Racial and ethnic survival disparities in primary liver cancer in the United States: a National Cancer Database propensity score-matched study Authors: Muhammet Ozer, Suleyman Y Goksu, Amy L Jones, Timothy J Brown, Nina Sanford, Nilesh Verma, Salwan Al Mutar, David Hsieh, Syed M Kazmi Published: 2026 ABSTRACT ABSTRACT:IntroductionPrimary liver cancer is a leading cause of cancer mortality worldwide. We evaluated whether racial and ethnic survival disparities among US patients with primary liver cancer persist after adjustment for clinical, treatment, socioeconomic, and access-related factors.MethodsWe performed a retrospective cohort study using the National Cancer Database. Race was analyzed as White versus Black, and race/ethnicity groups were defined as non-Hispanic White (NHW), non-Hispanic Black (NHB), and Hispanic of any race. Overall survival (OS) was assessed using Kaplan–Meier analysis, log-rank testing, multivariable Cox regression, and 1:1 propensity score matching for the White-versus-Black comparison.ResultsAmong 122,049 patients diagnosed between 2004 and 2016, Black patients more often presented with advanced disease and were less likely to receive surgery, chemotherapy, or radiotherapy. In the unmatched analysis, White patients had better OS. After matching, race was not an independent predictor of survival (HR 0.98, p = 0.11). Hispanic ethnicity remained associated with improved survival compared with NHW and NHB patients (HR 1.32 and 1.29; both p < 0.001).ConclusionThe unadjusted survival difference between White and Black patients was attenuated after adjustment for clinical, treatment, socioeconomic, and access-related factors. Hispanic ethnicity remained associated with improved survival in secondary analyses, warranting further investigation.Keywords: Primary liver Cancer, Racial disparities, Ethnic disparities, Propensity score matched study Introduction Primary liver cancer is a major cause of cancer-related mortality worldwide and represents a growing public health challenge in the United States [1]. Although the global burden of primary liver cancer reflects broad differences in viral hepatitis prevalence, metabolic disease, alcohol use, and access to early detection, the US population provides a unique setting in which to evaluate how race, ethnicity, socioeconomic status, insurance coverage, treatment access, and facility-level factors contribute to survival disparities. Hepatocellular carcinoma (HCC) is the most common type of primary liver cancer, followed by intrahepatic cholangiocarcinoma (ICC) [2]. In the US, early-stage liver cancer can be effectively treated with surgical resection, liver transplantation, and locoregional therapies [3]. However, many patients are diagnosed at advanced stages and are only candidates for systemic or palliative treatments. Although survival has improved with the introduction of novel systemic therapies, these advances have not benefited all patient populations equally, highlighting persistent disparities in liver cancer outcomes within the US [4–8]. Previous studies have indicated that, like many other solid tumors, liver cancer exhibits disparities in mortality and incidence based on race, ethnicity, geographic region, and socioeconomic status in the US population [9–11]. Addressing racial and ethnic disparities in cancer care remains challenging in the United States [12]. The underlying reasons for disparities in survival are multifaceted and not always readily apparent. Despite comprising the largest minority group in the US, Black patients have long been known to have higher mortality for most cancers than White patients [13]. Racial disparities in liver cancer have been evaluated in several previous studies ([11,12,14,15]. Most of these studies reported the highest survival rates among Asian patients, while Black patients had the lowest [14,16]. These disparities are often attributed to differences in the tumor stage, socioeconomic status (SES), the prevalence of underlying risk factors, access to healthcare, and early detection of liver cancer [14–18]. Although the mechanisms underlying liver cancer mortality disparities remain incompletely understood, examining patterns of disparity is a critical first step in addressing them. In this study, we analyzed data from the National Cancer Database (NCDB) to assess how race and ethnicity are associated with overall survival among patients with primary liver cancer in the United States. We hypothesized that unadjusted survival differences between White and Black patients would be attenuated after adjustment for clinical, socioeconomic, access-related, facility-level, and treatment-related factors. We further hypothesized that race/ethnicity-based differences, particularly among Hispanic patients, may persist as secondary prognostic associations. Materials and methods This was a retrospective cohort study of adult patients diagnosed with primary liver cancer between 2004 and 2016 using the National Cancer Database (NCDB), which provides de-identified clinical cancer data for approximately 70% of newly diagnosed cancer cases in the United States. The NCDB contains information on patient demographics, cancer stage, treatments, and survival outcomes. We followed the STROBE guidelines to ensure the quality of our study [19]. This study was deemed exempt by the Institutional Review Board of the University of Texas Southwestern Medical Center because it used de-identified data. Access to the NCDB Participant User File and permission to use the data for research were obtained through the NCDB Participant User File application process and were governed by the applicable data use agreement. Study population Adult patients (≥18 years) diagnosed with primary liver cancer between 2004 and 2016 were identified using ICD-O-3/WHO 2008 primary site/topography codes C22.0 and C22.1, corresponding to liver and intrahepatic bile duct primary sites, respectively. These site codes were used to define a broad primary liver cancer cohort in the NCDB, consistent with the objective of this study, rather than to define a histology-confirmed HCC-only cohort. Patients were excluded if they had unknown survival data, more than one primary tumor, incomplete first-course treatment at the reporting facility, unknown race, or race categorized as other than White or Black. Because the primary objective was to evaluate racial and ethnic disparities in primary liver cancer overall, patients were selected based on primary site/topography codes. Histology-specific morphology codes were not used as additional inclusion or exclusion criteria. The primary outcome was overall survival, defined as the time from diagnosis to death from any cause or last follow-up. Covariates were selected a priori based on clinical relevance and availability in the NCDB. Patient-level covariates included age at diagnosis, sex, race, Hispanic ethnicity, Charlson–Deyo comorbidity score, insurance status, tumor stage, primary site, surgery, chemotherapy, and radiotherapy. Facility-level covariates included facility type and facility location. Area-level covariates included median household income and education level, which were derived from US Census data based on the patient’s ZIP code of residence at diagnosis. Geographic access was represented by travel distance from the patient’s residence to the reporting facility. Covariates of interest included age at diagnosis, sex, race, ethnicity, Charlson-Deyo comorbidity score, facility type, facility location, rurality, median income quartiles based on US ZIP code, insurance status, education level, NCDB analytic stage group, primary site, definitive surgery, radiotherapy, and chemotherapy. Race and Hispanic ethnicity were defined using separate NCDB variables. For the primary race-based analysis, race was classified as White or Black according to the NCDB race variable, regardless of Hispanic origin. Hispanic ethnicity was treated as a separate baseline variable and categorized as Hispanic, non-Hispanic, or unknown. For race/ethnicity subgroup analyses, mutually exclusive groups were created as non-Hispanic White (NHW), non-Hispanic Black (NHB), and Hispanic of any race. Patients with unknown Hispanic origin were included as “unknown” in descriptive analyses and handled consistently in multivariable modeling. The chemotherapy variable was classified into any type of chemotherapy receipt, none, and unknown, and stratified into single-agent and multi-agent categories. Socioeconomic factors were defined as median income quartiles, insurance status, and education level. Education was derived from US Census data at the ZIP code level and categorized into quartiles based on the percentage of adults without a high school diploma: ≥17.6%, 10.9–17.5%, 6.3–10.8%, and <6.3%. Also, income was not individual-level measures but were derived from US Census data at the ZIP code level based on the patient’s residence at diagnosis. To assess healthcare access, we evaluated travel distance, calculated as the great circle distance between the patient’s residence at diagnosis and the reporting facility, based on the geographic centroid of ZIP codes. Travel distance was categorized into four groups: <12.5 miles, 12.5–49.9 miles, 50–249 miles, and ≥250 miles. These groupings have been used previously to assess geographic and healthcare access issues [20]. The NCDB analytic stage group is typically determined based on pathological staging, but when this information is unavailable, the clinical-stage group is used instead. Our study considered stage III or IV as indicative of a higher stage. To account for comorbid conditions, we employed the Charlson-Deyo comorbidity score, which assigns scores ranging from 1 to 6 for 19 diseases based on their impact on one-year survival rates [21]. Statistical analyses We employed the chi-square or Fisher’s exact test to assess differences in baseline characteristics between White and Black participants. Any variables with missing data were treated as “unknown” and included in the multivariable analyses. Overall survival in the unmatched cohort was estimated using Kaplan–Meier methods and compared using log-rank tests. A multivariable Cox proportional hazards model was performed in the unmatched cohort to adjust for a range of factors, including demographics, ethnicity, facility type, rurality, socioeconomic factors, travel distance, comorbidity score, tumor stage, primary site, and treatments including surgery, chemotherapy, and radiotherapy. Primary site was included to account for potential outcome differences between tumors arising from the liver primary site, C22.0, and the intrahepatic bile duct primary site, C22.1. Hazard ratios (HRs) and 95% confidence intervals (CIs) were reported. We conducted propensity score matching to address potential selection bias from measured covariates in the primary race-based comparison between White and Black patients. Race, defined as White versus Black, was the exposure variable for the matched analysis. A 1:1 nearest-neighbor matching algorithm was used for the primary White-versus-Black race comparison, resulting in a matched cohort of 41,360 patients, including 20,680 White and 20,680 Black patients. The propensity score model included sex, Hispanic ethnicity, facility type, facility location, insurance status, median income, education level, rurality, travel distance to treatment facility, Charlson–Deyo comorbidity score, primary site, tumor stage, surgery, chemotherapy, and radiotherapy. Hispanic ethnicity was included as a matching covariate because it differed substantially between White and Black patients in the unmatched cohort and was associated with overall survival; however, it was not the primary exposure in the propensity score matched race analysis. Hispanic ethnicity was categorized as Hispanic, non-Hispanic, or unknown for the matching procedure. We utilized the R software version 3.6.2 with the MatchIt package (Nonparametric Preprocessing for Parametric Causal Inference) to perform the matching. Following propensity score matching, we assessed the balance of covariates using standardized differences, with covariates considered well-balanced if the standardized differences were less than 0.1 [22]. Kaplan–Meier survival analyses were repeated in the matched cohort, and survival curves were compared using log-rank tests. Cox proportional hazards regression was then performed in the matched cohort to estimate the association between race/ethnicity and overall survival. All statistical tests were two-sided, and a p-value <0.05 was considered statistically significant. Statistical analyses were performed using SPSS version 22 and R version 3.6.2. Results Baseline characteristics In the full unmatched cohort, 122,049 individuals with primary liver cancer met the inclusion criteria; 101,369 patients (83.1%) were classified as White, and 20,680 patients (16.9%) as Black. After 1:1 propensity score matching for the primary race-based comparison, the matched cohort included 41,360 patients, with 20,680 White and 20,680 Black patients. The consort diagram is shown in Figure 1. Demographic characteristics were assessed before propensity score matching. Older age was more common among White patients (44% vs. 28%, p < 0.001), as was Hispanic ethnicity (15% vs. 1%, p < 0.001), in comparison to Black patients. White patients exhibited increased rates of surgery (26% vs. 22%, p < 0.001), chemotherapy (39% vs. 38%, p < 0.001), and radiotherapy (10% vs. 8%, p < 0.001) relative to their Black counterparts. Moreover, Black patients were more likely to have advanced-stage disease (40% vs. 37%, p < 0.001) but demonstrated a lower comorbidity score (59% vs. 57%, p < 0.001) than White patients. While there were more males than females included in the analysis, the gender distribution was balanced across both study groups. Tumors coded to the liver primary site, C22.0, represented the majority of cases and were more common among Black patients compared with White patients (94% vs. 88%, p < 0.001), whereas tumors coded to the intrahepatic bile duct primary site, C22.1, were less common (Table 1). CONSORT diagram illustrating patient selection and analytic cohort construction from the NCDB database for primary liver cancer diagnosed between 2004 and 2016. After exclusions, the full unmatched cohort included 122,049 patients. A 1:1 propensity score matched cohort was then created for the primary White-versus-Black race comparison, yielding 20,680 White and 20,680 Black patients. Abbreviations: NCDB, National Cancer Database; PSM, propensity score matching. Table 1.Baseline characteristics.CharacteristicsWhite (%)Black (%)p-value101,369 (83.1)20,680 (16.9)Sex 0.33 Male73,415 (72.4)14,908 (72.1) Female27,954 (27.6)5,772 (27.9) Age at Diagnosis <0.001 <50 years8,416 (8.3)2,205 (10.7) 50-6448,800 (48.1)12,708 (61.5) ≥65 years44,153 (43.6)5,767 (27.9) Ethnicity <0.001 Hispanics14,950 (14.7)266 (1.3) Non-Hispanics82,518 (81.4)19,554 (94.6) Unknown3,901 (3.8)860 (4.2) Comorbidity Score <0.001 058,029 (57.2)12,241 (59.2) 121,848 (21.6)4,470 (21.6) 2+21,492 (21.2)3,969 (19.2) Facility Type <0.001 Academic/Research55,988 (55.2)12,957 (62.7) Non-academic43,849 (43.3)7,123 (34.4) Others1,532 (1.5)600 (2.9) Facility Location <0.001 New England5,759 (5.7)525 (2.5) Middle Atlantic15,376 (15.2)3,705 (17.9) South Atlantic19,029 (18.8)5,881 (28.4) East North Central14,229 (14.0)3,147 (15.2) East South Central6,628 (6.5)1,540 (7.4) West North Central7,449 (7.3)1,053 (5.1) West South Central12,841 (12.7)2,854 (13.8) Mountain4,589 (4.5)192 (0.9) Pacific13,937 (13.7)1,183 (5.7) Others1532 (1.5)600 (2.9) Insurance status <0.001 Uninsured5,553 (5.5)1,859 (9.0) Private33,658 (33.2)5,824 (28.2) Government59,640 (58.8)12,418 (60.0) Unknown2,518 (2.5)579 (2.8) Income <0.001 <$40,22720,548 (20.3)10,159 (49.1) $$40,227 – $50,35324,247 (23.9)4,110 (19.9) $50,354 – $63,33224,036 (23.7)2,949 (14.3) ≥$63,33230,698 (30.3)3,118 (15.1) Unknown1,840 (1.8)344 (1.7) Education <0.001 ≥17.6%26,087 (25.7)9,219 (44.6) 10.9–17.5%26,663 (26.3)6,279 (30.4) 6.3–10.8%26,694 (26.3)3,398 (16.4) <6.3%20,282 (20.0)1,477 (7.1) Unknown1,643 (1.6)307 (1.5) Rurality <0.001 Metropolitan83,167 (82.0)18,958 (91.7) Non-metropolitan15,526 (15.3)1,385 (6.7) Unknown2,676 (2.6)337 (1.6) Travel distance <0.001 <12.5 miles46,428 (45.8)14,673 (71.0) 12.5-49.9 miles31,525 (31.1)3,760 (18.2) ≥5023,025 (22.7)2,180 (10.5) Unknown391 (0.4)67 (0.3) Primary Site <0.001 Liver primary site (C22.0)89,570 (88.4)19,362 (93.6) Intrahepatic bile duct primary site (C22.1)11,799 (11.6)1,318 (6.4) Stage <0.001 I26,339 (26.0)5,043 (24.4) II17,491 (17.3)3,210 (15.5) III18,676 (18.4)4,489 (21.7) IV18,452 (18.2)4,066 (18.4) Unknown20,411 (20.1)3,872 (18.7) Surgery <0.001 No74,370 (73.4)16,137 (78.0) Yes26,790 (26.4)4,488 (21.7) Unknown209 (0.2)55 (0.3) Radiotherapy <0.001 No91,031 (89.8)18,905 (91.4) Yes9,837 (9.7)1,666 (8.1) Unknown501 (0.5)109 (0.5) Chemotherapy <0.001 No59,196 (58.4)12,170 (58.8) Yes39,362 (38.8)7,846 (37.9) Unknown2,811 (2.8)664 (3.2) Survival (months) <0.001 Median9.9 (9.7–10.1)8.5 (8.2–8.8) Baseline characteristics. Socioeconomic and healthcare access variables also differed between groups. White patients were more likely to have private insurance (33% vs. 28%, p < 0.001). By contrast, Black patients were more likely to travel shorter distances for treatment (<12.5 miles, 71% vs. 46%, p < 0.001), to live in ZIP code areas with lower income (<$40,227, 49% vs. 20%, p < 0.001) and lower education levels (45% vs. 26% without a high school diploma, p < 0.001). Regarding care setting, Black patients were more often residents of metropolitan areas (92% vs. 82%, p < 0.001) and more frequently received treatment at academic or research facilities (63% vs. 55%, p < 0.001) (Table 1). After propensity score matching, standardized mean differences were assessed for all covariates included in the matching model. All standardized mean differences were <0.1, indicating adequate balance between White and Black patients across demographic, socioeconomic, facility-level, access-related, tumor-related, and treatment-related covariates (Table S1). Survival analyses Unmatched cohort In the unadjusted analysis, White patients exhibited better overall survival (OS) than Black patients (median OS 9.9 vs. 8.5 months, HR 95% CI 1.08 (1.06–1.10), Black reference, p < 0.001) (Figure 2). However, this difference was not significant in a multivariable Cox regression analysis after adjusting for key clinical factors, such as age at diagnosis, gender, insurance, and income status, facility type, rurality, travel distance, comorbidity score, treatment status (surgery, chemotherapy, radiotherapy), primary site, and stage (HR = 1.01, 95% CI: 0.99–1.03, p = 0.18) (Table 2). Kaplan–Meier survival curve comparing overall survival between White and Black patients in the unmatched cohort. Median overall survival was 9.9 months for White patients and 8.5 months for Black patients; log-rank p < 0.001. Abbreviations: OS, overall survival. Table 2.Multivariable cox regression analysis for overall survival by race and race/ethnicity group.CharacteristicsUnmatched (White N: 101,369; Black N: 20,680)Propensity matched (White N: 20,680; Black N: 20,680)HR (95% CI)p-valueHR (95% CI)p-valueRace group WhiteRef Ref Black1.01 (0.99–1.03)0.180.98 (0.96–1.00)0.11Race/Ethicity Group HispanicsRef Ref Non-Hispanic White1.26 (1.23–1.28)<0.0011.32 (1.25–1.41)<0.001 Non-Hispanic Black1.22 (1.19–1.25)<0.0011.29 (1.22–1.37)<0.001Note: Multivariable Cox proportional hazards models were adjusted for age at diagnosis, sex, race or race/ethnicity group as applicable, facility type, facility location, insurance status, median household income quartile, education quartile, rurality, travel distance to treatment facility, Charlson–Deyo comorbidity score, tumor stage, primary site, surgery, chemotherapy, and radiotherapy. Abbreviations: CI, confidence interval; HCC, hepatocellular carcinoma; HR, hazard ratio; NCDB, National Cancer Database; NHB, non-Hispanic Black; NHW, non-Hispanic White; OS, overall survival; PSM, propensity score matching; SMD, standardized mean difference. Multivariable cox regression analysis for overall survival by race and race/ethnicity group. Note: Multivariable Cox proportional hazards models were adjusted for age at diagnosis, sex, race or race/ethnicity group as applicable, facility type, facility location, insurance status, median household income quartile, education quartile, rurality, travel distance to treatment facility, Charlson–Deyo comorbidity score, tumor stage, primary site, surgery, chemotherapy, and radiotherapy. Abbreviations: CI, confidence interval; HCC, hepatocellular carcinoma; HR, hazard ratio; NCDB, National Cancer Database; NHB, non-Hispanic Black; NHW, non-Hispanic White; OS, overall survival; PSM, propensity score matching; SMD, standardized mean difference. Subgroup analysis was conducted among different ethnic groups. This analysis revealed that Hispanic patients experienced significantly better OS compared to non-Hispanic White and non-Hispanic Black patients (median OS 12.0 vs. 9.8 vs. 8.5 months, respectively, p < 0.001) (Figure 3). This significance persisted in the multivariate analysis (Hispanics as a reference, NHW HR: 1.26 [1.23–1.28], NHB 1.22 [1.19–1.25], p < 0.001) (Table 2). Kaplan–Meier survival curve comparing overall survival across mutually exclusive race/ethnicity groups in the unmatched cohort: Hispanic of any race, non-Hispanic Black, and non-Hispanic White. Median overall survival was 12.0 months for Hispanic patients, 8.5 months for non-Hispanic Black patients, and 9.8 months for non-Hispanic White patients; log-rank p < 0.001. Abbreviations: NHB, non-Hispanic Black; NHW, non-Hispanic White; OS, overall survival. Matched cohort Following propensity score matching for the primary White-versus-Black race comparison, multivariable analysis demonstrated that race was not independently associated with survival (HR 0.98 [0.96–1.00], p = 0.11). In a secondary analysis within the race-matched cohort, Hispanic ethnicity remained associated with improved overall survival compared with non-Hispanic White and non-Hispanic Black patients, using Hispanic patients as the reference group (NHW HR: 1.32 [1.25–1.41]; NHB HR: 1.29 [1.22–1.37]; both p < 0.001). Median OS was 14.3 months for Hispanic patients, 8.6 months for NHW patients, and 8.5 months for NHB patients. Because Hispanic ethnicity was included as a covariate in the race-focused matching model, these ethnicity-related findings should be interpreted as secondary associations within the matched cohort. Discussion This study used the NCDB to examine a large, registry-defined cohort of patients with primary liver cancer and compared survival outcomes according to race and race/ethnicity. We identified that White patients had better OS than Black patients. The higher stage of the disease, lower rates of treatment receipt, and lower SES among Black patients may explain this. However, this difference did not prevail after adjusting for clinical confounding factors. On the other hand, when comparing ethnicity, we found that Hispanic patients had better OS than NHW and NHB patients. This survival benefit persisted in the multivariate analysis. After propensity score matching for the primary White-versus-Black race comparison, race was not independently associated with survival. Hispanic ethnicity remained associated with improved OS compared with NHW and NHB patients in secondary analyses within the matched cohort. Because Hispanic ethnicity was included as a balancing covariate in the race-focused propensity score model, these ethnicity-related findings should be interpreted as secondary prognostic associations rather than as the primary estimand of the matched analysis. This is known as the Hispanic paradox, which refers to the observed phenomenon in which Hispanic individuals tend to have better health outcomes, including longer OS, than NHW and NHB individuals despite often facing lower socioeconomic status and limited access to healthcare [23–27]. In the context of liver cancer, this paradox holds as Hispanic patients tend to have better survival outcomes, which may be attributed to a combination of cultural, behavioral, and possibly biological factors that mitigate some health risks [28]. In the present analysis, the propensity score matched model was primarily designed to evaluate racial differences between White and Black patients. Hispanic ethnicity was included in the propensity score model as a covariate to balance the race comparison because it was unevenly distributed across racial groups and was associated with survival. Therefore, the matched analysis should be interpreted primarily as a race-focused analysis. The observed association between Hispanic ethnicity and improved survival in the matched cohort should be considered secondary and hypothesis-generating. Notably, Hispanic ethnicity was more frequent among White than Black patients, including after propensity score matching, and the survival difference between Hispanic and non-Hispanic patients was larger than the White–Black survival difference. Therefore, the unequal distribution of Hispanic ethnicity may have influenced the unadjusted White–Black comparison and supports separate interpretation of the Hispanic findings. Our findings are consistent with prior population-based studies demonstrating that Black patients with liver cancer are more likely to present with advanced disease, are less likely to receive cancer-directed therapy, and experience worse unadjusted survival compared with White patients. However, our study extends this literature by showing that the Black-White survival difference was attenuated after multivariable adjustment and propensity score matching for clinical, socioeconomic, facility-level, access-related, and treatment-related factors. This suggests that race alone may not independently drive survival differences once measurable structural and care-delivery factors are accounted for. In contrast, the persistent association between Hispanic ethnicity and improved survival is consistent with prior descriptions of the Hispanic paradox and highlights the need for additional research into unmeasured cultural, behavioral, biological, migration-related, and healthcare-access factors. The etiologies of this disparity are likely multifactorial, including SES, insurance status, income, education, and access to timely and high-quality cancer care [12,29–31]. Our study found that White patients were more likely to have private insurance and higher income compared to Black patients. Meanwhile, Black patients had lower education levels and lower rates of treatment utilization. These findings are consistent with prior reports, such as Steward et al. who reported that, among patients with HCC, Black patients were less likely to receive treatment and had poorer survival outcomes than their White counterparts [32]. Similar to our study, Sloane et al. found that Black patients tended to have higher tumor stages and to be younger than White patients [33]. Black patients also received fewer surgical treatments. Multiple studies reported demographic and socioeconomic factors as significant contributors to cancer survival [34,35]. Commonly accepted socioeconomic status indicators are education, income status, access to healthcare, and occupation [36]. From a larger perspective, higher mortality rates have been previously found in various cancers among NHB patients than among NHW patients [13,37–39]. National Health and Nutrition Examination (NHANES) III analyses showed that overall GI cancer-related mortality is significantly higher among NHB than NHW patients [40]. Notably, and consistent with our propensity-adjusted findings, subgroup analyses in NHANES III did not demonstrate a significant racial difference in primary liver cancer mortality overall. However, increased mortality was observed among NHB men, a disparity attributed in part to differences in socioeconomic status and access to preventive and specialty healthcare services. Together, these findings support our results that racial disparities in liver cancer outcomes are largely mediated by socioeconomic and healthcare access factors rather than race alone [36,41]. As one of the independent risk factors for primary liver cancer mortality, race and ethnicity were studied in several previous population-based studies [42–44]. A study using the Surveillance Epidemiology and End Results (SEER) database (2000–2012) reported that Black patients were less likely to receive cancer-directed treatment and had higher all-cause mortality risks [45]. Another SEER-Medicare study (1994–2007) found that diabetes and obesity were the main risk factors for HCC in Hispanics and Whites, while HCV was predominant among Blacks. It is important to note, however, that SEER-Medicare analyses are restricted to patients aged ≥65 years, whereas most patients in the NCDB cohort were younger than 65. This difference in age distribution may affect comparisons, particularly given competing causes of mortality in older populations and the tendency for Black patients in our cohort to present at younger ages and with more advanced disease. Alcohol-related HCC was most prevalent in Hispanics and Blacks [18,34,46]. Black patients were more vulnerable to alcohol-related liver damage [47]. In a recent study, Shaltiel et al. reported that Black patients with HCV develop HCC at an earlier stage of liver disease and show a more aggressive phenotype than patients from other racial groups [48]. These observations do not necessarily contradict our propensity-adjusted findings because the NCDB does not capture alcohol exposure, HCV status, cirrhosis severity, or liver function; consequently, these factors could not be included in the matching model and may still contribute to residual survival differences. Although most genetic variations occur among individuals rather than representing a racial group, biological differences may play a role in racial disparities [49]. Another recent study reported that Non-Caucasian HCC patients are more likely to have worse survival than their Caucasian counterparts [50]. Therefore, it’s generally accepted that racial disparities in the United States likely have multiple components [51]. Insurance status and household income significantly affect healthcare access and outcomes, with Black patients being less likely to have adequate insurance coverage and more likely to lose insurance after a cancer diagnosis [52,53]. International studies outside the United States report similar ethnicity-related differences in liver cancer survival, largely driven by disparities in stage at diagnosis, disease etiology, and access to care rather than ethnicity alone [54]. European data demonstrate worse outcomes among ethnic minority and migrant populations, while East Asian countries with established surveillance programs show more uniform survival across groups [55,56]. Healthcare-system implications Our findings also have important healthcare-system implications. Although race was not independently associated with survival after adjustment and propensity score matching, Black patients in the unmatched cohort presented with more advanced disease and had lower rates of surgery, chemotherapy, and radiotherapy. These findings suggest that healthcare systems should focus on modifiable access and care-delivery factors, including earlier identification of at-risk patients, improved linkage to liver cancer surveillance, timely referral to multidisciplinary liver tumor programs, equitable access to surgical and liver-directed therapies, patient navigation, transportation support, insurance counseling, and culturally tailored education. Standardized referral pathways and treatment algorithms may also help reduce facility-level variation and ensure that eligible patients receive guideline-concordant care regardless of race, ethnicity, income, insurance status, or geographic location. Limitations Our study has several important limitations. As a retrospective registry-based study, there is a potential risk of selection bias and misclassifications of racial and ethnic categories because of coding errors [57]. In addition, the cohort was defined using ICD-O-3/WHO primary site/topography codes C22.0 and C22.1 rather than restricted by histology-specific morphology codes. Therefore, although this approach captures a broad NCDB-defined primary liver cancer population and is consistent with the study objective, rare non-HCC or non-ICC histologic subtypes arising in the liver or intrahepatic bile duct may have been included. To mitigate this limitation, the primary site was included as a covariate in the multivariable Cox regression and propensity score matching analyses. Nevertheless, future studies using histology-specific morphology codes may further refine these findings for HCC-specific or ICC-specific populations. NCDB doesn’t have information regarding certain variables that could affect socioeconomic determinants, including social support, etiology of HCC, and treatment adherence. In addition, because Hispanic ethnicity was included as a covariate in the race-focused propensity score matching model, the matched cohort was not specifically designed to estimate the independent causal effect of ethnicity. Accordingly, ethnicity-related findings from the matched cohort should be interpreted cautiously and as secondary prognostic associations. Missing data were handled by categorizing missing values as “unknown” and retaining these patients in the analysis. This approach preserved sample size and avoided excluding patients with incomplete registry data; however, it may introduce bias if missingness is not random. For example, missingness in stage, treatment, insurance, income, or education variables may differ by race, ethnicity, facility type, or access to care, potentially obscuring or exaggerating true associations. The stage variable had substantial missingness, which is a recognized limitation of NCDB-based research and may affect the validity and generalizability of survival estimates. Alternative approaches, including multiple imputation or complete-case sensitivity analyses, may provide complementary estimates in future studies by evaluating the robustness of findings under different missing-data assumptions [58,59]. The NCDB also does not capture several clinically relevant variables, including liver disease etiology, hepatitis B or C status, alcohol-associated liver disease, cirrhosis severity, MASLD, social support, treatment adherence, and liver function measures, all of which may influence treatment selection and survival. Conclusion In conclusion, race was not independently associated with OS after multivariable adjustment and propensity score matching among patients with primary liver cancer in the NCDB. The unadjusted survival disadvantage observed among Black patients appeared to be largely explained by differences in stage, treatment receipt, socioeconomic factors, healthcare access, and other measured covariates. Hispanic ethnicity remained associated with improved survival in secondary analyses, a finding that should be interpreted cautiously and warrants further investigation. Supplementary Material Supplementary_Material_Hepatic_Oncology_R3.docx Acknowledgment No generative artificial intelligence (AI) tools were used in the preparation, writing, editing, analysis, or submission of this manuscript. This research was presented at the 2023 ASCO Gastrointestinal Cancers Symposium, January 19–21, 2023, San Francisco, CA, USA. Funding Statement This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Ethical declaration This retrospective study used de-identified National Cancer Database data and was deemed exempt by the Institutional Review Board of the University of Texas Southwestern Medical Center. The requirement for informed consent was waived because the study used de-identified registry data. Reviewer disclosure Peer reviewers on this manuscript have no relevant financial or other relationships to disclose. Disclosure statement The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties. No writing assistance was utilized in the production of this manuscript. Data availability statement The data that support the findings of this study were obtained from the National Cancer Database Participant User File. These de-identified patient-level data are subject to third-party access restrictions and cannot be redistributed by the authors. Eligible investigators may request access through the National Cancer Database Participant User File application process and applicable data use agreement. References Papers of special note have been highlighted as either of interest (*) or of considerable interest (**) to readers. Associated Data Supplementary Materials Supplementary_Material_Hepatic_Oncology_R3.docx Data Availability Statement The data that support the findings of this study were obtained from the National Cancer Database Participant User File. These de-identified patient-level data are subject to third-party access restrictions and cannot be redistributed by the authors. Eligible investigators may request access through the National Cancer Database Participant User File application process and applicable data use agreement.