Quantifying Sex-Based Disparities in Liver Allocation

This cohort study assesses whether sex disparities in wait list mortality and deceased donor liver transplant are associated with geographic location, candidate anthropometric and liver measurements, and Model for End-stage Liver Disease score.

Key Points

Question

What proportion of sex-based disparities in liver allocation is associated with geographic location, candidate anthropometric and liver measurements, or Model for End-stage Liver Disease score?

Findings

In this cohort study of 81 357 participants, women were 8.6% more likely to die while waiting for a liver transplant and were 14.4% less likely to receive a deceased donor liver transplant compared with men. Candidate anthropometric and liver measurements and creatinine level had stronger associations than geographic location with sex disparities in wait list mortality and the likelihood of deceased donor liver transplant.

Meaning

The findings suggest that mitigating sex-based disparities in liver allocation may require a comprehensive approach that extends beyond geographic factors currently being considered in the transplant community.

Abstract

Importance

Differences in local organ supply and demand have introduced geographic inequities in the Model for End-stage Liver Disease (MELD) score–based liver allocation system, prompting national debate and patient-initiated lawsuits. No study to our knowledge has quantified the sex disparities in allocation associated with clinical vs geographic characteristics.

Objective

To estimate the proportion of sex disparity in wait list mortality and deceased donor liver transplant (DDLT) associated with clinical and geographic characteristics.

Design, Setting, and Participants

This retrospective cohort study used adult (age ≥18 years) liver-only transplant listings reported to the Organ Procurement and Transplantation Network from June 18, 2013, through March 1, 2018.

Exposure

Liver transplant waiting list.

Main Outcomes and Measures

Primary outcomes included wait list mortality and DDLT. Multivariate Cox proportional hazards regression models were constructed, and inverse odds ratio weighting was used to estimate the proportion of disparity across geographic location, MELD score, and candidate anthropometric and liver measurements.

Results

Among 81 357 adults wait-listed for liver transplant only, 36.1% were women (mean [SD] age, 54.7 [11.3] years; interquartile range, 49.0-63.0 years) and 63.9% were men (mean [SD] age, 55.7 [10.1] years; interquartile range, 51.0-63.0 years). Compared with men, women were 8.6% more likely to die while on the waiting list (adjusted hazard ratio [aHR], 1.11; 95% CI, 1.04-1.18) and were 14.4% less likely to receive a DDLT (aHR, 0.86; 95% CI, 0.84-0.88). In the geographic domain, organ procurement organization was the only variable that was significantly associated with increased disparity between female sex and wait list mortality (22.1% increase; aHR, 1.22; 95% CI, 1.09-1.30); no measure of the geographic domain was associated with DDLT. Laboratory and allocation MELD scores were associated with increases in disparities in wait list mortality: 1.14 (95% CI, 1.09-1.19; 50.1% increase among women) and DDLT: 0.87 (95% CI, 0.86-0.88; 10.3% increase among women). Candidate anthropometric and liver measurements had the strongest association with disparities between men and women in wait list mortality (125.8% increase among women) and DDLT (49.0% increase among women).

Conclusions and Relevance

Our findings suggest that addressing geographic disparities alone may not mitigate sex-based disparities, which were associated with the inability of the MELD score to accurately estimate disease severity in women and to account for candidate anthropometric and liver measurements in this study.

Introduction

In 2002, allocation in liver transplantation evolved from a model based on waiting time to a model defined by objective medical urgency. A metric for identifying patients at highest risk of death within 90 days of being placed on a wait list, known as the Model for End-stage Liver Disease (MELD) score, was introduced.^1,2 MELD scores rank order candidates such that the sickest patients receive highest priority on the match run (priority list for receiving a transplant). Concerns for organ viability further dictated that livers be allocated locally first, then by region, and finally nationally, with the only mandatory regional share of livers required for wait-listed candidates with MELD scores of at least 35.^3,4 Observed disparate median allocation MELD scores and wait list mortality across US transplant centers have reflected differences in local organ supply and demand.^5,6 These geographic disparities mean that at any given moment, the sickest patient in the US as defined by MELD score may not actually be at the top of the local match run when a liver becomes available, prompting ongoing national debate and patient-initiated federal lawsuits regarding the need for geographic equity in liver allocation.⁷ Debate surrounding geographic disparities in liver allocation is rooted in 2 key assumptions: (1) that the MELD score accurately identifies the sickest patients^8,9 and (2) that patients at the top of the match run actually receive the liver transplant.¹⁰

The MELD score is a mathematical algorithm that uses objective laboratory data, including sodium level, serum total bilirubin level, serum creatinine (sCr) level, and the international normalized ratio (INR) for prothrombin time, to estimate wait list mortality within 90 days.¹¹ Inclusion of sCr level as a measure of kidney function has been considered a major advantage of the MELD score because impaired kidney function is an established negative prognostic marker for cirrhosis.^12,13 However, sCr level is merely a surrogate for kidney function, and sCr concentrations have been shown to vary widely based on differences in muscle mass; therefore, reported sCr values can have wide variation across individuals with the same kidney function (eg, glomerular filtration rate) but different baseline characteristics, such as age, sex, race, and ethnicity.^14,15 Multiple studies^16,17 have shown that sCr level overestimates kidney function in women compared with men such that women with the same glomerular filtration rate as men have a lower sCr level and resulting lower calculated MELD scores.^16,17 Sex-based differences in sCr level are not accounted for by the MELD score, suggesting that the MELD score may not accurately identify and prioritize the sickest patients for liver transplant.

Moreover, 30% of livers are not allocated to candidates within the top 3 positions on any given match run, suggesting that being ranked highest does not ensure receipt of a liver for transplant.^10,18 Organs can be declined for reasons such as donor-recipient size mismatch, which has been shown to contribute significantly to sex-based disparities in access to liver transplant.¹⁰ Studies^10,18 have shown that women at the top of the priority list are more likely to have donor livers declined on their behalf, with the reason for decline 4-fold more commonly because of donor-recipient size mismatch. Women with at least 1 organ decline have been shown to be 26% more likely to die on the waiting list than men.¹⁰ Sex-based differences in size are not accounted for within the existing MELD allocation system.

Differences in local organ supply and demand are associated with geographic disparities in liver allocation.⁵ However, discussions surrounding equity in allocation require critical assumptions to be met, including that the MELD score accurately defines disease severity and patients prioritized as the sickest actually receive the liver transplant. Previous studies^10,16,17,18 examining sex-based differences in wait list mortality and liver transplant rates have shown that these assumptions are routinely violated. To our knowledge, no study has attempted to quantify the associations of geographic location, MELD score, and candidate anthropometric and liver measurements with sex-based disparity in liver allocation. Without better understanding the degree of disparity associated with various factors, efforts made and resources used to inform the design of a more equitable liver allocation system may be misplaced. We estimated the proportion of the sex disparity in wait list mortality and deceased donor liver transplant (DDLT) associated with clinical and geographic characteristics.

Methods

Data Source

This cohort study used data from the Scientific Registry of Transplant Recipients¹⁹; these data are submitted by members of the Organ Procurement and Transplantation Network on all donors, wait-listed candidates, and transplant recipients in the US. The Health Resources and Services Administration of the US Department of Health and Human Services provided oversight for Organ Procurement and Transplantation Network and Scientific Registry of Transplant Recipients activities. This study was approved by the institutional review board of the University of Alabama at Birmingham in Birmingham. All transplant candidates, recipients, and donors nationally consent to their data being collected and made publicly available for research purposes. The institutional review board of the University of Alabama at Birmingham granted a waiver of informed consent for this reason and because data were deidentified.

Study Population

Adults (aged ≥18 years) on the liver-only transplant waiting list from June 18, 2013, through March 1, 2018, were identified (n = 81 357) and categorized by sex (self-reported) at the time they were added to the liver transplant waiting list (eMethods and eFigure in the Supplement).

Outcome Ascertainment

Primary outcomes were wait list mortality and likelihood of DDLT. Wait list mortality was defined as time from addition to the waiting list to death, being censored for transplant, waiting list removal, or administrative end of study (March 1, 2019). DDLT was defined as time from addition to the waiting list to transplant, being censored for death, waiting list removal (defined as too sick for transplant, improvement in health, or medically unsuitable), or administrative end of study.

Statistical Analyses

Exploratory Data Analyses

Waiting list characteristics were compared by sex. Continuous variables were reported as median and interquartile range because of their distribution and were analyzed using Wilcoxon rank sum tests, and categorical variables were examined using χ² tests.

Survival Analyses

To permit inclusion of prevalent listings, Cox proportional hazards regression models were built using counting process data structure to accommodate delayed entry.²⁰ Base models for wait list mortality and DDLT included all covariates from Scientific Registry of Transplant Recipients risk-adjustment models^21,22 and waiting list status. Self-defined race was categorized as white, black, and other, and all continuous variables were modeled as such. Different functional forms for continuous variables were explored using Martingale residuals,²³ and categorizations of continuous variables were assessed for their association with the effect size for sex. The proportional hazards assumption was assessed using Schoenfeld residuals. If 5% or less of any variable was missing, missing observations were coded as missing or unknown to permit incorporation in adjusted analyses.

Inverse Odds Ratio Weighting

To estimate the proportion of sex disparity in wait list mortality and DDLT associated with clinical and geographic characteristics, we used inverse odds ratio (OR) weighting, which gives decomposition of an association into direct and indirect associations.²⁴ We first generated the predicted probability of being female with covariates in the base model and a potential mediating variable. Resulting estimated probabilities were used to create weights. An unweighted Cox proportional hazards regression model adjusted for sex and base model covariates was fit for each outcome, providing an estimate of total disparity between men and women. Then, a weighted model estimated residual disparity between men and women after accounting for potential mediation through inclusion of the inverse OR weighting. The resulting estimate for disparity in this model represented the direct association, and the remaining unexplained disparity represented the indirect association. We calculated the proportion of disparity associated with the mediator as follows:

where β₁ is the unweighted coefficient for sex and β* is the weighted coefficient for sex. SEs were calculated using bootstrap analysis. We examined potential mediators in 3 domains: (1) geographic (United Network for Organ Sharing [UNOS] region, listing organ procurement organization, and listing center), (2) MELD score (laboratory MELD score, allocation MELD score, and all components of the MELD score), and (3) candidate anthropometric and liver measurements (height, weight, body mass index, body surface area, estimated liver volume, and estimated liver weight^{10,25,26,27,28}). A factor was considered to mediate the association between sex and the outcome if the proportion of disparity associated with the mediator was statistically significant (2-sided, P < .05). The association could be positive or negative depending on the directionality of the difference between unweighted and weighted hazard ratios (HRs).

Sensitivity Analyses

Additional analyses were conducted and are described in eMethods in the Supplement. Statistical analyses were performed in SAS, version 9.4 (SAS Institute Inc) and Stata, version 15.1 (StataCorp).

Results

Study Population

Of 81 357 adults with liver-only listings, 36.1% were women (mean [SD] age, 54.7 [11.3] years; interquartile range, 49.0-63.0 years) and 63.9% were men (mean [SD] age, 55.7 [10.1] years; interquartile range, 51.0-63.0 years) (P < .001). Women more commonly had previous abdominal surgery (55.6% vs 36.6%, P < .001), less commonly had hepatocellular carcinoma (7.3% vs 13.5%, P < .001), and had consistently lower anthropometric and liver measurements, as measured by body surface area, estimated liver volume, and estimated liver weight. No clinically meaningful differences in other baseline characteristics by sex were observed (Table 1).^{27,28,29,30,31,32,33,34}

Table 1. Baseline Demographics by Sex^a.

Characteristic	Female (n = 29 384)	Male (n = 51 973)	P value
Age, mean (SD), y	54.7 (11.3)	55.7 (10.1)	<.001
Race			<.001
White	24 927 (84.8)	45 012 (86.6)	<.001
Black	2766 (9.4)	3842 (7.4)
Other	1691 (5.8)	3119 (6.0)
Ethnicity
Hispanic	4932 (16.8)	8020 (15.4)	<.001
Body mass index, median (IQR)b	28 (24-33)	28 (25-32)	<.001
Previous abdominal surgery	16 345 (55.6)	19 011 (36.6)	<.001
Portal vein thrombosis	1959 (6.8)	3507 (6.9)	.69
Encephalopathy	17 877 (60.8)	29 750 (57.2)	<.001
Ascites	21 175 (72.1)	35 592 (68.5)	<.001
Diabetes	8175 (27.8)	15 209 (29.3)	<.001
Hepatocellular carcinoma	2131 (7.3)	6968 (13.5)	<.001
Bacterial peritonitis	2008 (6.8)	3818 (7.4)	.02
INR, median (IQR)	1.5 (1.2-2.1)	1.4 (1.2-2.0)	<.001
Albumin level, median (IQR)	3.2 (2.7-3.7)	3.2 (2.7-3.7)	.40
Creatinine level, median (IQR)	1.0 (0.7-1.5)	1.1 (0.8-1.6)	<.001
Sodium level, median (IQR)	138 (134-140)	137 (134-140)	<.001
Bilirubin level, median (IQR)	2.8 (1.3-8.3)	2.4 (1.2-6.2)	<.001
Hospitalization	1773 (6.0)	2538 (4.9)	<.001
Hemodialysis	4935 (16.8)	7375 (14.2)	<.001
MELD
Laboratoryc	15 (11-22)	15 (10-21)	<.001
Allocation	16 (11-23)	16 (11-22)	<.001
No functional assistance	8314 (29.0)	17 398 (34.4)	<.001
Blood type
A	10 964 (37.3)	19 630 (37.8)	.02
B	3457 (11.8)	6400 (12.3)
AB	1063 (3.6)	1871 (3.6)
O	13 900 (47.3)	24 072 (46.3)
Body surface area, median (IQR)
DuBois and DuBois29	1.8 (1.6-1.9)	2.1 (1.9-2.2)	<.001
Mosteller30	1.8 (1.7-2.0)	2.1 (1.9-2.3)	<.001
Estimated liver volume, median (IQR)
Urata et al27	1259 (1165-1363)	1451 (1350-1560)	<.001
Heinemann et al31	1598 (1438-1778)	1885 (1718-2069)	<.001
Vauthey et al28	1461 (1292-1647)	1805 (1624-2001)	<.001
Estimated liver weight, median (IQR)
Yoshizumi et al32	1374 (1271-1487)	1584 (1473-1703)	<.001
Choukèr et al33	1824 (1640-2048)	2100 (1885-2357)	<.001
DeLand and North34	1596 (1460-1745)	1872 (1727-2030)	<.001
Work for income	5334 (18.2)	2035 (26.4)	<.001
Insurance
Medicaid	5182 (17.6)	8083 (15.6)	<.001
Medicare	7800 (26.6)	12 496 (24.1)
Private	15 412 (52.8)	28 150 (54.6)

Abbreviations: INR, international normalized ratio; IQR, interquartile range; MELD, Model for End-stage Liver Disease.

^aData are presented as number (percentage) of participants unless otherwise indicated.

^bCalculated as weight in kilograms divided by height in meters squared.

^cBiologic MELD.

Wait List Mortality

Of 8827 individuals who died on the waiting list, 3615 (41.0%) were female and 5212 (59.0%) were male (P < .001). After adjustment, women had 8.6% greater risk of wait list mortality compared with men (adjusted HR [aHR], 1.09; 95% CI, 1.05-1.14) (Table 2 and Figure 1).

Table 2. Association of Waiting List Characteristics With Risk of Wait List Mortality Among Women vs Men.

Model	Risk of wait list mortality
	Adjusted HR (95% CI)a	Risk of wait list mortality, %b		Adjusted HR (95% CI)c	Risk of wait list mortality, %b
		Increase	Decrease		Increase	Decrease
Base modeld	1.09 (1.05-1.14)	NA	NA	1.11 (1.08-1.14)	NA	NA
Candidate anthropometric and liver measurements
Weight	1.19 (1.13-1.25)	94.5e	NA	1.13 (1.03-1.24)	19.6e	NA
Height	1.13 (1.16-1.48)	58.2e	NA	1.22 (1.11-1.33)	88.2e	NA
Body surface area
DuBois and DuBois29	1.20 (1.10-1.31)	106.5e	NA	1.24 (1.13-1.35)	105.6e	NA
Mosteller30	1.22 (1.09-1.30)	125.8e	NA	1.24 (1.15-1.33)	104.3e	NA
Estimated liver volume
Urata et al27	1.16 (1.10-1.23)	66.9e	NA	1.23 (1.15-1.33)	103.3e	NA
Heinemann et al31	1.18 (1.12-1.24)	88.8e	NA	1.23 (1.15-1.32)	101.2e	NA
Vauthey et al28	1.19 (1.10-1.31)	98.8e	NA	1.23 (1.15-1.33)	102.0e	NA
Estimated liver weight
Yoshizumi et al32	1.20 (1.13-1.26)	103.2e	NA	1.23 (1.15-1.32)	101.1e	NA
Choukèr et al33	1.20 (1.14-1.25)	102.3e	NA	1.22 (1.14-1.30)	89.8e	NA
DeLand and North34	1.20 (1.12-1.29)	108.6e	NA	1.22 (1.13-1.32)	92.6e	NA
Body mass indexf	1.10 (1.05-1.15)	4.2	NA	1.12 (1.07-1.18)	10.3	NA
MELD score and MELD components
Laboratory MELD score	1.14 (1.09-1.19)	50.1e	NA	NA	NA	NA
Serum creatinine level	1.13 (1.08-1.18)	35.5e	NA	NA	NA	NA
INR	1.10 (1.03-1.19)	11.4	NA	NA	NA	NA
Bilirubin level	1.12 (1.07-1.17)	26.8e	NA	NA	NA	NA
Sodium level	1.13 (1.08-1.19)	38.8e	NA	NA	NA	NA
Dialysis in previous week	1.11 (1.06-1.16)	17.4e	NA	NA	NA	NA
Allocation MELD	1.14 (1.09-1.20)	53.0e	NA	NA	NA	NA
Exception points	1.11 (1.06-1.16)	16.8e	NA	NA	NA	NA
Hepatocellular carcinoma	1.11 (1.06-1.15)	16.0	NA	NA	NA	NA
Ascites	1.11 (1.04-1.18)	19.5e	NA	NA	NA	NA
Albumin level	1.12 (1.05-1.20)	27.0e	NA	NA	NA	NA
Encephalopathy	1.10 (1.04-1.17)	12.4	NA	NA	NA	NA
Geographic domain
Listing OPO	1.10 (1.05-1.16)	10.6	NA	1.12 (1.02-1.23)	7.4	NA
Listing center	1.11 (1.04-1.18)	15.7	NA	1.12 (1.11-1.13)	11.0	NA
UNOS region	1.11 (1.06-1.17)	22.8e	NA	1.13 (1.04-1.24)	19.7e	NA

Abbreviations: DDLT, deceased donor liver transplant; HR, hazard ratio; INR, international normalized ratio; MELD, Model for End-stage Liver Disease; NA, not applicable; OPO, organ procurement organization; UNOS, United Network for Organ Sharing.

^aAdjusted for age, race, ethnicity, blood type, history of abdominal surgery, bacterial peritonitis, hepatitis C virus infection, life support, transjugular intrahepatic portosystemic shunt, working for income, insurance type, educational level, history of malignant tumor, ventilator use, diabetes type, previous pancreas transplant, previous lung transplant, disease origin, and active status.

^bInverse odds ratio weighting, calculated as the difference between the base model and the weighted model divided by the base model. If positive, the variable was considered to be associated with increased wait list mortality. If negative, the variable was considered to be associated with decreased wait list mortality.

^cAdjusted for age, race, ethnicity, blood type, history of abdominal surgery, bacterial peritonitis, hepatitis C virus infection, life support, transjugular intrahepatic portosystemic shunt, working for income, insurance type, educational level, history of malignant tumor, ventilator use, diabetes type, previous pancreas transplant, previous lung transplant, disease origin, active status, and MELD score.

^dInterpreted as the likelihood of DDLT or wait list mortality for women compared with men.

^eP < .05.

^fCalculated as weight in kilograms divided by height in meters squared.

Figure 1. Change in Excess Risk of Wait List Mortality Among Candidates for Liver Transplant Across Geographic Factors, Model for End-stage Liver Disease (MELD) Score, and Candidate Anthropometric and Liver Measurements.

Adjusted hazard ratio (HR) of 1.00 indicates statistically equal likelihood of wait list mortality for men and women.

After weighting for UNOS region, women had 11.1% greater risk of wait list mortality compared with men (aHR, 1.11; 95% CI, 1.06-1.17), representing a 22.8% increase in the disparity compared with the unweighted model. After weighting for listing center, women had 10.5% greater risk of wait list mortality compared with men (aHR, 1.11; 95% CI, 1.04-1.18), representing a 10.6% increase in disparity, but this was not statistically significant (Table 2 and Figure 1).

After weighting for the laboratory MELD score, women had 14.4% greater risk of wait list mortality compared with men (aHR, 1.14; 95% CI, 1.09-1.19), corresponding to a 50.1% increase in the disparity between men and women. Allocation MELD score was associated with a 53.0% increase in the disparity between men and women, such that women had 14.4% greater risk of wait list mortality than men (aHR, 1.14; 95% CI, 1.09-1.20). When weighted by sCr level, women had a 13.6 greater risk of wait list mortality compared with men (aHR, 1.13; 95% CI, 1.08-1.18), representing a 35.5% increase in the disparity between men and women vs the unweighted model. Encephalopathy and INR were not significant mediators of the association between sex and wait list mortality (Table 2 and Figure 1).

After weighting for body surface area, women had a 22.1% greater risk of wait list mortality compared with men (aHR, 1.22; 95% CI, 1.09-1.30), corresponding to a 125.8% increase in risk compared with the unweighted model. Estimated liver volume (aHR, 1.19; 95% CI, 1.10-1.31) significantly increased the disparity between men and women by 98.8% and estimated liver weight (aHR, 1.20; 95% CI, 1.12-1.29) increased the disparity between men and women by 108.6%. Body mass index was not a substantial mediator of the association between female sex and wait list mortality (Table 2 and Figure 1).

In an unweighted model adjusted for laboratory MELD score and other clinical characteristics, female sex was associated with a 10.9% greater risk of wait list mortality compared with men (aHR, 1.11; 95% CI, 1.08-1.14). After weighting for body surface area, this risk increased among women compared with men by 23.7% (aHR, 1.24; 95% CI, 1.13-1.35), with an associated increase in disparity of 105.6%. After weighting for estimated liver volume, women had 23.1% increased risk of wait list mortality compared with men (aHR, 1.23; 95% CI, 1.15-1.32), with an increase in disparity of 101.2% compared with the unweighted model. A similar increase in the disparity between men and women was found after weighting for estimated liver weight (aHR, 1.23; 95% CI, 1.15-1.32). Body mass index did not mediate the association between female sex and wait list mortality (Table 2 and Figure 1).

In an unweighted model adjusting for laboratory MELD score, anthropometric and liver measurements, and UNOS region, female sex was associated with 25.2% greater risk of wait list mortality compared with male sex (aHR, 1.25; 95% CI, 1.21-1.29) (Figure 1).

Likelihood of DDLT

Of 37 114 individuals who received a DDLT, 12 370 (33.3%) were female and 24 744 (66.7%) were male (P < .001). After adjustment, women were 14.4% less likely to receive a DDLT compared with men (aHR, 0.86; 95% CI, 0.84-0.88) (Table 3 and Figure 2).

Table 3. Association of Waiting List Characteristics With Likelihood of DDLT Among Female Registrants.

Model	Likelihood of deceased donor liver transplant
	Adjusted HR (95% CI)a	Likelihood of DDLT, %b		Adjusted HR (95% CI)c	Likelihood of DDLT, %b
		Increase	Decrease		Increase	Decrease
Base modeld	0.86 (0.84-0.88)	NA	NA	0.92 (0.93-0.97)	NA	NA
Candidate anthropometric and liver measurements
Weight	0.89 (0.87-0.90)	22.2e	NA	NA	NA	NA
Height	0.92 (0.87-0.97)	38.2e	NA	NA	NA	NA
Body surface area
DuBois and DuBois29	0.92 (0.89-0.94)	44.2e	NA	NA	NA	NA
Mosteller30	0.90 (0.88-0.92)	34.1e	NA	NA	NA	NA
Estimated liver volume
Urata et al27	0.92 (0.90-0.95)	49.0e	NA	NA	NA	NA
Heinemann et al31	0.92 (0.89-0.94)	44.3e	NA	NA	NA	NA
Vauthey et al28	0.91 (0.88-0.94)	40.5e	NA	NA	NA	NA
Estimated liver weight
Yoshizumi et al32	0.92 (0.89-0.94)	44.1e	NA	NA	NA	NA
Choukèr et al33	0.90 (0.88-0.92)	29.7e	NA	NA	NA	NA
DeLand and North34	0.89 (0.86-0.93)	28.1e	NA	NA	NA	NA
Body mass indexf	0.85 (0.83-0.86)	NA	6.8e	NA	NA	NA
MELD score and MELD components
Laboratory MELD	0.87 (0.86-0.88)	10.3e	NA	0.92 (0.88-0.97)	NA	4.4
Serum creatinine level	0.86 (0.84-0.88)	5.4e	NA	0.95 (0.94-0.96)	35.5e	NA
INR	0.84 (0.82-0.86)	NA	12.7e	0.89 (0.88-0.90)	NA	43.4e
Bilirubin level	0.85 (0.83-0.87)	NA	3.2e	0.92 (0.92-0.94)	NA	9.3
Sodium level	0.86 (0.84-0.87)	0.7	NA	0.94 (0.92-0.98)	25.7	NA
Dialysis in previous week	0.84 (0.83-0.86)	NA	9.3e	0.93 (0.87-0.98)	6.7	NA
Allocation MELD	0.87 (0.85-0.89)	13.4e	NA	0.92 (0.87-0.98)	NA	3.1
Exception points	0.85 (0.84-0.87)	NA	2.5	0.93 (0.89-0.98)	12.0	NA
Hepatocellular carcinoma	0.85 (0.84-0.87)	NA	2.6	0.93 (0.89-0.98)	11.6	NA
Ascites	0.85 (0.83-0.87)	NA	6.0	0.92 (0.92-0.94)	0.07	NA
Albumin level	0.85 (0.83-0.86)	NA	7.0e	0.92 (0.92-0.93)	NA	3.4
Encephalopathy	0.84 (0.82-0.86)	NA	11.6e	0.91 (0.90-0.92)	NA	14.9e
Geographic domain
Listing OPO	0.86 (0.85-0.87)	2.4	NA	0.92 (0.87-0.97)	NA	8.1
Listing center	0.86 (0.83-0.88)	0.9	NA	0.92 (0.90-0.94)	NA	13.0
UNOS region	0.85 (0.84-0.86)	NA	3.9	0.91 (0.87-0.96)	NA	16.7

Abbreviations: DDLT, deceased donor liver transplant; HR, hazard ratio; INR, international normalized ratio; MELD, Model for End-stage Liver Disease; NA, not applicable; OPO, organ procurement organization; UNOS, United Network for Organ Sharing.

^aAdjusted for age, race, ethnicity, blood type, history of abdominal surgery, bacterial peritonitis, hepatitis C virus infection, life support, transjugular intrahepatic portosystemic shunt, working for income, insurance type, educational level, history of malignant tumor, ventilator use, diabetes type, previous pancreas transplant, previous lung transplant, disease origin, and active status.

^bAdjusted for age, race, ethnicity, blood type, history of abdominal surgery, bacterial peritonitis, hepatitis C virus infection, life support, transjugular intrahepatic portosystemic shunt, working for income, insurance type, educational level, history of malignant tumor, ventilator use, diabetes type, previous pancreas transplant, previous lung transplant, disease origin, active status, and size.

^cInverse odds ratio weighting calculated as the difference between the base model and the weighted model divided by the base model. If positive, the variable was considered to be associated with increased DDLT. If negative, the variable was considered to be associated with decreased DDLT.

^dInterpreted as the likelihood of DDLT or wait list mortality for women compared with men.

^eP < .05.

^fCalculated as weight in kilograms divided by height in meters squared.

Figure 2. Change in Disparity in Likelihood of Deceased Donor Liver Transplant (DDLT) Among Liver Transplant Candidates Across Geographic Factors, Model for End-stage Liver Disease (MELD) Score, and Candidate Anthropometric and Liver Measurements.

HR indicates hazard ratio.

After weighting for UNOS region, women were 15.0% less likely to have a DDLT compared with men (aHR, 0.85; 95% CI, 0.85-0.86), with an increase in disparity of 3.9%, which was not statistically significant. Similarly, wait listing center and organ procurement organization did not mediate the association between sex and DDLT (Table 3 and Figure 2).

After weighting for laboratory MELD score, women were 13.1% less likely to receive a DDLT compared with men (aHR, 0.87; 95% CI; 0.86-0.88), representing a 10.3% decrease in the disparity between men and women. Allocation MELD was associated with a 13.4% increase and sCR level with a 5.4% decrease in the disparity between men and women. After weighting for INR, the disparity between men and women increased by 12.7% such that women were 16.1% less likely to receive a DDLT (aHR, 0.84; 95% CI, 0.82-0.86) (Table 3 and Figure 2).

After weighting for estimated liver volume, women were 8.5% less likely to receive a transplant than men (aHR, 0.92; 95% CI, 0.90-0.95), corresponding to a 49.0% decrease in disparity. Similarly, body surface area was associated with a 44.2% reduction in the disparity and estimated liver weight with a 44.1% reduction in the disparity between men and women (aHR, 0.92 [95% CI, 0.89-0.94] vs. 0.92 [95% CI, 0.89-0.84]). Body mass index was also a substantial mediator of the association between sex and DDLT; it was associated with an increased disparity between men and women of 6.8% such that women were 15.3% less likely to receive a transplant (aHR, 0.85; 95% CI, 0.83-0.86) (Table 3 and Figure 2).

In an unweighted model adjusting for size and other clinical characteristics, women were 8.4% less likely to receive a DDLT (aHR, 0.92; 95% CI, 0.91-0.94). Weighting for sCr level was associated with a 35.5% reduced disparity between men and women such that women were only 4.7% less likely to receive a DDLT compared with men (aHR, 0.95; 95% CI, 0.94-0.96). Weighting for INR associated with a 43.4% increased disparity between men and women, and weighting for encephalopathy was associated with a 14.9% reduced disparity. Geographic location did not mediate the association between sex and DDLT when controlling for candidate anthropometric and liver measurements (Table 3 and Figure 2).

In an unweighted model adjusting for laboratory MELD score, anthropometric and liver measurements, and UNOS region, women were 4.8% less likely to receive a DDLT compared with men (aHR, 0.95; 95% CI, 0.92-0.97) (Figure 2).

Discussion

In this, to our knowledge, first national study, we performed mediation analyses through inverse OR weighting to examine sex differences in allocation across 3 domains, including geographic location, MELD score, and candidate anthropometric and liver measurements, and assessed their associations with the current sex-based disparities in liver allocation. We found that women had 8.6% greater risk of wait list mortality and were 14.4% less likely to receive a transplant compared with men. Geographic location was strongly associated with increased disparities in wait list mortality (22.8%), but candidate anthropometric and liver measurements and laboratory MELD scores had more statistically significant associations (representing 125.8% and 50.1% of the sex-based disparity, respectively); thus, size mismatch between the donor and intended recipient and incorrect assessments of liver disease severity were more strongly associated with the observed sex disparity in wait list mortality than local supply of organs. For DDLT, the associations with geographic differences were not statistically significant (only 3.9% of sex-based disparity), whereas metrics of candidate anthropometric and liver measurements (49.0%) and liver disease severity (MELD score) (10.3%) had the strongest associations with inequities in DDLT between women and men.

Ensuring equity in settings with limited resources is not without challenges. Because there are too few organs for those in need, priority for liver transplant is given to those with the greatest medical urgency. Equity is achieved when the sickest patients are allocated and receive livers first. Although introduction of MELD-based allocation has been associated with significant reductions in wait list mortality,^35,36 studies^8,9,17 have shown that as a metric of medical urgency, MELD score is imperfect and has exacerbated certain inequities. There is inherent bias in sCr level as a measure of kidney function, and use of glomerular filtration rate to replace sCr level in the MELD score calculation would be a step toward rectifying this bias. Higher than expected organ declines for women with small body stature could be addressed by preferential allocation of small donor livers or left lateral segments from mandated split livers.^{10,17,37,38,39} Despite objective data showing sex-based disparities in liver allocation, currently no change in policy has occurred. Whether these measures would mitigate sex-based disparities remains unclear because UNOS, which holds the Organ Procurement and Transplantation Network contract and is responsible for deceased donor organ allocation in the US, has never modeled such changes to our knowledge.

Currently, the transplant community is considering geographic redistribution, the most significant proposed allocation change since the introduction of the MELD score and Share-35 (which prioritized regional allocation of deceased donor livers to candidates with a MELD score exceeding 35), to redefine local organ supply by replacing donor service areas with fixed concentric circles around donor hospitals.⁴⁰ However, newly proposed geographic models rely on the same metric for medical urgency, the MELD score,^{5,6,41,42,43,44,45,46} and offer no solution for candidates with small body stature who may appear at the top of the match run yet are routinely skipped secondary to discrepancies in donor-recipient size.¹⁰ On the basis of our data, geographic redistribution may not ameliorate the sex disparities in transplant access. Although geographic factors matter, examining geographic access alone may be insufficient. We believe that efforts should simultaneously focus on ensuring that the definition of medical urgency, the MELD score, is equitable and that those who are prioritized for transplant, even those with small body stature, actually receive the transplant. We propose that a better course of action is to simultaneously address the attributes of the existing allocation system that were most strongly associated with increased sex disparities in wait list mortality and DDLT in our study: the MELD score and candidate anthropomorphic and liver measurements. Findings from our study support such process improvement in liver allocation.

Strengths and Limitations

A strength of our study is the novel application of inverse OR weighting to characterize factors associated with sex-based disparities in liver allocation. Weighting created independent associations between treatment and mediators, eliminating indirect pathways for mediators. This framework, however, is limited by the reliability and accuracy of variables captured by the Organ Procurement and Transplantation Network. Moreover, it is plausible that other factors not routinely captured by the Organ Procurement and Transplantation Network may be associated with disparities in allocation or may confound our findings specific to sex. In addition, variances of estimates derived from inverse OR weighting can be wider than in other mediation methods, and we may not have detected smaller mediating factors, such as geographic factors.

Conclusions

Our findings suggest that the MELD score does not accurately estimate disease severity in women and that the lack of consideration of candidate anthropomorphic and liver measurements in the current allocation system may have a greater association with the sex disparity in liver allocation than geographic factors. These data further suggest that proposed policies designed to mitigate geographic disparities alone in liver allocation may not mitigate existing sex-based inequities.

Supplement.

eMethods. Inverse Odds Ratio Weighting, Competing Risks Modeling, Sensitivity Analyses

eFigure. Cohort Construction