Stage of Cirrhosis Predicts Risk of Liver Related Death in Low MELD Patients with Cirrhosis Awaiting Liver Transplant

Abstract

MELD has reduced predictive ability in cirrhotic patients with MELD ≤ 20. We aimed to assess whether a 5-stage clinical model identifies liver transplant (LT) candidates with low MELD who are at increased risk for death. We conducted a case-control study of cirrhotic subjects with MELD ≤ 20 awaiting LT at a single academic medical center from 2/2002-5/2011. Conditional logistic regression was used to evaluate the risk of liver-related death based on cirrhosis stage. We identified 41 case subjects who died from liver-related causes with a MELD ≤ 20 within 90 days of death while waiting for LT. Cases were matched with up to 3 controls, totaling 66 controls, based on listing year, age, gender, liver disease etiology, hepatocellular carcinoma, and MELD. Cirrhosis stage (1=no varices or ascites, 2=varices, 3=variceal bleed, 4=ascites, 5=ascites and variceal bleed) was assessed for all subjects. MELD scores were similar for cases and controls. Clinical states contributing to death in cases were: sepsis 48%, spontaneous bacterial peritonitis 42%, variceal bleeding 24%, and hepatorenal syndrome 21%. In univariate analyses, variceal bleed (OR 5.6, p= 0.003), albumin (OR 0.5, p=0.041), increasing cirrhosis stage (p=0.003) and reaching stage 2, 3, or 4 cirrhosis versus lower stages (OR 3.6, p=0.048; OR 7.4, p<0.001; OR 4.1, p=0.008; respectively), sodium < 135 (OR 3.4, p=0.006), and hepatic encephalopathy (OR 2.3, p=0.082) were associated with liver-related death. In a multivariable model including cirrhosis stage, albumin, sodium, and hepatic encephalopathy, increasing cirrhosis stage (p=0.010) was independently associated with liver-related death. In conclusion, assessing the cirrhosis stage in low MELD patients awaiting LT may help select candidates for more aggressive monitoring or for living donor or extended criteria donation.

Introduction

Predicting risk of death on the liver transplant (LT) waiting list is the goal of risk stratification tools like the model for end stage liver disease (MELD) score, which has been used as the backbone for liver graft allocation in the United States since 2002. There are limitations in the MELD score’s predictive ability for certain subpopulations of transplant patients^1-5. The clinical sequelae of portal hypertension such as ascites, varices, and variceal bleeding have been identified as risk factors for death independent of the MELD score, and these clinical parameters may be particularly important in patients with low MELD scores^6-8.

D’Amico et al described discrete stages in which cirrhotic patients can be classified^9,10. This unidirectional model was originally created with four stages of cirrhosis and then expanded to five. Each stage predicts the risk of 1-year mortality. Stage 1 represents patients with cirrhosis who have never had varices, ascites, or variceal bleeding. Stage 2 patients have had documented varices in the past but have not developed variceal bleeding or ascites. Stage 3 patients have had variceal bleeding but have never had ascites. Stage 4 patients have had ascites with or without varices but have never had a variceal bleed. Stage 5 patients have had ascites and variceal bleeding. The novelty and benefit of this staging model over the Childs-Turcotte-Pugh (CTP) score^1-5 is its cumulative nature. Once a patient develops a clinical scenario that places him or her in a particular stage, that patient will always be in that stage or a more advanced stage until transplant or death. This helps mitigate some of the subjective nature of clinical scores like CTP, which only show a snapshot of a patient’s liver disease severity and are thus more likely to vary from one assessment to another. One-year mortality rates for D’Amico cirrhosis stages 1-5 were 1%, 3.5%, 15%, 26%, and 30%, respectively. The D’Amico cirrhosis stage paradigm also describes the risk of advancing from one stage to another.

The importance of this classification tool is one of risk stratification for patients who may be underserved by the MELD score. To our knowledge, this paradigm of 5 cirrhosis stages has never been examined exclusively in low MELD patients. The aim of this study was to analyze liver-related mortality risk in low MELD patients on the LT waiting list with particular attention to cirrhosis stage and its constituent variables. We hypothesize that certain patients with low MELD scores may, as a function of the reduced accuracy of the MELD score in the low MELD range^8,11, require more frequent monitoring or more aggressive risk-reduction strategies.

Patients and Methods

Selection of cases and controls

We used a nested case-control design with 1:n matching for cases:controls for this study. In order to identify cases of low MELD liver-related death, we queried the University of Colorado Hospital liver transplant database for patients who died after being listed for LT but who had not yet undergone LT during the period of February 2002 to May 2011. We included all adults aged 18-75 years with clinical, histologic, or radiographic evidence of cirrhosis who were listed for primary LT and had a laboratory MELD score of ≤ 20 within 90 days of death.

Controls were identified from the same time period and database as the cases. We identified all patients listed for primary LT who did not die within 90 days of being placed on the waiting list. For each case, up to 3 control subjects were selected at random from the pool of all potential controls for a case subject matched on following criteria: 1) year of listing (within 4 years) 2) age at listing (within 5 years), 3) gender 4) liver disease etiology (HCV, alcohol, HCV with alcohol, other viral, cholestatic/autoimmune, NASH/cryptogenic, or other metabolic/genetic), 5) presence of hepatocellular carcinoma and 6) MELD at listing (within 2 points of the lowest MELD score within 90 days of the case subject death). No control subjects were used more than once.

Data extraction and definitions

We performed identical chart reviews of cases and controls. The observation period for cases was defined as the period between the lowest MELD within 90 days of death and the death of the subject. For controls, the observation period started at the time of listing and ended at 90 days after listing or at liver transplant, whichever happened first. All MELD scores generated for a patient during the observation period were collected. We established baseline MELD scores for both cases and controls. For cases, the baseline MELD score was the lowest MELD score within 90 days of death. For controls, we used the listing MELD as the baseline. Two independent reviewers assessed charts for the history of varices, variceal bleeding, ascites, hepatic encephalopathy, and transjugular intrahepatic portosystemic shunt (TIPS) prior to the baseline MELD for cases and at the end of the study period for controls (to allow for full assessment of prevalent conditions). The top three causes of liver-related death in cases were also collected by independent chart review. Subjective variables were assessed by two independent chart reviewers with discordant cases adjudicated by an expert reviewer.

The presence of ascites was ascertained as defined by the International Ascites Club^12,13 and was graded as follows: Grade 1 (radiographic evidence only), Grade 2 (clinically apparent mild to moderate), Grade 3 (large ascites and/or paracentesis greater than 3 liters), or refractory (resistant or intractable to medical therapy and requiring repeated large-volume paracentesis or TIPS). Large ascites was defined by physical exam findings describing tense, massive, large, or similarly described ascites or requiring a paracentesis of at least 3 liters. For the purpose of data analysis, at least Grade 3 or refractory ascites was required for dichotomous variables. The presence of varices was collected as any grade documented on endoscopy and including esophageal, gastric, rectal, or small/large bowel varices. Variceal bleeding was defined as any chart or procedural documentation implicating varices as the source of gastrointestinal bleeding. Hepatic encephalopathy was retrieved from clinical notes or from documentation of encephalopathy-modifying medications. Medication use was collected, including non-selective beta-blockers, at the time of nadir MELD for cases and at the end of the follow-up period for controls.

Cirrhosis stage ¹⁰ was determined as of the date of the baseline MELD score for cases and at the end of the observation period for controls. It was determined using any history of varices, variceal bleeding, and ascites as defined above. The definition of ascites for assessment of cirrhosis stage required the history of at least grade 3 or refractory ascites.

Two independent reviewers conducted detailed chart reviews to examine cause of death in cases, which was coded as liver-related or non-liver-related using a priori defined criteria as follows: bleeding, infection, sepsis, pulmonary complications including pneumonia and pulmonary edema, acute renal failure, multiorgan failure, and neurological complications resulting from hepatic encephalopathy. Designation as a non-liver-related death required the absence of a dominant liver-associated cause and a clear non-liver-related cause of death such as: acute myocardial infarction, accidental trauma-related deaths, stroke, and complications of procedures not directly related to care of the liver disease. Indeterminate causes of death were adjudicated by a third independent reviewer. Patients with a non-liver-related death or insufficient information to determine the cause of death were excluded.

All albumin and sodium lab draws were collected during the observation periods. The albumin and sodium at the time of baseline MELD score for cases and the lowest albumin and sodium during the controls’ observation period were identified and used for analysis.

Statistical Analysis

The baseline MELD scores for cases and controls were compared using a mixed effect model to allow for the matching of cases to controls. Conditional logistic regression (to account for the matching of cases to controls) was conducted on all potential predictors of liver-related death. Variables that were significant in univariate modeling (p≤0.10) were then entered into a multivariable model. Significance in the multivariable model was defined as a p<0.05. Cirrhosis stage was modeled both as a binary variable (e.g. stage 5 vs. stages 1-4) and as an ordinal variable (stage 1 to stage 5) tested for trend. Serum sodium was also modeled as both a binary variable (< or ≥ 135 mEq/L) and as a continuous variable. Baseline characteristics of cases and patients excluded for death outside of our hospital system were compared using the Chi Squared test or Fisher’s Exact test as appropriate, the Cochran-Armitage trend test for Stage comparisons, and the Wilcoxon rank sum test for the comparison of ages and MELD scores between cases and excluded patients. SAS 9.3 was used for statistical analysis (SAS Institute, Inc. Cary, NC).

Results

Characteristics of the Study Cohort

Figure 1 shows our patient acquisition flow diagram. We identified 1676 patients who were listed for LT from February 2002 to May 2011. The number of individuals who died while listed was 321, and 117 of those who died had a MELD ≤ 20 within 90 days of death. We excluded 71 patients because they died outside the University of Colorado Hospital system and had insufficient records to assess cause of death. Two additional patients were excluded because they experienced non-liver-related deaths (1 from accidental trauma and 1 from post-sphincterotomy bleeding after ERCP). Three patients were excluded because they lacked a suitable control meeting a priori matching criteria. Therefore our study population included 41 case subjects, each matched to up to 3 control subjects, for total of 66 controls.

Figure 1.

Patient Acquisition Flow Diagram

Because cases could have up to three matched controls, there were some expected imbalances in demographics of cases and controls. Median (Interquartile Range, IQR) age at death for cases and median (IQR) age at listing for controls was 58 (55-60) and 55 (52-57) years, respectively. Cases were 51% male, whereas controls were 61% male. More cases had reached higher stages of cirrhosis compared to controls (66% of cases in stage 3 or higher versus 24% of controls) [Table 1]. Infection of any source was in the top 3 causes of death in 61% of cases [Table 2]. Sepsis, spontaneous bacterial peritonitis (SBP), and pulmonary infection were implicated in the deaths of 49%, 15%, and 20% of cases, respectively. Gastrointestinal bleeding of any source and variceal bleeding were implicated in the deaths of 29% and 24% of cases, respectively. Acute renal failure of any cause and hepatorenal syndrome were implicated in the deaths of 32% and 22% of cases, respectively.

Table 1.

Cirrhosis stage of cases and controls.

Stage	1	2	3	4	5
Varices	−	+	+	+/−	+
Variceal Bleeding	−	−	+	−	+
Ascites	−	−	−	+	+
Cases n (%)	5 (12)	9 (22)	11 (27)	11 (27)	5 (12)
Controls n (%)	19 (29)	31 (47)	6 (9)	8 (12)	2 (3)

Table 2.

Disease States Contributing to Liver-Related Death in Cases^*

Contributing Disease State		N	%
Any Infection		25	61
	Sepsis	20	49
	Spontaneous Bacterial Peritonitis	6	15
	Pneumonia/Empyema/Pulmonary Abscess	8	20
Any Gastrointestinal Bleeding		12	29
	Variceal Bleeding	10	24
Any Acute Renal Failure		13	32
	Hepatorenal Syndrome	9	22

^*41 cases, multiple contributing disease states may occur in each individual patient.

To examine the potential impact of excluding cases of death outside our hospital system with incomplete detail of cause of death, we compared the demographic, laboratory, and clinical data of such patients to our identified cases. Table 3 compares the demographics of the case subjects in our study population to the 71 patients excluded for dying outside of our hospital system. As these patients all had routine care prior to their death, similar data was available for nearly all other data fields as the cases. There were no significant differences in baseline characteristics between case subjects and patients who were excluded for death outside of our hospital system.

Table 3.

Demographic comparison of cases and patients excluded from the study due to death occurring outside of the University of Colorado Hospital

Characteristic	Cases n=41	Non-UCH Deaths n=71	p value
Gender
Male (%)	21 (51)	48 (68)	0.086
Age
Median Age at Death	58	55	0.271
(Interquartile Range)	(55-60)	(52-61)
Etiology
Alcohol (%)	10 (24)	14 (20)	0.562
Hepatitis C (%)	12 (29)	22 (31)	0.849
Hepatitis C + Alcohol (%)	7 (17)	20 (28)	0.186
Cholestatic/Autoimmune (%)	8 (20)	5 (7)	0.066
Other (%)	4 (10)	10 (14)	0.505
HCC
HCC (%)	4 (10)	12 (17)	0.298
MELD
Median Baseline MELD*	17	16	0.310
(Interquartile Range)	(15-18)	(13-18)
Cirrhosis Stage
1/2/3/4/5 n	5/9/11/11/5	10/21/13/20/7	0.541**
(%)	(12/22/27/27/12)	(14/30/18/28/10)

^*Lowest MELD within 90 days of death

^**Ordinal variable with test of trend

In order to evaluate controls for significant improvement or deterioration after listing, we collected their lowest and highest MELD scores over the observation period. Case and control median baseline MELD scores (IQR) were 17 (15-18) and 16 (15-18), respectively (p=0.187), consistent with effective matching. The median (IQR) change between controls’ baseline MELD scores and their highest and lowest MELD scores during the observation period were 0 (0-2) and 0 (0-2), respectively, indicating minimal change in controls’ MELD scores throughout the observation period.

Analysis of Predictors of Low MELD Deaths

Using our 1:n case-control study design, we assessed predictors of liver-related death in univariate then multivariable analyses. The results of the univariate analysis are shown in Table 4. Statistically significant univariate predictors included: prior variceal bleed (OR 5.6, 95% CI 1.8-17.7, p=0.003), prior or current Grade 3 or refractory ascites (OR 4.1, 95%CI 1.5-11.4, p=0.008), serum albumin (OR 0.5, 95% CI 0.2-1.0, p=0.041), serum sodium (OR 0.9, 95% CI 0.8-1.0, p=0.051), serum sodium < 135 mmol/dL (OR 3.4, 95% CI 1.4-8.3, p=0.006), cirrhosis stages 4-5 vs. stages 1-3 (OR 4.1, 95%CI 1.5-11.4, p=0.008), cirrhosis stages 3-5 vs. stages 1-2 (OR 7.4, 95%CI 2.5-21.8, p=0.000), cirrhosis stages 2-5 vs. stage 1 (OR 3.6, 95%CI 1.0-13.1, p=0.048), and increasing stage of cirrhosis as a categorical test of trend (p=0.003). Prior or current hepatic encephalopathy (OR 2.3, 95% CI 0.9-6.0, p=0.082) also met our threshold for statistical significance on univariate analysis. Prior or current non-bleeding varices (OR 1.5, 95% CI 0.6-3.7, p=0.361), cirrhosis stage 5 vs. cirrhosis stage 1-4 (OR 4.0, 95% CI 0.7-21.4, p=0.108), and history of TIPS placement (OR 2.3, 95% CI 0.6-8.4, p=0.215) did not have a significant association with liver-related death. In a multivariable model including cirrhosis stage as an ordinal variable tested for trend, serum albumin, serum sodium, and hepatic encephalopathy, only cirrhosis stage remained significant (p=0.010) [Table 5 and Figure 2].

Table 4.

Univariate Analysis. Risk Factors for Liver-Related Mortality Among Low MELD Patients Listed for Liver Transplant

Variable	Case n=41	Control n=66	OR	p value
Cirrhosis Stage 1-5*	-	-	-	0.003†
Cirrhosis Stage 5 (vs 1-4) %	12.2	3.0	4.0	0.108
Cirrhosis Stage 4-5 (vs 1-3) %	39	15	4.1	0.008†
Cirrhosis Stage 3-5 (vs 1-2) %	66	24	7.4	0.000†
Cirrhosis Stage 2-5 (vs 1) %	88	71	3.6	0.048†
Albumin (SD**) g/dL	2.5 (0.7)	2.8 (0.6)	0.5	0.041†
Sodium (SD**) mmol/L	133.0 (4.4)	134.9 (5.2)	0.9	0.051†
Sodium Dichotomous < 135mmol/L %	71	41	3.4	0.006†
Varices %	76	67	1.5	0.361
Variceal Bleeding %	39	12	5.6	0.003†
Ascites %	39	15	4.1	0.008†
Hepatic Encephalopathy %	83	70	2.3	0.082†
TIPS %	17	7.6	2.3	0.215

^*Ordinal variable with test of trend

^**Standard Deviation

^†Statistically significant on univariate analysis (p ≤ 0.10)

Table 5.

Multivariable analysis. Risk Factors for Liver-Related Mortality Among Low MELD Patients Listed for Liver Transplant

Variable	OR	95% CI	p value
Cirrhosis Stage 1-5*			0.010
Hepatic Encephalopathy	1.8	0.6-5.9	0.306
Continuous Sodium	1.0	0.9-1.1	0.607
Albumin	0.6	0.3-1.4	0.247

^*Ordinal variable with test of trend

Figure 2.

Odds Ratio (95% Confidence Interval, p-value) for liver-related mortality by Cirrhosis Stage (Reference is Stage 1) in Multivariable Analysis.

Our findings were robust in three sensitivity analyses. First, conditional logistic regression was repeated using refractory ascites as the threshold for stage progression, and cirrhosis stage as an ordinal value tested for trend remained statistically significant (p=0.046). Second, conditional logistic regression was repeated again after removing cases/controls with HCC from the analysis. Lastly, in a subgroup analysis of patients who had variceal bleeding, having more than one bleed, ever being on non-selective beta blockers, and undergoing TIPS were not associated with liver-related death (data not shown).

In order to investigate non-clinical variables that may explain our findings, we also analyzed available markers of socioeconomic and insurance status in our study population. Two variables of education, whether the subject graduated high school or obtained a Graduate Equivalency Diploma (GED) and whether the subject attended at least some college or technical school, were not significantly associated with death (high school graduation OR >999, 95% CI <0.001->999, p=0.9932; attending college OR 0.5, 95% CI 0.2-1.2, p=0.120). Insurance status coded as private versus public insurance showed a statistically significant protective effect in univariate analysis (OR 0.4, 95% CI 0.2-0.9, p=0.026); however, when included in the multivariable model with cirrhosis stage, hepatic encephalopathy, serum sodium, and serum albumin, it failed to reach statistical significance (OR 0.3, 95%CI 0.1-1.0, p=0.052) while cirrhosis stage remained statistically significant (p=0.047). While insurance status did not meet our threshold for statistical significance, the numerical value of the OR suggested a protective effect.

Discussion

This study examines the role of a new and evolving paradigm for predicting risk of liver-related death in patients listed for LT with low MELD scores (≤ 20). Our data suggests that increasing cirrhosis stage is the major risk factor for 90-day mortality in our study population. After controlling for several factors through our study design (patient age, gender, liver disease etiology, presence of hepatocellular carcinoma and MELD score) and developing a multivariable model including cirrhosis stage, serum sodium, serum albumin, and hepatic encephalopathy, only increasing cirrhosis stage remained significant as a predictor of death in patients listed for LT with MELD scores ≤ 20.

The uniqueness of the D’Amico model is the ability to prognosticate not just mortality, but also the risk of transition to more advanced stages. Our current study expands on that work by assessing cirrhosis stage among patients listed for liver transplant with low MELD scores. The importance of our data is twofold. First, our findings have implications for surveillance for low MELD patients while awaiting LT. Current UNOS policy requires updated MELD scores every three months for patients with MELD scores from 11 to 18. Of the 1676 patients listed during our study period, 117 (7%) died with a MELD score ≤ 20 within 90 days of death, representing 36% of patients who died on the transplant list during that period. Infection contributed to the deaths of most of the subjects in our study. This argues for increased surveillance practices (laboratory draw, clinic visits, imaging, variceal surveillance) and strict adherence to recommendations on antibiotic prophylaxis and early detection of infection¹⁴ for low MELD patients who have reached higher D’Amico cirrhosis stages and who therefore are at high risk of experiencing a serious decompensation or liver related death despite their low MELD score. Investigation regarding the effect of appropriate prophylactic antibiotics on outcomes may help define the risk of acute decompensation and death in relatively well compensated patients.

Furthermore, our results question the idea of a low MELD patient being “too well to transplant”, which has been a frequently referenced reason to not transplant. This in part comes from a seminal paper questioning the transplant benefit in patients with a MELD 15 or less¹⁵, but is based on a heterogeneous population not taking into account clinical factors outside the MELD. Our results suggest that for a subpopulation of low MELD transplant candidates, the transplant benefit may extend to lower MELDs than previously reported. Second, our data have implications for selecting patients for alternative donor sources and suggest that a population of patients with low MELD scores but high cirrhosis stage may benefit from living donor or extended criteria donors because of their higher risk of mortality despite low priority for deceased donation liver transplant.

Our data show a mortality risk plateau effect between cirrhosis stage 3 and 4 [Figure 2]. The difference between these stages and cirrhosis stage 2 is the development of either variceal bleeding (stage 3) or ascites (stage 4). The similar effect on mortality of ascites and bleeding, which causes the plateau effect, is consistent with other contributions to the literature citing equal mortality risks for patients with ascites and patients with variceal bleeding¹⁶.

The MELD score is the primary mortality risk assessment tool used in patients with cirrhosis who are listed for LT, as well as the backbone for the liver allocation system in the United States. However, other clinical variables such as presence of ascites and hyponatremia have been shown to improve upon the predictive ability of the MELD score, especially among patients with MELD ≤ 20^8,11,17-19. Our results support that assessing cirrhosis stage improves the prediction of liver-related death in patients with MELD ≤ 20. Our study identifies a subpopulation of patients whose mortality risk appears to be underestimated by the MELD score and identifies a risk factor (increasing cirrhosis stage) that identifies patients at higher risk of 90-day mortality. It is important to recognize the cirrhosis stage paradigm as a unidirectional and cumulative assessment of risk as opposed to the snap-shot provided by the CTP score. For example, once a patient develops ascites, he or she is at least cirrhosis stage 4 regardless of how well the ascites is medically managed until transplant or death.

One of the strengths of the MELD score is its use of only objective variables, and other authors have discussed the problems with using subjective scoring systems like the Child Pugh score^1-5. Cirrhosis stage may have some of these same subjectivity problems; however, they are mitigated by the clinically progressive nature of the cirrhosis stages. Furthermore, our sensitivity analysis examining the use of refractory ascites and still reaching a significant result argues against subjectivity playing a major role in our results. We are not suggesting the use of the cirrhosis staging system as a modification to the current liver transplant allocation system, but rather as an adjunct to clinical decision making in transplant candidates with low MELD scores.

The relative infrequency of death with a low MELD score made a prospective or retrospective cohort study impractical. For this reason, we chose a nested case-control study design to account for the relatively rare outcome. In order to increase the power of our study, we employed a 1:n matching strategy. While the baseline demographics were imperfectly balanced as a result, this was accounted for by conditional logistic regression in our statistical analysis.

We made study design choices to limit the influence of bias on our study. We chose the outcome of liver-related death because patients with low MELD scores may be more likely to have a non-liver death than patients with high MELD score. Despite this, we found that of the 46 patients dying in our hospital with detailed records available, only 2 had a non-liver death. Patients dying outside our hospital were excluded because of insufficient data and uncertainty of cause of death; however, these patients were listed and followed by our transplant center until the event that lead to death outside our hospital, and their characteristics did not differ considerably from those dying within our hospital system. We selected a case-control design with matching to mitigate the influence of measured (patient age, gender, liver disease etiology, presence hepatocellular carcinoma, MELD score) and associated unmeasured factors that may be related to our outcome. While cases and controls were chosen from different time points in their courses within our institution (cases after listing and controls at listing), they were matched on MELD score suggesting they had similar disease severity apart from the variable of interest (cirrhosis stage). For our control subjects, we allowed 90 days of observation after the date of listing to completely assess for prevalent conditions that define the cirrhosis stage. Similarly, we assessed the cirrhosis stage in our case subjects by observations made up to the time of their baseline MELD score. The baseline MELD scores were similar between cases and controls, suggesting that the MELD is not able to accurately assess the severity of liver disease and risk of death in patients with MELD ≤ 20. Cirrhosis stage could further stratify low MELD patients into different mortality risk categories.

In D’Amico’s cirrhosis stage model, the degree of ascites used as a threshold was not explicitly stated. We created an a priori assessment tool for retrospective review that graded ascites into Grade 1 (radiographic evidence only), Grade 2 (clinically apparent up to moderate), Grade 3 (large ascites and/or paracentesis greater than 3 liters), or refractory (resistant or intractable to medical therapy and requiring repeated large-volume paracentesis or TIPS)^12,13. We felt that the use of large ascites as the threshold for cirrhosis stage advancement represented a clinically meaningful threshold of ascites and likely discriminates severe liver disease from mild liver disease better than a less severe threshold (radiographically apparent only or mild/moderate ascites).

We also analyzed the significant variable of interest, progression of cirrhosis stage, using refractory ascites as the threshold as a sensitivity analysis, and it remained significant. We investigated the impact of multiple episodes of variceal bleeding, use of beta blockers, or TIPS in the subgroup of patients who had variceal bleeding. While our analysis showed no significant difference, we were underpowered to detect a difference. With regard to TIPS, while our study was underpowered to detect an association with mortality, previous contributions to the literature support a lack of mortality benefit from TIPS²⁰.

Finally, we analyzed the effect of three variables of socioeconomic status. One variable, insurance status, met statistical significance on univariate analysis, but was no longer significant in the multivariable analysis. Despite not meeting the threshold for statistical significance, insurance status showed a trend toward significance with a strong odds ratio. This may be an important association worthy of further investigation in future studies.

Our results support the use of a five stage clinical model to predict mortality in patients with low MELD scores listed for LT. These results support the need for increased attention to surveillance and prophylaxis in this population. Changes in monitoring parameters could be made using this information. In addition, cirrhosis stage could be used to make appropriate patient selection choices for living donation or extended criteria organs. This application could be quite meaningful for this patient population whose low MELD score prohibits deceased donation LT in most U.S. regions but whose disease severity informed by their cirrhosis stage predicts a higher risk of mortality than their calculated MELD alone.

Abbreviations

LT

Liver transplant