The Intention-to-Treat Effect of Bridging Treatments in the Setting of Milan Criteria–In Patients Waiting for Liver Transplantation

Liver transplantation (LT) is the best curative treatment of hepatocellular carcinoma (HCC) for a cirrhotic liver.1 In oncology, LT is considered a successful procedure when a longterm posttransplant tumor-free survival is obtained. Conversely, failure is equivalent to pretransplant delisting, posttransplant tumor recurrence, or death.

Because of allograft scarcity, patients with HCC awaiting LT are most often treated using neoadjuvant locoregional therapies (LRTs) to delay tumor progression and minimize the risk of wait-list delisting.2 When the tumor burden meets the Milan criteria (MC) at the time of the first referral, this approach is called “bridging toward LT.” Two recent international guidelines have emphasized the importance of the bridging strategy, for it has the potential to reduce the risk of pretransplant delisting and posttransplant recurrence. This is especially valid in cases where a partial/complete tumor response is achieved before LT.

Accordingly, LRTs have become a standard of care for treating the listed patients awaiting transplantation. However, although bridging therapies are considered a routine approach in the clinical practice, the reported quality of evidence regarding their use is poor because randomized controlled trials (RCTs) have not yet been completed.3, 4 In this setting, RCTs are untenable due to logistical and, even more, ethical reasons. Consequently, the majority of reported studies compare posttransplant outcomes of treated and untreated patients, thereby failing to analyze the clinical course from an intention-to-treat point of view.5, 6 Moreover, even in studies that include the wait-list period, substantial differences in tumor burden exist between initially bridged and untreated HCC patients.7

To compensate for these drawbacks, we conducted a retrospective analysis of a large European population of HCC patients meeting the MC at first referral and listed for LT. After “balancing” the cohort for possible confounders with an inverse probability of treatment weighting (IPTW), we investigated the risk factors for tumor-specific failure of LT, focusing on the role of LRT.

Patients and Methods

A total of 1083 HCC patients meeting MC at first referral and enlisted for transplant during the period from January 1, 2001, to December 31, 2015, were considered for the IPTW correction. Following the correction for several possible confounders, a pseudo-population of 2019 patients was obtained. The study protocol received a priori approval by the Institutional Review Commitee of the Université Catholique de Louvain. The study was registered at www.clinicaltrials.gov (identification number NCT03723304).

Results

Failure Rates in the Pre- and Post-IPTW Populations

In the original population of 1083 patients, 455 (42.0%) patients presented a failure event from the first referral period to the last follow-up; the failure was HCC-related in 188 (17.4%) and unrelated to HCC in 267 (24.7%) patients. The failure caused by a delisting event was observed in 156 (14.4%) patients. HCC-dependent delisting was observed in 97 (9.0%) patients. There were 38 (3.5%) patients who died during the waiting time (WT) due to non HCC-dependent causes; the remaining 21 patients were delisted due to worsened liver function unrelated to the LRT. The failure caused by post-LT recurrence or death was reported in 299 (27.6%) patients. A total of 91 (8.4%) recurrences were reported after a median time from LT to recurrence of 22 months (IQR, 11-47 months), and 32 relapsed patients were still alive at the last follow-up. In addition, 208 (19.2%) patients died after LT due to HCC-unrelated causes (Supporting Table 2).

In the pseudo-population of 2019 post-IPTW patients, 813 (40.3%) patients presented a failure event from the first referral period to the last follow-up; the failure was HCC-related in 350 (17.3%) and unrelated to HCC in 463 (22.9%) patients. The failure caused by a delisting event was observed in 228 (11.3%) patients. HCC-dependent delisting was observed in 146 (7.2%) patients, and 45 (2.2%) patients died during the WT due to non HCC-dependent causes. The remaining 37 patients were delisted due to worsened liver function unrelated to the LRT. The failure caused by post-LT recurrence or death was reported in 585 (29.0%) patients, and 204 (10.1%) recurrences were reported after a median time from LT to recurrence of 22 months (IQR, 12-47 months). At the last follow-up, 74 relapsed patients were still alive; 381 (18.9%) patients died after LT due to HCC-unrelated causes (Supporting Table 2).

When the number of LRTs was investigated in terms of HCC-related failure rates, we performed 2 separate analyses on the pre- and post-IPTW populations. In the pre-IPTW population, patients receiving no LRT presented similar results compared with patients treated with 1 LRT (log-rank P = 0.48) or 2-3 treatments (P = 0.16). However, when the number of treatments was ≥4, the failure rates grew accordingly (5-year failure rate: 30.8% versus 12.9%; P = 0.001; Fig. 1A).

When the post-IPTW population was investigated, we observed similar results. Patients receiving no LRT presented similar results compared with patients treated with 1 LRT (log-rank P = 0.77) or 2-3 treatments (P = 0.10). However, when the number of treatments was ≥4, the failure rates grew accordingly (5-year failure rate: 31.6% versus 15.9%; P < 0.001; Fig. 1B).

Discussion

In oncology, establishing the superiority of one therapeutic strategy over another one requires RCTs, which aim to identify proportions of therapeutic failure (ie, progressive disease [PD], recurrence, or death) between the 2 approaches.11 However, as is usually the case, basic oncological principles are overlooked in the field of transplant oncology. Given the shortcomings of statistical methodology, 3 different reasons might explain the lack of clarity about the LRT effect as a neoadjuvant treatment in LT.

First, an RCT that compares patients receiving upfront transplants with patients receiving LRT first as a bridge to LT is difficult to support from an ethical point of view. Thus, we developed a sophisticated statistical approach with the intent to balance a historical population of no-LRT and LRT patients on the basis of information available at the first referral for LT. The IPTW strongly affected sample size, by “artificially” increasing the number of no-LRT patients. Nevertheless, such methodology was the only key to offset the important, otherwise unresolvable, initial selection bias. The balancing effect should be noted observing how the results of the competing risk analyses changed in the pre- and post-IPTW population: This phenomenon was the consequence of the limitation of the initial biases presented in the original population. The IPTW is prone to conceptual drawbacks, but this methodology represents the most rigorous way to re-equilibrate the sample to test.12 Consistently, all the tests used to check the successful balancing of the 2 study groups confirmed the validity of our method.

Second, studies comparing no-LRT and LRT patients focus only on posttransplant data, thereby failing to obtain intention-to-treat results. It is only recently that the importance of intention-to-treat analyses in the setting of LT has been recognized.13-16 For the first time, the present study has investigated the intention-to-treat effect of LRT against upfront LT in MC-in HCC patients.

Third, the overlapping effects of several risk factors might lead to inaccurate results because of the absence of a competing risk analysis. A competing risk is an event that either hides the observation of the event in the study (ie, HCC-related outcomes) or alters the chance that this event occurs. Recently, the statistical analysis of competing risks has also been introduced in the setting of HCC and LT.17 Indeed, the competing risk analysis brings about the definition of real-world probabilities of a specific event by breaking down specific causes.

In our study, 2 risks that compete with curative treatment were considered: failure caused by tumor-related reasons (ie, pretransplant delisting caused by disease progression and posttransplant recurrence) and failure caused by nontumor-related events. The conceptual evaluation of pretransplant and posttransplant adverse events through the same “failure approach” represents a novelty in the LT set. In this analysis, disease progression represented the worst independent risk factor with a 5.62-fold increased risk for HCC-dependent failure. This observation is in line with many studies showing the detrimental role of poor radiological response on delisting, intention-to-treat death, transplant benefit, posttransplant tumor recurrence, and posttransplant death.5, 7, 13, 14, 18, 19 The diameter of the largest lesion, AFP levels at first referral, and MELD scores were also risk factors for HCC-dependent failure by previous reports 2, 7, 13-23

This study revealed that up to 3 pretransplant LRTs were protective compared with no LRT. One LRT reduced the risk of intention-to-treat failure by 49%, and 2 to 3 treatments decreased this risk by 34%. These findings are in line with the recent international guidelines, which suggest that bridging LRTs are appropriate in a LT project, despite the heterogeneity of the reported data.3, 4

Interestingly, when we investigated the risk factors for HCC-related failure in the pre-IPTW population, LRT number failed to be statistically significant. This result suggests that the investigation of the LRT effect on delisting and posttransplant recurrence should be markedly influenced by the initial heterogeneity of the investigated population. A recent meta-analysis, focusing on LRT and LT, specifically pointed out the heterogeneity biases among different studies, which are caused by variable demographics (ie, the great variability of WT), types of LRT, HCC stages (T1 versus T2), and treatment dynamics (frequency and interval between treatments).24 Despite these limitations, that meta-analysis partially hinted at our results: LRT proved beneficial in terms of global delisting rates (relative risk, 0.19; 95% CI, 0.15-0.24) and HCC-dependent delisting (relative risk, 0.11; 95% CI, 0.07-0.17).24 When studies comparing treated and untreated patients were tested, the relative risk seemed protective (0.32; 95% CI, 0.06-1.85). Nonetheless, the effect was not statistically significant, probably because of biases, imprecision, and inconsistency in the included studies.24 The beneficial effect of LRT on the risk of delisting has also been reported in the Western and Eastern literature.7, 14, 22, 23

The positive effect of upfront LRT was described in the recent multicenter US experience comprising 3601 patients reported by Agopian et al.5 One LRT was protective for the risk of recurrence compared with a direct transplant (HR, 0.86); conversely, an increased number of treatments raised the risk (≥4 LRT: HR, 2.5; P < 0.001).

Our results concerning LRT in MC-in HCC patients might explain the discrepancy within previous reports. The different number of bridge treatments determines different pretransplant and posttransplant outcomes. The decision to perform a direct transplant shifts the risk of pretransplant delisting into the risk of posttransplant recurrence by eliminating LRT as a selection criterion. This phenomenon is also shown in Supporting Table 4, in which a higher percentage of no-LRT patients experienced recurrence with respect to treated patients, whereas LRT patients presented more cases of PD and longer WTs (namely, selection by time and LRT). This argument has also been clearly shown in “fast-track” living donor LT and in studies about WT as a possible selector for the risk of posttransplant recurrence.25-27 In all of these studies, the patients presenting shorter WTs also received a lower number of LRTs.

Our study shows that performing a pretransplant LRT strategy gives a beneficial effect on post-LT results as long as it does not exceed 3 treatments. In other terms, when the number of required LRTs for stabilizing the tumor is ≤3, the positive effect (namely, reducing the posttransplant recurrence) is statistically and significantly superior with respect to the negative one (namely, increasing the delisting rates). When ≥4 bridging treatments are necessary with the intent to stabilize the tumor before LT, this positive phenomenon disappears. This negative course is a clear demonstration of a tumor selection by treatment: the higher the number of LRTs required, the higher the aggressiveness of the tumor and the worse the overall results.

Very low 5-year HCC-dependent failure rates can be achieved in patients who initially received 1 LRT and did not exhibit disease progression after the treatment. Similar data were also observed in a large retrospective US analysis performed on 2794 LT patients, in which a lower posttransplant recurrence rate was reported in patients undergoing LRT, whereas AFP and tumor burden were independent risk factors for recurrence.28 In light of these results, we can postulate that tumor characteristics prevail over the treatment in influencing the ultimate therapeutic results. Still, using LRT as a selection tool strongly discriminates between patients in terms of posttransplant clinical course.

In an era in which great pressure exists on health care quality improvement and cost reduction, our study suggests the opportunity to frame a shift in standard practice toward LRT in MC-in HCC patients. In our study, we observed a risk reduction from 34% to 49% in patients receiving ≤3 LRTs, possibly resulting in a substantial, cost-effective benefit. However, large prospective cost-effective analyses are needed to confirm this effect. Response to LRT seems to be a rather rudimentary but valuable predictive tool, being able to unveil tumor biology and, as a test of time, to select patients for LT. However, the decision to incorporate LRT as a standard approach in MC-in patients should also be implemented in light of different local philosophies. As an example, it could be argued that LRT could be offered as a standard approach in patients with a predicted WT ≥6 months. On the opposite side, studies clarifying the benefit of LRT in centers with shorter median times or with living donor programs require further evaluation.

The specific role of the LRT method used in our series has been only marginally explored in the study. In Supporting Table 5, we reported the different risks of delisting, recurrence, and overall HCC-dependent failure according to the use of an embolic versus ablative versus combined approach. Preliminary evidence exists on the necessity of treating tumors with a combinatory approach during the WT, and how it should be connected with an increased risk of overall tumor-related failure. However, we should consider these preliminary results with caution. Further analyses that focus on this aspect are required.

Our study has some limitations. The retrospective nature of the study impaired our possibility to specify in detail the reasons justifying a repetitive treatment approach (ie, LRT refractoriness of the target lesion versus initial multimodal approach versus treatment of new tumors).

Moreover, the study, which covers a long period, implies an evolution in the technical aspects of LRT. Hence, the study period was limited to the last 2 decades, and the period was introduced as a variable in the multivariate model to correct the results also for this possible confounder. The analysis showed that patients listed during the first era (2001-2009) fared worse, possibly because therapeutic approaches improved during the second era. One can postulate that excluding the cases of hepatic resection, transarterial radioembolization, or external radiotherapy should reduce the impact of our intention-to-treat analysis. On the other hand, we think that considering salvage LT after resection or very uncommon strategies, such as radioembolization or radiotherapy, should represent a bias, mainly due to their neglectable number in the present study.

Another possible limitation concerns intercenter differences in the length of WT and dynamics of LRT. Although these discrepancies are difficult to resolve, the centers belonging to the EurHeCaLT Study Group adopt similar approaches and policies in the HCC management before LT. Moreover, the difference across centers can be statistically beneficial because it enriches the variability within patient cohorts and brings about more solid statistical results.

Lastly, the use of the IPTW can be criticized for the artificial increase in the sample size. However, this sophisticated statistical approach is the only way forward in removing the initial selection bias. We are unable to assert that any residual confounding may occur in the study because of an imperfect measure of some confounder initially used for the construction of the IPTW model. Unfortunately, when this phenomenon is observed, an adjustment done using this imperfect measure does not completely remove the effect of the confounding variable. We should honestly underline that this latter limit, which derives from the great initial difference among LRT and no-LRT patients, makes it very difficult to construct a balanced model and that it is probably impossible to eliminate in this specific setting.

In conclusion, LRT for MC-in HCC patients is valuable when considering an intention-to-treat LT approach. When comparing treated and untreated patients, 1 single and 2-3 LRTs confer a 49% and 34% reduction, respectively, in the risk of HCC-dependent transplant failure. This beneficial effect disappears with ≥4 LRTs because of more aggressive tumor biology. Patients who show a poor response to LRT have a predictably greater risk for pretransplant tumor-related delisting or posttransplant recurrence.