatosis and nonalcoholic steatohepatitis (NASH) to fibrosis and cirrhosis, which may eventually progress to HCC (13, 14).

The mechanism for the disease progression from NAFLD to HCC remains unclear. One of the potential contributing factors may be iron overload. On average, healthy well-nourished adults have 1 to 3 grams of iron in their bodies (15). Dietary iron is the major source for humans. Under normal condition, approximately 10% of dietary iron is absorbed through intestine (16). Iron is not secreted through urine but naturally through the loss of blood (16). Most of the iron in the body is contained in the hemoglobin of matured red blood cells (17). The rest of body iron is present in ferritin complexes stored in bone marrow, liver and spleen (15). The liver ferritin is the primary iron reservoir for physiologic supply when needed (18, 19). Two clinical studies in Iran and Egypt found that high ferritin level may accelerate the liver injury or liver fibrosis in people with metabolic disease such as NAFLD (20, 21). However, the role of serum iron biomarkers on the progression of NAFLD to HCC are unknown.

Iron overload occurs when there are excess stores of iron in the body, which is defined as the total amount of the body iron surpasses 5 grams (22). The primary iron overload or hereditary hemochromatosis is caused by genetic variants in the HFE gene, resulting in excess absorption of iron from the intestine (16, 23) Up to 90% of individuals with hemochromatosis diagnosis carry the HFE mutations (23, 24). The secondary iron overload would result from excess absorption of iron from the food, repeated blood transfusions or excess intake of iron-containing substances (25). In the United States, one in 200 Caucasians is in the iron overload status (26). Regardless of the underlying cause, iron overload may cause hepatotoxicity (27) and lead to liver cirrhosis (28). Numerous studies examined and found a statistically significant association for hereditary hemochromatosis with risk of cirrhosis and HCC (29–33). However, epidemiologic data on elevated iron levels without hemochromatosis in relation to HCC risk are sparse.

In the current study, we examined the association between systemic biomarkers of iron status and risk of HCC among a large cohort of NAFLD patients without diagnosis of hemochromatosis, alcoholic and viral liver disease, and other liver conditions in a large health care network system in the Commonwealth of Pennsylvania (PA).

We conducted a retrospective cohort study of participants with NAFLD of the University of Pittsburgh Medical Center (UPMC) Health Insurance Plan (referred as the UPMC NAFLD Cohort Study). UPMC is a major medical service provider with 40 hospitals and 700 doctors' offices and outpatient sites that serves more than 3 million patients annually throughout western PA in the United States. Data were requested through the Health Record Research Request (R3) provided by the University of Pittsburgh Biomedical Informatics Services. For the current analysis, we requested deidentified electronic health records (EHR) of all UPMC Health Plan participants, 40 to 89 years of age when they had any medical service at the UPMC Healthcare Network System during January 1, 2004, through December 31, 2018. The study was approved by the University of Pittsburgh Internal Review Board, and conducted in accordance with the recognized ethical guidance.

Eligibility criteria for the UPMC NAFLD Cohort Study were patients with any of the following diagnoses according to the International Classification of Diseases 9th revision, clinical modification (ICD-9-CM) or 10th revision, clinical modification (ICD-10-CM) codes: non-alcoholic fatty liver (NAFL), NASH, cirrhosis of liver without mention of alcohol, hepatic failure, or hepatic encephalopathy (see ICD codes in Supplementary Table S1). Exclusion criteria for the UPMC NAFLD Cohort were patients with any of the following diagnoses based on their ICD-9-CM or ICD-10-CM codes on the EHRs: alcoholic liver disease, alcohol use disorder, somatic consequences of alcohol, autoimmune liver disease, alpha-1-antitrypsin deficiency, secondary or unspecified biliary cirrhosis, drug use disorder except nicotine/caffeine, hemochromatosis, Budd-Chiari syndrome, viral hepatitis, unspecified chronic hepatitis, or Wilson's disease (see ICD codes in Supplementary Table S2). These inclusion and exclusion criteria were based on the recent consensus statement by an expert panel for the definition of NAFLD using administrative coding in electronic health care records (34). The UPMC NAFLD Cohort Study consisted of 47,165 unique patients.

The primary exposure variables for the present analysis were serum biomarkers of iron status – iron, transferrin saturation, total iron binding capacity (TIBC) and ferritin quantified at routine clinical chemistry laboratories. Patients who did not have any one of the 4 iron biomarker measurements were excluded from the present analysis (n = 27,429 excluded). We further excluded 7 patients with extreme values of iron-related measurements, which deemed as physiologically impossible values (Fig. 1). The covariates extracted from the EHRs included age, sex, race, body mass index, smoking status, and histories of type II diabetes, dyslipidemia, and hypertension.

The primary outcome of the current analysis was incident HCC with the initial diagnosis date from January 1, 2004, through December 31, 2018. All HCC cases were identified based on ICD-9-CM code 155.0 or ICD-10-CM codes C22.0 and C22.8.

Any subject whose earliest measurement of any iron biomarker was performed within 30 days prior to or after the latest date of follow-up were excluded from the present analysis to minimize the potential impact of diagnostic and treatment procedures for HCC on the levels of serum iron biomarkers (n = 1,104 excluded). In addition, we excluded patients whose earliest measurement of any iron biomarker was done less than 180 days after the occurrence of gastrointestinal bleeding (n = 56 excluded), which might have resulted in reduced serum iron level (Fig. 1). Fifty-six patients with NAFLD (3 HCC cases and 53 non-HCC cases) who had a history of gastrointestinal bleeding more than 180 days before serum iron biomarker measurement were included in the analysis. Their average time interval from the gastrointestinal bleeding to the serum irone test was 32.2 months, which would have sufficient time to produce mature red blood cells. The final analysis included 18,569 NAFLD patients including 244 incident HCC cases (Fig. 1). As expected, a small percentage of NALFD patients were NASH (n = 2,517, 13.6%) because its diagnosis required an invasive liver biopsy procedure.

The distributions of serum iron, transferrin saturation, and ferritin were skewed. Thus, we used the nonparametric Wilcox test to examine the differences in median values of these iron-related biomarkers between patients with incident HCC and those without HCC. X² and pooled two-sample t test were used to compare the differences in the distributions of nominal or categorical variables and continuous variables, respectively, between HCC cases and non-HCC subjects. The Cox proportional hazard regression method was used to assess the associations between serum levels of iron biomarkers and HCC risk. Person-years at risk for each study subject was counted from the date of iron measurements to the date of HCC diagnosis, death, last encounter with any UPMC Healthcare Network facility, whichever occurred first. All study subjects were grouped into low (below lower limit of normal), normal, and high levels (above upper limit of normal) according to the recently recommended ranges for clinical use: the normal range was 75 to 175 μg/dL for serum iron (35, 36), 25% to 35% for transferrin saturation (37, 38), and 240 to 450 μg/dL for TIBC (38, 39) for both men and women. The normal range of serum ferritin was 3 to 30 μg/dL for men and 1 to 20 μg/dL for women (40, 41).

Statistical analyses were carried out using SAS software version 9.4 (SAS Institute, Cary, NC). All P values reported are two-sided. The P values of less than 0.05 were considered being statistically significant.

The curent analysis included 18,569 patients with NAFLD. After an average 4.34 years of follow-up, 244 developed HCC. Men were more likely to develop HCC than women. Patients who developed HCC were older, ever cigarette smokers, or more likely to have a history of type II diabetes, or hypertension, but less likely to have dyslipidemia (all P values < 0.05; Table 1). A stepwise Cox regression model identified the following independent variables that were statistically significantly associated with HCC risk: age (years), sex, BMI (kg/m²), race (white vs. non-white), history of type II diabetes (yes vs. no), and smoking status (never vs. ever smoker). These variables were included as covariates in the final multivariate Cox regression models examining the associations between serum iron biomarkers and HCC risk.

The median values of the four iron biomarkers for patients with and without HCC are shown in Table 2. Serum iron was slightly higher whereas serum ferritin and TIBC were lower in patients with HCC than those without HCC. However, their differences were not statistically significant (P > 0.05).

Higher level of serum iron was associated with significantly elevated risk of developing HCC (Table 3). Compared with the normal range, the risk of HCC for patients with NAFLD with iron level above the upper limit of normal was increased by a statistically significant 191% (HR = 2.91; 95% CI, 1.34–6.30) after adjusting for age, sex, race, BMI, history of type II diabetes, and smoking status. Similarly, HCC risk was doubled (HR = 2.02; 95% CI, 1.22–3.32) for patients with NAFLD with above normal range of transferrin saturation (>35%) as compared with the normal range. No statistically significant association was observed for TIBC or serum ferritin with HCC risk.

Serum iron levels were highly correlated with transferrin saturation [the correlation coefficient (r) = 0.83, P < 0.00], but not correlated with TIBC (r = 0.04) or serum ferritin (r = 0.11). The correlation of serum ferritin with other iron biomarker measurements was low or moderate (Supplementary Table S3).

Besides the NAFLD, iron deficiency anemia was the most common indication for serum iron test. As expected, a higher proportion of patients with NAFLD tested as having low serum iron level had a history of anemia (17.8%) than those with high serum iron (8.7%; Supplementary Table S4). Low serum iron was also associated with higher BMI, female gender, non-white race/ethnicity, and history of type II diabetes and dyslipidemia. To eliminate the potential confounding effect of the iron test indication, we performed a sensitivity analysis after excluding all subjects with a history of iron deficiency anemia, the results remained the same. The adjusted HR of HCC for the high serum iron was 3.19 (95% CI, 1.47–6.94; P = 0.004) compared with the normal range of serum iron.

We investigated the associations between four serum biomarkers of iron status and HCC risk in a retrospective cohort study of 18,569 patients derived from the UPMC NAFLD Cohort Study with clinical diagnosis of NAFLD without hemochromatosis or any other major underlying causes of chronic liver disease. We found that elevated levels of both serum iron and transferrin saturation were significantly associated with 2- to 3-fold increased risks of HCC than normal range of these iron biomarker measurements. There were no statistically significant association for serum ferritin and TIBC with the risk of HCC.

NAFLD has become the major cause of chronic liver disease in the United States (42). However, only a minority of the patients with fatty liver progress to cirrhosis, at 1.2% after 20 years of follow-up (43). Among NAFLD patients with cirrhosis, the HCC incidence rate was around 1% per year (44). The additional risk factors, such as iron overload, may accelerate the progression of NAFLD to cirrhosis and HCC. Liver iron deposits have been frequently observed in patients with NAFLD (45). A hospital-based case-control study involving 51 HCC cases and 102 matched controls free of HCC demonstrated that iron deposits (corrected total iron score > 0) were significantly more frequent in patients with NASH-related cirrhosis with HCC than in HCC-free controls (45). A recent meta-analysis summarized 9 previous studies published before 2019 (46). These studies used the design of a prospective cohort or nested case-control study within a prospective cohort in various populations in Asia, Europe, and the United States (46). Of the 9 studies, three investigated the serum iron and risk of liver cancer and found a pooled HR of 2.47 (95% CI, 1.31–4.63) for the highest level of serum iron compared with the lowest iron, although different cut-offs for high or low serum iron were used. Our findings with an HR of 2.91 (95% CI, 1.34–6.30) for HCC with clinically defined high level of serum iron were consistent with the pooled HR of this meta-analysis. In addition, a statistically significant increased risk of HCC with elevated transferrin saturation in our study was also consistent with similar findings from a previous study (47). Given the high correlation between serum iron and transferrin saturation, we were not able to separate their effect on HCC risk.

Serum ferritin, a biomarker for body iron reservoir, has been associated with increased risk of HCC. In a hospital-based case-control study of 24 HCC cases and 48 age-, race-, and sex-matched controls of blacks in South Africa, the iron-loaded subjects (defined as a raised serum ferritin concentration in combination with a transferrin saturation ≥60% due to high intake of iron from homemade alcoholic beverage) were at significantly increased risk of HCC (odds ratio = 10.6; 95% CI, 1.5–76.8) compared with normal ferritin level (48). The meta-analysis described above also found a statistically significant, 49% increased risk of HCC associated with higher levels of serum ferritin, although there was a significant heterogeneity among the 6 studies included (46). These studies used different cut-off values for serum ferritin, different study populations, the patients with different underlying causes of HCC or chronic liver disease. These differences may explain the different results of the meta-analysis from ours that used a clinically defined sex-specific cut-off values for serum ferritin and NAFLD patients only.

Our current study has several strengths. First, we retrospectively constructed a cohort of more than 18,000 patients with clinical diagnosis of NAFLD and iron measurements who were free of other chronic liver diseases including hemochromatosis and alcoholic and viral hepatitis. The exposure to iron was defined by four commonly used clinical measurements of serum iron-related biomarkers and quantified on average 4.34 years prior to the occurrence of HCC, which minimized the potential impact of diagnostic and treatment procedures for HCC on serum biomarker levels. All cut-off values of the iron biomarkers included in the present study were derived from most recently defined values with clinical relevance.

Out study also has several limitations. First, random errors of the four iron measurements due to incorrect data entry or at single time point may not be reflective of their change over time. However, measurement errors due to random variations usually attenuate the observed iron biomarker-HCC risk association towards the null. Second, although the present study included more than 18,000 patients with NAFLD with more than 4 years of follow-up, the number of HCC cases was relatively small given its low incidence rate, which did not allow for subgroup analysis stratified by gender, obesity status, or history of type II diabetes. Third, we cannot completely rule out potential residual confounding by relatively low levels of alcohol consumption on the observed iron-HCC risk association due to incomplete recording of moderate or lower alcohol intake history in patients' EHRs, although patients with heavy alcohol consumptions or related alcohol use disorders were excluded from the study population (49). Finally, as with any observational studies, confounding and/or selection bias might play a role on the observed association between serum iron and HCC risk.

In conclusion, in a large retrospective cohort study of 18,569 patients with NAFLD, we found that the elevations of prediagnostic serum iron and transferrin saturation were significantly associated with increased risk of HCC. Our findings provide supporting evidence for the detrimental role of iron overload on the progression of NAFLD to the development of HCC. Further prospective studies with large samples in diverse study populations are warranted to confirm our findings.

J. Behari reports grants from Gilead Sciences, Endra Life Sciences, and Pfizer Inc outside the submitted work, as well as institutional clinical trial contracts with Intercept, Pfizer, Genentech, Celgene, and Galectin. J.-M. Yuan reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

Y.-C. Yu: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. H.N. Luu: Formal analysis, writing–review and editing. R. Wang: Conceptualization, data curation, formal analysis, writing–review and editing. C.E. Thomas: Writing–review and editing. N.W. Glynn: Formal analysis, writing–review and editing. A.O. Youk: Formal analysis, writing–review and editing. J. Behari: Formal analysis, writing–review and editing. J.-M. Yuan: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing.

This research project was partially supported by an NIH grant (No. R01CA255809 to J. Behari and J.-M. Yuan), the University of Pittsburgh Medical Center Hillman Cancer Center Start-up fundings (to H.N. Luu and J.-M. Yuan). C.E. Thomas was supported by U.S. NIH training grant No. T32CA186873 (to J.-M. Yuan). We acknowledge the University of Pittsburgh Biomedical Informatics Services for providing de-identified electronic health records.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.