The rising prevalence of hepatitis C virus (HCV) and nonalcoholic fatty liver disease has led to a marked increase in the incidence of cirrhosis and hepatocellular carcinoma (HCC).1 For patients with early-stage HCC, there are 3 curative options—ablation, hepatic resection, and liver transplant (LT).2-4 Since the inception of model for end-stage liver disease (MELD)-based allocation in 2002, waitlisted patients with HCC within Milan criteria have been eligible for automatic MELD exception points to facilitate LT and decrease the risk of waitlist removal due to tumor progression.5-8 At the level of the individual patient, advocating for transplantation has been considered a strategy that maximizes survival. However, recent data suggest that due to donor organ scarcity, prioritization of HCC patients over those with high laboratory MELD scores has unintentionally resulted in a large reduction in LT-related survival benefit,9 given that nearly 80% of the US patients transplanted for HCC have minimal evidence of liver synthetic dysfunction and/or portal hypertension and thus may be candidates for another curative treatment.5-9 However, these data relied solely on a cohort of waitlisted patients, for whom long-term estimates of survival without LT may have been overestimated by the inclusion of long-term waitlist survivors.10 There are limited data on the incremental survival benefit of LT versus other curative options (resection or ablation) for HCC patients with preserved liver function, especially in the context of utilizing donor livers for patients with HCC over patients with decompensated cirrhosis without other curative options.
In order to definitively determine the survival benefit of LT versus resection or ablation, one would need to perform a randomized controlled trial of patients eligible for all 3 treatment modalities. This, however, is not feasible. The Veterans Health Administration (VHA) is unique in that it is a natural environment enriched with patients receiving nontransplant treatment due to various limitations in access to LT.11 Thus, the objective of this study was to leverage VHA data to compare the overall survival from HCC diagnosis for those receiving LT, resection, or ablation to estimate the incremental survival benefit of LT.
MATERIALS AND METHODS
Study Design and Data Sources
We performed a retrospective cohort study of VHA patients, collected as part of the Veterans Outcomes and Costs Associated with Liver Disease (VOCAL) study group dataset. The VHA is the largest single provider of liver care in the United States, and due to limitations in access to LT among this population, the dataset is enriched with patients who received nontransplant HCC therapies.11 The creation of the VOCAL dataset has previously been described, but in brief it is a well-characterized cohort of patients with incident cirrhosis identified between 2008 and 2016.12,13 Transplantation data were obtained from the United Network for Organ Sharing (UNOS) Standard Transplant Analysis and Research data file,12 and death data were obtained using the Medicare Vital Status File.14
The VOCAL database provided demographic (age, sex, race), clinical (body mass index), comorbidity (hypertension, diabetes, chronic kidney disease, coronary artery disease, congestive heart failure, cerebrovascular accident, atrial fibrillation, pulmonary embolism, decompensated cirrhosis), and laboratory data (sodium, creatinine, international normalized ratio, liver-associated enzymes, platelet count, alpha-fetoprotein [AFP]). Comorbidities including prior cirrhosis decompensation were obtained using international classification of disease (ICD)-9 coding algorithms detailed previously.12 Where relevant, all data were screened to identify values closest to the date of HCC diagnosis. AFP was modeled as both a continuous and categorical variable (<50, 50–99, 100–499, and ≥500 ng/mL), adapted from literature cut points.15-17 MELD scores were calculated from baseline laboratory parameters, and baseline alcohol use disorders identification test scores were obtained to identify patients with hazardous drinking patterns (positive score ≥4 for males and ≥3 for females).18-20 ICD-9 codes, current procedure terminology codes, and pharmacy data were also obtained for subsequent exposure ascertainment. Etiology of liver disease was established using a previously validated algorithm21 with the following final classifications: HCV, hepatitis B, alcoholic liver disease, nonalcoholic fatty liver disease, HCV + alcoholic liver disease, and other.
Ascertainment of Hepatocellular Carcinoma and Patient Selection
Patients with HCC were identified using a validated algorithm based on the presence of 1 inpatient or 2 outpatient ICD-9, Clinical Modification codes for malignant neoplasm of the liver (155.0 or 155.2) as a primary or secondary diagnosis.22 We included patients aged 50–69 with cirrhosis and HCC, given that UNOS data demonstrate that only 3.5% of HCC LT recipients from the VHA during the study period were aged ≥70 years. Importantly, we focused on patients with a calculated MELD score <15 at the time of HCC diagnosis, because data have demonstrated that this is the cutoff for survival benefit, and many consider a calculated MELD score ≥15 as an indication for LT independent of other factors.23 We excluded patients with missing baseline bilirubin, creatinine, and/or international normalized ratio values as a calculated MELD score was required for inclusion. Because HCC staging was not available for the entire cohort, we restricted our analyses to patients who received LT, resection, or ablation, as these are the only 3 treatment options with a curative intent for HCC.
Ascertainment of Nontransplant Hepatocellular Carcinoma Therapies (Exposures)
We identified nontransplant therapies through a combination of relational database queries and manual expression searches using STATA 15.1/IC (College Station, TX), adapted from prior literature.22 Resection/partial hepatectomy events were identified using structured query language searches for relevant current procedure terminology codes, excluding orthotopic liver transplantation and liver biopsy (Table S1, SDC, http://links.lww.com/TP/B757). Ablation therapy included microwave ablation, radiofrequency ablation, cryoablation, and ethanol ablation. Treatment group (LT, resection, or ablation) was initially assigned based on a ranking of highest survival rate in the following order: transplant > resection > ablation. That is, if a patient received resection followed by ablation, their treatment group was classified as “resection” throughout the duration of follow-up.
Treatment group was subsequently coded as a time updating variable, where patients could be reclassified if they received multiple therapies (eg, ablation followed by LT).
The primary outcome was time-to-event survival, measured from the date of HCC diagnosis. The time horizons we focused on were 1–5 years from diagnosis. We chose to begin follow-up from the time of diagnosis, as literature demonstrates that beginning follow-up at the time of treatment leads to biased estimates of risk in favor of treatment exposures with longer time intervals from diagnosis (immortal time).24,25 Rather, we elected to measure survival time from HCC diagnosis and model time-updating treatment status through Cox regression, as detailed below. This approach also reflects an “intention to treat” analysis whereby the decision to pursue a curative treatment path was made at the time of HCC diagnosis, accounting for multiple treatments and crossovers as described.
Descriptive statistics were computed based on the highest level of treatment received. Pearson chi-squared and Wilcoxon rank-sum tests were performed for categorical and continuous variables, respectively. We used univariate Cox regression analysis to select predictors for testing in multivariable regression. We considered all variables shown in Table 1 and used an alpha = 0.15 threshold. Subsequently, we performed multivariable Cox regression analysis using elements of reverse stepwise and clinician-driven modeling approach, with an alpha = 0.05 threshold used for variable retention. We used minimized Bayesian Information Criterion to select final models. Because of potential immortal time bias induced by establishing time of HCC diagnosis as the start of follow-up, we evaluated several models. The full model included age, aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase, albumin, platelet count, AFP category, and decompensation status. The immortal model adjusted for the same covariates, but also incorporated time-updated treatment exposure variables to mitigate potential immortal time bias, consistent with literature recommendations.24-26 This allowed for accurate risk adjustment for unexposed time periods. It also properly accounted for patients who received multiple treatments, including possible neoadjuvant or “bridging” therapies. For example, a patient might have an unexposed period which was followed by ablation and later LT. Importantly, owing to a violation in the proportional hazards assumption caused by AFP level, we treated this as a continuously time-varying exposure. We produced adjusted coefficient plots and survival curves for each model, stratified by treatment group, and also reported predicted probabilities of survival. In order to determine relative survival benefit of transplant, we integrated the adjusted survival curves to calculate expected survival over time as a function of treatment group and subsequently took the difference in expected survival between LT and resection or ablation, respectively, similar to previously described methods.27,28 This was done for each model, with the results plotted and point estimates reported at 1–5 years.
Hepatocellular Carcinoma Staging Subgroup Analysis
A subgroup of ~20% of the analytic cohort was randomly chosen for manual chart review in order to ascertain HCC staging data at the time of HCC diagnosis, as previously described.22 This involved review of clinical documentation and radiology reports. Adherence to the Milan criteria was defined as 1 lesion ≤5 cm or no >3 lesions all ≤3 cm.29 We obtained tumor number, largest tumor diameter (cm), and total tumor diameter (cm), and patients were also classified according to the Barcelona Clinic Liver Cancer (BCLC) staging system (stage 0, A, B, C, or D).4 We used this staging to confirm that this cohort primarily consisted of BCLC stage patients for whom hepatectomy, ablation, or transplant would be potentially reasonable pathways. Recent data suggest that BCLC-B (intermediate stage) candidates are a heterogeneous group with potentially comparable outcomes with ablation or resection as compared to chemoembolization.30-34 As such, we performed a subgroup analysis restricted to patients with BCLC stage 0, A, or B at HCC diagnosis (early/intermediate stage). This staging model followed the same variable selection process as described previously and adjusted for albumin and platelet count. Of note, tumor characteristics were tested but not found to be significant predictors in this model.
After selection criteria, 2129 veterans with cirrhosis and HCC were included in the analytic cohort (Figure 1), with median follow-up time 2.90 years (interquartile range [IQR] 1.82–4.59 y). The patient population was primarily male, with a racial distribution reflective of the overall population (Table 1). Of the 3 curative options, ablation was most common as the highest level of therapy (61.9%), followed by LT (26.7%) and hepatic resection (11.5%). From HCC diagnosis, the median time to therapy was longest for LT (1.25 y, IQR 0.70–1.98 y), followed by ablation (median 0.25 y, IQR 0.11–0.75 y) and resection (median 0.17 y, IQR 0.10–0.41). LT was most common among patients aged 50–64, with less invasive treatments (ablation and resection) more common in the 65–69 age group. LT patients were less likely to have isolated alcohol-induced liver disease and/or active alcohol use (measured by alcohol use disorders identification test), while they were more likely to have a lower platelet count and decompensated cirrhosis. Patients who received ablation or resection more cardiovascular comorbidities (ie, coronary artery disease, congestive heart failure, hypertension, and atrial fibrillation) relative to those who underwent LT.
Primary Cox Regression Analysis
In the full model, resection and ablation had significantly higher hazards of death relative to LT (hazard ratio [HR] 5.42, 95% confidence interval [CI] 4.15-7.08; and HR 5.50, 95% CI 4.51-6.71, respectively; Table 2 and Figure 2A). When adjusting for time-varying treatment exposures (immortal model), the hazards for each increased slightly (resection HR 6.18, 95% CI 4.80-7.97; ablation HR 5.79, 95% CI 5.27-6.37); however, overall survival was improved in all groups (Figure 2B). In both models, higher age, baseline MELD, alkaline phosphatase, aspartate aminotransferase, AFP category, and decompensated cirrhosis status were positively associated with poorer survival.
Staging Subgroup Analysis
Of the 2129 patients in the analytic cohort, a total of 432 (20.3%) were manually reviewed for staging data. Of these patients, 78.5% were within Milan criteria at the time of HCC diagnosis. A total of 16.4% were BCLC stage 0, 62.0% stage A, 17.4% stage B, 2.3% stage C, and 1.9% stage D. When isolating the cohort to only those with early/intermediate stage HCC (as defined by BCLC stage 0, A, or B), the increased hazard of death relative to LT with resection or ablation was somewhat attenuated (resection HR 4.54, 95% CI 2.80-7.36; ablation HR 4.88, 95% CI 3.44-6.92; staging model in Table 2). Tumor characteristics of this subcohort are shown in Table 3, where patients who received resection had a significantly larger maximum tumor diameter and total tumor diameter versus LT or ablation patients (median 3.45 cm vs 2.5–2.6 cm, and 3.45 cm vs 2.7 cm, respectively; P < 0.001 and P = 0.031). However, neither of these factors were retained in the multivariable staging model. Survival distributions by treatment (Figure 2C) were similar to those estimated in the full model. HRs for resection and ablation relative to LT, for each model, are summarized in Figure 3.
Survival Benefit of Transplant
Predicted probabilities of survival at 1–5 years, stratified by treatment modality, are reported in Table 4. By contrast, the expected survival over time and survival benefit relative to LT for each treatment group are shown in Figure 4, with plots produced for each model. Although LT yielded the highest expected survival in all models, the degree of survival benefit was small, especially at short follow-up intervals. For example, in all models, the survival benefit of LT over resection or ablation at 1 year was only 0.02–0.03 years. At 5 years, the survival benefit was between 1.04 and 1.24 years (Table 5).
Unfortunately, due to a scarcity of donor organs, only 1 in 6 patients in need of an LT is waitlisted each year, and up to 20% die on the waitlist each year.5 Because of the persistent mismatch between organ supply and demand, transplant physicians must balance the responsibility of advocating for individual patients with the overarching need to equitably and efficiently allocate scarce donor organs to those in greatest need who derive the greatest benefit from transplantation. In this analysis of patients with cirrhosis and HCC with low MELD scores, we demonstrated that LT is associated with significantly increased survival compared to resection and ablation.11,12 However, the incremental survival benefit in absolute terms is small at 5 and 3 years and negligible at 1 year. Through multiple strategies of adjusted modeling, at 5 years from HCC diagnosis, the survival benefit of LT was uniformly <1.25 years over resection or ablation. Depending on one’s perspective (individual patient vs population health), this survival benefit of LT may or may not be considered sufficient to justify the use of a deceased donor liver that could otherwise be used for a patient with decompensated cirrhosis who lacks other curative options.
The concept of survival benefit of transplant was first introduced in the context of MELD score thresholds.23 More recently, due to the increasing number of patients transplanted for HCC, the issue of transplant-related survival benefit for HCC has come into question.9 In 2015, Berry et al evaluated waitlisted patients from 2002 to 2013 to estimate the survival benefit of LT for HCC and non-HCC patients by calculating the difference of estimated 5-year post-LT and pre-LT survival.9 Based on this, the authors suggested there was a net negative survival benefit for HCC LT recipients with a calculated MELD score <14, which included 60% of all LT recipients with HCC during the study period. For patients with a MELD score of 14–21, the estimated 5-year survival benefit of LT was <1 year.9 Thus, the authors suggested that for the 90% of LT recipients with HCC with a MELD score ≤21, the survival benefit of LT was negligible and potentially associated with a net negative survival benefit.10,23 Critics argued that this underestimated the survival benefit of LT because waitlist data are insufficient to estimate long-term survival without LT, as the natural history of a patient’s clinical course cannot be evaluated when >80% of the patients with HCC are transplanted within 2 years of being waitlisted. It is for this reason that we sought to compare LT survival with resection and ablation in the VHA population, in which LT is less frequently achieved.
In the United States, HCC patients are more likely to be waitlisted than patients with decompensated cirrhosis, despite data suggesting a smaller relative survival benefit. Consideration of transplant for these patients should be viewed relative to the expected survival for patients undergoing other curative options (resection and ablation) and also relative to non-HCC patients with decompensated cirrhosis without other curative options. The relative survival benefit of transplant 5 years after HCC diagnosis compared to resection and ablation was between 1.04 and 1.24 years but was <0.5 years over a 3-year time horizon and <0.05 years over a 1-year time horizon. This reflects the fact that while all 3 options can be curative for a tumor or tumors at a specific time point, resection and ablation leave a diseased liver in place, with a risk of recurrent and/or de-novo HCC, while LT removes the liver, serving to dramatically reduce the risk of recurrence and mortality. Importantly, our estimates of survival benefit were similar when accounting for time-varying treatment exposure to address immortal time bias and isolating the cohort to early/intermediate stage HCC patients through manual chart review. In fact, when accounting for different times to therapy (immortal model), the survival benefit of LT was even lower (ie, 1.04–1.10 y of survival benefit at 5 y). This intuitively makes sense, as the median time to transplant (1.25 y) was much longer than the median times to ablation (0.25 y) or resection (0.17 y).
A key consideration is whether a survival benefit of 1.04–1.24 years over 5 years clinically justifies transplant over other curative options. We must consider the increased risks and resources required for LT, as well as the long-term health risk of immunosuppression, and quality of life after transplant and other procedures. When cost is added to the equation, data suggest that resection is the most cost-effective approach for early-stage HCC, with LT as the least cost-effective.35 Additionally, the magnitude of survival benefit is dependent on the availability of an organ; if the waiting period is too long, depending on the tumor growth pattern, the survival benefit provided by transplantation may be dampened by the risks that patients face while waiting.36 Importantly, these considerations are for patients without a potential living liver donor, who are competing against patients with decompensated cirrhosis for a potential deceased donor organ. In the setting of living donor liver transplantation, our data would suggest that undergoing a living donor LT is the preferred strategy, as it maximizes the chance for long-term survival without impacting the deceased donor organ pool. However, in the context of deceased donation, which mandates maximal utilization of a scarce resource,37,38 our data suggest that deceased donor LT for a low MELD HCC patient, rather than a patient with decompensated cirrhosis, fails to optimally utilize this scarce resource.
There are several limitations that we acknowledge in this work. First, the VHA system may not represent the overall population in terms of sex or socioeconomic conditions. However, the VHA is the largest single provider of liver care in the United States, and importantly, we would not expect estimates of the survival benefit of LT relative to resection or ablation to differ substantially as a function of sex or socioeconomic status (indeed, these covariates were not retained in any multivariable models). Thus, this should not have biased our results. Second, it has been shown for the VOCAL database that among patients with potentially curable HCC (based on BCLC staging; n = 1419), only 25% received potentially curative therapies such as resection, transplantation, or ablative therapy; even among patients with potentially curable HCC and good performance status (European Cooperative Oncology Group 1–2), 13% of patients did not receive active HCC therapy.22 Thus our analyses included a restricted cohort, which may limit external validity but enhance internal validity by addressing a highly-selected population. Third, HCC staging data were not available for the entire cohort, which would have enabled more precision in isolating a similarly-staged cohort in addition to narrowing CIs for parameter estimates. However, in our subgroup analysis of patients with staging data, the overall study results were not substantively different. This implies that the analytic cohort that we identified was primarily comprised of patients with early/intermediate-stage HCC who could conceivably be candidates for LT, resection, or ablation. Fourth, we did not look beyond a 5-year survival horizon, largely due to the available follow-up data. However, the focus of most survival benefit studies in LT range from 1 to 5 years, so our study is consistent with published literature. That being said, the benefit of LT over a longer term is likely to be even greater. Fifth, we measured survival time from diagnosis rather than from treatment. As noted before, this was an intentional decision to minimize bias, and we accounted for different times to treatment through time-updating modeling techniques. Additionally, our goal was to account not only for the time a patient undergoes treatment but for the transplant recipient, the time they were evaluated and waitlisted, which reflects a potentially more important measure. Sixth, there is likely some degree of residual confounding in our regression models. For example, we could not account for differing tumor locations, where certain modalities might not be a viable option. However, in circumstances where resection is infeasible based on location, ablation is typically an option. Thus, our goal of informing decisions when >1 curative option is available is still reflected by the data presented. Finally, we only included patients that were waitlisted and transplanted due to the nature of the data merge and could not ascertain which patients were waitlisted but not transplanted. This is a small fraction of waitlisted patients, as 86.6% of veterans waitlisted patients with HCC in the MELD era are transplanted (according to United Network for Organ Sharing data as of 30 April 2018), and this exclusion would serve to overestimate the survival benefit of pursuing LT as a treatment option by excluding those who died without an LT.
In summary, these data demonstrate that over the short-term, LT offers minimal survival benefit compared to resection or ablation but over a 5-year time horizon confers a survival benefit of 1.04–1.24 years. These data and others presented herein are important for contextualizing how to best allocate deceased donor organs, as well as counseling patients about their treatment options when diagnosed with early-stage HCC. Further data are needed to validate these results outside the VHA; however, this work represents an important step in optimizing the care of patients with HCC while also remaining responsible stewards of the scarce resource of deceased donor organs.
1. Goldberg D, Ditah IC, Saeian K, et al. Changes in the prevalence of hepatitis C virus infection, nonalcoholic steatohepatitis, and alcoholic liver disease among patients with cirrhosis or liver failure on the waitlist for liver transplantation. Gastroenterology. 2017; 15251090–1099.e1
2. Marrero JA, Kulik LM, Sirlin CB, et al. Diagnosis, staging, and management of hepatocellular carcinoma: 2018 practice guidance by the American Association for the Study of Liver Diseases. Hepatology. 2018; 682723–750
3. European Association for the Study of the Liver. EASL clinical practice guidelines: management of hepatocellular carcinoma. J Hepatol. 2018; 691182–236
4. Bruix J, Sherman M; American Association for the Study of Liver Diseases. Management of hepatocellular carcinoma: an update. Hepatology. 2011; 5331020–1022
5. Goldberg D, French B, Newcomb C, et al. Patients with hepatocellular carcinoma have highest rates of wait-listing for liver transplantation among patients with end-stage liver disease. Clin Gastroenterol Hepatol. 2016; 14111638–1646.e2
6. Goldberg D, French B, Abt P, et al. Increasing disparity in waitlist mortality rates with increased model for end-stage liver disease scores for candidates with hepatocellular carcinoma versus candidates without hepatocellular carcinoma. Liver Transpl. 2012; 184434–443
7. Bittermann T, Niu B, Hoteit MA, et al. Waitlist priority for hepatocellular carcinoma beyond Milan criteria: a potentially appropriate decision without a structured approach. Am J Transplant. 2014; 14179–87
8. Bittermann T, Hoteit MA, Abt PL, et al. Waiting time and explant pathology in transplant recipients with hepatocellular carcinoma: a novel study using national data. Am J Transplant. 2014; 1471657–1663
9. Berry K, Ioannou GN. Comparison of liver transplant-related survival benefit in patients with versus without hepatocellular carcinoma in the United States. Gastroenterology. 2015; 1493669–80. quiz e15
10. Mehta N, Heimbach J, Hirose R, et al. Minimal transplant survival benefit for hepatocellular carcinoma: is it real or an overestimation of waitlist life expectancy? Gastroenterology. 2016; 1502533–534
11. Goldberg DS, French B, Forde KA, et al. Association of distance from a transplant center with access to waitlist placement, receipt of liver transplantation, and survival among US veterans. JAMA. 2014; 311121234–1243
12. Kaplan DE, Dai F, Aytaman A, et al; VOCAL Study Group. Development and performance of an algorithm to estimate the Child-Turcotte-Pugh score from a national electronic healthcare database. Clin Gastroenterol Hepatol. 2015; 13132333–41.e1
13. Mahmud N, Kaplan DE, Taddei TH, et al. Incidence and mortality of acute-on-chronic liver failure using two definitions in patients with compensated cirrhosis. Hepatology. 2019; 6952150–2163
14. Sohn MW, Arnold N, Maynard C, et al. Accuracy and completeness of mortality data in the department of veterans affairs. Popul Health Metr. 2006; 4:2
15. Mehta N, Heimbach J, Harnois DM, et al. Validation of a Risk Estimation of Tumor Recurrence After Transplant (RETREAT) score for hepatocellular carcinoma recurrence after liver transplant. JAMA Oncol. 2017; 34493–500
16. Mahmud N, Shaked A, Olthoff KM, et al. Differences in posttransplant hepatocellular carcinoma recurrence by etiology of liver disease. Liver Transpl. 2019; 253388–398
17. Ioannou GN, Perkins JD, Carithers RL Jr. Liver transplantation for hepatocellular carcinoma: impact of the MELD allocation system and predictors of survival. Gastroenterology. 2008; 13451342–1351
18. Eyawo O, McGinnis KA, Justice AC, et al; VACS Project Team. Alcohol and mortality: combining self-reported (AUDIT-C) and biomarker detected (peth) alcohol measures among HIV infected and uninfected. J Acquir Immune Defic Syndr. 2018; 772135–143
19. Justice AC, Smith RV, Tate JP, et al; VA Million Veteran Program. AUDIT-C and ICD codes as phenotypes for harmful alcohol use: association with ADH1B polymorphisms in two US populations. Addiction. 2018; 113122214–2224
20. McGinnis KA, Tate JP, Williams EC, et al. Comparison of AUDIT-C collected via electronic medical record and self-administered research survey in HIV infected and uninfected patients. Drug Alcohol Depend. 2016; 168:196–202
21. Beste LA, Leipertz SL, Green PK, et al. Trends in burden of cirrhosis and hepatocellular carcinoma by underlying liver disease in US veterans, 2001-2013. Gastroenterology. 2015; 14961471–1482.e5. quiz e17
22. Serper M, Taddei TH, Mehta R, et al; VOCAL Study Group. Association of provider specialty and multidisciplinary care with hepatocellular carcinoma treatment and mortality. Gastroenterology. 2017; 15281954–1964
23. Merion RM, Schaubel DE, Dykstra DM, et al. The survival benefit of liver transplantation. Am J Transplant. 2005; 52307–313
24. Liu J, Weinhandl ED, Gilbertson DT, et al. Issues regarding “immortal time” in the analysis of the treatment effects in observational studies. Kidney Int. 2012; 814341–350
25. Liu J, Weinhandl E, Peter WS. Immortal time bias must be considered in observational studies. Paper presented at: American Society of Nephrology Conference, November 4–9, 2008, Philadelphia, PA
26. Jones M, Fowler R. Immortal time bias in observational studies of time-to-event outcomes. J Crit Care. 2016; 36:195–199
27. Uno H, Claggett B, Tian L, et al. Moving beyond the hazard ratio in quantifying the between-group difference in survival analysis. J Clin Oncol. 2014; 32222380–2385
28. Sung RS, Zhang M, Schaubel DE, et al. A reassessment of the survival advantage of simultaneous kidney-pancreas versus kidney-alone transplantation. Transplantation. 2015; 9991900–1906
29. Mazzaferro V, Regalia E, Doci R, et al. Liver transplantation for the treatment of small hepatocellular carcinomas in patients with cirrhosis. N Engl J Med. 1996; 33411693–699
30. Lin CT, Hsu KF, Chen TW, et al. Comparing hepatic resection and transarterial chemoembolization for Barcelona Clinic Liver Cancer (BCLC) stage B hepatocellular carcinoma: change for treatment of choice? World J Surg. 2010; 3492155–2161
31. Ciria R, López-Cillero P, Gallardo AB, et al. Optimizing the management of patients with BCLC stage-B hepatocellular carcinoma: modern surgical resection as a feasible alternative to transarterial chemoemolization. Eur J Surg Oncol. 2015; 4191153–1161
32. Zhong JH, Xiang BD, Gong WF, et al. Comparison of long-term survival of patients with BCLC stage B hepatocellular carcinoma after liver resection or transarterial chemoembolization. PLOS One. 2013; 87e68193
33. Galle PR, Tovoli F, Foerster F, et al. The treatment of intermediate stage tumours beyond TACE: from surgery to systemic therapy. J Hepatol. 2017; 671173–183
34. Bolondi L, Burroughs A, Dufour JF, et al. Heterogeneity of patients with intermediate (BCLC B) hepatocellular carcinoma: proposal for a subclassification to facilitate treatment decisions. Semin Liver Dis. 2012; 324348–359
35. Shaya FT, Breunig IM, Seal B, et al. Comparative and cost effectiveness of treatment modalities for hepatocellular carcinoma in SEER-medicare. Pharmacoeconomics. 2014; 32163–74
36. Sarasin FP, Giostra E, Mentha G, et al. Partial hepatectomy or orthotopic liver transplantation for the treatment of resectable hepatocellular carcinoma? A cost-effectiveness perspective. Hepatology. 1998; 282436–442
37. Ioannou GN. Transplant-related survival benefit should influence prioritization for liver transplantation especially in patients with hepatocellular carcinoma. Liver Transpl. 2017; 235652–662
38. Schaubel DE, Guidinger MK, Biggins SW, et al. Survival benefit-based deceased-donor liver allocation. Am J Transplant. 2009; 94 Pt 2970–981