|Home | About | Journals | Submit | Contact Us | Français|
Donor livers are offered to patients with the highest risk of death. How ascites may inform risk models to reduce liver transplant wait-list mortality is unclear. All adult candidates for primary liver transplantation for cirrhosis without exception points who were registered with the Organ Procurement and Transplantation Network from 2005 to 2007 comprised our study cohort. Using Cox models and advanced discriminative metrics, we evaluated the additional risk discrimination of moderate ascites over that of MELD and MELD plus serum sodium (MELDNa) alone to predict 90-day wait-list mortality with attention to geographic disparities. Additional analyses examined lower mortality risk candidates and those listed in high-demand, low-supply UNOS regions where accounting for ascites may impact wait-list mortality most significantly. Between 2005 and 2007, 18,124 subjects were listed for liver transplantation. Mortality was higher in patients with moderate ascites (15.5% vs. 6.0%, p<0.0001), a risk that persisted despite adjusting for MELD (HR 1.58, 95% CI 1.42–1.76) and MELDNa (HR 1.42, 95% CI 1.28–1.58). The effect of moderate ascites was more prominent when MELD<21, equating to 4.7 MELD units, or when MELDNa<21, equating to 3.5 MELDNa units. Wait-list mortality was higher in patients with moderate ascites listed in high-demand, limited-supply regions (25.8% vs. 17.5% at one-year, p<0.01). There was improvement in the overall risk model, particularly when MELD<21, with the addition of moderate ascites as measured by the C-index and IDI. Moderate ascites informs risk prediction, particularly when MELD<21 and in high-demand limited-supply regions. Under the MELD system, presence of moderate ascites should prompt clinicians to consider strategies to expand access to transplantation such as extended donor liver grafts.
Patients with cirrhosis and ascites have increased mortality risk.(1–3) Traditionally ascites has been regarded as a subjective measure that is often transient in nature, limiting its effectiveness as a reliable marker for risk prediction models; moreover, with the advent of the model for end-stage liver disease (MELD) and MELD plus serum sodium (MELDNa) score for risk prediction, ascites as a risk parameter has further fallen out of favor.(4–9) However, multiple groups have identified ascites as an important and independent predictor of short term mortality despite adjustment for MELD and MELDNa, but these studies have been limited to single center or small multicenter data sets.(4, 10–12) The impact of ascites on survival in this dataset offers a unique opportunity to examine its independent relationship without the transplant selection bias since the adoption of MELD-based scoring system.
Even with the advent of MELD and MELDNa for accurate risk profiling, there is substantial mortality on the transplant wait list. Although MELD and MELDNa captures short-term mortality risk well especially at high values, ascites may represent an important marker of liver decompensation not captured by low MELD and MELDNa scores. Therefore, we hypothesized that inclusion of ascites may be beneficial to patients with low MELD and MELDNa scores to identify when extended criteria donors may be warranted; furthermore, we hypothesize that ascites becomes more important in UNOS Regions with scarcity of organ donors leading to a high mean MELD score at transplantation. Within these high-demand, limited-supply regions, presence of moderate ascites may provide guidance of when to seek high-risk donor or extended criteria donors because the mortality risk is underestimated by current MELD allocation.
Data on all patients registered for liver transplantation in the United States were available through the Standard Transplant Analysis and Research file, provided by the Organ Procurement and Transplantation (OPTN) network. The file, created February 20, 2009, included listings for liver transplantation, laboratory data to calculate MELD and MELDNa, as well as outcomes while waiting (transplantation, death, or other withdrawal from the waitlist). Degree of ascites at registration, provided to OPTN by transplant coordinators, was coded as absent, slight or controlled by diuretics, or at least moderate despite diuretic use.
Our target population was adults with cirrhotic liver disease listed for first transplantation in the United States. We included new registrations during calendar years 2005 to 2007 (n=33,103) with complete laboratory and clinical data including 90 day follow up. For patients with multiple registrations in the study period, we used the earliest registration (n=2,760). Patients were excluded if had prior liver transplantation (n=1,159), were under age of 18 (n=2,231), had primary diagnoses other than cirrhotic liver disease (e.g. acute liver failure or hepatocellular carcinoma) (n=3,266), received hepatocellular carcinoma exception points (n=2,332), or received a simultaneous transplant other than a kidney (lung, heart, intestine, or pancreas) (n=73). We excluded an additional 3,164 patients for whom complete laboratory data were not available within 7 days of registration.
Etiologies were determined using diagnosis code fields and supplemented by diagnosis text boxes keyword search. The outcome of waiting for each patient (transplantation, death, or withdrawal from the list for another reason) was used to define our primary outcome variable, 90-day mortality while waiting for a transplant. Patients who were deemed too sick for transplantation and removed from the list and who later died were included as having died while waiting.
Patient demographics, baseline chemistries, etiologies, blood group, MELD, and MELDNa scores were tabulated and compared using t- or nonparametric tests for continuous variables, and chi-square or Fisher’s exact test for categorical variables. The MELD score was calculated using the standard formula: 11.2*ln(INR) + 9.57*ln(creatinine, in mg per deciliter) + 3.78*ln(bilirubin, in mg per deciliter) + 6.43, with a lower limit of 1 for all variables, upper limit of creatinine of 4 and a maximum MELD score of 40.(4) For patients who received dialysis twice within a week prior to the creatinine test, MELD was calculated with creatinine set to 4. A MELDNa value was calculated for each individual using the following formula: MELDNa = MELD + (140-Na) −0.025*MELD*(140-Na) where Na is capped between 125 mmol per liter to 140 mmol per liter.(6) Extreme laboratory values suggestive of error were dropped (INR > 25, bilirubin > 70, sodium > 160), which posed less than 0.1% of the values.
Cox proportional-hazard models were used to predict 90-day mortality. Proportional-hazard assumptions were examined using plots of Schoenfeld residuals. During model development, the effect of increasing severity of ascites was evaluated using a test of trend across categories, and the joint effects of ascites and Na were evaluated graphically and using Cox models. Categorizing ascites into none and small versus moderate most effectively discriminated ascites into binary risk groups. Graphical methods were also used to assess non-linearities as well as interaction between ascites and the MELD and MELDNa score.
We compared the discrimination for 90-day mortality of the predictors MELD, MELDNa, and the inclusion of moderate ascites using three metrics. First, we computed the C-index both for the individual indices and the addition of moderate ascites to the models. Second, we computed the integrated discrimination improvement (IDI), a novel analog of the change in the C-index.(13) Third, we calculated the net reclassification index (NRI).(13) In computing the NRI, we classified predicted 90-day mortality as less than 5%, 5–10%, and greater than 10%. Sensitivity analyses were performed comparing the hazards ratio between cohorts with moderate versus small ascites and when TIPS patients were excluded.
Since moderate ascites is subjective and may be used clinically for selecting candidates for high-risk donors, additional analyses were performed at MELD scores less than 21. The three metrics to analyze discrimination using the addition of moderate ascites to MELD and MELDNa were repeated. In addition, 90 day and one-year cumulative mortality rates among patients with MELD < 21 and moderate ascites were compared between regions of high and low MELD scores at transplant. UNOS region 1 (MA, VT, NH, CT, RI), region 5 (CA, NV, AZ, NM, UT), and region 9 (NY) were clustered as high MELD at transplant (mean MELD=27 at transplant) whereas the remaining regions were considered low MELD at transplant (mean MELD=23 at transplant). Proportions were compared using chi-square statistics.
STATA Version 11 (Stata Corp., College Station, TX) was used for all analyses. This study was approved by the committee on human research at the University of California, San Francisco.
A total of 18,124 patients were included in our final cohort. The majority (64%) were male and the median age was 53 years (range, 18 to 83). The most common etiologies of liver disease were HCV (41.7%) and alcohol (30.2%). At registration, the median MELD score was 16 (range 6 to 40) and the median MELDNa score was 19.3 (range, 6 to 40). At registration, 18.5% did not have ascites, 57.2% had small ascites, and 24.4% had at least moderate ascites. Demographic characteristics of the overall group and according to ascites status are shown in Table 1.
At the end of the 90 days following registration, 1498 (8.3%) patients had died. Mortality was higher in patients with moderate ascites (683/4417, 15.4%) as compared to those with small ascites (681/10358, 6.6%, p<0.0001) and those without ascites (138/3349, 4.1%, p<0.0001). The Kaplan Meier curve graphically demonstrates the adverse survival outcomes associated with moderate ascites (Figure 1). When compared to patients with none or small ascites, those with at least moderate ascites fared significantly worse (HR 3.59, 95% CI 3.24–4.0, Table 2). Moreover, moderate ascites remained an important predictor for short term mortality after adjustment for MELD (HR 1.58, 95% CI 1.42–1.76) and MELDNa (HR 1.42, 95% CI 1.28–1.58) (Table 2). Graphical methods demonstrated that ascites was associated with increases in mortality risk across the range of MELD and MELDNa scores (Figure 2a, b). Restricting the analyses to only patients with ascites (moderate versus small) and the patients without TIPS did not significantly alter the measure of association.
The hazard ratio associated with moderate ascites remained significant elevated (HR 3.22, 95% CI 2.64–3.94) when the cohort was restricted to those with MELD < 21 (Table 3). Similarly, after adjusting for MELD or MELDNa, moderate ascites remained an independent predictor of short term mortality (HR 2.62, 95% CI 2.14–3.21; HR 2.03, 95% CI 1.65–2.49, respectively). These hazards ratios translate to approximately 4.7 MELD units and 3.5 MELDNa units; that is, for an individual with a MELD score of 14, the adjusted MELD score in the presence of moderate ascites is approximately 18.7.
We also graphically explored the effect of ascites on the mortality risk independent of serum sodium (Figure 3). Moderate ascites was strongly associated with waitlist mortality (HR 2.00, 95% CI 1.85–2.17), an effect that persisted across the spectrum of serum sodium levels especially those in the normal range (>135mEq/dl) (HR 2.64, 95% CI 2.38–2.93).
The 90-day mortality was predicted with good discrimination by the MELD score (C-index 0.859), and more powerfully by the MELDNa score (C-index 0.869; Table 2). This modest improvement in the C-index was matched by the integrated discrimination improvement (IDI) of 0.5% (p<0.05) and 0.4% (p<0.05), respectively.
In the entire cohort, the addition of moderate ascites improved risk reclassification in approximately 3% of individuals (supplementary Tables S1, S2); that is, ascites more accurately categorized patients who died into a high risk profile while in individuals who did not die, it lowered their risk profile. These findings were seen in both models, that is, the addition of moderate ascites to MELD and MELDNa.
This effect of moderate ascites was larger when the cohort was restricted to MELD less than 21. In this setting, moderate ascites reclassified 8.6% of individuals more accurately compared to MELD and 5.8% of individuals more accurately when compared to MELDNa (supplementary Tables S3, S4). Other metrics of risk discrimination, such as the improvement in the C-index and IDI, were also observed (Table 3).
The mean MELD at the time of transplantation in high-demand, limited-supply UNOS regions (regions 1, 5, 9) was 27; in contrast, the remainder of the nation had a mean MELD at transplant of 23. Comparing the high to low MELD regions at time of transplantation, the presence of moderate ascites at listing was associated with a significantly elevated mortality rate at the end of one year (25.8% vs. 17.5%, RR=1.47, 95% CI 1.31–1.66). The regional variation was also evident when mortality rate was assessed at 90 days (19.2% vs. 13.8%, RR=1.39, 95% CI 1.21–1.60). Among those with moderate ascites and MELD < 21 who were listed at high-demand limited-supply regions, 10.6% of the individuals died at 90 days without a transplant and 20.8% died at one year. This compares unfavorably to individuals who are listed at the other regions associated with low MELD at transplantation, where 7.1% and 13.4% died at 90 days and one year, respectively. Between the regions, patients with moderate ascites and low MELD at listing died more frequently, an excess of 7.4% after 1-year (95% CI 3.5–11.3%, p<0.0001) and relative risk of 1.56 (95% CI 1.26–1.92). For every 13.5 patients with moderate ascites listed in high-demand, limited-supply regions, an additional death occurs while waiting compared to being listed at the other regions.
In this study of the US population of cirrhotic patients listed for liver transplantation, the MELD and MELDNa scores underestimate the mortality risk among 24.4% of patients with moderate ascites, an effect that persisted across the spectrum of MELD, MELDNa and serum sodium. Despite the subjective nature of ascites measurement, moderate ascites informed risk models beyond that of the MELD and MELDNa with a substantially improved waitlist mortality risk discrimination assessed by the C-statistic, IDI and NRI. It is not the intention of the authors to replace MELD or increase its subjectivity for liver allocation; instead, these results offer an opportunity to refine the risk prediction and interpretation of MELD, especially in the low-MELD spectrum. Clinically, the increased risk associated with moderate ascites can inform clinical decision-making and identify low-MELD patients who may benefit most from extended criteria donors.
The impact of ascites is most prominent at low scores, such that moderate ascites equates to the addition of 4.7 MELD points or 3.5 MELDNa points. Improvement to prognostic models is often limited by the “first-mover advantage” that is commonly witnessed in market competition. That is, the addition of a parameter to currently accepted prognostic models, whether that be MELD or MELDNa, is often at best incremental; however, we demonstrate here that the addition of this parameter is clinically significant under certain conditions. These conditions are those patients with low MELD/MELDNa and particularly among patients who are listed in high-demand, low-supply UNOS regions. In those regions, ignoring the presence of ascites underestimates the mortality risk and is associated with increased 90-day and 1-year wait list mortality. The clinical importance of this parameter is mirrored by the statistical significance (risk reclassification and NRI, improvement in the receiver operator curve and the C-statistic, and the IDI). These statistical tests convey that incorporating ascites into the risk model more accurately categorized patients who died into a high-risk profile while in individuals who did not die, it lowered their risk profile. While similar findings have been reported previously, this work extends beyond small retrospective studies and single institutions.(4, 10–12)
In geographic areas with heightened scarcity of available liver grafts, patients with ascites may require more aggressive monitoring and stronger consideration of extended criteria donors. Under the current MELD-based liver allocation, MELD score alone is used for prioritization of urgency for liver transplantation. Because ascites is a well known predictor of mortality with cirrhosis,(10, 14–16) it is a common, yet often unsuccessful, justification cited in petitions for exceptional priority outside the standard MELD-based allocation.(17–19) The MELD score required to obtain a liver graft offer varies dramatically by UNOS region and the designated service areas within UNOS region.(20, 21) Without an approved petition for additional exceptional priority, our study and a smaller study of 211 cirrhotic veterans suggest that patients with ascites may be further disadvantaged in their access to transplant by approximately 4.7 MELD points when the laboratory MELD was < 21.(10) Additionally, our study demonstrates that not accounting for the risk associated with moderate ascites results in the 8.6% misclassification in this population with limited access to liver transplant because of a MELD <21. Schaubel et al. have shown that with an increasing severity of illness as measured by MELD, there is an increasing benefit of transplantation with less ideal liver grafts as measured by the donor risk index (DRI). (22) Given the increased risk of dying without the opportunity for liver transplantation, patients with moderate ascites could benefit from a close assessment of the relative risks and benefits of extended donor liver graft options.
Concern about the objectivity of quantifying ascites has cast doubt on inclusion of ascites in risk prediction models. While it is not the objective of the authors to incorporate ascites to an allocation system, the risk-classification data particularly in the low-MELD cohort suggests that incorporating it into clinical decision making such as improving patient selection for extended donor liver grafts may improve outcomes. Given the frequent use and availability of radiologic studies in cirrhotic patients, ascites can be easily assessed by a relatively unbiased provider, the radiologist. However, limitations exist regarding the volumetric quantification of ascites as there is no currently accepted standard or no documented clinical benefit of precise measurements. The current variability in the assessed volume (none, small, moderate) between and within each individual radiologist can lead to the misclassification of ascites volume; however, this inaccuracy in coding small ascites and moderate ascites and vice versa would underestimate the impact of this parameter on mortality. In addition, the UNOS database lacks data on medication use and precludes examination of the contribution of diuretics or salt-restriction. It is likely that most patients with ascites who are listed and cared for by transplant hepatologists would be on these medications. Regardless, the inability to control ascites such that it is absent or small portends poor short-term outcomes. Data on the cause-specific mortality is also lacking, which limits our ability to identify interventions to decrease mortality in this cohort. Future research may quantify the utility of volumetric measurements or identify more objective measures of the hemodynamic derangements in cirrhosis not accounted for by current models, such as plasma levels of aldosterone, renin, or norepinephrine, that may be collinear with ascites and importantly delineate the mortality risk.(23)
In conclusion, moderate ascites informs mortality risk prediction in cirrhotic patients beyond MELD and MELDNa. Under the current MELD based liver allocation system, the presence of moderate ascites should prompt clinicians and patients to consider strategies to expand access to liver transplantation such as extended donor liver grafts.
Financial Support: This work was funded in part by grants from the American Society for Gastrointestinal Endoscopy Career Development Award and P30 DK026743 (UCSF Liver Center) (MS), from the National Institute of Diabetes and Digestive and Kidney Diseases (DK076565, SWB) (K24 DK080941, JMI), and from Agency for Healthcare Research and Quality (DK076565, SWB).
Disclosure: The authors of this manuscript have no conflicts of interest to disclose.