Social determinants of health impact mortality from HCC and cholangiocarcinoma: a population-based cohort study

Background and Aims: The social determinants of health can pose barriers to accessing cancer screening and treatment and have been associated with cancer mortality. However, it is not clear whether area deprivation is independently associated with mortality in HCC and cholangiocarcinoma when controlling for individual-level social determinants of health. Approach and Results: The cohort included individuals over 18 years old diagnosed with HCC (N=3460) or cholangiocarcinoma (N=781) and reported to the Indiana State Cancer Registry from 2009 to 2017. Area disadvantage was measured using the social deprivation index (SDI). SDI was obtained by linking addresses to the American Community Survey. Individual social determinants of health included race, ethnicity, sex, marital status, and insurance type. The primary outcome was mortality while controlling for SDI and individual social determinants of health by means of Cox proportional hazard modeling. In HCC, living in a neighborhood in the fourth quartile of census-track SDI (most deprived) was associated with higher mortality (HR: 1.14, 95% CI, 1.003–1.30, p=0.04) than living in a first quartile SDI neighborhood. Being uninsured (HR: 1.64, 95% CI, 1.30–2.07, p<0.0001) and never being married (HR: 1.31, 95% CI, 1.15–1.48, p<0.0001) were also associated with mortality in HCC. In cholangiocarcinoma, SDI was not associated with mortality. Conclusions: Social deprivation was independently associated with mortality in HCC but not cholangiocarcinoma. Further research is needed to better understand how to intervene on both area and individual social determinants of health and develop interventions to address these disparities.


INTRODUCTION
In 2020 HCC was the third most frequent cause of cancer-related deaths worldwide and is the fastest growing cause of cancer-related death in the US. [1,2] Despite the availability of effective screening and curative treatments, the 5-year survival is only 35% for localized disease. [3] Although cholangiocarcinoma (CCA) is relatively rare, the incidence in the US is rising. Furthermore, the outcomes for localized disease remain poor, with a 5-year survival of only between 2%-30% and 2%-15% for extrahepatic and intrahepatic disease, respectively. [4] Despite the overall poor survival rates, both HCC and CCA have the potential to be cured when diagnosed early and curative therapies can be accessed.
The ability to access treatment and navigate cancer care impacts survival and is greatly influenced by the social determinants of health (SDOH). [4][5][6] The SDOH are the conditions and the places where people live, learn, work, and play and are often grouped into domains that include economic stability, health care quality and access, social and community context, and neighborhood and built environment. [7,8] In 2020, 37.2 million (12.3%) of Americans lived in poverty. [9] However, this disadvantage is not spread evenly throughout the country's population. Impoverished individuals are often clustered within specific communities, contributing to an uneven distribution of morbidity and mortality. Composite measures of area disadvantage, like the social deprivation index (SDI), seek to capture multiple components of an area's socioeconomic status into 1 composite measure. These measures, including SDI, are becoming important tools for targeting cancer control interventions and for adjusting quality measures and payments. [10,11] Indiana is the representative of other states in the country and has 12% of census tracts classified as persistently poor. [12] Area-level SDOH have begun to be explored in HCC and CCA, [13][14][15][16] . Specifically in HCC, in a Surveillance, Epidemiology, and End Results (SEER) database analysis, area-level poverty was associated with mortality (HR: 1.23, 95% CI, 1.16-1.31, p < 0.001). [13] In CCA, area-level poverty was associated with poor survival (HR: 0.80, 95% CI, 0.696-0.921). [16] Notably, these studies measure a single area-level measure of disadvantage, whereas more sensitive, composite measures have not yet been explored. In addition, in HCC the relationship between both area-level and individual SDOH have not been studied. Therefore, in this analysis, we seek to measure the association between SDI and mortality while controlling for individual SDOH in a cohort of patients with HCC and CCA. We hypothesize that SDI is independently associated with mortality when controlling for individual SDOH.

Study population and data source
Adults aged 18 years and older diagnosed with HCC or CCA between January 1, 2009, and December 27, 2017, were included in this study. Demographic, clinical, and individual SDOH data were collected from the Indiana Sate Cancer Registry (ISCR). Indiana mandates reporting of all cases to the cancer registry; therefore, this data set is likely to represent the full burden of HCC and CCA across the state. Furthermore, the ISCR has been designated a US Cancer Statistics Registry for Surveillance by the Centers for Disease Control's National Program of Cancer Registries. This recognition is based on an annual review of data quality, completeness, and timeliness. All research was conducted in accordance with both the Declarations of Helsinki and Istanbul. All research was approved by the appropriate ethics and/or institutional review committee(s) of Indiana University School of Medicine.

Individual SDOH
Exposures of interest included race, ethnicity, sex, insurance type, and marital status. Race, ethnicity, and sex were considered for this analysis as social constructs, [7] which may be associated with a patient's ability to access care and thus their overall cancer mortality. Race, sex, and ethnicity are defined by selfreport in hospital systems, which report cases to the ISCR. For this analysis, race was classified as Black, White, and other (including but not limited to American Indian, Asian, and unknown race) and ethnicity as Hispanic and non-Hispanic. Sex was classified as men and women. Insurance type is a robust measure of both individual income and health care access. [17] Insurance was classified as private, Medicare, Medicaid (included Medicare patients with Medicaid supplement), uninsured, and other (ie, Veterans, Indian Public Health Service, and military benefits). Marital status has been associated with cancer outcomes and is hypothesized to be one measure of social support. [18] Marital status was classified as married (including common law), divorced, separated, single (never married), widowed, and unknown. intervention, we wanted to test the hypothesis that county SDI would be associated with mortality in this cohort. County and neighborhood disadvantage indicators, as well as rural versus urban residence were obtained by linking patients addresses to the 2015 American Community Survey (ACS). The ACS provides data about social, economic, housing, and demographic characteristics for multiple geographic areas. [19] Census tracts are small areas of~1200-8000 (average 4000) persons. To measure county and census-tract disadvantage, the SDI was calculated from the ACS. The SDI was developed by Butler et al. [10] to quantify social inequities across small geographical locations and provides an approach to evaluate their association with health outcomes. The SDI has been used to develop community health interventions and guide adjustments to quality measures and payments. It is a composite measure of 7 demographic characteristics collected at the area level from the ACS: percent living in poverty, percent with <12 years of education, percent singleparent household, percent living in the rented housing unit, percent living in the overcrowded housing unit, percent of households without a car, and percent nonemployed adults under 65 years of age. The SDI was chosen because it is a more sensitive measure than area income alone, but less sensitive than other measures like the area deprivation index, which includes up to 15 ACS variables, offering a balance of sensitivity and specificity. However, we also explored whether neighborhood income below the federal poverty level (FPL) alone was associated with HCC and CCA mortality when controlling for individual SDOH and rurality in this data set. Census-tract income below the FPL was also obtained from the ACS data set.
We used the rural-urban commuting area (RUCA) codes obtained from the ACS to define rural versus urban residence. [20] All metro census tracts, with RUCA codes 4-9, are considered rural by the Federal Office of Rural Health Policy and is consistent with our previously published work. [20][21][22] We further classified rural counties into rural-urban adjacent (RUCA codes 4, 6, and 8) and rural not metro adjacent (RUCA codes 5, 7, and 9). Counties with populations of > 250,000 were classified as urban (RUCA codes 1, 2, and 3).

Covariates
Other covariates captured from the ISCR included patient age at the time of diagnosis, date of diagnosis, date of death or last contact, and Surveillance, Epidemiology, and End Results (SEER) cancer stage. SEER stage was classified as localized, regional, distant, and unstaged. Localized stage defines tumors involving a single lobe of the liver. Regional stage defines tumors involving > 1 lobe through contiguous growth of a single lesion, extension to adjacent structures, or spread to regional lymph nodes. Distant stage includes tumors that have metastasized, extension of tumor growth to nearby organs, or spread to distant lymph nodes. [23] Outcomes The primary outcome was time to death in HCC and CCA when controlling for census-tract level SDI and individual SDOH. Mortality in this analysis is all-cause mortality. In a secondary analysis, mortality was explored while controlling for: (1) county level SDI in HCC and CCA and (2) census-tract level income below the FPL and individual SDOH in HCC and CCA. A conceptual framework for a potential relationship between the exposures, covariates, and outcomes is provided in Supplemental Figure 1 (http://links.lww.com/ HC9/A136) and is adapted from our previous work, as well as Alcaraz et al. [6] model on addressing social determinants in cancer care. [21] Statistical methods Individual patient characteristics and the SDOH were summarized by mortality status and SDI quartile as means and SDs, medians and quartile ranges, and frequencies and percentages, as appropriate. For the comparison of demographic and clinical characteristics by SDI quartile, continuous variables were tested for differences between the 4 quartiles using ANOVA. Categorical variables were tested for the association between variable and SDI quartile using chi-squared testing.
For the primary and secondary outcome of mortality status controlling for SDI and income below the FPL, univariable and multivariable time-to-event Cox proportional hazard models were performed. Because of the relationship between the exposures, covariates, and mortality proposed in our conceptual framework, all variables were considered for inclusion in the multivariable models if there was no evidence of collinearity. A variance inflation factor of 4 was used as the threshold for suspected collinearity. Because of the suspected collinearity with SDI and multiple other covariates, rurality was not included in the final multivariable models. Ethnicity was not included in the multivariable models for CCA mortality due to sample size (Hispanic subgroup n = 3). Interactions were explored between SDI, income below the FPL, and race for all models.
Descriptive and subgroup analysis in HCC Although rurality had the evidence of collinearity with SDI and was not included in multivariable models, we remained interested in exploring the geographic SOCIAL DETERMINANTS OF HEALTH IMPACT MORTALITY | 3 distribution of deaths in Indiana of those individuals in the fourth quartile of SDI given its association with HCC mortality. The relationship between rurality and SDI was explored visually by mapping quartile 4 SDI HCC mortality rates per 100,000 persons.
In HCC, over 70% of the cohort had at least 1 SDOH that was significantly associated with higher mortality. To identify the population most in need of cancer control interventions, subgroup analysis was performed to explore the cumulative burden of 1, 2, or 3 SDOH on HCC mortality. We focused on uninsured/Medicare, never married/widowed, and fourth quartile SDI SDOH as these were significantly associated with mortality in the primary analysis. Only patients with data on all SDOH were included in this analysis. Overall survival was calculated using the Kaplan-Meier method and logrank testing was performed to compare mortality between patients having zero and 3, 1 and 2, and 2 and 3 SDOH. Because of the statistical concern for multiple comparisons, only 1 comparison was made in each category and the decision was made a-priori.
Data summaries were produced using R Statistical Software (version 4.1.0). All analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC). A 0.05 significance level was used for all multivariable analyses.

Demographic and clinical characteristics
The study included 3460 patients with HCC and 781 patients with CCA. The median age of the HCC cohort was 64 years [interquartile range (IQR): 57-73] and included 25.2% women and 13.5% individuals of Black race ( Table 1). The median age of the CCA cohort was 68 years (IQR: 59-76) and included 52.5% women and 6.4% individuals of Black race (
For HCC, individuals residing in quartile 4 SDI neighborhoods (most deprived) were more likely to be Black, never married, or insured by Medicaid than individuals residing in quartile 1 neighborhoods (p-values <0.0001) ( Table 3). There was no difference in cancer stage between the SDI quartiles (p = 0.310) ( Table 3).

Cholangiocarcinoma
From the Cox proportional hazard model, mortality in CCA was associated with older age at diagnosis and more advanced SEER stage (Table 6). Regarding the SDOH, being never married was associated with mortality (HR: 1.34, 95% CI, 1.008-1.79, p = 0.0499). Neither county nor neighborhood SDI was not associated with time to death (Table 6).
In the univariable analysis, neighborhood income below the FPL quartile was also significantly associated with time to death (Supplemental Table 2, http://links. lww.com/HC9/A137).

Geographic mortality clusters in HCC
Over the course of the study period 83.4% of individuals with HCC were deceased. Mortality from HCC clustered

Cumulative burden of SDOH in HCC
To determine the cumulative burden of SDOH on HCC mortality, patients with 1 SDOH that was significantly associated with mortality [uninsured/Medicare (category 1), never married/widowed (category 2), or fourth quartile SDI (category 3)] were compared with patients with 2 SDOH (categories 5, 6, and 7) or 3 SDOH (category 4) ( Figure 2A). Among 2542 patients with complete data on SDOH 23.9% had none of the 3 SDOH associated with mortality and were grouped in category 0. By the end of the study period, there was a significant difference in survival between categories 0 and 4 (p < 0.0001) and between categories 3 and 4 (p = 0.003) (Figure 2).

DISCUSSION
In the last 5 years, HCC mortality has begun declining among US adults. [24] However, disparities in survival are known to still exist by race, ethnicity, insurance status, and geography. [25] Furthermore, area disadvantage has been associated with mortality in other cancers. [26] Yet, the association between area and individual-level SDOH and mortality has not been evaluated in HCC and CCA. Here, we demonstrate that in HCC, both area and individual SDOH matters. However, in CCA where treatment options are limited and the mortality rates are quite high, area-level SDOH may be less important. Finally, although rural geography was not significantly associated with mortality, high SDI urban mortality "hotspots" were identified in HCC. Individuals with HCC living in a neighborhood in the fourth quartile for SDI were 14% more likely to be deceased than those living in neighborhoods in the first quartile. Income below the FPL was also found to be associated with mortality. Area disadvantage has been associated with incidence and mortality in other cancers. [27,28] Specifically, in a recent SEER database analysis, area deprivation was associated with worse survival among patients with nonmetastatic breast, prostate, lung, and colorectal cancers after accounting for individual SES. [26] The role of neighborhood disadvantage, until now, had not been evaluated in HCC or CCA. Notably however, census-tract level poverty was also associated with mortality in HCC. [13] Therefore, although a composite measure may be more sensitive, independent measures of neighborhood poverty may be sufficient for identifying cohorts to target for interventions.
In addition to area deprivation, having Medicare or being uninsured, as well as having a marital status of never being married or widowed, were associated with mortality in HCC. This is consistent with what has previously been seen in the literature. In a retrospective SEER Medicare-linked database study when stratified by neighborhood poverty, Black patients living in high-poverty neighborhoods had worse survival (HR: 1.13; 95% CI, 1.02-1.25). [14] In an analysis by Wong and colleagues, patients from rural regions and lower-income households    had more advanced tumor stage at diagnosis and significantly higher HCC mortality. In CCA, never married marital status was the only SDOH associated with mortality. There have been very few studies looking at SDOH in CCA. [16,29] In a study exploring individual income and CCA outcomes, low income was associated with poor survival (< $37,999 vs. ≥ $63,000 annually; HR: 1.07, 95% CI, 1.01-1.14, p = 0.032). [29] However, this analysis did not include marital status, a surrogate marker for social support. Notably, SDI was not associated with time to death in CCA suggesting that in some high mortality cancers, stage is the strongest predictor of survival. Rurality and SDI had evidence of collinearity in our analysis, suggesting a strong relationship between neighborhood deprivation and rural-urban residence in this cohort. Mortality from neighborhoods in the fourth quartile of SDI was mostly clustered into 3 urban counties in HCC, including Marion and Lake County, which contain Indianapolis and Gary, IN, respectively. In 2019, Gary, IN, was ranked worse economically than any city in the US based on the percentage of people in the workforce, median household income, percentage of people without health insurance, median commute times, and the number of people living in poverty. [30] A SEER database study found that patients who lived in large metropolitan areas of over 1 million people had a better overall survival than other cohorts. [13] None of the 3 urban hotspots we identified has populations over 1 million residents. Therefore, these findings may be consistent with previous work.
Finally, we explored the cumulative burden of the SDOH on mortality in HCC and found that there was indeed an association. Multiple SDOH has been shown to increase mortality in adults with diabetes and chronic kidney disease; however, its effect on HCC mortality has not been previously explored. [31,32] Steps toward cancer control and prevention Taken together, area measures of deprivation may allow the identification of neighborhoods with the highest levels of disadvantage to better allocate time, effort, and resources. Combined with the relevant individual SDOH targeted evidence-based cancer control policies and interventions can be implemented in the populations with the highest risk for mortality. Community-based cancer control interventions that have demonstrated efficacy include culturally and demographically tailored cancer screening programs, informed decision-making tools, and community-based genomic, obesity, and sun safety programs. [33] Successful interventions to reduce disparities are often multilevel and resource intense. [34] In our own research group, these data are being used to identify the highest risk groups for intervention development and testing. Furthermore, there are a growing number of international (UK and New Zealand) and US regional programs (Massachusetts and Hennepin Health Center in Minnesota Medicaid Demonstration Project) that are using a blend of individual SDOH and neighborhood measures of deprivation to adjust payments for primary and social services. [11] For other states to follow this lead, tumor-specific, target communities need to be identified.
Our study included over 4000 patients spanning over 9 years making it the largest study to explore the role of SDI while controlling for individual determinants in these cancer types. Indiana, with its large rural population, uneven distribution of area wealth, and high burden of chronic liver disease, is uniquely suited to explore these questions. Our population was over 12% Black and nearly approximates the proportion of Black Americans living in the US but included fewer individuals of Hispanic and Asian ancestry than the US population. Interestingly, Black race was not associated with mortality in this cohort as it has been in other data sets. [14,25] This likely reflects the high mortality rate in both Black and White patients with HCC experienced in Indiana. The CIs for the association with some of the SDOH and mortality were near one. However, our cumulative burden subgroup analysis suggests that indeed these factors do impact mortality. This analysis, however, assumes that each SDOH is equally weighted. Further work is needed to understand the relative weight of each SDOH on cancer mortality. On the basis of our previous work that includes some patients from this cohort, we hypothesize that the mechanism by which the SDOH impacts mortality is through access to treatments like liver transplantation. [35] Prospective studies are needed to further clarify this relationship. Finally, for this analysis, we did not have data on liver disease severity and comorbidity, which may also impact mortality. However, in an analysis that did include measures of comorbidity, neighborhood poverty remained independently associated with mortality. [14] We hypothesize that the strength of the association would also remain significant.
In conclusion, both area-level and individual-level SDOH are associated with mortality in HCC. These associations are not as clear in CCA. If we are to develop effective cancer control and prevention interventions that improve health equity, both area and individual determinants of health should be considered.

FUNDING INFORMATION
Lauren Nephew is funded by NMIHD 1K23MD018090-01.

CONFLICT OF INTEREST
Naga Chalasani reports no relevant conflicts of interest. For full disclosure, he has paid consulting agreements with Madrigal, Lilly, GSK, Boehringer-Ingelheim, Zydus, Galectin, Altimmune, and Foresite. He receives research grant support from DSM and Exact Sciences. Lauren Nephew reports no relevant conflicts of interest. For full disclosure, she has ongoing research collaboration with DelFi diagnostics. The other authors report no conflicts of interest.