Disparities in Access to Liver Transplant Referral and Evaluation among Patients with Hepatocellular Carcinoma in Georgia Open Access

Hepatocellular carcinoma (HCC) is a major cause of cancer death in the United States (1). According to the American Association for the Study of Liver Diseases practice guidelines, liver transplantation is an effective therapy for patients with HCC meeting the Milan criteria (single tumor ≤5 cm or no more than three tumors each ≤3 cm, no vascular invasion, and no extrahepatic involvement; refs. 2, 3). In some cases, patients with tumors outside these criteria can receive therapies to reduce tumor burden and achieve transplant eligibility, also called downstaging (3). Prompt access to liver transplantation is of particular importance for patients with HCC, as patients can become ineligible for transplant if their disease progresses.

Receiving a transplant is a multistep process requiring referral to a transplant center, completion of a medical and psychosocial evaluation, and selection for placement on the deceased donor waiting list (listing) or identification of an appropriate living donor. Prior studies have demonstrated substantial racial (4–7), socioeconomic (8, 9), and geographic (10) disparities in rates of liver transplantation among patients with HCC. However, little is known about how patients navigate the early parts of the process, including referral and evaluation, in part because data prior to listing are not systematically reported (11). Understanding the steps at which patients experience barriers to care is critical to informing effective interventions to improve equity in access to transplant for patients with HCC.

In this study, we linked data on a subset of patients with HCC who either met the Milan criteria or downstaging criteria in Georgia from the Georgia Cancer Registry (GCR) to transplant center data from the only two liver transplant centers in the state. We describe rates of transplant referral, evaluation initiation, and evaluation completion in this population for the first time, and identify demographic factors associated with completion of each step.

Data on incident HCC diagnosed in the state of Georgia were obtained from the GCR, part of a coordinated system of cancer registries funded by the NCI (Surveillance, Epidemiology, and End Results; SEER registries). SEER registries are noted for their data quality and completeness; prior studies have estimated over 97% completeness of hospital cancer case reporting (12). Data on transplant referral, evaluation, and listing were obtained from the two liver transplant centers in the state of Georgia: Emory Transplant Center and Piedmont Transplant Institute. Probabilistic linkage of these data sources was conducted by the GCR. This study was reviewed and approved by the Emory Institutional Review Board.

Our study population included incident, adult (18 ≤ age ≤ 75) patients with HCC in Georgia diagnosed between January 1, 2010 and December 31, 2019 and followed for outcomes (transplant, death, loss to follow-up) through December 31, 2020. Patients were excluded if they had a diagnosis date that occurred after they initiated evaluation with a transplant center (n = 184) or occurred more than 1 month after liver transplant referral (n = 32).

Cancer registry data formatting precluded full use of the most common liver transplant criteria, the Milan criteria, as inclusion parameters for our cohort. Cases were coded using the clinical tumor—node—metastasis (TNM) staging system. For inclusion, patients must have had no spread to lymph nodes (N0) or metastasis (M0). For T staging, single tumors were coded as T1, while cases coded as T2 could have either multiple tumors with none greater than 5 cm or a single tumor with vascular involvement. Because T2 patients may not fall within Milan criteria, we decided to only include T1 cases in our main analysis (Fig. 1). We conducted a sensitivity analysis including T2 patients with multiple tumors. Our size limit for T1 patients was set at ≤8 cm in accordance with downstaging protocols (13). Patients with tumors that were >5 but ≤8 cm were classified as “downstaging required for transplant.”

Primary outcomes for this analysis were completion of each step of the transplant process prior to waitlisting: referral, evaluation initiation, and evaluation completion. Time to event was calculated as the time between the event date and the date of the prior event in the process (for referral, this was diagnosis). For patients referred to both transplant centers in our study, only the first referral following HCC diagnosis was included.

For 27 patients whose recorded diagnosis date occurred after referral but within 30 days of referral and prior to initiating transplant evaluation, referral was assumed to occur contemporaneously with diagnosis. These patients were retained in the analysis and assigned a referral time of 1 day. We conducted sensitivity analyses setting referral time to the midpoint value between referral and diagnosis and the absolute value between referral and diagnosis and determined our findings were robust to this decision.

Demographic factors hypothesized to be associated with access to transplant included age at diagnosis, gender, patient race (White, Black, or Asian), patient ethnicity (non-Hispanic or Hispanic), urbanicity (living in an urban commuting area or a non-urban commuting area), proportion of the census tract living in poverty (<5%, 5%–<10%, 10%–<20%, and 20%–100%), and insurance at cancer diagnosis [private, Medicaid (including patients dually enrolled in Medicare), Medicare, military insurance, uninsured, and unknown; the last three categories were collapsed for analyses]. Requirements for tumor downstaging were also hypothesized to be associated with transplant access.

Demographic variables were obtained from the GCR at the time of HCC diagnosis and include both patient- and provider-reported variables. Geographic characteristics (census tract poverty, urbanicity) were assigned by the GCR from patient addresses at the time of diagnosis.

We described demographic characteristics of our patient cohort and evaluated bivariate associations using χ² tests, Fisher exact tests, or t tests as appropriate. To evaluate multivariable associations between demographic factors and completion of the selection process, we estimated the time to each event accounting for death as a competing risk using a cause-specific hazard model.

In sensitivity analyses, we reassessed multivariable associations between patient-level factors and step completion accounting for non-transplant cancer treatment (e.g., surgical resection, local tumor destruction) in addition to death as competing risks. In the first iteration of these analyses, patients with a history of resection or local tumor destruction were censored prior to transplant referral. Event dates for these treatments were not collected by transplant centers or the cancer registry; therefore, these patients were assigned the median referral time as their time to censorship. Among referred patients who received a non-transplant curative treatment, these patients were censored prior to evaluation initiation and assigned the median time to evaluation start as their time to censorship.

In a final sensitivity analysis, we described characteristics of patients who were reported as receiving a transplant to the GCR but who were not referred to one of our two transplant centers to evaluate potential selection bias from patients leaving the state to receive care. All analyses were conducted in R.

The data analyzed in this study are available from GCR. Restrictions apply to the availability of these data, which were used under license for this study. Data are available upon reasonable request with the permission of the Georgia Department of Public Health.

We identified 1,379 patients that met our primary cohort inclusion criteria. Approximately 75% of patients were male, with an average age at diagnosis of 62.3 years (Table 1). The most common insurance type was Medicare (46%), followed by private insurance (26%). Nearly two-thirds of patients (64%) were White, 31% were Black, and 6% were Asian.

Among included patients, 358 (26%) were ever referred to one of the two liver transplant centers in Georgia (Fig. 2). Of those referred, 89% (n = 320) initiated the evaluation and 75% of patients who initiated the evaluation completed the process (n = 240).

Table 2 provides bivariate associations between patient demographic characteristics and liver transplant referral. Age (P < 0.001), insurance type (P < 0.001), and requirements for tumor downstaging (P < 0.001) were associated with referral status. Specifically, referred patients were younger at diagnosis compared with patients not referred (mean age: 60.4 vs. 63.0 years). Referred patients were also more likely to have private insurance (37%) compared with patients who were not referred for transplant (22%). Patients who required downstaging made up 8% of those referred for transplant compared with 19% of those not referred.

In a multivariable model (Table 4), patients with Medicaid [cause-specific hazard ratio (csHR): 0.60, 95% confidence interval (CI): 0.41–0.86], Medicare (csHR: 0.72, 95% CI: 0.56–0.93), and other insurance (csHR: 0.42, 95% CI: 0.29–0.61) had a significantly lower cause-specific hazard of referral to transplant compared with patients with private insurance. Increasing age was also associated with lower cause-specific hazard of referral (csHR per year: 0.97, 95% CI: 0.96–0.98), as was requiring downstaging prior to transplant (csHR: 0.42, 95% CI: 0.29–0.62).

Table 3 provides bivariate associations between demographic characteristics and the transplant evaluation process. Age at diagnosis (P = 0.03), patient race (P = 0.002), and census tract poverty (P = 0.002) were associated with evaluation initiation. Among referred patients, those who initiated the evaluation were younger compared with those who did not (mean age: 60.0 vs. 63.0 years). Black patients made up 27% of evaluation initiators compared with 55% of non-initiators, and 30% of patients who initiated the evaluation lived in census tracts with ≥20% of residents in poverty, compared with 63% of those who did not initiate the evaluation.

Among those referred to transplant, Black patients had a 24% lower cause-specific hazard of initiating the evaluation (csHR: 0.76, 95% CI: 0.58–1.00) than White patients (Table 4). Patients living in census tracts with ≥20% of residents in poverty had a 38% lower cause-specific hazard of evaluation initiation compared with patients living in census tracts with <5% of residents in poverty (csHR: 0.62, 95% CI: 0.42–0.94). Increasing age was also associated with lower cause-specific hazard of evaluation initiation (csHR per year: 0.98, 95% CI: 0.97–1.00).

Health insurance type was the only variable associated with completion of the transplant evaluation (Table 3). Among patients who completed the evaluation, 8% of patients had Medicaid, compared with 19% of those who did not complete the evaluation (P = 0.04). In multivariable analyses (Table 4), Medicaid patients had a nearly 50% lower cause-specific hazard of evaluation completion compared with patients with private insurance (csHR: 0.53, 95% CI: 0.32–0.89).

In our expanded cohort, we identified 1,471 adult patients diagnosed with HCC in Georgia between 2010 and 2019. Demographic characteristics of the additional 92 T2 patients in this cohort are described in the Supplementary Table S1. On average, T2 patients were older at HCC diagnosis compared with T1 patients (mean age: 64.0 vs. 62.3 years) and were more likely to require tumor downstaging to be eligible for transplant (100% vs. 16.2%). Rates of transplant referral (26%), evaluation initiation (89%), and evaluation completion (76%) were virtually unchanged in the expanded cohort (Supplementary Fig. S1).

Supplementary Table S2 provides multivariable associations between demographic factors and completion of each step of the transplant process among patients in the expanded cohort. Similar to the primary cohort, increasing age at diagnosis, requiring tumor downstaging, and non-private health insurance were associated with decreased cause-specific hazard of transplant referral. Notably, patients diagnosed with multiple tumors had a cause-specific hazard of transplant referral three times higher than patients diagnosed with a single tumor (csHR: 3.12, 95% CI: 1.84–5.27).

Also similar to the primary cohort, race and age were associated with cause-specific hazard of evaluation initiation. Living in a census tract with ≥20% of residents in poverty was associated with a lower cause-specific hazard of evaluation initiation compared with living in a census tract with <5% of residents in poverty but the association was not statistically significant in this cohort (csHR: 0.67, 95% CI: 0.45–1.01). Among those who initiated with evaluation, patients requiring tumor downstaging for transplant eligibility had 44% lower cause-specific hazard of evaluation completion compared with those not requiring downstaging (csHR: 0.56, 95% CI: 0.32–0.98).

In multivariable models accounting for non-transplant treatment and death as competing risks, findings were consistent with those from our primary analyses. In models where patients with history of resection, local tumor destruction, or other treatments (n = 375) were censored prior to transplant referral, increasing age (csHR: 0.97, 95% CI: 0.96–0.98), required downstaging (csHR: 0.41, 95% CI: 0.27–0.63), and non-private health insurance [Medicaid (csHR: 0.56, 95% CI: 0.37–0.83), Medicare (csHR: 0.65, 95% CI: 0.49–0.86), other (csHR: 0.43, 95% CI: 0.29–0.63)] were still associated with lower cause-specific hazard of transplant referral (Table 5).

In models where patients with history of non-transplant cancer treatments were censored following transplant referral (n = 63), cause-specific hazard estimates of referral were unchanged from primary analysis (Table 5). Similar to primary analyses, Black patients had a 29% lower cause-specific hazard of evaluation initiation compared with White patients (csHR: 0.71, 95% CI: 0.52–0.96), and patients living in census tracts with ≥20% of residents in poverty had a 43% lower cause-specific hazard of initiating the evaluation compared with patients living in census tracts with <5% of residents in poverty (csHR: 0.57, 95% CI: 0.36–0.89).

In our final sensitivity analysis (Table 6), we identified 45 patients recorded as receiving a transplant in the GCR but not referred to one of our two transplant centers. We compare demographic characteristics of these 45 patients with those who received a transplant in the state of Georgia (n = 146). Black patients made up 25% of those who received a transplant in Georgia compared with 16% outside the state. Patients living in census tracts with ≥20% poverty made up 30% of those who received a transplant in Georgia compared with 20% of those outside the state, and patients with Medicaid made up 6% of those who received a transplant in Georgia compared with 11% of those receiving one outside the state. None of these differences were statistically significant.

Among patients diagnosed with a single HCC tumor that was either within Milan or downstaging criteria in the state of Georgia, 26% were referred to one of the two liver transplant centers in the state. Age, insurance type, and downstaging requirements were associated with referral to transplant, race, and census tract poverty were associated with evaluation initiation, and Medicaid was associated with evaluation completion. Our findings highlight opportunities to improve access to transplant among T1 patients with HCC in Georgia, including improving referring provider knowledge about transplant insurance requirements, identifying barriers to evaluation initiation among Black patients and patients living in high poverty neighborhoods, and streamlining evaluation completion for Medicaid patients.

The optimal proportion of patients with HCC that should be referred to transplant is unknown. In our study, we found that 26% of patients diagnosed with a single HCC tumor that was either within Milan or downstaging criteria were referred to transplant, and an additional 23% received another form of curative therapy (i.e., resection or local tumor destruction), leaving 51% of the cohort with no curative treatment. Some of the patients who did not receive either form of therapy may have had an absolute contraindication to transplant, such as severe cardiac disease (3). However, most transplant criteria, including length of sobriety and degree of social support, are subjective and vary across transplant programs. Encouraging referring providers to err on the side of referral may result in an increased burden on transplant center staff by increasing the number of patients requiring evaluation. However, promoting referral also ensures that patients are not inappropriately ruled out for transplant by their referring provider before having the opportunity to be considered by the transplant team.

To our knowledge, this is the first study of referral to liver transplant among patients with HCC, although two previous studies have evaluated referral to treatment. One study using cancer registry data linked to Medicare from 1998 to 2007 found that referral to a “specialist” (including hepatology, oncology, surgery, or interventional radiology) was less likely for Black patients and more likely for patients living in the Northeast (14). However, this study did not link to any transplant center data. The second study linking data from the Pennsylvania Cancer Registry to hospital claims from 2006 to 2011 found that for patients diagnosed with HCC, referral to surgery (including ablation, resection, and transplantation) was less likely for older patients and patients with Medicaid (10). Neither of these studies examined referral to transplant specifically.

Similar to the Pennsylvania study, we found that insurance type was associated with likelihood of transplant referral. This is consistent with the findings of a recent scoping review of inequities in transplant access (15). While both transplant centers in our study accept transplant candidates with Medicaid or Medicare, patients with these insurance types were markedly less likely to be referred to transplant than privately insured patients. Referring providers may not be aware of transplant insurance requirements and decline to refer patients that may otherwise be eligible for transplantation due to their insurance coverage.

Previous single-center studies have examined completion of evaluation steps among referred patients. Our study found a notably higher proportion of patients initiated the evaluation (91%) than a prior study in Indiana (57%; ref. 16) or a survey of patients in North Carolina (49%; ref. 17). This may be because we focused on patients with early-stage HCC, as opposed to all end-stage liver disease patients, who may be more able or willing to initiate the transplant process. While not statistically significant, we observed an association in our study between race, socioeconomic status, and evaluation initiation. This is in contrast to the Indiana study, which did not find any differences in the proportion of patients from vulnerable populations who progressed to evaluation initiation. Similarly, while the Indiana study observed disparities in listing rates among patients with Medicaid and living in high poverty neighborhoods, race was not associated with listing in their study (16). This discrepancy indicates the potential for variation in disparities across the country and across transplant programs, and highlights the need for additional studies in this area.

While cancer registry data are a powerful tool for population-based analyses, there are limitations to its use for clinical research. The primary focus of cancer registries is cancer surveillance, and some variables that are important for clinical hepatology research, such as comorbidities or degree of liver dysfunction, are not systematically collected. In addition, staging methodologies used by cancer registries such as TNM staging may not perfectly map to clinical staging methods such as the Milan criteria. Although there have been major improvements in probabilistic record linkage (18), our observed referral rates may be underreported if patients were unable to be linked between registry and transplant center data. We were also unable to estimate referral and evaluation for patients who sought transplant care outside of the state of Georgia. In a sensitivity analysis, we determined that patients who received a transplant outside of Georgia were less likely to live in high poverty census tracts. This analysis indicates that the magnitude of our findings on inequities may be underestimated, and is consistent with prior research demonstrating that patients with lower socioeconomic status are less likely to travel outside of their own region to pursue transplantation (19). National sharing of referral and evaluation data would facilitate these types of studies in the future to gain a full understanding of the landscape of transplant care (11).

In summary, we found that among patients with HCC with a single tumor meeting either Milan or downstaging criteria in Georgia, non-privately insured patients were significantly less likely to be referred for liver transplant than those with private insurance. Black patients and patients living in high poverty neighborhoods were less likely to initiate the evaluation, and Medicaid patients were less likely to complete the evaluation. Our findings emphasize unique barriers to liver transplant access at each step in the process and the need for targeted interventions to improve equity in access.

No disclosures were reported.

K. Ross-Driscoll: Conceptualization, data curation, formal analysis, funding acquisition, methodology, writing-original draft, project administration, writing-review and editing. A.-T. Ayuk-Arrey: Data curation, formal analysis, visualization, project administration, writing-review and editing. R. Lynch: Conceptualization, funding acquisition, writing-review and editing. L.E. McCullough: Funding acquisition, methodology, writing-review and editing. G. Roccaro: Resources, funding acquisition, methodology, writing-review and editing. L. Nephew: Methodology, writing-review and editing. J. Hundley: Resources, writing-review and editing. R.A. Rubin: Resources, writing-review and editing. R. Patzer: Conceptualization, supervision, funding acquisition, writing-review and editing.

K. Ross-Driscoll is supported by the National Center for Advancing Translational Sciences of the NIH under award numbers UL1TR002378 and KL2TR002381. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.