PATIENTS AND METHODS. Data Source

Transcription

1 LIVER TRANSPLANTATION 20: , 2014 ORIGINAL ARTICLE Waiting Time Predicts Survival After Liver Transplantation for Hepatocellular Carcinoma: A Cohort Study Using the United Network for Organ Sharing Registry Barry Schlansky, 1 Yiyi Chen, 2 David L. Scott, 3 Donald Austin, 2 and Willscott E. Naugler 1 1 Division of Gastroenterology and Hepatology, Department of Medicine; 2 Department of Public Health and Preventive Medicine; and 3 Division of Transplantation, Department of Surgery, Oregon Health and Science University, Portland, OR Recipients of liver transplantation (LT) for hepatocellular carcinoma (HCC) have an 8% to 20% risk of HCC recurrence. Single-center studies suggest that a period of waiting after HCC therapy may facilitate the selection of patients at low risk for post-lt HCC recurrence and mortality. We evaluated whether a longer waiting time after Model for End-Stage Liver Disease (MELD) prioritization for HCC predicts longer post-lt survival. From the United Network for Organ Sharing registry, we selected 2 groups registered for LT between March 2005 and March 2009: (1) HCC patients receiving MELD prioritization and (2) non-hcc patients. Patients were stratified by their MELD status at LT (a marker of time on the wait list after HCC MELD prioritization) and were followed from LT until death or censoring through October By comparing post-lt survival to intention-to-treat (ITT) survival from registration, we assessed predictors of post-lt survival and estimated the benefit of LT. The median MELD scores at LT were 22 (HCC) and 24 (non-hcc). A higher MELD score at LT was independently associated with lower post-lt mortality in the HCC group [hazard ratio (HR) , 95% confidence interval (CI) ] and higher post-lt mortality in the non-hcc group (HR , 95% CI ). Compared with the HCC group, the non-hcc group had lower post-lt mortality [relative risk (RR) , log-rank P < 0.01] but higher ITT mortality (RR , log-rank P < 0.01) because of a 33 percentage point lower probability of undergoing LT. In conclusion, a longer waiting time before LT for HCC predicted longer post-lt survival in a national transplant registry. Delaying LT for HCC may reduce disparities in ITT survival and access to LT among different indications and thereby improve system utility and organ allocation equity for the overall pool of LT candidates. Liver Transpl 20: , VC 2014 AASLD. Received January 2, 2014; accepted May 12, Additional Supporting Information may be found in the online version of this article. Abbreviations: AFP, alpha-fetoprotein; BMI, body mass index; CI, confidence interval; DMF, Death Master File; HBV, hepatitis B virus; HCC, hepatocellular carcinoma; HCV, hepatitis C virus; HR, hazard ratio; ITT, intention-to-treat; LDLT, living donor liver transplantation; LT, liver transplantation; MELD, Model for End-Stage Liver Disease; NASH, nonalcoholic steatohepatitis; OPTN, Organ Procurement and Transplantation Network; PBC, primary biliary cirrhosis; PSC, primary sclerosing cholangitis; RR, relative risk; STAR, Standard Transplant Analysis and Research; UNOS, United Network for Organ Sharing. The authors report no conflicts of interest. This work was supported in part by the Health Resources and Services Administration (contract C). The content is the responsibility of the authors alone and does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government. Address reprint requests to Barry Schlansky, M.D., M.P.H., Division of Gastroenterology and Hepatology, Department of Medicine, Oregon Health and Science University, 3181 Southwest Sam Jackson Park Road, Mail Code L-461, Portland, OR Telephone: ; FAX: ; schlansk@ohsu.edu DOI /lt View this article online at wileyonlinelibrary.com. LIVER TRANSPLANTATION.DOI /lt. Published on behalf of the American Association for the Study of Liver Diseases VC 2014 American Association for the Study of Liver Diseases.

2 1046 SCHLANSKY ET AL. LIVER TRANSPLANTATION, September 2014 The incidence of hepatocellular carcinoma (HCC) is rising in both developing and Western nations. 1,2 Unique among HCC therapies, liver transplantation (LT) affords an opportunity both to cure HCC that remains confined to the liver and to prevent future de novo HCC through the removal of the background field defect of cirrhosis, the main risk factor for hepatocarcinogenesis. 3 Although LT for early HCC results in acceptable survival, 4-6 recipients nevertheless have an 8% to 20% risk of HCC recurrence at a median of 23 to 25 months after LT. 7-9 Retransplantation for recurrent HCC is discouraged because of the poor survival and organ scarcity, and the median survival is less than 1 year. 10,11 Although increases in the size and number of HCC tumors are associated with lower survival after LT, 3,12 patients with HCC exceeding the Milan criteria have demonstrated low rates of post-lt HCC recurrence and mortality with down-staging protocols when a tumor response has been achieved with liver-directed HCC therapy A shared feature of these studies is a period of monitoring during and after HCC therapy: candidates experienced either cancer progression (which resulted in wait-list dropout or death) or cancer stability (which resulted in LT). The extremely favorable results observed in these studies have prompted speculation that, instead of the favorable outcomes being derived solely from the pre-lt cancer therapy, the time period between the cancer treatment and LT may have facilitated the selection of LT candidates at low risk for post-lt HCC recurrence. 16,17 This theory is analogous to the length bias observed in cancer screening: indolent tumors spend more time in the preclinical disease phase and are, therefore, more likely to be screen-detected than aggressive tumors that rapidly present with symptoms. Similarly, an observation period before LT may allow the exclusion of patients with biologically aggressive cancers, who will derive minimal survival benefit from LT because of the high risk of post-lt cancer recurrence. Despite modifications of Organ Procurement and Transplantation Network (OPTN) policy intended to deemphasize Model for End-Stage Liver Disease (MELD) prioritization for HCC, patients with HCC continue to have more opportunities to undergo LT in comparison with patients with other indications. 18 Although an increased waiting time for candidates with HCC predicts wait-list death and dropout due to cancer progression, these events are more frequent for candidates who do not have HCC. 18 Strategies to equalize the probabilities of wait-list dropout and death for patients with HCC and patients with non-hcc LT indications have been proposed. 22 However, allocation policies exclusively focusing on pre-lt outcomes ignore differences in post-lt survival that contribute to the overall utility of LT. 17 The consideration of the intention-to-treat (ITT) survival from wait-list registration, which incorporates the survival of patients waiting for LT and the survival of those patients undergoing LT, may inform a policy that balances key ethical principles for the allocation of scarce resources In this study of a national transplant registry, we hypothesized that increased time on the LT wait list after MELD prioritization for HCC would predict longer post-lt survival. We also hypothesized that this association would not be observed in non-hcc patients. Finally, by comparing post-lt survival to ITT survival for each indication, we sought to estimate the benefit of LT for MELD-prioritized HCC patients versus non-hcc patients. PATIENTS AND METHODS Our aims were to (1) evaluate the data for an association between the time on the LT wait list after MELD prioritization for HCC (expressed as the MELD status at LT) and post-lt survival, (2) compare the association of the MELD status at LT and post-lt survival between MELD-prioritized HCC patients and non- HCC patients without MELD prioritization, and (3) compare ITT survival, wait-list death and dropout, and receipt of LT between MELD-prioritized HCC patients and non-hcc patients without MELD prioritization. To achieve the first 2 aims, we performed a survival analysis of persons registered for LT during the current MELD prioritization era (in which patients with HCC are granted 22 MELD points per OPTN policy ) 26 from the time of LT. The wait-list duration varies considerably by center and region, so it is a poor indicator of a patient s status at the time of LT. Consequently, we elected to use the MELD status at LT, which is useful both as a marker of increasing time from MELD prioritization to LT for MELDprioritized patients with HCC (a 10% increase in the MELD status is awarded for every 3 months on the wait list) and as a marker of increasing liver disease severity for patients without MELD prioritization; it thus allows patients with and without MELD prioritization to be compared. In comparison with low MELD status centers, high MELD status centers performed transplantation for patients who either waited longer after they had received MELD prioritization for HCC or had more severe liver dysfunction without MELD prioritization. To achieve the third aim, we performed an ITT analysis of persons registered for LT in the current HCC MELD prioritization era from the time of wait-list registration. The primary outcome was the ITT survival time, which took into account deaths on the wait list, wait-list dropouts, patients remaining on the wait list without LT, and patients surviving or dying after undergoing LT. Data Source We analyzed the Standard Transplant Analysis and Research (STAR) registry of the United Network for Organ Sharing (UNOS), which includes all US solid organ donation and transplant events reported to the OPTN since October 1, The STAR version was current through October 19, We linked the

3 LIVER TRANSPLANTATION, Vol. 20, No. 9, 2014 SCHLANSKY ET AL STAR registry to the Death Master File (DMF) to supplement deaths already reported to UNOS by individual centers. The DMF is maintained by the National Technical Information Service of the US Department of Commerce, and it contains more than 86 million records of persons with Social Security numbers whose deaths were reported to the Social Security Administration. 28 The DMF version was current through November 1, Study Population We selected 2 groups who were registered on the LT wait list between March 1, 2005 and March 1, The primary group was composed of patients receiving MELD prioritization for HCC with 22 MELD points, regardless of whether MELD prioritization was obtained according to standardized criteria for T2 HCC or through a specific determination by a regional review board. Patients with HCC exceeding the Milan criteria were included if MELD prioritization had been granted. The second group was composed of non-hcc patients who did not receive MELD prioritization for a non-hcc indication. The ITT cohort included all patients selected for these 2 groups, whereas the post-lt cohort included only the subset that underwent LT. Thus, a patient who underwent LT was included in both the ITT analysis and the post-lt analysis, but a patient who did not undergo LT was included only in the ITT analysis. For inclusion in the HCC group in the post-lt cohort, subjects were required to have had an active MELD exception for HCC at the time of LT, whereas for inclusion in the HCC group in the ITT cohort, subjects were required to have had an active MELD prioritization for HCC at any time while they were on the LT wait list (because patients receiving HCC MELD prioritization did not necessarily undergo LT). March 1, 2005 was selected as the initial date of enrollment because the current MELD prioritization protocol was enacted then by UNOS. The end date for enrollment (March 1, 2009) was selected to allow sufficient atrisk time for post-lt HCC recurrence and mortality; because most MELD-prioritized patients with HCC undergo LT within 1 year, the cohort was anticipated to have a minimum at-risk time of 30 months (LT was anticipated by March 1, 2010, and data collection ended on October 19, 2012), which exceeded the estimated median time to post-lt HCC recurrence by 5 to 7 months. We excluded patients < 18 years at the time of LT, Pediatric End-Stage Liver Disease registrants, patients with OPTN status 1 indications, living donor liver transplantation (LDLT) recipients, split or partial liver recipients, multiorgan recipients, prior LT recipients, and patients receiving MELD prioritization for non- HCC indications. Subjects with incidental HCC identified in their explants were excluded because the HCC diagnosis was unrecognized until LT. We excluded patients receiving MELD prioritization for alphafetoprotein (AFP) levels 500 ng/ml without radiographically identifiable liver tumors because few patients with this indication have HCC in their explants. 29 We excluded recipients of grafts from expanded criteria donors from the post-lt analysis because these organs may be associated with worse outcomes for the lower MELD status recipients to whom they are preferentially offered; thus, an expanded criteria donor status represents an effect modifier on the association of liver disease severity and post-lt mortality. 30 We adjusted for the donor risk index to account for residual differences in donor quality. 31 Predictors and Outcomes For the post-lt analysis, variables were recorded at the time of LT. Peak values of certain HCC-specific variables were recorded because testing was serially performed every 3 months on the LT wait list (peak AFP level, peak aggregate tumor diameter, and peak tumor number). The receipt of liver-directed therapy was defined as a patient having received locoregional HCC therapy at any time on the wait list. The outcome was the time from LT to death (the first death date reported by UNOS or the DMF) or the last reported living status. For the ITT analysis, variables were recorded at the time of wait-list registration. The outcome was the time from wait-list registration to death (the first death date reported by UNOS or the DMF) or the last reported living status. For subjects who dropped off the wait list but did not have a reported death, the date of wait-list dropout was used as the censoring date because these patients did not have additional documented follow-up; however, we also performed a sensitivity analysis under the assumption that waitlist dropouts lived until the end of data collection. Statistical Analysis Normally distributed variables were expressed as means and standard deviations, and nonnormally distributed variables were expressed as medians and interquartile ranges. Baseline variables were compared with the v 2 test if they were categorical, with the Student t test if they were normally distributed, and with the rank-sum test if they were nonnormally distributed. With the Kaplan-Meier method, 1-, 3-, and 5-year actuarial survival rates and associated 95% confidence intervals (CIs) were calculated for the overall post-lt and ITT cohorts and for each group. Mortality rates were compared with relative risks (RRs), and estimated survival functions were compared with the log-rank test. For the post-lt analysis, we performed univariate Cox proportional hazards regressions for all candidate predictors of survival. We selected variables for inclusion in the multiple regression model with stepwise methods (with P value thresholds of 0.05 for inclusion and 0.10 for elimination), and we secondarily evaluated a broader model that incorporated all potential

4 1048 SCHLANSKY ET AL. LIVER TRANSPLANTATION, September 2014 TABLE 1. Demographic and Clinical Characteristics of Patients Registered on the Wait List for LT Between March 1, 2005 and March 1, 2009 Posttransplant Cohort (n 5 10,653)* ITT Cohort (n 5 30,992) HCC Characteristics (n ) Non-HCC (n ) P Value Characteristics HCC (n ) Non-HCC (n 5 24,541) P Value Age (years) 57 (52-61) 53 (48-59) <0.01 Age (years) 56 (52-62) 54 (48-59) <0.01 Sex: male (%) <0.01 Sex: male (%) <0.01 Race/ethnicity (%) Race/ethnicity (%) White <0.01 White <0.01 Black 8 9 Black 8 8 Hispanic Hispanic Asian 10 3 Asian 11 3 Other 1 1 Other 1 1 BMI (kg/m 2 ) 28 (25-31) 28 (24-32) 0.65 BMI (kg/m 2 ) 28 (25-32) 28 (25-32) 0.31 Listing diagnosis (%) Listing diagnosis (%) HCV <0.01 HCV <0.01 Alcohol <0.01 Alcohol <0.01 HBV 8 4 <0.01 HBV 9 3 <0.01 NASH or cryptogenic 8 17 <0.01 NASH or cryptogenic 8 17 <0.01 PBC or PSC 2 10 <0.01 PBC or PSC 2 9 <0.01 Laboratory MELD score at LT 12 (9-15) 23 (18-30) <0.01 Laboratory MELD score 11 (8-14) 16 (12-22) <0.01 at registration MELD status at LT 22 (22-25) 24 (18-31) 0.47 LT (%) <0.01 Donor characteristics Dropped off wait list (%) <0.01 Donor risk index 1.44 ( ) 1.45 ( ) 0.03 Died on wait list (%) 5 18 <0.01 Donor sex: male (%) Died or dropped off wait list (%) Donor BMI (kg/m 2 ) 26 (23-30) 26 (23-30) 0.38 Died or dropped off wait list because of medical deterioration (%) HCC-specific variables HCC-specific variables Within Milan criteria (%) 95 Within Milan criteria (%) 94 Peak aggregate tumor 3.2 ( ) Peak aggregate tumor 3.3 ( ) Peak tumor number 1 (1-2) Peak tumor number 1 (1-2) Liver-directed therapy (%) 60 Liver-directed therapy (%) 58 Peak AFP (ng/ml) 15 (6-68) Peak AFP (ng/ml) 16 (6-82) < <0.01 *Variables were recorded at the time of LT. Variables were recorded at the time of wait-list registration. The data are presented as medians and interquartile ranges. Percentages may not sum to 100 due to rounding to the nearest 1%. Percentages may sum to greater than 100 due to patients having >1 listing diagnosis.

5 LIVER TRANSPLANTATION, Vol. 20, No. 9, 2014 SCHLANSKY ET AL STATA/IC 11.0 for Macintosh OS X (StataCorp, College Station, TX). The Oregon Health and Science University institutional review board approved the study protocol. Figure 1. Posttransplant survival of (A) MELD-prioritized HCC patients and (B) MELD-prioritized non-hcc patients registered on the LT wait list between March 1, 2005 and March 1, For HCC patients, a higher MELD score (>24), representing a longer waiting time before transplantation, identified a subgroup with longer posttransplant survival. In contrast, for non-hcc patients, a higher MELD score (>24), representing more severe liver dysfunction, identified a subgroup with shorter posttransplant survival. confounders. We tested for interactions between the MELD status and all other variables, for multicollinearity with variable inflation factors, and for violations of the proportional hazards assumption with an approximate score statistic of the linear correlation between the rank order of failure times in the cohort and Schoenfeld partial residuals. 32 Covariates that violated the proportional hazards assumption were stratified in the final models. We created HCC group specific and non-hcc group specific multiple regression models with the same methods. For the ITT analysis, we performed univariate Cox proportional hazards regressions for all candidate predictors of survival. We evaluated the mortality risk associated with wait-list dropout or the receipt of LT through the expression of these dichotomous exposure variables as time-dependent covariates in the Cox regression analysis. We estimated the benefit of LT for the HCC and non-hcc groups through the comparison of the mortality risk in the ITT cohort from registration to the mortality risk in the post-lt cohort with RRs. A 2-sided P value < 0.05 was considered to be statistically significant. Calculations were performed with RESULTS In all, 10,653 subjects were selected for the post-lt cohort: 3256 in the HCC group and 7397 in the non- HCC group (Table 1). HCC patients were more often male and nonwhite and more often had hepatitis B virus (HBV) or hepatitis C virus (HCV) in comparison with non-hcc patients (P < 0.01). The median MELD scores at LT were similar for the groups (22 for HCC patients and 24 for non-hcc patients, P ). Although the MELD status mirrored the calculated MELD score for non-hcc patients (median 5 23), the calculated MELD score was considerably lower than the MELD status for HCC patients (median 5 12, P < 0.01). The median times on the wait list were broad: 3.3 months (range months) for the HCC group and 1.6 months (range months) for the non-hcc group. Most HCC patients had tumors within the Milan criteria and peak AFP levels less than 100 ng/ml and received liver-directed HCC therapy. In all, 30,992 patients were selected for the ITT cohort: 6451 in the HCC group and 24,541 in the non-hcc group (Table 1). Seventy-five percent of the subjects in the HCC group underwent LT, and 23% died without LT or dropped off the wait list; the corresponding values were 42% and 48% for the non-hcc group. Most wait-list dropouts were due to medical deterioration, and 46% of the patients who were removed from the wait list died within a median of 1.7 months. Association of the MELD Status at LT With Post-LT Survival We first stratified post-lt survival according to the MELD status at LT: high (>24) or low (24). The MELD score cutoff of 24 was chosen from the sample distribution to allow a balanced comparison because the majority of the HCC patients had a MELD score- 24 at LT. The high MELD category identified an HCC subgroup with significantly longer post-lt survival, whereas the opposite relationship was observed for the non-hcc group (Fig. 1). In the fitted multiple regression model, higher MELD scores at LT, indicating progressively longer waiting times before LT, independently predicted a lower hazard of post-lt mortality in the HCC group [hazard ratio (HR) , 95% CI , P trend ; Table 2]. This association persisted in subanalyses of an aggregate tumor diameter < 3 cm (HR , 95% CI ), an aggregate tumor diameter 3 cm (HR , 95% CI ), and tumors outside the Milan criteria (HR , 95% CI ), although the HRs were nonsignificant for individual tumor diameter categories. In contrast, a higher MELD score at LT

6 1050 SCHLANSKY ET AL. LIVER TRANSPLANTATION, September 2014 TABLE 2. Fitted Multiple Regression Models of Posttransplant Survival for Patients Registered on the Wait List for LT Between March 1, 2005 and March 1, 2009 HCC Non-HCC Adjusted HR (95% CI)* Adjusted HR (95% CI)* MELD status at LT Overall 0.84 ( ) 1.20 ( ) <22 Referent Referent 1.23 ( ) ( ) 1.38 ( ) > ( ) 1.74 ( ) Laboratory MELD score at LT 1.13 ( ) BMI (kg/m 2 ) 1.02 ( ) 0.99 ( ) Race/ethnicity White Referent Referent Black 1.25 ( ) 1.28 ( ) Hispanic 0.87 ( ) 0.86 ( ) Asian 0.64 ( ) 0.69 ( ) Other 1.06 ( ) 0.94 ( ) Listing diagnosis HCV 1.35 ( ) PBC or PSC 0.77 ( ) Donor characteristics Donor risk index 1.34 ( ) 1.80 ( ) Donor sex: male 1.12 ( ) Donor BMI (kg/m 2 ) 1.01 ( ) 1.01 ( ) HCC-specific characteristics Within Milan criteria 0.94 ( ) Liver-directed therapy 1.21 ( ) Peak AFP Overall 1.31 ( ) <10 ng/ml Referent ng/ml 1.30 ( ) ng/ml 1.78 ( ) >1000 ng/ml 2.14 ( ) *Age-stratified regression models (<55 or 55 years). Sex, listing diagnosis, donor sex, peak aggregate tumor diameter, and peak tumor number did not significantly contribute to the regression model. Sex did not significantly contribute to the regression model. The laboratory MELD score was excluded from the non-hcc model because of collinearity with the MELD status at LT. was independently associated with shorter post-lt survival in the non-hcc group, and this also occurred in a dose-dependent fashion (HR , 95% CI , P for the interaction between HCC and non-hcc groups ; Fig. 2). The same significant associations were observed in regression models that included all potential confounders (Supporting Table 1). As expected, the wait-list time from registration was not associated with post-lt survival; this was a result of the broad variation in the wait-list times from the variable threshold at which centers registered patients on the LT wait list. No significant interactions were identified between the predictor variables and the association of the MELD status at LT with post-lt survival in the HCC group specific or non-hcc group specific multiple regression models. In the overall post-lt cohort, the HCV status significantly interacted with the association between group and post-lt survival (P ), and this reflected the association between the HCV status and post-lt survival in the non-hcc group but not in the HCC group. Comparison of Post-LT and ITT Survival Survival data for both cohorts are summarized in Table 3. In the post-lt cohort, the HCC group had greater survival in the first post-lt year; however, the survival of the non-hcc group then eclipsed that of the HCC group, with the non-hcc group having superior post-lt survival overall (log-rank P < 0.01; Fig. 3). In the ITT cohort, the HCC group had greater survival throughout the study period (log-rank P < 0.01), with the non-hcc group having deaths occur more frequently soon after wait-list registration. To estimate the mortality benefit obtained from LT, we compared the mortality rate after LT to the ITT mortality rate after wait-list registration. For the non-hcc group,

7 LIVER TRANSPLANTATION, Vol. 20, No. 9, 2014 SCHLANSKY ET AL Figure 2. A higher MELD score at LT was independently associated with a lower HR for mortality for MELD-prioritized HCC patients in a dose-dependent fashion. An inverse dose-response relationship was observed for non-hcc patients (P for interaction ). TABLE 3. Posttransplant and ITT Survival of Patients Registered on the Wait List for LT Between March 1, 2005 and March 1, 2009 Posttransplant Cohort (n 5 10,653) ITT Cohort (n 5 30,992) HCC (n ) Non-HCC (n ) HCC (n ) Non-HCC (n 5 24,541) Deaths (n) Person-years (31000) Mortality rate per ( ) 4.6 ( ) 8.4 ( ) 10.6 ( ) person-years RR Referent 0.85 ( ) Referent 1.25 ( ) Survival 1 year 0.92 ( ) 0.91 ( ) 0.88 ( ) 0.80 ( ) 3 years 0.81 ( ) 0.84 ( ) 0.72 ( ) 0.67 ( ) 5 years 0.73 ( ) 0.77 ( ) 0.62 ( ) 0.57 ( ) Log-rank P value <0.01 <0.01 NOTE: Data in parentheses are 95% CIs. mortality was reduced 57% with LT (RR , P < 0.01), whereas mortality was reduced only 36% with LT in the HCC group (RR , P < 0.01); this suggested a larger mortality risk reduction from LT for non-hcc patients versus HCC patients. In the ITT analysis, the time-dependent receipt of LT conferred a 61% reduction in the hazard of death with respect to the wait-list period (Supporting Table 2). As expected, wait-list dropout strongly predicted death (HR for HCC group , HR for non-hcc group ). The direction and magnitude of the associations for other predictors were similar to those observed in the univariate post-lt analysis. Post Hoc Analyses Although we used the MELD status at LT as a marker of the waiting time in our main analysis, we also evaluated the association of the time on the wait list from the first successful application for HCC MELD prioritization to LT with post-lt survival (we excluded the lead time from wait-list registration to MELD prioritization). According to the multiple regression, a longer time on the wait list after initial MELD prioritization independently predicted longer post-lt survival in both a fitted model and a model including all potential confounders (HR , 95% CI ; Supporting Table 3). However, this method precluded a comparison of the HCC and non- HCC groups. We could not evaluate the non-hcc group because the use of the time on the wait list from the calculated MELD score first reaching 22 to LT would have introduced a selection bias on account of the exclusion of a large number of non- HCC patients who had undergone LT with MELD scores less than 22.

8 1052 SCHLANSKY ET AL. LIVER TRANSPLANTATION, September 2014 Figure 3. Comparison of posttransplant and ITT survival for MELD-prioritized HCC patients and non-hcc patients registered on the LT wait list between March 1, 2005 and March 1, The mortality rate for HCC patients was reduced 36% after LT in comparison with the ITT period from wait-list registration, whereas non-hcc patients had a larger, 57% reduction in the mortality rate with LT. We also evaluated transplant center specific variations in our post-lt analysis of the HCC cohort. Given the known disparity in 1-year outcomes after LT at different centers, 33 we theorized that centers with longer waiting times before LT might also represent centers offering higher quality HCC care; our observed association between an increased MELD score at LT and lower post-lt mortality could, therefore, reflect high-quality care rather than an HCC-specific phenomenon. To estimate the overall quality of transplant care at LT centers, we categorized centers into high, medium, and low tertiles of 1-year post-lt survival for non-hcc patients. Stratifying post-lt survival in the HCC group by this center-quality variable, we observed a nonsignificant pattern of longer survival for LT recipients from centers in the high-quality category (log-rank P ; Supporting Fig. 1). Center quality was a nonsignificant predictor of post-lt mortality according to the multiple regression (HR for medium-quality centers versus low-quality centers , 95% CI ; HR for highquality centers versus low-quality centers , 95% CI , P trend ). An increased MELD score at LT remained independently predictive of lower post-lt mortality after the incorporation of the center-quality variable into the multiple regression (HR , 95% CI , P trend ), and this suggested that the variation in the center-specific quality of care did not explain this association. We evaluated HCC recurrence as a secondary outcome in the post-lt analysis. Data on the recurrence of pretransplant malignancies are supplied to UNOS by individual centers; unfortunately, records are not linked to cancer registries, and prior studies have noted an extremely low HCC recurrence rate in the STAR registry versus the rate observed in singlecenter studies. 16,33 Because of this unreliability, we considered an analysis of HCC recurrence using STAR to be exploratory. We identified 241 HCC recurrences in the HCC group, and this yielded a rate of 1.9 recurrences per 1000 person-years (95% CI ). The probabilities of 1-, 3-, and 5-year HCC recurrence-free survival were 97%, 92%, and 90%, respectively. According to the multiple regression, an increasing MELD status was a nonsignificant predictor of HCC recurrence (HR , 95% CI ; Supporting Table 4). Finally, to test the validity of our decision to censor wait-list dropouts in the ITT analysis at the time of dropout, we performed a sensitivity analysis assuming instead that wait-list dropouts lived until the end of the data collection (October 19, 2012). This resulted in the censoring date being changed for 4523 wait-list dropouts (14.6% of the ITT cohort) who had no record of death after dropout. Although the mortality rates in each group were slightly reduced because of the increase in person time with an unchanged number of deaths, the relative differences in ITT survival between the groups were similar and remained significant (Supporting Table 5), and this indicated that the ITT analysis was robust to the method of censoring wait-list dropouts. DISCUSSION In this observational cohort study of a national transplant registry, we found that a longer waiting time predicted longer post-lt survival for MELD-prioritized

9 LIVER TRANSPLANTATION, Vol. 20, No. 9, 2014 SCHLANSKY ET AL HCC patients. In contrast, a longer waiting time predicted shorter post-lt survival for non-hcc patients. We infer that a longer waiting time for HCC patients selects LT for candidates with a favorable tumor biology and a consequently lower risk of post-lt mortality, whereas the severity of liver dysfunction, rather than the waiting time, drives post-lt mortality for non-hcc patients. Furthermore, MELD-prioritized HCC patients demonstrated longer ITT survival and shorter post-lt survival than non-hcc patients; this discrepancy reflects HCC patients enhanced access to LT despite their inferior post-lt outcomes (a 33 percentage point advantage in the probability of undergoing LT). These novel findings highlight the need for changes to an inequitable organ allocation policy. The association between the waiting time and post- LT survival in MELD-prioritized HCC patients could stem from differences in the clinical behavior of their cancers or from differences in patient characteristics that predict survival and are unrelated to their cancers. The former scenario describes the tendency of an increasing waiting time to select for indolent HCC that lingers within transplantable criteria for long durations and confers a low risk of post-lt HCC recurrence and mortality. The latter scenario, a survivorship bias, describes the tendency of an increasing waiting time to select for an overall healthier pool of patients with longer posttransplant survival due to their superior health status before LT. Regardless of which phenomenon dominates, there is a tradeoff between the longer post-lt survival gained from the selection of a favorable sample for LT and the greater proportions of wait-list dropout and death for patients not undergoing LT. Several features of our analysis argue against a survivorship bias as the explanation for the observed association. The association persisted after adjustments for covariates associated with favorable survival on the wait list, such as the body mass index (BMI), race/ethnicity, center quality, and liver disease severity. The association was also internally consistent in strata of these covariates (strata associated with a better health status did not reveal stronger associations with post-lt survival). Finally, the non- HCC group did not demonstrate an association between an increasing waiting time and greater post- LT survival, yet this group should have been vulnerable to the same survivor bias proposed for the HCC group (because this phenomenon is independent of cancer characteristics). We conclude that the association is a manifestation of differences in cancer behavior rather than the health status of LT candidates. Misclassification bias due to diagnostic error is a second alternative hypothesis for the association of the waiting time and survival in the HCC group. Freeman et al. 34 reported the pathological absence of HCC in more than 20% of liver explants from patients undergoing transplantation for HCC from 2003 to 2006 who had not received liver-directed HCC therapy. At that time, MELD prioritization was available for T1 HCC and an AFP level > 500 ng/ml without HCC on imaging studies, and misdiagnosis has likely decreased since their analysis. Nevertheless, patients misdiagnosed with HCC would be expected to have longer waiting times because of a lower probability of wait-list dropout from tumor progression beyond transplantable criteria and longer post-lt survival because of their lack of HCC recurrence risk and good liver function at LT. We did not evaluate explant pathology in our analysis because these data were incompletely reported and the majority of our HCC patients had received liver-directed therapy (no standardized method for the histological evaluation of treated HCC is available). However, we found that the association of a longer waiting time with longer post- LT survival persisted in the subgroup of patients with larger tumors, who were more apt to have accurate HCC diagnoses, including HCC outside the Milan criteria. Because of this finding and trends of increasingly accurate imaging and additional UNOS oversight since the study by Freeman et al., poor diagnosis is unlikely to explain the association of a longer waiting time with longer post-lt survival for HCC. The association fulfills epidemiological criteria for validity and causality. It is independent of confounders, statistically significant, and consistent across strata of covariates and exhibits a gradient of effect, with increasing categories of MELD status more strongly associated with post-lt survival. The post hoc analysis using the time from first HCC MELD prioritization to LT also identified a significant association between a longer waiting time and longer post-lt survival; although the effect size was smaller than that observed in the main analysis, the inability of the STAR registry to adequately track the time from when Milan criteria is met to LT is a recognized deficiency in the data source (and justifies our use of the MELD status at LT as the primary marker of the waiting time). 17 The association is coherent with the findings of small studies of liver-directed HCC therapy with or without tumor down-staging that incorporated a waiting period before LT, 13-15,35 national studies that compared cadaveric LT and LDLT and found higher HCC recurrence rates for LDLT recipients with short waiting times, 36,37 and an analysis of a recent HCC cohort from the UNOS registry that found an independent association between a tumor status outside the Milan criteria and lower post-lt mortality. 38 Finally, the purported causal pathway is plausible; occult metastatic disease becomes increasingly evident with increasing time and leads to wait-list dropout and the selection of a residual cohort with less aggressive cancers for LT. 16 This relationship is analogous to the overdiagnosis observed in screening studies for nonliver cancer; for example, frequent mammography was shown to detect a greater proportion of indolent breast cancers that would not have resulted in major morbidity or mortality and thus undermined the benefit of screening. 39 Accordingly, patients with HCC within transplantable criteria who remain on the LT wait list for long periods are less likely to have aggressive cancers and have a

10 1054 SCHLANSKY ET AL. LIVER TRANSPLANTATION, September 2014 consequently reduced risk of cancer recurrence and death after LT. Pathophysiological differences underlie the contrasting behaviors of the HCC and non-hcc groups. In the non-hcc group, an increased MELD score at LT predicted poor post-lt survival. Although the MELD score was inversely correlated with the waiting time, the MELD score was associated with post-lt survival, but the waiting time was not. Hence, non-hcc patients derived a post-lt survival benefit from having better liver function at LT, a factor independent of the waiting time. In contrast, the HCC group had uniformly good liver function at LT and derived a post-lt survival benefit from increasing time on the wait list because of the selection of patients with indolent HCC for LT. HCV infection predicted only shorter post-lt survival for non-hcc patients. Prior studies noted the same lack of an association between HCV and post-lt outcomes in patients undergoing transplantation for HCC We theorize that HCV reinfection after LT primarily imparts a post-lt mortality risk by increasing the risk of recurrent cirrhosis rather than recurrent HCC. Because recurrent cirrhosis and recurrent HCC are competing risks in patients undergoing transplantation with both HCV and HCC (HCC recurrence does not require HCV-related graft fibrosis), it follows that HCV would be a weaker predictor of post- LT survival in patients with both HCC and HCV versus HCV alone. An inadequate ascertainment of post- LT HCC recurrence in the STAR registry precluded our testing of this hypothesis. In the ITT analysis, the HR for death was 61% lower for LT recipients versus comparable candidates. The magnitude of this reduction was lower than that reported by Merion et al. 43 for a non-hcc cohort (79%) or that reported by Pelletier et al. 21 for a cohort with HCC (77%). These discrepancies likely reflect methodological differences. The period of the former study was only 2 years from wait-list registration to post-lt death or censoring and was potentially inadequate for capturing subjects who had survived for long periods on the LT wait list and thus derived less mortality risk reduction from LT with respect to their lengthy wait-list period. The latter study included HCC patients registered on the wait list from 1998 to 2006 and thus represented a heterogeneous group of patients registered before MELD-based allocation was enacted and patients registered with historical MELD prioritization protocols for HCC (with up to 29 points awarded for T2 cancers); this limited the comparability to the HCC cohort in our study. Our results support imposing a delay on LT for HCC, and this proposal is currently under consideration by UNOS. 44,45 We found that MELD-prioritized HCC patients had much lower rates of wait-list dropout and higher rates of LT than non-hcc patients. Furthermore, MELD-prioritized HCC patients had markedly better ITT survival from wait-list registration. We agree with recent arguments that equalizing ITT survival across different indications for LT, rather than by the probability of wait-list dropout or death, may better optimize LT utility for the overall candidate pool. 23,24 Because MELD-prioritized HCC patients had shorter post-lt survival than non-hcc patients and post-lt survival was improved with a prolonged waiting time, it follows that a strategy of delaying LT for MELD-prioritized HCC patients would increase their post-lt survival while simultaneously increasing the probability of LT for non-hcc patients, who are more likely to experience wait-list dropout and death despite their greater post-lt survival. Thus, we anticipate that a strategy of delaying LT for MELDprioritized HCC would equalize ITT survival in the overall pool of LT candidates. This strategy would also better align LT allocation with ethical practices by maximizing system utility (ITT survival would improve) while simultaneously equalizing individual access to a scarce resource (the probabilities of undergoing LT for HCC and non-hcc indications would be more similar). Although STAR provided a large sample size with national representation, prospective data collection, and excellent patient ascertainment with few missing data, we recognize several limitations of this study. First, the ascertainment of mortality was compromised by the lack of a DMF linkage to STAR for the final 11 months of the study period, but this represented a minority of the follow-up time (14%), and the inclusion of deaths reported to STAR by individual centers allowed some degree of mortality ascertainment in the period for which DMF data were unavailable. When DMF data were available, 90% to 96% of deaths in the National Death Index were likely ascertained. 46 Second, the individual HRs within each MELD status category were not statistically significant, and this likely reflected insufficient power to individually evaluate these subgroups due to an inadequate number of events within each MELD category; however, the trend of lower post-lt mortality with a longer waiting time in the HCC group was statistically significant. Third, despite our attempt to demonstrate that the variation in post-lt mortality for HCC patients by the waiting time was a manifestation of differences in cancer biology rather than a survivorship bias, the poor ascertainment of HCC recurrence in the UNOS registry precluded a definitive validation of the idea that post-lt mortality in the HCC group was due to cancer recurrence rather than other factors associated with the waiting time. Our observed 1- year post-lt HCC recurrence rates of 3% to 5% are well below the incidences reported in single-center studies, and we were unable to link STAR with cancer registries to elucidate whether recurrent HCC was the dominant explanation for post-lt mortality. With the recent UNOS requirement for the standardized collection of explant pathology, it may soon be possible to correlate the histological characteristics of hepatectomy specimens with both the waiting time and the post-lt outcomes with contemporary data; we recognize that explant pathology will likely remain an incomplete predictor of post-lt outcomes because

11 LIVER TRANSPLANTATION, Vol. 20, No. 9, 2014 SCHLANSKY ET AL HCC may recur in the absence of adverse histological findings, possibly because of unrecognized extrahepatic metastasis. In conclusion, we report an independent, statistically significant trend for increasing MELD scores at LT in MELD-prioritized HCC patients and longer post- LT survival in a national transplant database, and we present evidence that this association is causally mediated by the selection of patients with indolent cancers for LT with increasing waiting time. Patients with MELD prioritization for HCC benefit from enhanced access to LT that results in less wait-list dropout and superior ITT survival in comparison with patients with non-hcc indications, and this overwhelms their lower survival after LT. The burden of HCC is rising, but donor organs remain scarce. Delaying LT for MELD-prioritized HCC patients may improve utility and fairness for the overall pool of LT candidates. ACKNOWLEDGMENT The authors thank Atif Zaman, M.D., M.P.H., and Amnon Sonnenberg, M.D., M.S., for their assistance with a critical review of this article. REFERENCES 1. Ferlay J, Shin HR, Bray F, Forman D, Mathers C, Parkin DM. Estimates of worldwide burden of cancer in 2008: GLOBOCAN Int J Cancer 2010;127: Altekruse SF, McGlynn KA, Reichman ME. Hepatocellular carcinoma incidence, mortality, and survival trends in the United States from 1975 to J Clin Oncol 2009; 27: Mazzaferro V, Chun YS, Poon RT, Schwartz ME, Yao FY, Marsh JW, et al. Liver transplantation for hepatocellular carcinoma. Ann Surg Oncol 2008;15: Mazzaferro V, Regalia E, Doci R, Andreola S, Pulvirenti A, Bozzetti F, et al. Liver transplantation for the treatment of small hepatocellular carcinomas in patients with cirrhosis. N Engl J Med 1996;334: Hayashi PH, Trotter JF, Forman L, Kugelmas M, Steinberg T, Russ P, et al. Impact of pretransplant diagnosis of hepatocellular carcinoma on cadaveric liver allocation in the era of MELD. Liver Transpl 2004;10: Bruix J, Fuster J, Llovet JM. Liver transplantation for hepatocellular carcinoma: Foucault pendulum versus evidence-based decision. Liver Transpl 2003;9: Clavien PA, Lesurtel M, Bossuyt PM, Gores GJ, Langer B, Perrier A; for OLT for HCC Consensus Group. Recommendations for liver transplantation for hepatocellular carcinoma: an international consensus conference report. Lancet Oncol 2012;13:e11-e Zimmerman MA, Ghobrial RM, Tong MJ, Hiatt JR, Cameron AM, Hong J, Busuttil RW. Recurrence of hepatocellular carcinoma following liver transplantation: a review of preoperative and postoperative prognostic indicators. Arch Surg 2008;143: Sharma P, Welch K, Hussain H, Pelletier SJ, Fontana RJ, Marrero J, Merion RM. Incidence and risk factors of hepatocellular carcinoma recurrence after liver transplantation in the MELD era. Dig Dis Sci 2012;57: Biggins SW. Futility and rationing in liver retransplantation: when and how can we say no? J Hepatol 2012;56: Schwartz M, Roayaie S, Llovet J. How should patients with hepatocellular carcinoma recurrence after liver transplantation be treated? J Hepatol 2005;43: Ioannou GN, Perkins JD, Carithers RL Jr. Liver transplantation for hepatocellular carcinoma: impact of the MELD allocation system and predictors of survival. Gastroenterology 2008;134: Otto G, Herber S, Heise M, Lohse AW, M onch C, Bittinger F, et al. Response to transarterial chemoembolization as a biological selection criterion for liver transplantation in hepatocellular carcinoma. Liver Transpl 2006;12: Yao FY, Hirose R, LaBerge JM, Davern TJ III, Bass NM, Kerlan RK Jr, et al. A prospective study on downstaging of hepatocellular carcinoma prior to liver transplantation. Liver Transpl 2005;11: Yao FY, Kerlan RK Jr, Hirose R, Davern TJ III, Bass NM, Feng S, et al. Excellent outcome following down-staging of hepatocellular carcinoma prior to liver transplantation: an intention-to-treat analysis. Hepatology 2008;48: Roberts JP, Venook A, Kerlan R, Yao F. Hepatocellular carcinoma: ablate and wait versus rapid transplantation. Liver Transpl 2010;16: Pomfret EA, Washburn K, Wald C, Nalesnik MA, Douglas D, Russo M, et al. Report of a national conference on liver allocation in patients with hepatocellular carcinoma in the United States. Liver Transpl 2010;16: Washburn K, Edwards E, Harper A, Freeman R. Hepatocellular carcinoma patients are advantaged in the current liver transplant allocation system. Am J Transplant 2010;10: Wiesner RH, Freeman RB, Mulligan DC. Liver transplantation for hepatocellular cancer: the impact of the MELD allocation policy. Gastroenterology 2004;127(suppl 1): S261-S Yao FY, Bass NM, Nikolai B, Davern TJ, Kerlan R, Wu V, et al. Liver transplantation for hepatocellular carcinoma: analysis of survival according to the intention-to-treat principle and dropout from the waiting list. Liver Transpl 2002;8: Pelletier SJ, Fu S, Thyagarajan V, Romero-Marrero C, Batheja MJ, Punch JD, et al. An intention-to-treat analysis of liver transplantation for hepatocellular carcinoma using Organ Procurement Transplant Network data. Liver Transpl 2009;15: Toso C, Dupuis-Lozeron E, Majno P, Berney T, Kneteman NM, Perneger T, et al. A model for dropout assessment of candidates with or without hepatocellular carcinoma on a common liver transplant waiting list. Hepatology 2012; 56: Keller EJ, Kwo PY, Helft PR. Ethical considerations surrounding survival benefit-based liver allocation. Liver Transpl 2014;20: Neuberger J. Liver allocation for patients with hepatocellular carcinoma. Liver Transpl 2010;16: Persad G, Wertheimer A, Emanuel EJ. Principles for allocation of scarce medical interventions. Lancet 2009;373: Organ Procurement and Transplantation Network. Policy 3.6: allocation of livers. PoliciesandBylaws2/policies/pdfs/policy_8.pdf. Accessed July Organ Procurement and Transplantation Network. Citing data. Accessed July Social Security Death Master File. FolderID/54/SessionID/%7BA32A139E-72F B508- B3E00C201793%7D/PageVars/Library/InfoManage/Guide. htm. Accessed July 2013.