Regional Social Vulnerability is Associated with Geographic Disparity in Waitlist Outcomes for Patients with Non-HCC MELD Exceptions in the United States

Structured Abstract

Background:

This study was undertaken to evaluate the role of regional social vulnerability in geographic disparity for patients listed for liver transplant with non-HCC MELD exceptions.

Methods:

Adults listed at a single center for a first time liver only transplant without HCC after June 18, 2013 in the SRTR database as of March 2021 were examined. Candidates were mapped to hospital referral regions (HRRs). Adjusted likelihood of mortality and liver transplant were modeled. Advantaged HRRs were defined as those where exception patients were more likely to be transplanted, yet no more likely to die in adjusted analysis. The Centers for Disease Control’s Social Vulnerability Index (SVI) was used as the measure for community health. Higher SVIs indicate poorer community health.

Results:

There were 49,494 candidates in the cohort, of whom 4,337 (8.8%) had MELD exceptions. Among continental US HRRs, 27.3% (n=78) were identified as advantaged. The mean SVI of advantaged HRRs was 0.42 vs. 0.53 in non-advantaged HRRs (p=0.002), indicating better community health in these areas. Only 25.3% of advantaged HRRs were in spatial clusters of high SVI vs. 40.7% of non-advantaged HRRs, while 44.6% of advantaged HRRs were in spatial clusters of low SVI vs.38.0% of non-advantaged HRRs (p=0.037).

Conclusion:

Advantage for non-HCC MELD exception patients is associated with lower social vulnerability on a population level. These findings suggest assigning similar waitlist priority to all non-HCC exception candidates without considering geographic differences in social determinants of health may actually exacerbate rather than ameliorate disparity.

Mini Abstract

In this retrospective review of national level data, liver transplant candidates with non-HCC MELD exceptions held a significant advantage in terms of access to transplant over candidates without exceptions in 27.8% of continental United States Hospital Referral Regions (HRRs). Advantaged HRRs had significantly lower social vulnerability compared to non-advantaged HRRs.

Introduction

Since 2002, priority on the liver transplant waitlist in the United States (US) has been based on the model for end stage liver disease (MELD) score^{1, 2}. In most cases, the MELD score and its updated version including sodium (MELD-Na) have been shown to accurately predict waitlist mortality^2–4. There exist a number of conditions, however, where a patient’s risk of death or progression of disease to a point excluding transplant is underestimated by the MELD score. The most notable of these MELD exception diagnoses is hepatocellular carcinoma (HCC)^{5, 6}. While HCC is perhaps the most common diagnosis for which MELD exception points are granted, there are numerous other non-HCC exception diagnoses, including primary sclerosing cholangitis (PSC), portopulmonary hypertension (POPH), hepatopulmonary syndrome (HPS), and refractory hepatic hydrothorax^7–10. While standard criteria exist as to whether exceptions points should be granted or denied for some non-HCC diagnoses, many cases fall outside standard exception criteria and must be adjudicated by peer review. Prior to 2019, exception requests were handled independently by each United Network for Organ Sharing (UNOS) region¹¹. In order to address potential geographic disparities that could arise from variation in regional level exception point review^{12, 13}, a national liver review board (NLRB) has subsequently been implemented that handles all exception requests, regardless of the region from which the request originates¹⁴.

Excessive or inappropriate application of MELD exception scores has led to an overall increase in the MELD score a candidate needs to realistically receive a transplant, a phenomenon known as MELD inflation¹⁵. Put more plainly, MELD inflation means that patients without exceptions must be closer to death before they have sufficient priority on the waitlist to be transplanted. We have previously confirmed that, in aggregate, MELD exception scores granted to patients for diagnoses other than HCC do appear justified on the basis that these candidates are at greater risk of death than is reflected by their laboratory MELD score. On a more granular level, however, we found that in some UNOS regions exception patients were indeed given greater priority for transplant than was justified by their actual mortality risk¹⁶.

Social determinants of health are defined as “the conditions in which people are born, grow, work, live, age, and the set of forces and systems shaping the conditions of daily life”¹⁷. Previous research has noted numerous social determinants of health such as urbanization, education, housing availability, poverty, and overall community life expectancy as important factors in access to transplantation. The Institute of Medicine has also highlighted access to care and health insurance coverage as critical factors creating disparity in access to transplantation^18–23. Based on this previous work, our hypothesis was that geographic differences in social determinants of health could have played an important role in this observed disparity. The goal of this present study is thus to investigate a possible association between social determinants of health and the appropriateness of non-HCC MELD exceptions on a more granular and clinically relevant geographic scale.

Methods

Patient Population

This study used data from the Scientific Registry of Transplant Recipients (SRTR). The SRTR data system includes data on all donor, wait-listed candidates, and transplant recipients in the US, submitted by the members of the Organ Procurement and Transplantation Network (OPTN). The Health Resources and Services Administration (HRSA), U.S. Department of Health and Human Services provides oversight to the activities of the OPTN and SRTR contractors. Data from the SRTR standard analysis file (SAF) as of March 2021 supplemented with candidate zip code of residence were examined. The data reported here have been supplied by the Hennepin Healthcare Research Institute (HHRI) as the contractor for the Scientific Registry of Transplant Recipients (SRTR). The interpretation and reporting of these data are the responsibility of the author(s) and in no way should be seen as an official policy of or interpretation by the SRTR or the U.S. Government. The University of Alabama at Birmingham Institutional Review Board approved this study with waiver of informed consent (Protocol #161212003). Adults listed at a single center for a first time liver-only transplant without HCC after June 18, 2013 were included. The start date of the study was chosen to coincide with initiation of the previous system for allocation of donor livers in the United States. Candidates listed at multiple centers were excluded to reduce bias in the likelihood of transplant introduced by multi-listing.

Hospital referral regions (HRRs) as defined by the Dartmouth Atlas of Healthcare are defined as “regional health care markets for tertiary medical care,” and represent the geographic catchment area of referral hospitals most commonly used by residents in a given region²⁴. Given the tertiary nature of end stage liver disease care (ESLD), we believe that HRRs are the geographic unit which most appropriately capture the healthcare environment in which ESLD patients live. Individual candidates were assigned to HRRs utilizing the most recent zip code to HRR crosswalk files provided by the Dartmouth Atlas of Healthcare²⁵. We used the US Centers for Disease Control and Prevention (CDC) Social Vulnerability Index (SVI) as a measure of community level social determinants of health. The SVI is based on 16 factors (listed in Supplemental Table 1) across multiple domains, including poverty and unemployment, education, racial composition, housing, and access to a vehicle²⁶. Our group has previously demonstrated an association between living in a community characterized by high SVI (i.e., more vulnerable) and decreased access to living donor kidney transplantation^27–29. As the original SVI is calculated at the census tract level, the SVI methodology was utilized to create an index for the HRR level. Each individual component of the SVI was weighted by population ($\text{component}\text{score}\times \frac{censustractpopulation}{HRRpopulation}$). The population weighted average for each component was then calculated by summing all the tract level weighted scores and then ranked by percentile according to the CDC’s methodology to create an HRR level SVI²⁶. SVI scores thus represent percentiles of vulnerability relative to other HRRs, with higher percentile indicating greater vulnerability or worse social determinants of health.

Analytic Methods

The primary outcomes of interest were death or receipt of a liver transplant. The time to each outcome was measured from addition to the waitlist until the event of interest or the date of last follow-up for those still waiting at the end of the study period. Patient death was ascertained by report to the OPTN or the National Technical Information Services Death Master File. Deaths in the SRTR database are updated monthly with data from the Master Death File, ensuring that candidate survival is accurately tabulated even in the event of removal from the waitlist³⁰. Because a candidate could potentially survive for a prolonged period after waitlist removal for “clinical deterioration,” and what constitutes clinical deterioration sufficient for waitlist removal likely varies by center, we chose not to use waitlist dropout as an endpoint given the availability of accurate survival data as described above.

Cause-specific hazards regression for death and transplant, respectively, was performed with MELD exception status as the exposure of interest, adjusting for implementation of the National Liver Review Board and acuity circle based allocation in a time dependent manner. MELD exception status was treated as a time dependent covariate, so a candidate’s time on the waitlist prior to granting MELD exception points was counted in the non-exception group. Finally, as a patient’s MELD can also change over the course of follow-up, calculated MELD was also treated as a time dependent covariate to allow the model to reflect these changes. We chose a cause-specific hazards rather than competing risks framework as this approach has been shown to be more appropriate for studies that are more etiologic rather than prognostic in nature³¹. In keeping with previously published work, we did not adjust for additional patient factors in our models as such factors are not used in liver allocation policy to determine medical urgency, as mortality risk is all assumed to be captured by the MELD score¹⁶. Analysis was performed both on a national level and stratified by hospital referral region. The proportional hazards assumption for the global analysis was assessed by visual inspection of survival curves.

Having determined the risks for death or transplant associated with non-HCC MELD exception status in each HRR, we then sought to examine a potential association with social determinants of health as measured by the SVI as described above. After assigning a SVI to each hospital referral region, the presence of geographic clustering in HRR level social determinants was then investigated using the Optimized Outlier Analysis tool within ArcGIS Pro (ESRI, Redlands, CA)³². The Optimized Outlier Analysis tool uses the local version of Moran’s I, a local indicator of spatial association (LISA)³³ to identify whether each HRR is part of a larger geographic cluster as it relates to SVI. Four types of clusters may be identified: “High-High” (also known as “hot spots”) in which an area with high values is surrounded by other areas with high values, “Low-Low” (also known as “cold spots”) in which an area with low values is surrounded by areas with low values, “High-Low” (also known as high outliers) in which an area with high values is surrounded by areas with low values, and “Low-High” (also known as low outliers) in which an area with low vales is surrounded by areas with high values.

HRRs in which patients with MELD exceptions had a trend towards lower likelihood of death (defined as hazard ratio < 1 and p-value >0.05) and were significantly more likely to be transplanted (defined as hazard ratio > 1 and P<0.05) than their counterparts without exceptions in the cause specific hazards regression were labeled as “advantaged”, since exception patients hold an advantage over non-exception patients in these areas. HRRs not meeting these criteria were defined as having “no advantage”. The SVI between “advantaged” HRRs and HRRs with “no advantage” were compared using Student’s t-test. The association between SVI based clusters identified in the spatial analysis and advantaged status was tested using linear regression, with adjustment for multiple comparisons by Tukey’s HSD. Mapping and spatial analysis were performed using ArcGIS Pro (ESRI, Redlands, CA) while non-spatial analysis and data management were performed using SAS version 9.4 (SAS Institute, Cary, NC). P-values <0.05 were considered statistically significant.

Results

There were 49,494 patients forming the study cohort. The patient inclusion/exclusion flowchart is included in the supplemental material (supplemental figure 1). Candidates excluded for missing data as noted above comprised only 1.4% of the group meeting study inclusion criteria. Of the total cohort, 8.8% (n=4,337) had MELD exception points. Nonstandard exceptions were the most common, with “other” listed as the majority of MELD exception diagnoses (52.7%; n=2,285). Demographics of exception and non-exception patients are summarized in table 1.

Table 1:

Demographics of Exception vs. Non-Exception Patients. Continuous variables are summarized as mean (standard deviation) and categorical variables are summarized as count (percentage). P-values <0.05 highlighted in bold.

	Exception Patients (n=4,337)	Non-Exception Patients (n=45,158)
Age at Listing (yrs)	54.4 (12.1)	53.7 (11.2)	<0.001
Female Sex	1904 (43.9%)	18358 (40.7%)	<0.001
Race and Ethnicity			<0.001
Caucasian	3325 (76.7%)	32977 (73.0%)
African American	254 (5.9%)	3227 (7.2%)
Hispanic	540 (12.5%)	6690 (14.8%)
Asian	160 (3.7%)	1438 (3.2%)
Native American/Pacific Islander	35 (0.8%)	568 (1.3%)
Other	58 (1.3%)	258 (0.6%)
Insurance			<0.001
Private	2417 (56.4%)	23891 (53.3%)
Medicaid	607 (14.2%)	9269 (20.7%)
Medicare	1113 (26.0%)	9767 (21.8%)
Other	152 (3.6%)	1913 (4.3%)
Working for Income	1186 (27.7%)	9588 (21.4%)	<0.001

Global Cause Specific Hazards Analysis

In the global cause specific hazards analysis, patients with exception points were more likely to die compared to patients without exception. Increasing laboratory MELD was also associated with an increased hazard of death, while introduction of the NLRB was associated with a decreased hazard for death. Acuity circles based allocation was not significantly associated with likelihood of death. Patients with MELD exceptions were more likely to be transplanted compared to patients without exception points. Increasing laboratory MELD, introduction of the NLRB, and introduction of acuity circle based allocation were also associated with increased likelihood of transplant. The results of the global cause-specific hazards regression are summarized in table 2.

Table 2.

Global cause specific hazards regression. Hazard ratios (HR) expressed as point-estimate (95% confidence interval). P-values <0.05 are highlighted in bold.

	HR for Death	P-value for Death	HR for Transplant	P-value for Transplant
Exception Patient (vs. Non-Exception)	1.690 (1.544–1.850)	<0.001	3.455 (3.317–3.598)	<0.001
Post-NRLB Era (vs. Pre-NLRB)	0.787 (0.728–0.849)	<0.001	1.162 (1.120–1.205)	<0.001
Post-Acuity Circles Based Allocation (vs. Pre-Acuity Circles)	1.094 (0.991–1.207)	0.074	1.280 (1.1227–1.336)	<0.001
Laboratory MELD (each additional Point)	1.106 (1.104–1.107)	<0.001	1.109 (1.108–1.110)	<0.001

Analysis Stratified by Hospital Referral Region

When stratified by hospital referral region, patients with non-HCC MELD exceptions had a trend (HR > 1, p>0.05) towards decreased likelihood of death in 119 HRRs (39.1%), a trend towards greater likelihood of death in 120 (39.5%) HRRs, and a significantly greater likelihood of death in 56 (13.2%, Figure 1a). Likelihood of death was statistically inestimable in 25 (8.2%) of HRRs. Exception patients were significantly more likely to be transplanted in 202 (66.5%) HRRs, had a trend towards greater likelihood of transplant in 78 (25.7%) HRRs, and had a trend towards lower likelihood of transplant in 16 HRRs (5.3%, Figure 1b). Likelihood of transplant was statistically inestimable in 8 (2.6%) of HRRs. Combining these findings, there were 83 (27.3%) HRRs where exception patients had an advantage over patients without exceptions (Figure 1c).

Figure 1:

a.) Likelihood of death based on hazard ratio for non-HCC MELD exception patients vs. patients without exceptions, adjusted for laboratory MELD, NLRB implementation, and implementation of acuity circle based allocation. b.) Likelihood of transplant based on hazard ratio for non-HCC MELD exception patients vs. patients without exceptions, adjusted for laboratory MELD, NLRB implementation, and implementation of acuity circle based allocation. c.) Status of candidates with non-HCC MELD exceptions vs. patients without MELD exceptions by HRR. Regions where exception patients have a trend towards lower likelihood of death and a significantly higher likelihood of transplant on adjusted Cox regression analysis are defined as advantaged.

Association Between Advantage for Exception Patients and Social Determinants of Health

The mean SVI in advantaged HRRs was lower, indicating lower community vulnerability, than in HRRs without advantage (0.42 vs. 0.53; p=0.002). On optimized outlier analysis, there were 111 (36.5%) HRRs located within clusters of high SVI (either high-high or high-low clusters), 121 (39.8%) HRRs located within clusters of low SVI (either low-low or low-high), and 72 (23.7%) HRRs not located within a significant cluster (figure 2). There was a stepwise decrease in mean SVI going from areas of concentrated vulnerability (high-high clusters) to high outlier (high-low) clusters, to low outlier (low-high) clusters, and finally to low-low clusters (Table 3). A greater proportion of advantaged HRRs were in clusters of low vulnerability (either low-high or low-low, Table 4).

Figure 2:

a.) HRR level SVI. b.) Clustering of HRRs by SVI as determined by optimized outlier analysis. Clusters may be high-high (high SVI in a region of high SVI), high-low (high SVI in a region of low SVI), low-high (low SVI in a region of high SVI), or low-low (low SVI in a region of low SVI).

Table 3:

Mean hospital referral region (HRR) social vulnerability index (SVI) for geographic clusters identified by optimized outlier analysis.

	Number of HRRs	Mean SVI	P-value (vs. no cluster)
High-High	77	0.8	<0.001
High-Low	34	0.65	0.092
Low-High	18	0.34	<0.001
Low-Low	103	0.22	<0.001
No Cluster	72	0.56	n/a

Table 4:

Membership of advantaged and non-advantaged hospital referral regions (HRRs) in geographic clusters defined by social vulnerability index (SVI)

	Advantaged HRRs	Non-Advantaged HRRs
High Score Clusters	21 (25.3%)	90 (40.7%)
Low Score Clusters	37 (44.6%)	84 (38.0%)
Non-Cluster HRRs	25 (30.1%)	47 (21.3%)
		p=0.037

Discussion

In this work we have found hospital referral regions where waitlist candidates with non-HCC MELD exceptions were no more likely to die than was already reflected by their calculated MELD score, yet were more likely to be transplanted than candidates of similar acuity. This state of relative advantage for exception patients in these regions was associated with lower community level social vulnerability, indicating that social determinants of health may play a role in the disparity. Notably, these findings occur in the context of two major US policies intended to ameliorate geographic disparity, the NLRB and acuity circles based allocation. A novel and important feature of this analysis compared to our earlier work¹⁶ and that of others on this subject is utilization of hospital referral regions as the unit of geographic analysis. As opposed to UNOS regions or DSA, which represent arbitrary administrative boundaries, HRRs reflect actual patterns of care on a more granular level. Further key strengths of this study include use of the SVI to capture a broad range of social determinants of health rather than individual measures, and the use of spatial epidemiologic methods to identify unique geographic clusters where the disparities we discovered are most impactful.

Analysis of the relationship between HRR level social determinants of health, as represented by the SVI, confirmed an association between unfavorable social determinants of health and increased risk of death for exception patients. Conversely, HRRs in which non-HCC exception patients held an advantage were better off in terms of social determinants of health, represented by a lower SVI. Access to specialized care represents one possible explanation for the association between favorable waitlist outcomes for exception patients and lower community social vulnerability. Previous work has demonstrated that the number of gastroenterologists in the local community was as predictive of waitlist outcomes as the MELD score, with a 12% greater likelihood of transplant for each additional GI specialist per 100,000 population in a community³⁴. Ross and colleagues have demonstrated an association between poor access to care and a higher risk of waitlist mortality than was captured by the MELD score³⁵. Others have demonstrated that diverse factors such as urbanization, transplant center distribution, education, and housing availability were also associated with access to liver transplantation^{19, 20, 23}.

As an example of how two similar individual patients can have a very different risk of mortality while awaiting liver transplant based upon the prevailing conditions where they live, consider the case of primary sclerosing cholangitis (PSC). Patients with PSC may have compensated liver function with low calculated MELD, but are at risk of dying from cholangitis associated sepsis. The potential for death from sepsis in the absence of liver failure forms the basis for granting MELD exception status to patients suffering from this disease. A patient who develops sepsis in a community with ready access to specialty care including gastroenterology/hepatology, infectious disease, critical care, and interventional radiology would conceivably be less likely to succumb to the same infection than one who lives in an area without access to specialists, quality hospital care, and a well-developed referral network for transfer to a higher level of care when needed. It is worth noting that these favorable factors need not be markers for proximity to a transplant center, as high-quality specialist care can frequently be found in non-transplant hospitals. Furthermore, transplant centers can facilitate quality liver-related care in geographically distant regions by effective outreach programs.

Another important implication of the findings in this manuscript is the granting of exception points need not be a one-size fits all exercise. Prior to institution of the NLRB, MELD exception requests were adjudicated by individual regional review boards, which were also given discretion in the number of points to be awarded for each individual case. While such a system had drawbacks including the arbitrary nature of UNOS regional borders and the potential for wide variation in permissiveness towards exception requests, there were regions with MELD exceptions that were appropriately stratifying risk¹⁶. These findings imply that consideration of local factors in estimating mortality risk (and thus how many points to award) may be beneficial. While we believe the NLRB has in general been a step forward in promoting greater uniformity in the decision of whether to grant exception points, the practice of granting standardized scores (such as median MELD at transplant in the region minus three, for example) may in theory exacerbate geographic disparities in access to transplantation. Just as social determinants of health are being explored in determining transplant priority for non-exception patients in development of a geographically borderless distribution scheme for liver transplant in the US³⁶, factoring community level social determinants of health into how many MELD exception points to award may also be desirable in promoting geographic equity in access to liver transplant.

Despite our concerns about the equity of how MELD exception points are currently assigned, we are not suggesting a return to the previous system of independent regional review boards empirically assigning exception scores. Indeed, our data demonstrates an overall reduced hazard of death and greater likelihood of transplantation for all comers in the NLRB era. The hypothesis our current study raises is that the exception scores assigned by the NLRB can be more precise than the current standard of MMAT minus 3. Ideally, an assigned MELD exception scores should accurately reflect the risk of mortality or waitlist dropout (ie, a patient with an allocation MELD score of 26 should have the same mortality risk as a patient with a calculated MELD score of 26). The findings of this study demonstrate that where a patient lives, and by extension the prevailing social determinants of health in that region, have an impact on mortality risk and thus should be factored into how many exception points to award. Future research will need to determine what exception scores will most accurately reflect risk in each HRR, and whether an allocation system based on such scores can eliminate the relative advantage for certain HRRs that we have demonstrated in the current system. While such a policy could seemingly be at odds with the Final Rule’s mandate that allocation policy not be based on where a patient lives, we would argue that such a policy honors the spirit of the law by promoting equal access to transplant for exception and non-exception patients in all parts of the country.

Geographic imbalances in social determinants of health and their impact on access to transplantation is more than just an allocation policy issue, however. As we noted above, transplant centers can potentially provide a positive impact by implementing effective outreach programs in underserved communities. Such efforts can range from the simple, such as ensuring open lines of communication between community physicians and transplant physicians by providing easy means of contact, to more complex endeavors such as the opening of satellite clinics. The transplant community can also recognize the role of wider government policy in ensuring access to quality care and work to effectively advocate for measures that will improve care for vulnerable populations. Finally, we should recognize that our mandate extends beyond those on our waitlist to all patients suffering from organ failure.

There are several important limitations to this study that must be kept in mind. Small numbers of patients did not allow for estimation of hazard ratios in some HRRs. Similarly, there were not sufficient numbers of patients to allow for stratification by exception diagnosis within HRRs, so there remains a possibility that differences in waitlist mortality between HRRs could reflect differences in disease patterns. We would note that current practice in the US often gives a uniform exception score (such as median MELD minus three) regardless of the disease process. Thus, while our results do not discriminate between differing disease processes, neither does current clinical practice. Finally, it is also important to note that the results presented in this manuscript represent population-based outcomes and associations. To apply these findings to individual patients would be an example of ecological fallacy. Strengths of this study include use of more granular geographic boundaries based on actual patterns of care and use of specialized spatial epidemiologic statistical methods.

With these considerations in mind, the findings of this manuscript advance the case for the important association between community level social determinants of health and access to liver transplant for patients with MELD exception diagnoses and further the cause for the transplant community to take a greater role in community health.