Risk Factors for Hepatocellular Carcinoma (HCC) in the northeast of the United States: Results of a Case-Control Study

Abstract

Purpose:

HCC incidence has been continuously rising in the US for past 30 years. To understand the increase in HCC risk, we conducted a case-control study in Connecticut, New Jersey and part of New York City.

Methods:

Through rapid case ascertainment and random digit dialing, we recruited 673 incident HCC patients and 1,166 controls. Information on demographic and anthropometric characteristics, lifestyle factors, medical and family cancer histories, were ascertained through telephone interviews using a structured questionnaire. Saliva specimens were collected for testing hepatitis C virus (HCV) antibodies. Unconditional logistic regression models were utilized to calculate odds ratio (OR) and 95% confidence interval (CI) to determine HCC associations with risk factors.

Results:

The study confirmed that HCV infection and obesity were important risk factors for HCC, ORs=110 (95%CI: 59.2–204) and 2.13 (95%CI: 1.52–3.00), respectively. High BMI and HCV infection had synergy in association with elevated HCC risk. Patients both obese and infected with HCV had HCC detected on average nearly 10 years earlier than those with neither factor. Diabetes, cigarette smoking and heavy alcohol intake were all associated with increased risk of HCC, whereas aspirin and other NSAID use were associated with reduced risk. HCC cases tended to attain less education, with lower household incomes, unmarried, and to have had more sexual partners than the controls.

Conclusions:

Individuals at risk of HCC in the US comprise a unique population with low socioeconomic status and unhealthy lifestyle choices. Given the multifactorial nature, a comprehensive approach is needed in HCC prevention.

Introduction

HCC is a devastating disease; every year more than 750,000 people die from HCC worldwide [1]. HCC ranks third among all cancer deaths in the world [2–4]. The incidence of HCC varies widely from country to country. Although HCC is relatively infrequent in the US, the incidence of HCC has been steadily rising [5–8]. The rate has tripled over the last 30 years [9]. If this trend continues, HCC may become the 11^th most common cancer in the US by 2030 [10]. The increase in HCC incidence occurs across the US for both males and females [9, 11, 12]. The steady increase in HCC incidence indicates elevated exposure of American populations to HCC risk factors, which raises significant public health concern but also opportunities for prevention.

HCC has several established risk factors, including infection with hepatitis B and C viruses (HBV, HCV), consumption of food contaminated with fungal toxin aflatoxin B1 (AFB1), excessive alcohol intake, occupational exposure to chemical hazards, long-term obesity, diabetes mellitus, and hereditary hemochromatosis [5, 6, 13]. Of these factors, the first three are the most important because they are responsible for the majority of HCC cases worldwide [5]. In regions where HCC incidences are high, HBV infection and AFB1 exposure are the major risk factors, explaining more than 80% of the disease [5]. However, in countries where HCC rates are low, including the US, HBV and AFB1 are less common, and HCV infection represents the main HCC risk factor [5, 6, 14]. Studies also demonstrate that obesity and its consequence of diabetes mellitus and non-alcoholic fatty liver disease play an important role in the etiology of HCC [15–19]. The link between lifestyle and HCC has significant implications for public health because the increasing prevalence of obesity and diabetes in the US suggests that the incidence of HCC may continue to rise in the absence of effective interventions [10].

Most epidemiologic studies on HCC in the US were hospital-based investigation, which, to certain extent, has the limitation to generalize the study findings to a real population for consideration of HCC prevention and intervention. In order to assess HCC risk more closely in a general US population, we conducted a largely population-based epidemiologic study in the states of Connecticut and New Jersey, as well as in part of New York City. The study was designed to confirm the role of obesity in the etiology of HCC and to assess the synergy of obesity with other major risk factors of HCC such as HCV infection. The study also examined other known and suspected risk and lifestyle factors for HCC.

Methods and Materials

Study population

A case-control study was conducted in Connecticut (CT), New Jersey (NJ) and part of New York City (NYC). The study was approved by the IRBs at Yale University, Connecticut Department of Health, Rutgers Cancer Institute of New Jersey, New Jersey Department of Health, Columbia University, and hospitals involved in patient recruitment. HCC cases were defined as patients aged 35–84 years with a first diagnosis of HCC between January 2011 and February 2016, whose diagnoses were reported to the state tumor registries in CT and NJ, or who were treated in the liver transplant center at Columbia University Irving Medical Center (CUIMC), and resided in CT, NJJ or NYC for at least one year. Population-based HCC cases were identified through two rapid case ascertainment (RCA) systems, one in CT and one in NJ, operated by Yale Cancer Center and Cancer Institute of New Jersey, respectively. The Liver Transplant Center at CUIMC also recruited consecutive HCC patients receiving treatment at the center. After eligible patients were identified, patient physicians were approached for permission to contact their patients. Following approval, patients were called by the study interviewers to confirm study eligibility and to seek consent for participation. The median time interval between diagnosis and interview was 102 days. Patients were enrolled in the study after signing and returning written informed consent by mail. During the study recruitment, the RCA systems identified 1,251 potential HCC patients and CUIMC ascertained 116. Of these, 724 consented to participate and 673 completed the enrollment requirements, including 235 from CT, 333 from NJ and 105 from NYC. In comparison of their age, gender, and race distributions, HCC cases enrolled in the study did not differ substantially from those who were identified but not enrolled (Supplemental Table S1).

During the same time of case enrollment, control subjects were defined as individuals living in CT or NJ and without any form of cancer at the time of contact. Control individuals were selected through preletter-assisted random-digit dialing of residential telephone numbers listed in Genesys sampling blocks in CT and NJ according to standard sampling practices, with attempts to match to cases in frequency on categories of age, gender and state. In total, 3,388 potentially eligible control subjects were identified through random digit dialing. These control candidates were contacted by the study interviewers via telephone to confirm eligibility and willingness to participate in the study. Of the 3,388, 1,243 agreed to participate and 1,166 completed the study requirements, 416 from CT and 750 from NJ. Consented study participants, both cases and controls (673 cases + 1,166 controls = 1,839), underwent telephone interviews conducted by trained study interviewers using a structured questionnaire described below. The study subjects were also instructed to mail in two saliva samples collected with commercial kits, one for DNA genotyping and one for HCV antibody assay.

Questionnaire information

Information on characteristics of study participants was ascertained through telephone interviews. During the interviews, subjects answered structured, closed-end questions that elicited information on demographic and personal characteristics (date and country of birth, gender, race and ethnicity, years of education attained, household income, and marital status), anthropometric measures (current height, current body weight, body weight 5 years in the past and during decades of ages 20s, 30s, 40s, and 50s years), lifestyle factors (amount and duration of cigarette smoking, alcohol consumption, and physical activity, and numbers of sexual partners), disease and medical histories (self-report of hepatitis B and C virus infection, diagnoses of cirrhosis, fatty liver disease, diabetes mellitus), family history of cancer (any type), and regular use (at least once per week over a duration of 3 months or more) of aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs). The telephone interviews were conducted by experienced interviewers who were trained to ascertain information equally and identically from each study participant regardless of disease status. A self-administered semi-quantitative food frequency questionnaire (FFQ), containing questions about consumption of 160 food items and supplements and about eating styles, was mailed to each subject at enrollment along with the consent materials.

Saliva collection and HCV assay

Two oral fluid samples were collected from each study participant, one for genotyping with the Saliva Self-Collection Kit (OG-250, DNA Genotek, Ottawa, Ontario, Canada) and one for HCV assay using the OraSure® oral specimen collection device (OraSure Technologies, Inc., Bethlehem, PA). The collections were conducted following the manufacturers’ instructions. For collection using the OraSure kit, study individuals placed the OraSure pad between the lower cheek and gum for 2–5 minutes, and then dropped the pad into a vial containing preservatives and sealed it for transportation at ambient temperatures to our research laboratory. After centrifugation, the supernatants of oral fluid samples were liquated into 2 ml cryovials and stored at −80°C until assay. The OraQuick HCV Rapid Antibody Test kit (OraSure Technologies, Inc.) was used for HCV assay according to the manufacturer’s protocol.

Data analysis and statistical methods

Questionnaire data were reviewed and extracted to develop exposure variables for various risk factors. Body mass index (BMI) was calculated based on the formula: weight divided by height² (kg/m²), using self-reported height and age-specific weight, and BMI values were grouped into 3 categories: normal weight (BMI<25), overweight (25≤BMI<30), and obese (BMI≥30). For smoking, cumulative pack years was calculated using the formula, average number of packs of cigarettes smoked per day multiplied by number of years smoked. Kilogram-years was used to measure cumulative amount of alcohol consumed by subjects who drank alcohol regularly. Study subjects were asked about their habits of alcohol consumption in terms of type, frequency and amount of alcohol typically consumed during ages 30 to 50 years and under the age of 30 years. For each age period, estimated grams of alcohol per serving of each type of alcohol (15.3 gram/serving of beer, 18 gram/serving of wine, and 17.6 gram/serving of spirits or mixed drinks) were multiplied by the number of servings per year and then the number of years of consumption. These numbers from each type of alcohol were summed to obtain total gram-years of alcohol consumed, and then converted to kilogram-years by dividing the sum by 1000. For physical activity, metabolic equivalent task (MET) rate was assigned to each of 18 activity items specifically described, plus other reported activities if not explicitly enumerated. Total MET values were summed across all activities by adding the products of minutes of each time (MET_minutes), times per week and weeks per year to estimate total amount of physical activity. These MET variables were further divided by 60 to convert to MET_hours.

Information on HCV infection was ascertained through both questionnaire inquiry and saliva testing. Of the 1,839 subjects enrolled in the study, 1,827 (99.3%) had HCV data from either source or both, 1,791 providing questionnaire information on HCV and 1,429 having HCV test results. Of the 1,365 individuals who had HCV information from both sources, 1,304 had consistent HCV status, and the concordant proportion among those with both types of information was 95.5%. Among the individuals whose saliva tests were negative for HCV (n=1,143), 51 (4.5%) were HCV positive based on questionnaire information, and among those whose saliva tests were positive for HCV (n=222), 10 (4.5%) were questionnaire negative. To include as many subjects in data analysis as possible, we defined HCV status (positive or negative) first based on saliva test result, and if saliva test was not available, we used questionnaire information to supplement. For subjects who had information from both sources, but the status was inconsistent (n=61), we took the positive results in either source to minimize false-negative possibility.

Descriptive statistics on each variable of interest were calculated and compared between cases and controls. For continuous data, we used the Student’s t statistic to compare means, and the Mann-Whitney U statistic for median comparison. A chi-square statistic was used for comparing categorical and ordinal data. To measure the strength and significance of associations between HCC and suspected risk factors, unconditional logistic regression models were used to calculate odds ratios (OR) and their 95% confidence intervals (CI). Multivariate logistic regression analyses were performed to adjust for covariates and potential confounding factors, including age at interview or diagnosis, gender, race (white, black, other), education (no college, completed college, graduate or professional school), body mass index (BMI, kg/m²), smoking (pack years), and alcohol intake (kilogram years). Other variables considered in the study included current marital status (married, divorced, separated, widowed, single), household income (<$50,000, $50,000-$99,999, $100,000-$149,999, $150,000-$199,999, $200,000 or more), place of birth (US, non-US), physical activity (MET), lifetime number of sexual partners, status of viral infection (HCV saliva test and reported positivity; reported HBV status), medical history (cirrhosis, fatty liver disease, diabetes), use of aspirin and other NSAIDs, and family history of cancer.

To assess the strength of associations between lifestyle factors and HCC risk while controlling for HCV infection status, we performed logistic regression analyses in subgroups of HCC patients stratified by status of HCV infection (HCV positive or negative cases vs. HCV-negative controls). The heterogeneity of the associations by HCV status was evaluated using the chi-square statistics. To assess the possible synergy between high BMI and HCV infection, we created an ordinal variable in which BMI and HCV were classified into high versus low (BMI cutoff 25) and positive versus negative, respectively, and the categories were combined into 4 groups from low to high, negative/low, negative/high, positive/low, and positive/high. In a multivariate logistic regression model, we calculated the odds ratio for each group using the negative/low group as reference, and the calculation was done with and without adjustment for covariates and confounding variables. All statistical analyses were done with SAS software (version 9.4, SAS Institute), and p-values reported are two-sided.

Results

Case and control distributions of age, gender, race, education, marital status, household income and place of birth are shown in Table 1. Larger fractions of cases than controls were male, non-white, not currently married, and born outside of the US. Also, cases were more likely than controls to have low household income and no college education. Ages of the cases at diagnosis (63.5 years) were not appreciably different from those of the controls at interview (62.8 years). The distributions of age, gender, race, education, and viral status did not show substantial differences among cases identified from CT, NJ, and NYC (Supplemental Table S2)

Table 1.

Demographic and socioeconomic features of cases and controls in the study

	Cases (n=673)		Controls (n=1166)
Variable	Number	Mean (SD)	Number	Mean (SD)
Age at diagnosis/interview	671	63.51 (8.63)	1156	62.78(10.96)
	Number	Percent	Number	Percent
Gender
Female	137	20.54	407	35.70
Male	530	79.46	733	64.30
Ethnic Group
White	481	71.79	1085	93.45
Black	94	14.03	36	3.10
Others	95	14.18	40	3.45
Education
High School or lower	317	47.17	169	14.52
College	293	43.60	610	52.41
Graduate School	62	9.23	385	33.08
Marital Status
Married	394	59.61	776	67.30
Divorced or Separated	126	19.06	158	13.70
Widowed or Single	140	21.18	216	18.73
Do not know	1	0.15	3	0.26
Household Income ($)
<$50,000	249	41.29	177	16.28
$50,000-$99,000	229	37.98	397	36.52
$100,000 or higher	125	20.73	513	47.19
Birth Place
US	555	82.59	1075	92.43
Non-US	117	17.41	88	7.57

Table 2 shows the comparisons of several exposure factors between cases and controls. As expected, 57.5% of the HCC cases were infected with HCV compared to only 1.3% of the controls. Eighteen HCC cases reported that they were infected with HBV versus 8 of the controls. Twenty-five cases had both HCV and HBV infection, while only one control had both infections. More HCC cases than controls reported histories of cirrhosis, fatty liver disease and diabetes, whereas more controls than cases reported regularly using aspirin or other NSAIDs. All of these differences remained statistically significant after adjusting for age, gender, race, education, household income and marital status. Family history of cancer did not show a significant difference between cases and controls. HCC cases reported having higher BMIs than controls at most of the time points queried, including at 5 years before the study and at each decade of adulthood from ages 20s to 50s (Supplemental Table S3). Current BMI, however, was not different between the cases and controls, but had changed from 5 years previous by an average of −0.65 kg/m² for cases vs. +0.34 for controls. Cases also had less cumulative physical activity than controls. HCC cases smoked cigarettes and drank alcohol much more than controls. High BMI, cigarette smoking and alcohol intake were all significantly associated with HCC risk, and the associations remained significant after adjusting for covariates and confounding variables (Table 3). Cases also reported to have had more sexual partners than controls, and the number of sexual partners were positively associated with HCC risk (Table 3). The analyses in Tables 2 and 3 were repeated in the CT and NJ cases only, excluding those from NYC. The results did not show substantial changes (Supplemental Tables S4 and S5).

Table 2.

Associations of HCC with known or suspected risk factors in the study

	Cases		Controls
	No.	%	No.	%	OR	95%CL	OR*	95%CL
Viral Status
Non	239	35.99	1139	97.93	1		1
HCV	382	57.53	15	1.29	121.37	71.12–207.10	109.88	59.22–203.87
HBV	18	2.71	8	0.69	10.72	4.61–24.95	7.65	2.56–22.81
HCV/HBV**	25	3.77	1	0.09	119.14	16.07–883.55	81.77	10.75–622.06
Cirrhosis
No	217	32.39	1163	99.74	1		1
Yes**	453	67.61	3	0.26	808.37	257.52–2537.52	886.76	275.00–2859.43
Fatty Liver
No	555	83.58	1134	97.42	1		1
Yes	109	16.42	30	2.58	7.42	4.89–11.26	7.55	4.64–12.30
Diabetes
No	449	66.71	1020	87.48	1		1
Yes	224	33.29	146	12.52	3.48	2.75–4.41	3.00	2.24–4.02
Aspirin
No	466	71.47	663	57.50	1		1
Yes	186	28.53	490	42.50	0.54	0.44–0.66	0.39	0.30–0.52
NSAID
No	609	93.40	982	85.17	1		1
Yes	43	6.60	171	14.83	0.41	0.29–0.56	0.44	0.29–0.67
Family Cancer History
No	287	44.50	463	40.37	1		1
Yes	358	55.50	684	59.63	0.84	0.70–1.03	0.90	0.71–1.15

^*Adjusted for age, gender, race, education, household income, and marital status.

^**The number of subjects with the condition in the controls was small and the result was unstable due to the small number.

Table 3.

Associations of HCC with lifestyle risk factors in the study

	Cases		Controls
	No.	%	No.	%	OR	95%CL	OR*	95%CL
BMI (current)
normal (<25)	208	31.95	383	33.25	1		1
overweight (25–29)	265	40.71	469	40.71	1.04	0.83–1.31	0.90	0.68–1.19
obese (≥30)	178	27.34	300	26.04	1.09	0.85–1.40	0.90	0.66–1.23
BMI (5 years in past)
normal (<25)	120	24.84	344	34.85	1		1
overweight (25–29)	209	43.27	404	40.93	1.48	1.14–1.94	1.37	0.98–1.91
obese (≥30)	154	31.88	239	24.21	1.85	1.38–2.47	1.79	1.25–2.56
BMI in 20s
normal (<25)	411	64.83	834	73.67	1		1
overweight (25–29)	179	28.23	240	21.21	1.51	1.21–1.90	1.43	1.07–1.90
obese (≥30)	44	6.94	58	5.12	1.54	1.02–2.32	1.36	0.83–2.25
BMI in 30s
normal (<25)	320	50.40	706	62.59	1		1
overweight (25–29)	245	38.58	324	28.72	1.67	1.35–2.06	1.50	1.15–1.97
obese (≥30)	70	11.02	98	8.69	1.58	1.13–2.20	1.33	0.88–1.99
BMI in 40s
normal (<25)	224	35.50	547	49.37	1		1
overweight (25–29)	267	42.31	403	36.44	1.62	1.30–2.01	1.48	1.13–1.95
obese (≥30)	140	22.19	150	13.64	2.28	1.73–3.01	1.75	1.24–2.47
BMI in 50s
normal (<25)	153	24.96	376	37.26	1		1
overweight (25–29)	262	42.74	446	44.21	1.44	1.13–1.84	1.31	0.97–1.78
obese (≥30)	198	32.30	187	18.53	2.60	1.98–3.42	2.13	1.52–3.00
Physical Activity (MET_hours)**
Low	88	33.46	265	31.40	1		1
Mid	108	40.68	285	33.77	1.14	0.82–1.58	1.46	0.98–2.18
High	67	25.86	294	34.83	0.69	0.48–0.98	0.99	0.63–1.54
Physical Activity (MET_hours/week)**
Low	87	33.08	270	31.99	1		1
Mid	96	36.50	295	34.95	1.01	0.72–1.41	1.17	0.78–1.76
High	80	33.06	279	33.06	0.89	0.63–1.26	1.09	0.71–1.69
Physical Activity (MET_hours/year)**
Low	81	30.80	275	32.58	1		1
Mid	79	30.04	288	34.12	0.93	0.66–1.32	1.06	0.69–1.61
High	103	39.16	281	33.29	1.24	0.89–1.74	1.51	0.99–2.30
Alcohol intake under age 30 (kilogram years)**
Low	52	11.50	149	33.56	1		1
Mid	72	15.93	143	32.21	1.44	0.94–2.21	1.53	0.90–2.60
High	328	34.23	152	34.23	6.18	4.27–8.95	4.65	2.91–7.43
Alcohol intake ages 30–50 (kilogram years)**
Low	51	11.62	197	33.39	1		1
Mid	51	11.62	199	33.73	0.99	0.64–1.53	0.97	0.56–1.68
High	337	76.77	194	32.88	6.71	4.71–9.57	5.46	3.45–8.64
Alcohol intake life time (kilogram years)**
Low	55	11.36	207	33.01	1		1
Mid	65	13.43	213	33.97	1.15	0.77–1.73	1.04	0.63–1.71
High	207	75.21	207	33.01	6.62	4.70–9.32	5.12	5.33–7.89
Smoking (pack year)**
Low	84	17.72	148	32.96	1		1
Mid	111	23.42	144	32.07	1.36	0.94–1.96	1.36	0.87–2.12
High	279	58.86	157	34.97	3.13	2.25–4.36	2.61	1.73–3.95
Sexual Partners
<3	125	20.76	337	31.88	1		1
≥3	477	79.24	720	68.12	1.79	1.41–2.26	1.84	1.34–2.51

^*Adjusted for age, gender, race, education, household income, and marital status.

^**Grouped based on the tertile distribution in the controls.

In the subgroup analyses stratified by HCV infection, BMI associations with HCC risk were more evident in HCV negative patients than HCV positive ones, adjusted for age, gender, race, education, household income and marital status (Figure 1). Evaluation of heterogeneity across HCV strata indicated that the associations of BMI with HCC significantly differ by HCV status when BMI was assessed at ages 30s, 40s, and 50s, or 5 years prior to the study, p=0.042, <0.001, 0.005, and 0.005, respectively. Comparing BMI from ages 20s to 50s, controls always had the lowest mean BMI throughout the age periods. HCC cases without HCV infection had the highest BMI, and HCC cases with HCV infection were in the middle (Supplemental Figure S1). Compared to subjects whose BMIs over age were always under 25, individuals had over 100% higher HCC risk if they were always overweight or obese from ages 20s through 50s, and 43% higher risk if they were overweight or obese only after their age 40s (Table 4). The association with accumulative overweight/obesity was more substantial for HCV-negative cases than positive, OR=3.14 vs. 1.72 for individuals who were always overweight or obese, and OR=1.80 vs. 1.36 for those who were overweight or obese after their age 40s (Table 4).

Figure 1.

Odds ratios for HCC in association with BMI ascertained at different ages stratified by HCV infection and adjusted for age, gender, race, education, household income and marital status.

Table 4.

Associations of HCC with BMI change over time

	All Cases		HCV-negative Cases		HCV-positive Cases
	OR	95%CL	OR	95%CL	OR	95%CL
BMI change over time
Group 1 (n=469)	1		1		1
Group 2 (n=228)	1.70	1.22–2.38	1.81	1.07–3.06	1.74	1.18–2.57
Group 3 (n=429)	2.42	1.83–3.19	3.31	2.18–5.03	2.12	1.52–2.94
Group 4 (n=449)	1.83	1.39–2.42	2.23	1.45–3.42	1.74	1.26–2.42
BMI change over time*
Group 1 (n=450)	1		1		1
Group 2 (n=216)	1.43	0.93–2.18	1.80	0.97–3.31	1.36	0.812–2.26
Group 3 (n=406)	2.02	1.41–2.90	3.14	1.90–5.19	1.72	1.10–2.70
Group 4 (n=421)	1.60	1.13–2.28	2.06	1.25–3.40	1.56	1.01–2.40

^*Adjusted for age, gender, race, education, household income, and marital status.

Group 1: BMI<25 in ages 20s, 30s, 40s and 50s.

Group 2: BMI<25 in ages 20s and 30s and ≥25 in ages 40s and 50s.

Group 3: BMI≥25 in ages 20s, 30s, 40s and 50s.

Group 4: BMI changed over time other than Group 1–3.

In assessing the synergy between HCV infection and high BMI, we found that the estimated HCC risks were 82-fold for normal weight individuals with HCV infection compared to those without infection, and 2-fold for overweight or obese individuals without HCV infection compared to those with normal BMI. A possible synergy was suggested for these risk factors as the odds ratio for HCC was about 149-fold greater in subjects with both factors (overweight or obesity plus HCV infection) compared to those with neither (Table 5). HCC patients without either obesity or HCV infection had a median age of diagnosis at 66 years, compared to a median age of 64 years among the obese only, 60 years among HCV-positive patients only, and 57 years for patients with both obesity and HCV infection. A nearly 10-year difference in age at diagnosis of HCC was observed between patients with neither obesity nor HCV infection and those with both risk factors (Supplemental Figure S2).

Table 5.

Associations of HCC with high BMI and HCV infection individually and jointly

	Cases	Controls	OR	95%CI	OR*	95%CI*
Virus−/BMI<25	132	746	1		1
Virus−/BMI≥25	83	225	2.06	1.53–2.85	2.06	1.42–2.98
Virus+/BMI<25	265	18	83.20	49.86–138.84	81.66	45.18–147.60
Virus+/BMI≥25	118	4	166.72	60.51–459.34	149.41	52.50–425.16

^*Adjusted for age, gender, race, education, household income, and marital status.

Cigarette smoking (pack years) and alcohol intake (kilogram years under 30 years of age, between ages 30–50 years, and across the lifetime) were all associated with HCC risk, both in HCV-positive and -negative cases, but the associations were stronger in patients with than without HCV infection (data not shown). Histories of fatty liver disease and diabetes were also significantly associated with HCC risk in both HCV-positive and -negative patients, but the magnitudes of these associations were greater in HCV-negative than -positive cases (data not shown). The study further observed that aspirin use and other NSAID use were inversely associated with risks of HCC, and these associations did not appear to differ between patients with and without HCV infection.

Discussion

This study confirmed that HCV infection is strongly associated with HCC in the US, and that overweight and obesity are risk factors for HCC. Based on the analysis of HCV and BMI joint effect as well as comparison of patient age at diagnosis, we found that high BMI and HCV infection might have a synergistic interaction on HCC. We also observed that conditions linked to obesity, fatty liver disease and diabetes, were associated with increased risks of HCC. These associations were stronger for cases without than with HCV infection and are consistent with the literature [20, 21], supporting the relevance of obesity in the etiology of HCC. Our study confirmed that cigarette smoking and high cumulative alcohol intake were risk factors for HCC, regardless of HCV infection status. Our data also indicated that compared to the control subjects HCC cases attained less education, and were more likely to be in low-income circumstances and less likely to be married. These factors may influence their choice of unhealthy lifestyles involving cigarette smoking, heavy alcohol consumption and sexual related behaviors that increase risk of viral infections. Overall, the study found that lifestyle factors, education and socioeconomic status were associated with HCC risk in the US.

The association between obesity and HCC risk has been demonstrated in a number of human studies [22–28], and pathogenic pathways that link obesity to HCC have been delineated in human [29–33] and animal models [34–38] regardless of HCV infection. The process starts with insulin resistance and non-alcoholic fatty liver disease (NAFLD) that slowly progresses to non-alcoholic steatohepatitis (NASH) and fibrosis in some cases, which further advances to HCC with or without cirrhosis as an intermediate [15, 22, 39, 40]. This carcinogenic mechanism has a significant implication for public health because 25% to 45% of people in the world are affected by obesity, including 75 to 100 million Americans [41, 42]. From a population perspective, this high prevalence renders obesity and NAFLD important risk factors for HCC in the US in addition to HCV infection. Using data from the SEER registry, Makarova-Rusher et al. estimated population attributable fractions for risk factors of HCC and found that metabolic disorders (32%), which include obesity, diabetes, impaired glucose metabolism, metabolic syndrome and NAFLD, were higher than HCV (20.5%) and other risk factors, including alcohol (13.4%), smoking (9%), HBV (4.3%) and genetic disorders (1.5%) [43]. However, HCV status may not be well characterized in the SEER data, and based on an infection frequency of 1.29% in our controls and an estimated odds ratio of 110, HCV would account for 58.4% of HCC on a population basis using the estimates from our study. Our study also demonstrated that obese individuals infected with HCV had a much higher odds ratio than those with any of the factors and such patients had a nearly 10-year earlier diagnosis of HCC compared to those with neither obesity nor HCV, suggesting a possible synergy between the two major risk factors in the pathogenesis of HCC.

Most epidemiologic studies of HCC have been conducted in Asian and European populations; relatively few population-based studies have been carried out on HCC among Americans [44, 45]. Our study was a relatively large investigation in which incident HCC cases were identified and recruited mostly through statewide tumor registries in CT and NJ. Population control individuals in the study were also found and enrolled statewide through pre-letter-enhanced random digit dialing methods. Thus, the cases and controls in our study may be considered approximate representations of general populations of cases and controls in the northeast US. To make the cases and controls comparable, we tried frequency matching on age, gender and race, but the extremely skewed distributions of patient gender and race made it difficult to achieve adequate matching for the last two factors. Given these differences, we used unconditional logistic regression methods with adjustment for age, gender and race to minimize the remaining impact of these covariates on study results. We also performed sensitivity tests by limiting data analysis to males or to whites only, and the results did not show substantial changes. In addition to the matching issue, our response fractions for cases and controls were lower than desired, potentially allowing factors associated with participation to have affected the results. We also could not blind the disease status during interview due to the severity of the disease and protocol requirement to reveal the sources of study participants being identified. As a retrospective study, our case-control comparison inherits other limitations, such as potential recall bias between cases and controls as well as collection of largely questionnaire-based information without confirmation from other sources. In addition, some of the data analyses, such as the BMI and HCV synergy, were based on small numbers of subjects, which could generate unstable results. Thus, caution should be exercised when interpreting some of the results. Despite the limitations, our study findings are remarkably consistent with the literature.

Besides obesity and diabetes, the known major risk factors for HCC were confirmed in our study, including HCV infection, cigarette smoking and alcohol intake. Leisure-time physical activity was recently reported to be inversely associated with liver cancer risk [46]. Evidence in our study was not entirely clear although the cases had lower MET_hours than controls. As expected, HCV infection is the strongest risk factor for HCC in the US, and 58% of our cases tested positive for HCV antibodies, compared to only 1.3% of control subjects. This fraction of HCV infection in HCC is close to other estimates that 50% of HCC cases in the US are infected with HCV [14, 47]. The odds ratio for HCV was quite high in our study, and it did not change substantially after we excluded the cases from NYC. Hassan et al. reported an odds ratio of 79.2 for HCV in a similar case-control study [48]. For HBV infection, we only ascertained status information through questionnaire interview since a valid saliva assay was not available for testing. As anticipated, the proportion of HBV infection was quite low in the US, only 2.7% in the cases and 0.7% in the controls. In addition, 25 cases and 1 control reported having infection with both HCV and HBV (3.8% versus 0.09%). In our study, HCV status was determined based on information from saliva test and questionnaire inquiry. To exclude potential influence of recall bias, we also analyzed the data only among subjects whose HCV status was determined by saliva test. The results of these analyses did not differ from those of all the subjects, suggesting that including subjects with questionnaire data on HCV infection status did not materially affect the study findings.

Our study showed overall that HCC cases differed substantially from control subjects with respect to socioeconomic status and related behaviors. Compared to the controls, the cases were more likely to have attained only high school education or lower (47.2% vs. 14.5%), less likely to be married (59.6% vs. 67.3%) or to have high household income (20.7% vs. 47.2%). We also observed that more controls than cases reported the regular use of aspirin and of other NSAIDs, suggesting that anti-inflammatory drugs may reduce the risk of HCC development. An inverse association between aspirin use and HCC risk has been reported previously. In a liver-cancer pooling project involving 10 prospective cohort studies, Petrick et al. observed a 32% reduction in HCC risk in current aspirin users compared to non-users, with risk reduction more evident among daily, long-term and low-dose users [49]. The inverse association was independent of alcohol consumption, obesity and diabetes history, and not affected by hepatitis virus infection.

An earlier study showed a 41% reduction in HCC risk among aspirin users [50]. In our study, the estimated reduction in risk was 57% and was more substantial in HCV-positive than in HCV-negative cases (64% vs. 45%). We also found that regular NSAID users, though fewer than aspirin users, had a similar reduction in HCC risk. This association, however, was not observed in the previous studies [49, 50].

In summary, our study showed that individuals with HCC differed substantially from the general population in the US with respect to their backgrounds and lifestyles. HCC cases were more likely to have low socioeconomic status and limited education, and were more closely associated with unhealthy lifestyles, such as cigarette smoking, heavy alcohol consumption, multiple sexual partners, hepatitis viral infection, obesity and diabetes. Thus, to reduce the HCC burden in the US, we must focus on these factors collectively for prevention. Liver transplant and resection are the only curative therapies for HCC, but most patients do not qualify for these treatments and inadequate donor livers limit the ability to treat. Efforts at early detection and prevention are clearly necessary to improve outcomes and decrease the incidence of HCC. Despite effective therapies for hepatitis B and C, the incidence of HCC continues to rise. This study shows that a comprehensive public education strategy to promote healthy lifestyle choices including weight management, exercise, smoking cessation and limited alcohol use will be necessary to truly have an impact on prevention of liver cancer.

Supplementary Material

10552_2020_1277_MOESM2_ESM

Supplemental Figure S2. Median and mean ages of HCC patients grouped by HCV infection and obesity status.

10552_2020_1277_MOESM1_ESM

Supplemental Figure S1. Average BMI at ages 20s, 30s, 40s and 50s in HCC patients without HCV infection (top line), HCC patients with HCV infection (middle line) and population controls (bottom line).