Racial/Ethnic Disparities in Hepatocellular Carcinoma: The Role of Neighborhood Socioeconomic Deprivation

The public health burden of hepatocellular carcinoma (HCC) is substantial worldwide and in the United States (1). In 2020, HCC was the sixth most frequently occurring cancer and the third most common contributor to cancer death worldwide (2), and one of the most rapidly rising solid tumors in the United States (3). However, the burden of HCC is unevenly distributed across socioeconomic strata and racial/ethnic groups. Specifically, racial/ethnic minorities and those with low socioeconomic status (SES) are disproportionally impacted and have higher HCC incidence and mortality rates (4, 5).

Health disparities may arise from the uneven distribution of risk factors for disease, and/or from the uneven effect of these risk factors (Fig. 1). For example, the prevalence of obesity is higher among racial/ethnic minorities (6), while the harms associated with alcohol consumption are disproportionately high among low SES populations (7). Yet, while we have a more complete understanding of risk factors leading to HCC at the individual level, empirical evidence on the role of contextual factors (e.g., neighborhood built and social environment) in these disparities is extremely limited.

The study of neighborhoods effects on health and how neighborhoods environments contribute to social and racial/ethnic disparities has grown exponentially over the past three decades (8). Interest in studying neighborhoods effects on health outcomes has been fueled by, at least, four important reasons: (i) decades of observations of geographic and spatial variations in health; (ii) an increased recognition of the socioecological model and the social determinants of health to understand health outcomes and racial/ethnic disparities; (iii) interest in understanding the health effects of policies; (iv) an increasing availability of methods suited for integrating effects of exposures at multiple levels and domains, or capturing spatial patterns (9). Neighborhoods represent relevant contexts to understand population health because they encompass physical and social attributes that link broader social and economic factors and which could plausibly affect the health of individuals. Furthermore, many of these attributes represent intervention targets, with broader reach and more long-lasting effects (9, 10).

In this issue of Cancer Epidemiology, Biomarkers & Prevention, Oluyomi and colleagues conducted a Texas-wide study examining the contribution of neighborhood-level socioeconomic deprivation, proxied by the area deprivation index (ADI; ref. 11), to racial/ethnic disparities in HCC stage at diagnosis (local vs. regional/advanced). The authors used cancer registry data along with small area estimates from the American Community Survey to derive the ADI. The authors document important racial/ethnic and SES disparities, with Hispanics and those living in areas with higher socioeconomic deprivation being more likely to have advanced HCC stage at diagnosis compared with their counterparts. The results examining ADI as a contributor to higher rates among Hispanics and non-Hispanic Blacks (NHB), as compared with non-Hispanic Whites, show a heterogeneous contribution of the ADI. Specifically, around 13% of the total disparity between Hispanics and NHW was mediated by the ADI, compared with just 2% for the NHB versus NHW disparity.

For the interpretation of these findings, there are at least three key considerations to keep in mind: (i) cancer registry data, (ii) the nuances of mediation analyses, and (iii) the life course impacts of socioeconomic hardship and racism on health. First, cancer registries have been a great milestone in cancer surveillance. They constitute information systems designed for the collection, storage, and management of data on people with cancer. In the United States, the largest and most comprehensive cancer registry is called Surveillance, Epidemiology and End Results (12). Cancer registries have some advantages and disadvantages that are important to consider to contextualize research findings. Cancer registry routinely collects data on newly diagnosed cancer cases. They aim to collect a wide array of information related to the demographics, tumor characteristics and stage, treatment, and outcome. Potentially, these data would allow comparisons across different population subgroups, space, and over time. However, an important limitation that should not be overlooked stem from the lack of universal health care access in the United States, and suboptimal care experienced among racial/ethnic minorities and those from low socioeconomic backgrounds. For example, in the 2019, 20% of Hispanics were uninsured and 48% had Medicaid or other public insurance. In comparison, the rates of uninsured and public insurance among NHW were 8% and 19%, respectively (13). These differential access to care and type of care, may undermine the accuracy of these registries and may led to differential undercounting of cases, affecting people living in low socioeconomic neighborhoods or those with major health care access barriers. For example, in the case of this study, a sizable number of cases (20%) lacked information about their cancer stage and where excluded. We could argue that the lack of HCC staging data is informative and could itself serve as a proxy of poor health care access. However, if anything, this differential measurement error would operate by reducing the degree of disparities, so the results presented by the authors would be a conservative estimate.

Second, the approach used by the authors to estimate the mediation effect of ADI seems to be an extension of the classical Baron and Kenny approach (14), where three models are fit: two for the outcome, including the main exposure and confounders, and with and without the mediator, and one for the mediator, including the main exposure and confounders. These models allow for the estimation of the total effect (exposure in the outcome model without the mediator), the direct effect (exposure in the outcome model with mediator), and the indirect effect (product of the effect of the exposure in the outcome model with mediator and the effect of the exposure in the mediator model; ref. 14). This approach requires a series of strong assumptions, including the usual lack of exposure-outcome confounding, but also lack of exposure-mediator and mediator-outcome confounding (14). Moreover, to avoid collider-stratification bias, this approach also requires the absence of a confounder of the mediator-outcome relationship that is, in itself, affected by the exposure (14). This last assumption tends to be the most problematic, as it precludes any analysis of mediation exclusively using regression-based methods (14). These assumptions should be considered if a causal interpretation is desired, although results remain useful for hypothesis generation (14).

Third and last, we need to take into consideration the life course impact of neighborhood contexts, including especially the effect of structural racism and racial residential segregation (15), which complicates separating racial/ethnic disparities from neighborhood-based ones (16). On the basis of the conceptual framework in Fig. 1, although socioeconomic disadvantages indeed have some role buffering or magnifying the impact of the causes of HCC (i.e., as mediators), neighborhood deprivation is a more distal determinant of the causes of exposures leading to HCC. Future studies could incorporate mortality data to have a more complete picture of the impact of the socioeconomic context, and examine the role of health care access/quality as mediator affected by socioeconomic hardship. Moreover, and given the long lags in the development of HCC, a consideration of neighborhood effects in earlier life stages may also be warranted (17).

Despite these limitations, the results of this study are important, and should serve as an urgent call for studies to fill important gaps in the literature, so that we can begin to address the disparities in the incidence and prognosis of HCC.

No disclosures were reported.