Community-Level Factors Associated with Hepatocellular Carcinoma Incidence and Mortality: An Observational Registry Study

Hepatocellular carcinoma (HCC) accounts for over 75% of all primary liver cancers and is a growing concern worldwide (1, 2). In the United States, mortality from liver cancer was rising for decades and only recently stabilized (3). Despite this relative improvement, HCC continues to carry a poor prognosis, ranked alongside pancreatic and esophageal cancer for survival (3).

In the United States, HCC usually develops in the setting of chronic liver disease attributable to viral hepatitis B and/or C, chronic alcohol consumption, or metabolic dysfunction–associated steatotic liver disease (formerly nonalcoholic fatty liver disease), with additional risk associated with obesity and diabetes (1, 2, 4). Disparities in access to healthcare represent another risk factor for HCC and associated mortality (5–10). While there are established HCC screening and surveillance guidelines (4, 11, 12), only 43% of all HCC cases are diagnosed at early stages eligible for curative therapy (3, 13). This is in part due to inadequate screening among individuals at risk of HCC: only about 1 in 4 eligible persons receives the recommended screening (14, 15).

Most studies of socioeconomic disparities in HCC use national cancer registries including Surveillance, Epidemiology, and End Results (SEER; ref. 6), but there are geographic, political, and historical drivers of health trends in the United States (16, 17). For example, access to treatment for hepatitis C virus (HCV), liver transplant listing, and transplant waitlist mortality differ in states that expanded Medicaid under the Affordable Care Act compared with states that did not (18–20). There are also robust state by state differences in liver cancer incidence (21).

Interventions aimed at eliminating disparities depend upon understanding local context (22, 23). For example, Zhou and colleagues (10) identified geographic “hotspots” for late-stage HCC within Los Angeles County and mapped their relationships to nearby health facilities. Identifying high-risk communities has the potential to reduce HCC disparities because screening for HCC is associated with earlier diagnosis, curative treatment, and improved survival (24). Screening is cost-effective in populations with liver cirrhosis or chronic hepatitis B virus (HBV; refs. 24, 25). The cascade of HCC care therefore optimally begins with identifying at-risk individuals (4).

We undertook this study to understand HCC disparities in the state of Connecticut, where our medical center is located. Connecticut has an incidence of liver cancer around the national average (21) and encompasses a variety of rural, urban, and commuter communities with a diverse population. Connecticut is also among the states with the highest wealth inequalities in the nation (26). Our primary aims were to identify communities with the highest incidence of HCC, evaluate for disparities in HCC outcomes at the community level, and to assess community-level factors associated with observed disparities. Our overarching goal is to improve understanding of population health factors associated with HCC incidence to inform resource allocation for potential health system interventions. The results we present are deliberately specific to one state, but this methodology can be applied to other localities.

This is a retrospective observational study using a state tumor registry and publicly available data. This study was reviewed and approved by the State of Connecticut Department of Public Health Human Investigations Committee (Protocol #954) and the authors’ Institutional Review Board (Protocol ID 2000031228) and in accordance with the Declaration of Helsinki and the U.S. Common Rule.

New diagnoses of HCC were identified from the Connecticut Tumor Registry, a population-based registry of cancer incidence and outcomes (available from: https://portal.ct.gov/dph/Tumor-Registry/CTR-Home). The registry includes all reported cancers diagnosed in Connecticut residents since 1935, as well as treatment and survival data on reported cases. The Connecticut Tumor Registry has been one of nineteen SEER reporting sources since 1973 based on its high-quality data. The identities of all patients are protected by state confidentiality laws. The registry collects date of diagnosis, primary tumor site, method of diagnosis (e.g., imaging, clinical diagnosis, histology), histology codes, reporting source (e.g., medical provider, death certificate), stage at diagnosis, sex, age, race and Hispanic ethnicity, birthplace, primary payer (health insurance), census tract of residence, comorbid health conditions and secondary diagnoses per the North American Association of Central Cancer Registries (NAACCR; ref. 27), history of HBV and/or HCV, date of last contact, vital status, primary cause of death, and survival time from diagnosis.

A total of 3,542 cases of liver cancer captured by the Connecticut Tumor Registry between 2008 and 2019 were evaluated for inclusion in the analytic sample. The ICD-O-3 primary site code C220 (liver) and ICD-O-3 histology codes 8170–8175 (HCC) were used to identify HCC diagnosed using histology. Liver tumors with histology other than HCC (n = 362), occurring in persons under age 18 (n = 35), those reported by death certificate only (n = 69), or with no imaging or histologic confirmation (n = 295) were excluded. The final sample consisted of 2,781 HCC cases diagnosed by imaging and/or histologic confirmation. Demographic information including census tract of residence, age, sex, race, Hispanic ethnicity, and primary insurance, and clinical information including comorbid conditions, history of HCV and/or HBV, vital status, and date of death were obtained for each person diagnosed with HCC. Year of diagnosis, census tract of residence, and age at diagnosis were available for all HCC cases. Race was unknown for 3 cases (0.1%), primary insurance was missing for 117 (4.2%), and cancer stage at diagnosis was missing for 474 individuals (17.0%).

The registry records race and Hispanic ethnicity using a combination of medical records and a supplemental NAACCR algorithm that derives Hispanic ethnicity based on name, place of birth, and sex (28). Individuals were categorized as non-Hispanic White, non-Hispanic Black, Hispanic, Asian, American Indian/Alaskan Native, or Other. Other included those who identified as “some other race” (n = 51), whose race was unknown (n = 3), and Hawaiian/Pacific Islander (n = 1).

The American Joint Commission on Cancer (AJCC) manual stages liver cancers from I (least advanced) to IV (most advanced). Stage I indicates a solitary tumor without vascular invasion, stage II indicates a solitary tumor with vascular invasion or multiple tumors <5 cm in size, stage III indicates multiple tumors with at least one >5 cm or invasion of a tumor involving a major branch of the hepatic or portal vein, and stage IV indicates any tumor with distant metastasis (29). The AJCC6 stage was used for this study.

The U.S. Census Bureau publishes tables mapping census tracts to ZIP code tabulation areas (ZCTA; ref. 30). Using census tract of residence, we localized all HCC cases to the appropriate ZCTA at time of diagnosis. Cases were aggregated at the ZCTA level to align with the unit of geography employed by the Census (31).

Community descriptors at the ZCTA level were extracted from publicly available Census and Centers for Disease Control and Prevention (CDC) data. We used the 5-year estimates published in the 2013 American Community Survey (the midpoint of our observation period) to estimate the total adult population (32) and distributions of race and Hispanic ethnicity (33), educational attainment (34), poverty status (35), and health insurance (32) in each ZCTA. The prevalence of individual-level health behaviors at the ZCTA level was obtained using the 2022 CDC PLACES dataset, which published health information from the 2020 Behavioral Risk Factor Surveillance System. This was the first year that the PLACES dataset included all ZCTAs.

We classified each ZCTA as urban, commuter, micropolitan, or rural using the U.S. Department of Agriculture rural–urban commuting area (RUCA) codes (36). Metropolitan areas with population flow within an urbanized area were classified as urban. Metropolitan areas with population flow to an urban area were classified as commuter. Micropolitan areas were classified as micropolitan. Rural areas with primary population flow to an urban area or urban cluster were classified as rural. These codes and their classification for this analysis are shown in Supplementary Tables S1 and S2. We calculated the percent of adults identifying as non-Hispanic Black, non-Hispanic White, and Hispanic, which are provided as mutually exclusive categories in the American Community Survey estimates, the percent of adults living in poverty (defined as < 138% the federal poverty limit, the threshold set by the Department of Health and Human Services for Medicaid eligibility under the Affordable Care Act), the percent of adults with less than a high school education (grade completion less than 12), the percent of adults with Medicaid insurance, with private insurance, and the percent of adults age 18 to 64 (i.e., not of age to qualify for Medicare) without health insurance. For the description of trends in HCC over time, calendar time was classified as before (2008–2014) or after (2015–2019) introduction of direct-acting antivirals (DAAs) for HCV.

The percent of adults reporting binge drinking in the last 30 days (5 or more drinks for men or 4 or more drinks for women in a single occasion), current smoking (smoking some or all days), diagnosis with obesity (body mass index ≥30), diagnosis with diabetes (ever being told they have diabetes), and a preventative health visit within the past year were extracted from the PLACES dataset at the ZCTA level.

We summed the total number of HCC cases per ZCTA over the study period to calculate the average annual incidence of HCC per 100,000 adults by ZCTA. Those ZCTAs with fewer than 1,000 adults were excluded due to concern for unreliable incidence estimates based on small denominators. We described overall trends in HCC diagnosis rates over time and used interrupted time series analysis to examine associations with the introduction of DAAs to treat HCV. This was performed using an indicator variable for before versus after introduction of DAA therapy, time (in years), and time since DAA introduction (in years).

The average annual incidence of HCC and community-level factors significantly associated with average annual incidence of HCC (education, poverty, and rurality) were mapped at the ZCTA level to demonstrate spatial trends. In the map showing average annual incidence, adjacent ZCTAs with fewer than 10 total HCC cases over the study period were blended, combining cases and total adult population, to comply with confidentiality regulations. Maps are color-coded according to percentile: 0–25th percentile (lightest shade), 25–50th percentile, 50–75th percentile, 75–90th percentile, >90th percentile (darkest shade). The absolute values corresponding to these percentiles are displayed in each map's figure legend.

We classified all ZCTAs according to average annual incidence of HCC by quartile and identified those with the highest (top 25% and top 10%) and lowest (bottom 25% and bottom 10%) average annual incidence of HCC. We classified all ZCTAs by percent of adults with less than a high school education and percent of adults living in poverty, the two community-level factors that remained associated with average annual incidence of HCC in the multivariable model described below, using quartiles and top and bottom 10%. We compared age and stage at HCC diagnosis and overall survival after diagnosis across these groups of ZCTAs using ANOVA and χ² tests.

We used bivariate and multivariate linear regression to evaluate associations between community-level characteristics (independent variables) and average annual incidence of HCC (dependent variable) at the ZCTA level. Predictors included community type, racial and ethnic composition, health insurance, poverty, education level, and the percent of adults reporting binge drinking, smoking, diabetes, obesity, and a recent preventative health visit. These predictors were selected a-priori based on established associations with general health and HCC outcomes (1, 2, 6, 23, 37, 38). To allow for ease of comparison of relative effect size, we used standardized predictors for the multivariate model. The regression coefficients represent the change in average annual HCC cases per 100,000 adults per SD change in the predictor.

Analyses were performed using the statistical software SAS 9.5. Geocoding for maps was performed using the ArcGIS (Geographic information system).

Raw data for this study were generated by the Connecticut Tumor Registry. Derived data supporting the findings of this study are available from the corresponding author upon request.

A total of 2,781 cases of HCC was observed. The number of new diagnoses per year varied from 169 in 2009 to 289 in 2015 and are displayed in Fig. 1, with new HCC diagnoses per 100,000 adults on the Y-axis and calendar year on the X-axis. The interrupted time series analysis showed a significant positive trend in HCC incidence per 100,000 adults over time (0.3875 per calendar year, P < 0.001) and a significant negative trend in HCC incidence per 100,000 adults over time after DAAs for HCV were introduced (−0.7206 per year after introduction of DAAs; P < 0.001). Annual HCC incidence per 100,000 adults predicted by the interrupted time series model is shown in Fig. 2. Stratified by community type (urban, commuter, micropolitan, or rural), these trends were significant only in urban communities (Supplementary Table S3).

The overall average annual incidence of HCC during the study period was 8.9 cases per 100,000 adults and varied widely across ZCTAs, from 0 to 97.7 per 100,000 adults. The average annual incidence per 100,000 adults by ZCTA is displayed in Fig. 3A, with the lowest incidence ZCTAs shown in the lightest shade and the highest incidence ZCTAs in the darkest shade. The geographic distributions of the percent of adults with less than a high school education, the percent of adults living in poverty, and rural ZCTAs are shown in Figs. 3B–D.

Using the distribution of race and ethnicity in the state to calculate expected values, the total number of HCC cases diagnosed during the study period was lower than expected in the non-Hispanic White population (1,852 observed vs. 1,961 expected), and higher than expected in the Hispanic (450 observed vs. 386 expected), and Black (303 observed vs. 262 expected) populations (P value for χ² test < 0.001).

The sociodemographic characteristics of ZCTAs with the highest and lowest incidence of HCC differed along lines of race, ethnicity, education, poverty, and health insurance. Compared with communities with the lowest incidence of HCC, communities with the highest incidence of HCC had, on average, more Black and Hispanic residents, more residents with less than a high school education, without insurance or with Medicaid insurance, and more residents who lived in poverty. Over 60% of ZCTAs with the top 25% average annual HCC incidence were urban (Table 1).

Individuals diagnosed with HCC who lived in ZCTAs in the top quartile for annual HCC incidence were diagnosed on average 3 years younger than those living in ZCTAs in the lowest quartile (overall ANOVA P < 0.001). Comparing ZCTAs in the top 10% for annual HCC incidence to those in the bottom 10%, individuals living in the highest incidence ZCTAs were less likely to be alive at 1 year (47.0% vs. 64.7%; P = 0.02), 2 years (30.0% vs. 47.1%; P = 0.03), and 5 years (10.1% vs. 27.3%; P = 0.02) after diagnosis (Fig. 4A).

After diagnosis, people living in ZCTAs in the highest quartile for HCC incidence had an average survival of 24 months, while average survival for those living in ZCTAs in the lowest HCC incidence quartile was 27 months, a difference that was not statistically significant. However, individuals living in ZCTAs in the top 10% for HCC incidence survived an average of 23 months after diagnosis compared with 34 months for those living in ZCTAs in the lowest 10% (F = 4.23; P = 0.04).

Comparing ZCTAs by education, people living in ZCTAs with a lower percent of adults who had graduated high school were more likely to be diagnosed with stage IV disease (Fig. 4B). Individuals diagnosed with HCC living in ZCTAs where fewer adults had a high school education were less likely to be alive at 1 and 2 years after diagnosis. Comparing ZCTAs in the bottom 10% for high school graduation to those in the top 10%, individuals living in the ZCTAs with the lowest high school graduation rates were less likely to be alive at 2 years (28.7% vs. 43.9%; P = 0.02) and 5 years (10.7% vs. 21.4%; P = 0.01) after HCC diagnosis (Fig. 4B).

Comparing ZCTAs by percent of adults living in poverty, individuals living in ZCTAs with poverty levels in the top quartile were more likely to have stage IV HCC at time of diagnosis (12.1%) than those living in ZCTAs with poverty levels in the bottom quartile (7.3%). Although there was no difference in overall survival at 1 year after diagnosis, individuals living in ZCTAs in the top quartile for poverty were less likely to be alive at 2 and 5 years after diagnosis (Fig. 4C).

In bivariate models, almost all community-level factors were significantly associated with community-level HCC incidence: higher percent of residents identifying as White, with private insurance, and reporting binge drinking were associated with lower average annual HCC incidence while higher percent of residents identifying as Black or Hispanic, with less than a high school education, with no health insurance or Medicaid insurance, living in poverty, currently smoking, with diabetes, and with obesity were associated with higher average annual HCC incidence. Rural community type was also associated with higher average annual HCC incidence (Table 2, first column of results; Supplementary Figures S1–S14). Community-level predictors were highly correlated (Supplementary Table S4). In the multivariate model, percent with less than a high school education [increase of 6.73 cases/100,000 adults per 1 SD increase; 95% confidence interval (CI), 2.22–11.24], percent of adults living in poverty (increase of 3.45 cases/100,000 adults per 1 SD increase; 95% CI, 0.41–6.49), and rural community type (increase of 15.5 cases/100,000 adults; 95% CI, 6.26–24.70) remained significantly associated with HCC incidence. (Table 2, second column of results.)

In this study, we used a state Tumor Registry, U.S. Census, and CDC data to evaluate for disparities in HCC incidence and outcomes in Connecticut and to identify drivers of these disparities at the community level. Using a geo-localization approach, we identified wide variation in HCC incidence and strong associations between community-level education and poverty levels, community type, and HCC incidence. We also found that individuals diagnosed with HCC living in communities with the highest HCC incidence were diagnosed at a younger age and had poorer overall survival compared with those living in the lowest incidence ZCTAs, suggesting that allocating resources to improve identification of persons at increased risk of HCC and screening in high incidence communities may improve disparities in HCC outcomes.

A previous study using national-level SEER data found that individuals from rural regions were diagnosed with later stage disease and had greater HCC mortality than those in large metropolitan areas (39). In Connecticut, urban ZCTAs constituted the majority of those with high HCC rates (per 100,000 adults) in addition to representing the greatest absolute number of HCC cases. However, of the 16 rural ZCTAs with at least 1,000 adults, all were in the top 25% annual HCC incidence and 5 were in the top 10%. This finding underscores the heterogeneity of populations at increased risk of HCC and the importance of understanding local trends.

We observed that non-White racial and ethnic minorities were over-represented in diagnoses of HCC relative to their population in Connecticut, recapitulating disparities in HCC incidence previously observed (3, 5, 6, 8). A recent study in Texas found significantly higher HCC incidence in communities with greater representation of individuals who were Black and Hispanic (40). At the community level, we observed similar trends in unadjusted associations between community racial composition and HCC incidence, but after accounting for other community-level factors (education, poverty, community type), this association did not persist. In our state, associations between race, ethnicity, and HCC incidence at the community-level appear to be due to the clustering of persons belonging to minority racial groups in communities with high poverty and low education levels.

Risk factors for HCC segregate differentially according to race and ethnicity. For example, HCV is more prevalent among Black individuals and HBV among people from Asia, while smoking is more common among Native American persons (1, 2, 41). The information available on alcohol consumption was limited to recent binge drinking and did not fully capture the spectrum of alcohol use at the population level. Binge drinking is more common in young adults (42), and the association between binge drinking and HCC risk at the population level here is likely confounded by age. Future studies of HCC risk would benefit from combining pertinent information from both epidemiologic and clinical data sources to better characterize risk factors for HCC.

Our study has several limitations. We report the raw cumulative incidence of HCC by ZCTA as opposed to estimates weighted to account for variations in population risk factors. This is consistent with our aim to identify communities with the highest absolute rates of HCC per total adult population. Variation around population estimates from the U.S. Census may impact the reliability of incidence calculations. We used the ZCTA at time of diagnosis and did not account for migration into or out of ZCTAs during the study period, which aligned with our focus on incidence as the primary outcome. We were also limited to clinical data available in the Connecticut Tumor Registry. Although HCV and HBV history were recorded, these data relied on medical records; as a result, they were incomplete and those with a known history of viral hepatitis were not further classified with respect to whether these viruses were cleared (spontaneously or with treatment). Notwithstanding these limitations, the Connecticut Tumor Registry is one of the oldest in the nation and has been selected to contribute to SEER since its inception due to the high quality of the registry's surveillance data. The method we describe here may not be as reliable in states with less experience maintaining accurate cancer registries.

The study period (2008–2019) included the introduction of DAAs to treat HCV in 2014. We observed an increase in HCC diagnosis rates in 2015, likely due to increased HCV testing and liver imaging prompted by the availability of a novel, effective treatment for HCV. After 2015, the rate of new HCC diagnoses declined, which may be partially explained by reduced HCC risk among those who achieved HCV eradication with DAA therapy. Immunotherapies for advanced HCC were also introduced during the study period in 2017 (nivolumab), 2018 (pembrolizumab), and 2020 (ipilimumab/bevacizumab) and may have contributed to longer overall survival after diagnosis in later years. However, we did not have adequate follow up time after introduction of these immunotherapies to formally assess for trends.

The most prevalent risk factor for HCC in the United States is liver cirrhosis. Cirrhosis is often “silent” for many years, lacking clinical signs or symptoms until complications develop. Simple noninvasive tools like the FIB-4 score, which is calculated using age, platelet count, and hepatic transaminases, can be used to risk stratify patients for possible cirrhosis requiring further evaluation. This methodology has been used to identify previously undetected chronic liver disease, and decision support tools can be used to prompt further evaluation among those with high-risk values (43, 44). Only a minority of eligible patients in the United States, including those in regular medical care, undergo recommended HCC screening due to process failure in providers identifying patients at risk and ordering screening tests and patients completing the recommended testing (14, 15). Potential interventions to address these process failures include tools that leverage the electronic medical record to prompt further testing in patients likely to have cirrhosis and, among those with chronic HBV and cirrhosis, to track HCC screening. Similar electronic tools have been used to accurately track colorectal cancer screening status (45). Resources to implement such interventions can be targeted to communities with the highest risk of poor HCC outcomes identified by the associations measured here.

Our study shows that State Tumor Registries can be combined with community-level data to provide valuable information on the social determinants of HCC incidence and outcomes. Our study suggests that in Connecticut, the major drivers of disparities in HCC incidence and prognosis are socio-economic. While the greatest absolute concentration of HCC cases was in urban communities, we also observed high relative incidence of HCC in rural communities with small populations. These heterogeneous findings underscore the importance of understanding local trends to address unequal outcomes in HCC. Survival after HCC diagnosis was worse in communities with the highest incidence, suggesting that this geo-localization approach has the potential to improve HCC outcomes by identifying the communities most likely to benefit from targeted interventions to increase identification of persons at risk, screening, and early diagnosis. This relatively simple approach intentionally focuses on assessing local trends and could be applied in other states or localities to inform health system interventions.

M. Strazzabosco reports other support from ENGITIX outside the submitted work. No disclosures were reported by the other authors.

C. Mezzacappa: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. R. Rossi: Resources, data curation, software, writing–review and editing. A. Jaffe: Conceptualization, methodology, writing–review and editing. T.H. Taddei: Conceptualization, supervision, methodology, writing–review and editing. M. Strazzabosco: Conceptualization, resources, supervision, methodology, writing–review and editing.

C. Mezzacappa and R. Rossi were supported by NIH T32DK007356 and all authors were supported by the Yale Liver Center NIH P30 DK034989 Clinical-Translational Core.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.