Racial and Ethnic Disparities in Survival Among Patients with Hepatocellular Carcinoma in the United States: A Systematic Review and Meta-Analysis

Abstract

Background and Aims:

Hepatocellular carcinoma (HCC) is the fastest rising cause of cancer-related death in the United States; however, HCC incidence and mortality are not equally distributed among racial/ethnic groups. Our aim was to characterize the direction and magnitude of racial/ethnic disparities in overall survival and early tumor detection among patients with HCC.

Methods:

We searched MEDLINE, EMBASE and Cochrane databases from inception through August 2020 for studies reporting HCC outcomes (early stage presentation and overall survival) by race and ethnicity. We calculated pooled hazard ratios (HRs) and odds ratios (ORs) for each racial/ethnic group (White, Black, Hispanic, Asian) using the DerSimonian and Laird method for a random-effects model.

Results:

We identified 35 articles comprising 563,097 patients (53.0% White, 17.3% Black, 18.4% Hispanic, 5.0% Asian). Compared to Whites, Black patients had worse survival (pooled HR 1.08, 95%CI 1.05 – 1.12), whereas Hispanic (pooled HR 0.92, 95%CI 0.87 – 0.97) and Asian (pooled HR 0.81, 95%CI 0.73 – 0.88) patients had better survival. Among articles reporting tumor stage (n=20), Blacks had lower odds of early stage HCC compared to Whites (OR 0.66, 95%CI 0.54 – 0.78). Conversely, there was no difference in odds of early HCC detection for Asian (OR 1.01, 95%CI 0.97 – 1.05) or Hispanic patients (OR 0.87, 95%CI 0.74 – 1.01) compared to Whites. The most common limitation of studies was risk of residual confounding from socioeconomic status and liver dysfunction.

Conclusions:

There are significant racial and ethnic disparities in HCC prognosis in the United States, with Black patients having worse overall survival and Hispanic and Asian patients having better overall survival compared to White patients. Interventions are needed to reduce disparities in early HCC detection to improve HCC prognosis.

INTRODUCTION

Hepatocellular carcinoma (HCC) is the fourth leading cause of cancer-related death worldwide, with the highest burden in East Asia and Africa. In the United States, it is the fastest rising cause of cancer-related death, with a disproportionate impact on racial and ethnic minority populations.¹ Disparities in HCC incidence have been consistently documented over the past three decades², with non-Hispanic Black, Hispanic, and Asian populations having higher age-adjusted incidence rates compared to non-Hispanic whites.³ Disparities also appear to exist for HCC prognosis, though data conflict on the direction and magnitude of racial and ethnic differences in overall survival.^4–8

Disparities may occur at various points along the cancer care continuum, including: access to subspecialty care, adherence to screening programs, and receipt of timely diagnosis and treatment.^9–13 Further, the mechanisms underlying disparities in cancer risk and outcomes are complex and can be conceptualized across several domains (e.g. biological, behavioral, sociocultural/environmental, and healthcare system) and at individual, interpersonal, community, and societal levels.¹⁴ One of the strongest drivers of HCC prognosis is tumor stage, with patients diagnosed at an early stage eligible for potentially curative therapies whereas patients diagnosed at later stages have a median survival of 12–18 months.¹⁵ Therefore, HCC screening is recommended by professional society guidelines in at-risk individuals, such as those with cirrhosis; however, screening continues to be underutilized in clinical practice, particularly among racial/ethnic minorities and the socioeconomically disadvantaged.¹⁰ Disparities have similarly been demonstrated for HCC treatment receipt, with Black patients less likely than White patients to receive curative treatments even among those detected at an early stage.⁹

Improving our understanding of existing disparities in HCC prognosis is the crucial first step to identify targets and implement interventions to eliminate disparities and achieve equitable outcomes for all patients. Therefore, our aim was to conduct a systematic review and meta-analysis to characterize and quantify racial and ethnic disparities in early tumor detection and overall survival among patients with HCC in the United States.

METHODS

Data Sources and Search Strategy

We searched Ovid MEDLINE, EMBASE, and the Cochrane Library from inception to August 1, 2020 using the search terms detailed in Supplemental Material. We performed a manual search of American Association for the Study of Liver Diseases, Digestive Diseases Week, and American Society for Clinical Oncology meeting abstracts from 2017–2019, as well as recursive searches of reference lists from included articles to identify articles missed by the original search terms. This study was conducted in accordance with Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines.¹⁶

Outcomes and Study Selection

Our primary outcome of interest was overall survival; the secondary outcome of interest was detection of HCC at an early stage. Studies were selected for inclusion if they met the following criteria: 1) original data characterizing overall survival and/or tumor stage at diagnosis, stratified by race/ethnicity among patients with HCC in the United States; and 2) included at least two distinct racial/ethnic groups. Studies were excluded if they 1) had incomplete data to calculate hazard ratios; 2) evaluated outcomes among patients post-liver transplantation; 3) lacked original data (e.g. reviews, commentaries, letters to editor); 4) were not conducted in the United States; 5) were non-English language studies or non-human studies; or 6) used overlapping cohorts. For studies with overlapping cohorts, we selected the study with the most contemporary cohort and/or the most comprehensive cohort for inclusion. If studies used the same cohort but assessed different outcomes (e.g. Black-White vs. Hispanic-White comparisons), each study was included for the applicable analysis.

After duplicates were removed, all studies identified during the initial search were independently screened for inclusion and reviewed by two authors (N.E.R. and C.C.) in a multistep process with any disagreements resolved by discussion with a third author (A.G.S.). Articles were iteratively screened based on title, abstract, and finally full text review for inclusion into the study (Figure 1).

Figure 1A.

Risk of death among Black vs White (Ref) patients with HCC

Figure 1B. Risk of death among Hispanic vs White (Ref) patients with HCC

Figure 1C. Risk of death among Asian vs White (Ref) patients with HCC

Data Extraction and Quality Assessment

Next, two authors (N.E.R. and C.C.) extracted data from included studies using standardized forms, including study years, study design, characteristics of study cohort (including race, ethnicity, sex, age, liver disease etiology, proportion with Child Pugh C cirrhosis), proportion of HCC detected at an early stage (total and by race and ethnicity), HCC treatments received, overall survival (by race and ethnicity), and covariates included in survival analyses. All included studies reported race/ethnicity as mutually exclusive categories: non-Hispanic white (White), non-Hispanic black (Black), Hispanic, Asian, and “other” race/ethnicity. We assessed the risk of bias for each individual study using a modified Newcastle-Ottawa scale, which assesses selection of the study cohort, comparability of study groups, and adequacy of assessing the outcome of interest. We specifically assessed the following: 1) patient selection, 2) methods for ascertainment of the outcomes (i.e. early tumor detection and overall survival), 3) factors controlled for in analyses (e.g., tumor stage, HCC treatment, socioeconomic factors), and 4) length of follow-up. All discrepancies were resolved by consensus.

Data Synthesis and Statistical Analysis

To evaluate racial and ethnic disparities in HCC overall survival, we recorded unadjusted and adjusted hazard ratios (HR) and 95% confidence intervals (CI) from each study for each racial/ethnic group compared to Whites. When available, we preferentially used adjusted HRs in our analyses. To evaluate racial and ethnic disparities in early tumor detection, we recorded the total number of HCC diagnosed and proportion diagnosed at an early stage for each racial and ethnic group. We recorded the definition of early HCC stage used in each study, including 1) Milan criteria¹⁷; 2) BCLC stage 0/A¹⁸; 3) localized (SEER stage); or 4) AJCC stage I/II. For studies that recorded multiple definitions, we preferentially recorded BCLC stage or Milan Criteria, as these are the most clinically relevant and incorporated into professional society guidelines.¹⁹ For each study, we calculated an odds ratio (OR) for early detection for each racial/ethnic group (Black, Hispanic, Asian), using non-Hispanic white as the reference group.

We calculated pooled hazard ratio estimates (for overall survival) and pooled odds ratio (for early tumor detection) estimates using the DerSimonian and Laird method for a random effects model. We evaluated heterogeneity both by visual examination of forest plots and I² statistic. An I² >75% indicates a high level of heterogeneity, whereas I² values between 50–75% indicate moderate heterogeneity. We then performed sensitivity analyses, removing outlier studies with differences in study design to evaluate for undue effects of a single study on heterogeneity. We performed subgroup analyses by cohort type (i.e., academic or hospital-based vs population-based) and among the subset of studies that defined early stage HCC by BCLC system or Milan criteria.

Risk of publication bias was evaluated for each study outcome by visual inspection of funnel plots and Egger’s test. All analyses were performed using Stata version 16.1 (StataCorp, College Station, TX).

RESULTS

Literature Search

We identified 4310 unique citations, of which 576 abstracts were potentially relevant upon initial review. Among 238 full texts that were reviewed, 134 met inclusion criteria; we identified an additional 2 studies from recursive searches. After applying exclusion criteria including overlapping cohorts and lack of granular data for racial/ethnic disparities (Supplemental Material), 35 studies were included in the meta-analysis.^{4–8, 20–49} A flow diagram of study selection is outlined in Supplemental Figure 1. Visual inspection of funnel plots suggested asymmetry with small studies with small effect sizes potentially missing for Hispanic-white disparities in survival and small studies with large effect sizes potentially missing for Black-White disparities in stage (Supplemental Figures 2 and 3). However, we found no evidence of publication bias by Egger’s test except for studies examining Black-White disparities in stage (p=0.02; p>0.10 for all other comparisons).

Characteristics of Included Studies

Of the 35 studies (n=563,097 patients), 29 characterized racial and ethnic disparities in overall survival and 20 studies characterized disparities in early HCC detection. Study characteristics are summarized in Table 1. In brief, all studies were retrospective in design, and all were published after 2005. Twenty-one studies (n=16,913) assessed HCC outcomes among patients at academic centers, whereas 14 studies (n=571,787) were conducted using data population-based cohorts including SEER, state cancer registries and claims databases. Study populations were racially and ethnically diverse, with 53.0% White, 17.3% Black, 18.4% Hispanic, 5.0% Asian and 6.2% categorized as “Other”; and most studies reported HCV as the most common liver disease etiology.

Table 1.

Study Characteristics

Author, year	Study Period	Data Source	No. patients with HCC	No. patients with HCC by race and ethnicity (White/Black/Hispanic/Asian)	Liver disease etiology (%)	Cirrhosis (%)	Child Pugh C (%)
Davila, 2006	1987–2001	SEER 9	9919	5426/1278/746/1573	NR	NR	NR
Kemmer, 2008	2000–2005	University of Cincinnati	169	108/49/0/0	HCV - 73%, HBV - 4%, ETOH 11%, NASH - 8%	100.0	24% (NR by race)
Cubillas*, 2009	2002–2008	Dallas VA	179	109/51/19/0	NR	NR	NR
Ochner, 2010	1993–2008	University of Hawaii	157	85/0/0/72	HCV - 55%, HBV - 25%	NR	NR
Yu, 2010	2002–2008	Columbia University - NYP	462	184/63/139/47	HCV - 61%, HBV 19%, ETOH 16%, NASH 16%	NR	W - 25%, B - 35%, H - 46%
Zaydfudim, 2010	2004–2006	Tennessee Cancer Registry	680	513/127/0/0	NR	NR	NR
Jan, 2012	2003–2011	Tulane University	206	138/68/0/0	HCV - 67%	94.2	NR (median MELD W – 11.3, B - 10.8)
Aparo, 2014	2000–2011	Montefiore Medical Center	633	144/180/309/0	HCV - 65.9%, HBV - 10.4%, ETOH - 46.9%	80.9	NR
Al-Rajabi, 2015	2008–2013	UT San Antonio	107	26/2/77/2	HCV - 66%, HBV 4.7%, ETOH 58.9%, NAFLD 2.8%, other 18.7%	100.0	0%
Hoehn, 2015	1998–2011	National Cancer Database (NCDB)	143692	110547/21102/0/12043	NR	NR	NR
Aru*, 2016	1995–2005	University of Mississippi	421	233/170/0/0	NR	NR	NR
Chan*, 2016	2014–2016	Alameda Health System	191	33/46/23/84	NR	79.6	NR
Stewart, 2016	1988–2012	California Cancer Registry	33270	12710/2609/8500/9170	NR	NR	NR
Alkhalili, 2017	2000–2014	University of New Mexico	326	106/0/183/0	HCV - 60.1%, HBV - 5.2%, ETOH - 50.3%, NASH - 6.4%	92.0	53% Child B/C (W - 41.5%, H - 58.5%)
Chayanupatkul, 2017	2001–2013	VA Corporate Data Warehouse	317	158/106/0/29	HBV - 100% (some HIV/HCV coinfected)	79.8	(mean CP score: W - 7.0, B - 6.7, A - 6.1)
Mokdad, 2017	2001–2012	Texas Cancer Registry	15932	7143/2091/5912/741	NR	NR	NR
Venepalli, 2017	2005–2011	University of Illinois at Chicago	195	90/61/44/0	HCV - 68%, HBV - 8.1%, ETOH - 28%, NASH - 18%	NR	NR (median MELD 11.0); a higher proportion of Hispanics had ascites
Franco, 2018	2002–2012	SEER 18	43868	29279/5643/0/0	NR	NR	NR
Gabriel, 2018	1998–2012	National Cancer Database (NCDB)	4364	3172/478/0/578	NR	NR	NR
Atiemo, 2019	2006–2012	HealthLNK database (6 hospitals in Chicago)	2825	1386/538/398/131	NR (for HCC)	NR	NR
Dakhoul, 2019	2000–2014	Indiana University	1196	1032/164/0/0	HCV - 53%, NAFLD - 17%, Alcohol - 12%, HBV - 3%, autoimmune/other - 5%	88.5	W - 20%; B – 21%
Estevez, 2019	1991–2016	3 US centers (Stanford, Mayo, Sinai)	1156	578/578/0/0	HCV - 68%, HBV - 11%, 8% ETOH, 5% NASH	73.4	W - 7%, B - 7%
Jones, 2018	2004–2014	University of Miami / Jackson Memorial Hospital	901	414/135/310/22	HCV - 64%, HBV - 12%, ETOH - 26%, NAFL - 12%	86.9	NR (median MELD 8 (IQR 5 – 12.6)
Kim, 2018	1998–2015	Stanford University	2106	785/44/355/823	HCV - 57%, HBV - 22%	83.9	13.7% (NR by race)
Sobotka, 2018	2010–2013	Nationwide Inpatient Sample (NIS)	62604	32428/9726/8988/0	HCV - 17.3%, HBV - 5.5%, ETOH - 16.0%, NASH - 35.1%	NR	NR
Jones, 2019	2004–2013	Florida Cancer Data System	10852	7282/1420/1708/346	NR	NR	NR
Kuftinec, 2019	2008–2014	University of California Davis	325	255/0/70/0	HCV - 74%, HBV - 4%, ETOH −33%, NAFLD −6%	281	W - 11%, H - 12%
Rich, 2019	2008–2017	UT Southwestern and Parkland Hospital	1117	401/384/332/0	HCV - 65%, HBV - 5%, ETOH - 14%, NASH - 12%	NR	B - 9.4%, H - 18.45, A - 13.5%
Yang*, 2019	2004–2014	National Cancer Database (NCDB)	6261	3824/869/984/556	NR	NR	NR
Zhang, 2019	2005–2016	Montefiore Medical Center	991	106/242/332/18	HCV - 69%, HBV 7%	NR	13% (NR by race)
Brar, 2020	2000–2014	SEER-Medicare	11522	7405/823/1336/1846	HCV - 8%, Metabolic - 33%, ETOH - 5%, HBV - 2%, unknown - 24%, multiple - 27%	NR	NR
Kangas-Dick, 2020	2005–2015	Nationwide Inpatient Sample (NIS)	200163	90944/36172/31104/10301	HCV - 100%	NR	NR
Lee, 2020	2012–2014	US Safety Net Collaborative	1479	714/421/310/115	Viral - 41%, ETOH - 10%, Viral/ETOH - 17%, Unknown 25%	NR	NR (median MELD 10)
Pomenti, 2020	2000–2014	5 US centers (Atrium Health, Columbia University, Indiana University, MD Anderson, Vanderbilt University	4217	3693/0/521/0	HCV - 18%, HCV+ETOH - 15.9%, NAFLD - 13.8%, HBV - 2%, ETOH - 11.3%	NR	NR (median MELD W – 11.8; H - 11.3)
Scaglione, 2020	2012–2013	4 US centers (Michigan, Loyola, Parkland, Ben Taub)	379	200/75/74/0	HCV - 58%, NASH - 22%, ETOH - 16%	88.1	W - 15.4%, B 12.7%, H - 31.1%

NR – not reported; W – White; B – Black; H – Hispanic; A - Asian

Early stage HCC was defined by BCLC stage or Milan criteria in 14 studies, AJCC stage in 9 studies, SEER stage in 6 studies, unifocal HCC in 2 studies, and not reported in 4 studies (Table 2). Outside of one study only including very early stage and another only with advanced stage HCC, the proportion of patients with early stage HCC varied widely from 17.0% to 68.7% between studies.

Table 2.

Racial and ethnic disparities in early stage HCC and survival

Author, year	Study Period	Data Source	No. patients with HCC	Definition of Early Stage HCC	No. Early stage (%)	No. Early stage by race and ethnicity (White/Black/ Hispanic/Asian)	HCC treatment	Crude survival by race and ethnicity	Factors adjusted for in survival analyses
Davila, 2006	1987–2001	SEER 9	9919	SEER	2798 (28.2%)	1694/376/266/462	Surgical - 1044, Locoregional - 188, None - 8312	1 year survival: W - 22.4%, B - 21.4%, H - 21.9%, A - 31.5%	age, sex, SEER registry, time period, stage, treatment
Kemmer, 2008	2000–2005	University of Cincinnati	169	Milan	82 (48.5%)	61/14/0/0	NR	NR	NR
Cubillas*, 2009	2002–2008	Dallas VA Medical Center	179	BCLC (A/B combined)	103 (57.5%)	70/21/12/0	NR	NR	NR
Ochner, 2010	1993–2008	University of Hawaii	157	NR (%Tumor over 10 cm)	NR	NR	None - 68, Resection - 34, OLT - 21, Other - 75	Mean survival: W – 43.3 months vs A – 10.9 months	age, sex, smoking, etiology, symptoms, encephalopathy, screening
Yu, 2010	2002–2008	Columbia University - NYP	462	Milan	160 (34.6%)	76/15/40/21	OLT - 175, no OLT 287	5 year survival: W 38.0%, B 0.0%, H 16.7%, A 15.1%	age, sex, etiology, stage, comorbidities, tobacco, income
Zaydfudim, 2010	2004–2006	Tennessee Cancer Registry	680	AJCC	251 (36.9%)	NR	Any surgery - 181	NR	age, sex, rural residence, insurance status
Jan, 2012	2003–2011	Tulane University	206	Granular tumor factors only	NR	NR	OLT - 51, other treatments - 145	NR	demographic, socioeconomic, clinical factors including tumor size, insurance status
Aparo, 2014	2000–2011	Montefiore Medical Center	633	AJCC	320 (50.6%)	73/90/157/0	OLT - 34, resection - 45, LRT - 312, systemic - 186	NR	age, sex, MELD, tumor stage, AFP, treatment
Al-Rajabi, 2015	2008–2013	UT San Antonio	107	BCLC (all advanced HCC)	0 (0.0%)	0/0/0/0	Sorafenib - 107	Median OS: H – 11.2 months vs W - 8.7 months	unadjusted
Hoehn, 2015	1998–2011	National Cancer Database (NCDB)	143692	AJCC	24432 (17.0%)	18941/3398/0/2093	Resection - 15607, Transplant - 9308, RFA - 6889	NR	pathologic stage
Aru*, 2016	1995–2005	University of Mississippi	421	AJCC	NR	NR (HRs reported)	NR	NR	NR
Chan*, 2016	2014–2016	Alameda Health System	191	Milan	51 (26.7%)	NR	NR	NR	NR
Stewart, 2016	1988–2012	California Cancer Registry	33270	SEER	14188 (42.6%)	5355/987/3825/3912	OLT/Resection - 4344, Local - 2288, None 11653	NR	age, sex, socioeconomic status, region, stage, time period of diagnosis, treatment
Alkhalili, 2017	2000–2014	University of New Mexico	326	AJCC	81 (24.8%)	30/0/51/0	Resection - 84, Chemotherapy - 70, Local therapy - 91	Median OS: W – 14 months vs H – 11 months	unadjusted
Chayanupatkul, 2017	2001–2013	VA Corporate Data Warehouse	317	BCLC	128 (40.3%)	NR	NR	NR	age, presence of cirrhosis, MELD, tumor stage, AFP, HIV infection
Mokdad, 2017	2001–2012	Texas Cancer Registry	15932	SEER	6506 (40.8%)	NR	OLT - 835, Resection - 1198, Local therapy - 1230, Chemotherapy - 1527, Combination - 1527, None 6775	NR	age, sex, poverty index, year of HCC diagnosis, safety net designation, tumor stage, treatment
Venepalli, 2017	2005–2011	University of Illinois at Chicago	195	unifocal	99 (50.7%)	49/30/20/0	OLT - 34, resection - 2, LRT −154, systemic - 6, None/BSC - 7	NR	unadjusted
Franco, 2018	2002–2012	SEER 18	43868	SEER	21292 (48.5%)	14364/2576/0/0	Resection - 4208, OLT - 2877, local therapy - 4611	NR	age, sex, stage, grade, treatment
Gabriel, 2018	1998–2012	National Cancer Database (NCDB)	4364	AJCC	881 (20.2%)	NR	Resection - 4364	NR	unadjusted
Atiemo, 2019	2006–2012	HealthLNK database (6 hospitals in Chicago)	2825	NR	NR	NR	NR	NR	unadjusted
Dakhoul, 2019	2000–2014	Indiana University	1196	BCLC, Milan	564 (47.1%)	495/69/0/0	OLT - 291, resection - 181, LRT - 647, sorafenib - 124	1 year (B - 54% W - 56%), 3 year (B - 27% W - 32%), 5 year (B - 14% W - 20%)	NR
Estevez, 2019	1991–2016	3 US centers (Stanford, Mayo, Sinai)	1156	BCLC	376 (32.5%)	202/174/0/0	OLT - 26, resection - 191, ablation - 155, LRT - 537, systemic - 67	5 year (W - 37%, B – 38%)	age, sex, insurance, stage, alcohol use, Child Pugh, treatment, etiology
Jones, 2018	2004–2014	University of Miami / Jackson Memorial Hospital	901	BCLC, Milan	431 (47.8%)	225/41/145/9	Resection - 110, OLT 284, LRT - 555, systemic - 236	1 year (W – 70.5%, B – 54.8%, H – 61.6%, A – 59.1%); 5 year (W – 27.5%, B – 14.8%, H – 21.6%, A – 27.3%)	age, sex, insurance, hospital, alcohol, tobacco, family history, PVT, HE, BCLC stage, AFP, treatment, etiology
Kim, 2018	1998–2015	Stanford University	2106	Milan	866 (41.1%)	NR	Resection - 280, OLT 272, LRT - 1002, BSC - 497	5 year (W - 32%, B - 20%, H - 36%, A - 40%)	age, sex, diagnosis year, etiology, MELD, treatment, other factors
Sobotka, 2018	2010–2013	Nationwide Inpatient Sample (NIS)	62604	NR	NR	NR	OLT - 2045, Resection - 7501, LRT - 8754	NR	age, sex, geographic region, etiology, metastatic disease, comorbidity index, treatment
Jones, 2019	2004–2013	Florida Cancer Data System	10852	SEER	4875 (44.9%)	3264/532/774/166	OLT - 784, Resection - 722, None - 8402	1 year (W – 52%, B – 46%, H – 55%, A 58%); 5 year (W – 36%, B – 32%, H – 37%, A – 46%)	age, sex, payer, tumor stage, treatment
Kuftinec, 2019	2008–2014	University of California Davis	240	BCLC	165 (68.7%)	138/0/27/0	NR	NR	NR
Rich, 2019	2008–2017	UT Southwestern and Parkland Hospital	1117	Milan, BCLC	463 (53.7%)	185/149/129/0	Resection - 132, OLT 73, LRT - 417, Systemic - 120, BSC - 366	NR	Child Pugh, tumor stage, treatment, insurance status
Yang*, 2019	2004–2014	National Cancer Database (NCDB)	6261	AJCC	6261 (100%)	NR	Ablation, Resection, Transplant (NR)	NR	age, insurance status, facility type, region, AFP, comorbidity index, MELD
Zhang, 2019	2005–2016	Montefiore Medical Center	991	AJCC	329 (33.2%)	NR	Resection - 59, OLT - 15, LRT - 435, systemic - 107	NR	unadjusted
Brar, 2020	2000–2014	SEER-Medicare	11522	SEER	5661 (49.1%)	NR	Resection - 1034, OLT - 142, Radiation - 325, Ablation - 346, LRT - 1670	NR	age, sex, comorbidity index, etiology, alcohol use, tumor characteristics, treatment
Kangas-Dick, 2020	2005–2015	Nationwide Inpatient Sample (NIS)	200163	NR	NR	NR	Trends in procedures by year but no exact numbers	NR	sex, comorbidity, insurance status, hospital type, region, year of diagnosis
Lee, 2020	2012–2014	US Safety Net Collaborative	1479	AJCC	923 (21.9%)	NR	OLT - 229, Resection - 75, LRT - 683, systemic - 131, None - 361	NR	sex, hospital setting, insurance, etiology, MELD, tumor stage, treatment
Pomenti, 2020	2000–2014	5 US centers (Atrium Health, Columbia University, Indiana University, MD Anderson, Vanderbilt University)	4217	Milan	1530 (36.3%)	1393/0/137/0	Resection - 527, OLT - 744, None - 1031	Median OS: H – 1.4 years vs W – 1.3 years	age, sex, tumor stage, treatment
Scaglione, 2020	2012–2013	4 US centers (Michigan, Loyola, Parkland, Ben Taub)	379	Milan	173 (45.6%)	96/30/37/0	OLT/resection/ab lation - 102, Other - 163, None - 114	1 year survival - 62.4%	treatment, insurance

NR – not reported; W – White; B – Black; H – Hispanic; A - Asian

Racial and Ethnic Disparities in Overall Survival

Twenty-nine studies (n=560,701 patients) assessed disparities in overall survival (17 Black-White^{4–8, 20, 24, 25, 27, 31, 32, 34, 38–40, 47, 49}_; 18 Hispanic-White^{4, 5, 8, 20, 26, 30, 32, 33, 36, 39, 41, 43–49}; and 11 Asian-White^{4, 20, 23, 31, 32, 35, 39, 44–47}). Black patients had worse survival compared to Whites (pooled HR 1.08, 95%CI 1.05 – 1.12), whereas Hispanics (pooled HR 0.92, 95%CI 0.87 – 0.97) and Asians (pooled HR 0.81, 95%CI 0.73 – 0.88) had better survival than Whites (Figures 1A–C). All 17 studies comparing Black-White survival reported adjusted HRs, with 13 studies adjusting for tumor stage and 12 studies adjusting for treatment receipt (Table 2). Among 18 studies reporting Hispanic-White comparisons, 5 studies reported only unadjusted HRs^{26, 30, 33, 36, 44}; when these studies were removed, results were unchanged (pooled HR 0.92, 95%CI 0.87 – 0.97). Similarly, among the 11 studies reporting Asian-White comparisons, 2 reported only unadjusted HRs^{35, 44} and results were unchanged when these studies were removed (pooled HR 0.82, 95%CI 0.73 – 0.93).

We performed pre-planned subgroup analyses among the subset of studies that reported both early detection (using BCLC stage or Milan criteria) and survival (Black-White = 7; Hispanic-White = 7; Asian-White =3). Results were consistent, with Blacks having worse survival than Whites (HR 1.12, 95%CI 1.10 – 1.14), whereas Hispanic (HR 0.86, 95%CI 0.79 – 0.92) and Asians (HR 0.65, 95% CI 0.54 – 0.76) had better survival compared to Whites. We also performed subgroup analyses among studies performed in academic vs population-based cohorts. Among population-based cohort studies, Black patients had worse survival (HR 1.09, 95%CI 1.06 – 1.12), Hispanic patients had similar survival (HR 0.97, 95% CI 0.93 – 1.02), and Asian patients had better overall survival (HR 0.83, 95% CI 0.75 – 0.92) compared to Whites. In academic cohorts, Black patients had similar survival compared to Whites (HR 1.00, 95% CI 0.85 – 1.15), whereas Hispanic (HR 0.81, 95% CI 0.72 – 0.90) and Asian (HR 0.73, 95% CI 0.56 – 0.90) patients had better overall survival than Whites.

Racial and Ethnic Disparities in Early Tumor Detection

Twenty studies (n=209,622 patients) compared the proportion of patients with early stage HCC between two races or ethnicities (14 Black-White^{4–8, 20–22, 27–29, 37, 38, 49}; 13 Hispanic-White^{4–6, 8, 20, 22, 30, 33, 38, 41, 42, 48, 49}; 5 Asian-White^{4, 5, 20, 27, 38, 41}). Blacks (pooled OR 0.66, 95% CI 0.54 – 0.78) were less likely to be diagnosed with early stage HCC compared to Whites. Conversely, there was no significant difference in early stage detection among Asian (pooled OR 1.01, 95% CI 0.97 – 1.05) or Hispanic patients (pooled OR 0.86, 95%CI 0.70 – 1.02) compared to Whites (Figures 2A–C). On visual inspection of the Black-White disparities forest plot, the study by Chan et al appeared to be an outlier. A sensitivity analysis removing this study (Supplemental Figure 4) showed similar odds of early tumor detection for Blacks compared to Whites (pooled OR 0.68, 95%CI 0.56 – 0.80).

Figure 2A.

Odds of early stage HCC at diagnosis in Blacks vs Whites (Ref)

Figure 2B. Odds of early stage HCC at diagnosis in Hispanics vs Whites (Ref)

Figure 2C. Odds of early stage HCC at diagnosis in Asians vs Whites (Ref)

Correlates of Racial/Ethnic Disparities

Few studies evaluated factors associated with racial and ethnic disparities in survival (Supplemental Table 1). The two most common factors included in multivariable analyses were HCC treatment receipt (n=20) and insurance status (n=16), whereas other factors (e.g. healthcare utilization, geographic region, and socioeconomic status) were not routinely reported or included in multivariable models for most studies. Only three studies adjusted for healthcare utilization, with two adjusting for receipt of subspecialty care^{8, 49} and one study adjusting for receipt of other preventative care and age-appropriate cancer screenings.⁴¹.

Despite extensive literature highlighting the intersection of race and socioeconomic status, only seven studies to date assessed the impact of socioeconomic status (at either at the individual or neighborhood level) on disparities in HCC prognosis. For example, in one study analyzing data from the Florida Cancer Data System, neighborhood poverty level was highest among Hispanics compared to other races/ethnicities, although they experienced better overall survival compared to Blacks and Whites.⁴¹ At the individual level, Scaglione et al found race and ethnicity were no longer associated with survival after adjusting for insurance status.⁴⁹ None of the included studies performed more complex analyses used in health disparities research⁵⁰, such as formal mediation analysis or multi-level analyses including factors across patient-, provider-, health system-, and neighborhood-levels.

Quality Assessment

The quality assessment for included studies is outlined in Table 3. All studies were retrospective, raising the potential for missing data and measurement bias for important prognostic factors (e.g. ECOG performance status). In terms of patient selection, 21 studies were conducted at academic medical centers, introducing a selection bias of patients with adequate access and engaged in routine medical care. Given all studies were retrospective cohorts, there is risk of misclassification for race/ethnicity. Many studies used self-reported race/ethnicity collected for other purposes, which does not account for multi-racial individuals and may not have been reliably collected. For outcome ascertainment, overall survival (our primary outcome) was ascertained by chart review and/or linkage to the National Death Index or state registries in 22 studies. Studies only relying on chart review were prone to ascertainment bias and may not have accurately captured date of death for those who died elsewhere. Most included studies had a high risk of bias from residual confounding. Among the 29 studies reporting disparities in survival, 6 only reported unadjusted survival analyses. Most studies with adjusted analyses (19 of 23) controlled for tumor stage and/or treatment receipt, however only 6 studies adjusted for liver disease severity and only 14 adjusted for socioeconomic status. As discussed above, no studies to date have performed formal mediation or multi-level analyses to identify factors mediating racial and ethnic disparities in prognosis. Finally, duration of follow-up was only reported in 11 of 35 studies; among these, two studies had a high risk of bias related to short duration of follow-up <1 year.

Table 3.

Quality assessment and risk of bias of included studies

Author, year	Population-based cohort	Ascertainment of outcome (tumor stage)	Detailed information on ascertainment of outcome (survival)	Reported length of follow-up time	Analyses controlled for prognostic factors (e.g., tumor stage, treatment, liver function)	Adjusted for socioeconomic and/or geographic factors
Davila, 2006	Yes	Yes	Yes	No	Yes**	No
Kemmer, 2008	No	Yes	No	No	No	No
Cubillas § , 2009	No	Yes	No	No	No	No
Ochner, 2010	No	No	Yes	No	Yes	No
Yu, 2010	No	Yes	Yes	No	Yes*	Yes*
Zaydfudim, 2010	Yes	No	Yes	No	Yes	Yes**
Jan, 2012	No	No	No	No	Yes*	Yes*
Aparo, 2014	No	Yes	Yes	No	Yes**	No
Al-Rajabi, 2015	No	Yes	No	No	No	No
Hoehn, 2015	Yes	Yes	No	Yes	Yes*	No
Aru § , 2016	No	No	No	No	No	No
Chan § , 2016	No	No	No	No	No	No
Stewart, 2016	Yes	Yes	Yes	No	Yes**	Yes**
Alkhalili, 2017	No	Yes	Yes	No	No	No
Chayanupatkul, 2017	Yes	No	Yes	Yes	Yes*	No
Mokdad, 2017	Yes	No	Yes	No	Yes**	Yes*
Venepalli, 2017	No	Yes	Yes	Yes	No	No
Franco, 2018	Yes	Yes	Yes	Yes	Yes**	No
Gabriel, 2018	Yes	No	No	Yes	No	No
Atiemo, 2019	Yes	No	Yes	No	No	No
Dakhoul, 2019	No	Yes	Yes	No	No	No
Estevez, 2019	No	Yes	Yes	Yes	Yes**	Yes*
Jones, 2018	No	Yes	Yes	No	Yes**	Yes*
Kim, 2018	No	No	Yes	Yes	Yes*	No
Sobotka, 2018	Yes	No	No	No	Yes*	Yes*
Jones, 2019	Yes	Yes	Yes	Yes	Yes**	Yes*
Kuftinec, 2019	No	Yes	No	No	No	No
Rich, 2019	No	Yes	Yes	Yes	Yes**	Yes*
Yang § , 2019	Yes	No	No	No	Yes	Yes**
Zhang, 2019	No	No	Yes	Yes	No	No
Brar, 2020	Yes	No	Yes	No	Yes**	No
Kangas-Dick, 2020	Yes	No	No	No	Yes	Yes*
Lee, 2020	No	No	Yes	Yes	Yes**	Yes*
Pomenti, 2020	No	Yes	Yes	No	Yes**	No
Scaglione, 2020	No	Yes	Yes	No	Yes*	Yes*

^*One asterisk denotes study adjusted for either tumor stage or treatment; two asterisks denotes study adjusted for both tumor stage and treatment

^§denotes abstract only

DISCUSSION

To our knowledge, this is the first systematic review characterizing racial and ethnic disparities in HCC prognosis. We demonstrated significant racial and ethnic disparities in overall survival and early HCC detection in the United States. Black patients had worse overall survival compared to Whites, whereas Hispanics and Asians had better survival compared to Whites. While Black patients were less likely to be diagnosed with early stage HCC compared to Whites, we found no significant difference in early detection across studies in Hispanics or Asians compared to Whites.

We found Black patients with HCC had worse survival compared to all other racial and ethnic groups, which is consistent with a considerable body of evidence for many cancer types.⁵¹ The National Institute on Minority Health and Health Disparities (NIHMD) framework highlights that reasons for this finding are complex and may be attributed to many interrelated factors, including differences in access to medical care, screening receipt, stage at presentation, and provision of timely guideline-concordant treatment – which can each be attributed to failures at the patient-, provider-, and system-levels.¹⁴ Prior studies have demonstrated lower HCC screening and treatment rates among Black patients compared to other racial/ethnic groups, suggesting these processes may be intervention targets to reduce disparities.¹¹ However, we found a weak to moderate correlation (r=−0.39) between Black-White disparities in early tumor detection and overall survival. Further, Black-White disparities persisted even after adjusting for tumor stage and treatment receipt in most studies. Alternatively, individual- and neighborhood-level factors such as poverty, environmental exposures and other forms of adversity (e.g. discrimination, food insecurity) may also have deleterious epigenetic effects that accumulate during the lifespan and contribute to health disparities later in life.⁵² Notably, we found differences in survival were not observed among studies conducted in academic settings, suggesting that disparities may be mitigated among those who are able to access high-quality tertiary care. System- and societal-level interventions aimed at reducing economic inequalities and improving health care access for all patients, as well as targeted interventions addressing the additional impact of race, are needed to improve care across the continuum and reduce this survival disparity for patients with HCC.

In contrast, we found better survival in both Asians and Hispanics compared to Whites. Whereas multiple prior studies have reported better survival in Asian patients, better survival among Hispanics is less well recognized. Results from our meta-analysis support the idea of a “Hispanic paradox”, where Hispanics have better survival outcomes despite being more socioeconomically disadvantaged in the U.S., a phenomenon previously described in other cancers^53–56 but less so in HCC. Other studies have hypothesized this may be due to geography, with Hispanic patients more likely to live in close proximity to academic centers and therefore, have greater access to multidisciplinary care.⁴¹ The survival advantage seen in Hispanics with HCC may also be in part due to differences in demographics and underlying liver disease etiology, with a higher proportion of women and NAFLD-related HCC than their counterparts.⁵⁷ However, further prospective studies with granular clinical data are needed in diverse patient populations with HCC to confirm this finding and elucidate the underlying mechanism.

Though differences in HCC prognosis appear to extend beyond disparities in healthcare access, the contribution of racial and the ethnic differences in tumor biology appears to be minimal. Studies of HCC tumor doubling time have not demonstrated significant racial or ethnic differences in tumor biology or “aggressiveness”, with the exception of more rapid tumor growth in Asian populations with predominately HBV-related HCC.^{58, 59} There are no significant racial/ethnic differences in the frequency of common somatic mutations associated with HCC, such as mutations in CTNNB1.⁶⁰ However, the role of epigenetic effects (e.g. DNA methylation, which may be induced by environmental stressors), have been investigated in the context of disparities in other cancers (e.g. breast, prostate, colorectal) but have unknown effects on HCC risk and outcomes.⁶¹ Despite extensive data describing racial and ethnic disparities in HCC, fewer studies have attempted to identify root causes of these disparities, thereby limiting efforts to implement interventions. This is likely in part due to the limitations of the study cohorts used; for example, administrative datasets have limited information on social determinants of health and liver disease etiology. Further, there are also notable gender, geographic, and socioeconomic disparities in HCC, and few studies have investigated the intersectionality of these factors with race and ethnicity.^{8, 62} Social determinants of health (e.g. poverty, racism, environmental stress, health literacy, social support) are likely to heavily influence disparities in HCC incidence and mortality, but the impact of these factors remains understudied. Future studies conducted in diverse cohorts that collect multilevel data to characterize granular determinants of racial and ethnic disparities in HCC outcomes will be critical to expand our understanding of causal factors and mechanisms and identify specific, intervenable targets to reduce disparities and achieve equitable outcomes for all patients with HCC.

Though our study has several strengths, meta-analyses are inherently limited by the quality of the included studies. First, all studies were retrospective, with the potential for missing data and unmeasured confounders. Second, research in this area has historically been conducted using large administrative databases (e.g. SEER, state cancer registries), however these registries have important limitations for evaluating HCC prognosis. These registries have limited data on liver disease etiology, liver dysfunction, or treatments beyond the first HCC treatment, and utilize staging systems (e.g. localized, regional, distant) that are less clinically applicable to HCC compared to the widely used BCLC staging system. These registries are also prone to ascertainment bias as HCC is diagnosed radiographically, not histologically, in a majority of cases. Third, though we took care to exclude overlapping cohorts from this analysis, this may not represent completely unique patients given the inclusion of state cancer registries and SEER registries in this study. Fourth, data were limited on survival of Alaskan Natives and Native Americans with HCC – two groups with increasing HCC incidence.⁶³ Fifth, this was a study-level, rather than patient-level meta-analysis, limiting our ability to identify factors associated with the presence of disparities. Lastly, there are inherent challenges in studying cancer disparities, including the accurate ascertainment of race and ethnicity (e.g. based on self-report vs clinician assessment) and classification of multi- ethnic individuals.^{64, 65} It is increasingly believed that race is primarily a social construct⁶⁶, and this concept is supported by the fact that there is as much genetic variability within races than between races. In the U.S., the role of race and ethnicity in cancer disparities is also intertwined with and difficult to separate from socioeconomic disparities and lack of healthcare access, and these factors could not be specifically addressed in this study.

In conclusion, there are significant racial and ethnic disparities in survival among patients with HCC in the United States. Though considerable progress has been made in describing disparities in HCC, data are limited on the specific determinants driving these disparities. To achieve health equity, future studies must focus on taking the critical next steps to determine the root causes of disparities in HCC, identify actionable intervention targets, engage patient and provider stakeholders to implement interventions, and advocate for system-level change to eliminate disparities.

Supplementary Material

1

Supplemental Figure 1. Flow diagram for study selection

Supplemental Figure 2. Funnel plots (A) Black-White survival (Egger’s test p=0.19); (B) Hispanic-White survival (Egger’s test p=0.61); (C) Asian-White survival (Egger’s test p=0.59)

Supplemental Figure 3. Funnel plots (A) Black-White stage (Egger’s test p=0.02); (B) Hispanic-White stage (Egger’s test p=0.12); (C) Asian-White stage (Egger’s test p=0.59)

Supplemental Figure 4. Odds of early stage HCC at diagnosis in Blacks vs Whites (Ref) with Chan outlier removed

What You Need to Know.

Background and Context

Hepatocellular carcinoma (HCC) is the fastest rising cause of cancer-related death in the United States; however, the burden of HCC is not equally distributed among racial/ethnic groups.

New Findings

In this systematic review and meta-analysis of 35 studies comprising 563,097 patients with HCC, we found that compared to White patients, Black patients have worse survival and Hispanic and Asian patients have better survival. This is likely in part due to differences in tumor stage at diagnosis, with Black patients less likely to be diagnosed at an early stage compared to White patients.

Limitations

Meta-analyses are inherently limited by the quality of the included studies. The most common limitation was risk of residual confounding from socioeconomic status and liver dysfunction.

Impact

There are significant racial and ethnic disparities in prognosis among patients with HCC in the United States. Our findings highlight the need for targeted interventions in HCC screening and treatment to reduce disparities and promote health equity.

Lay Summary

There are significant racial and ethnic disparities in outcomes among patients with liver cancer, likely related to differences in tumor stage at time of diagnosis.

Abbreviations:

AJCC

American Joint Committee on Cancer

BCLC

Barcelona Clinic Liver Cancer

HR

hazard ratio

HBV

hepatitis B virus

HCC

hepatocellular carcinoma

HCV

hepatitis C virus

NAFLD

nonalcoholic fatty liver disease

OR

odds ratio

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses