Liver cancer understaging in liver transplantation in the current era of radiologic imaging and newer generation locoregional therapies

Imaging protocols

For HCC diagnosis, we utilized standardized protocols based on imaging criteria following UNOS guidelines. Patients on the transplant list underwent every 3-month imaging. The majority underwent MRI with gadolinium, with most examinations performed on a 1.5T Siemens Avanto MRI scanner. Until 2017, a minority of patients were scanned on a 1.5T GE Signa device. Between 2017-2019, patients were scanned with a 1.5T Siemens Aera or 3.0T Vida device. MRI protocol included multiplanar HASTE, axial fat-saturated T2, axial opposed-phase T1, MRCP, pre- and dynamic post-contrast fat-saturated 3D axial T1 and diffusion weighted images. Four post-contrast images were obtained in the arterial, portal venous, 3-minute and 5-minute delayed phases. Gadolinium contrast agents used over the studied time period included ProHance® (gadoteridol), Magnevist® (gadopentetate dimeglumine), MultiHance® (gadobenate dimeglumine), and occasionally Eovist® (gadoxetate disodium). CT scans used modern equipment. Between 2008- 2014, the CT protocol included unenhanced images followed by post-contrast images obtained in the arterial and portal venous phases. In 2014, an additional delayed venous phase was added to meet American College of Radiology (ACR) Liver Reporting and Data system (Li-RADS) recommendations. The above imaging protocols apply to patients scanned at VCU. Patients scanned within the VA system used different equipment with their own independent protocols.

Variables

Pre-transplant LRT included TACE, thermal ablation, TARE and SBRT. Other treatments included percutaneous ethanol infusion, chemoinfusion (infusion of Adriamycin/cisplatin into the artery feeding the tumor), and proton therapy. On the last pre-LT imaging, we collected data on tumor diameter in cm, the number of lesions, evidence of progression, and the time from imaging to LT in days. Patients were considered DS if initially outside of MC, received LRT, and then documented as within MC before LT. Patients already within MC received LRT only as a bridge therapy. Patients were within UCSF criteria (University of California San Francisco) if: (a) one liver lesion > 5 cm but ≤ 8 cm; (b) 2-3 lesions each less than 5 cm and total diameter of all lesions ≤ 8 cm; or (c) 4-5 lesions each less than 3 cm with total diameter of all lesions ≤ 8 cm. For explant data, tumor number, diameter, differentiation, vascular invasion, LN involvement, and extrahepatic spread were collected and categorized as within or outside MC. Analysis with all variables described above was also performed between the two transplant time periods of 2008-2013 and 2014-2019.

Outcomes

The primary outcome was defined as HCC that was discordant (understaged) based on comparisons of radiologic findings meeting MC on last imaging pre-LT and findings of explant histology. Secondary outcomes were post-LT recurrence of HCC and death.

Statistical analysis

Frequencies and percentages were presented for categorical variables and means (SD) or median (IQR) as appropriate for the full sample and stratified by concordance vs. discordance. Serum AFP was analyzed both as a continuous variable and within categories: < 20, 20-99, and >100 ng/mL. Comparisons between frequencies for concordance vs. understaged were made using chi-squared tests or exact tests if asymptotic methods were not appropriate. Similarly, t-tests or Wilcoxon Rank Sum tests were used for the comparison of continuous variables. Survival analysis was used to assess differences in survival or disease-free survival recurrence between those concordant or understaged. Kaplan-Meier curves were produced and a log-rank test compared groups. For an analysis focused on death due to HCC recurrence, other causes of death were considered censored. To determine whether the results were differed by transplant period (i.e., effect modifier for the association between concordance/discordance and variables assessed), the period was divided into two: 2008-2013 and 2014-2019. For binary variables, the Breslow-Day test or Zelen test (an exact test) were used to assess the significance of the transplant period as an effect modifier. For categorical variables with > 2 levels, logistic regression was used with the interaction term evaluated for effect modification, again allowing for an exact test when asymptotic assumptions were not valid. For continuous variables, logistic regression was used to assess for effect modification. Effect modification for survival and disease-free survival was assessed with an interaction term in a model using Cox-proportion hazards regression. For all analyses, when effect modification was detected, the association was assessed and reported separately for the periods 2008-2013 and 2014-2019. Significance was considered for P < 0.05. SAS version 9.4 was used for all analyses.

Demographics/ discordance rates

Of the 146 patients transplanted, the mean age was 61 years, with 83% male, 71% white and 14% black [Table 1], the majority (55%) having HCV. Overall, the discordance rate between pre-LT imaging stage and explant was 45%. All cases of discordance involved understaging on imaging. When evaluating sex, race or etiology of liver disease, there was no association between concordance or discordance.

Pre-transplant characteristics

At the time of HCC diagnosis, 85% of patients were within MC [Table 2]. Of the 17% (n = 25) who required DS prior to listing, 76% were beyond MC but within UCSF criteria, leaving six beyond both criteria. All patients undergoing LT were downstaged to within MC prior to LT. There was no association between discordance and disease burden within or outside MC (P = 0.14) or UCSF criteria (P = 1.00). In patients requiring DS, a higher rate of discordance was seen, although not statistically significant (P = 0.087). AFP values around the time of HCC diagnosis had no association with concordance/discordance whether considered continuous (P = 0.7723) or categorized (P = 0.1306). No association with concordance/discordance was found in the number of days from last pre-LT imaging to LT (P = 0.33), wait-list time (P = 0.59) or mean MELD/MELD Na at time of listing.

HCC characteristics

For patients with solitary lesions at diagnosis, 66% were in the discordant group and 67% in the concordant group [Table 3]. In those with multifocal lesions at diagnosis, patients with ≥ 3 lesions had 36% discordance vs. 11% concordance. This was statistically significant (P = 0.046). There was no association between concordance/discordance and location of the HCC lesion(s) (unilobar vs. bilobar).

Last imaging before transplant

The majority of patients (95%) underwent MRI vs. 5% with CT. Overall, 66% had a complete radiologic response to treatment on the last imaging pre-LT, with 68% in the discordant group and 64% in the concordant group. No statistical significance was seen between concordance and treatment response seen on the last pre-LT imaging (partial or complete). No association was seen between discordance and tumor multifocality, new HCC lesion, increase in tumor size or treatment response on last imaging [Table 3].

Pre-transplant HCC treatment

The majority of patients (97%) were treated with LRT. Most received TACE (71%), followed by thermal ablation at 43%. Only 3% received other traditional treatments including chemoinfusion, percutaneous ethanol injection and proton therapy. For new generation LRT, 17% received TARE and 6% received SBRT. While 75% received only traditional LRT with either TACE, RFA, and PEI, 12% received only new generation pre-LT LRT with either TARE and/or SBRT. Another 10% received a combination of traditional and new generation LRT. No association between concordance/discordance was seen between patients receiving traditional LRT only or new generation LRT only, or a combination of both new and old LRT. No correlation was seen between concordance/discordance and patients receiving multiple vs. single LRT [Table 4].

Explant characteristics

In the discordant group, 16% had well-differentiated HCC, 69% had moderately differentiated, and 15% had poorly differentiated. In the concordant group, 14% had well-differentiated, 69% had moderately differentiated, and 10% had poorly differentiated. There was no association between HCC histopathology and discordance. Only a trend was noted towards discordance with microvascular invasion on explant (P = 0.053). Of note, two patients likely had mixed HCC-cholangiocarcinoma (CC) with incidental discovery of CC on explant. One with a combination of HCC was unable to be differentiated. The second only had CC on histopathology with surrounding necrosis. This likely reflected treated HCC as the patient did have significantly high AFP prior to LRT and subsequent normalization of AFP after treatment [Table 5].

There was an association between patients with explant histology outside of MC and discordance (31%) (P < 0.0001). Of the 19 outside MC explant staging, 10 were within UCSF criteria, and all were in the discordant group. With tumor multifocality on explant, 32% had ≥ 2 HCC nodules and was associated with higher discordance rates (51% discordant group vs. 18% concordant group); P < 0.0001. The median size of the largest tumor diameter on explant was 2.0 cm, with no association between discordance. Single lesions on explant whether < 3 cm, 3-5 cm or > 5 cm had no correlation with concordance/discordance. There was an association between pathologic response to treatment on explant (partial, complete, progression) and concordance/discordance (P < 0.0001). While 74% of discordant patients and 27% of concordant patients had a partial response, no discordant patients had a complete response. Pathologic progression on explant was seen in 25% in the discordant group vs. 6% in the concordant group (P < 0.0001).

Transplant time periods

Concordance/discordance was compared between the two periods of transplantation: 2008-2013 vs. 2014- 2019, to determine the impact of improved imaging modalities and newer generation LRT. During the period of 2008-2013, 38% of patients were transplanted, with 34% in the discordant group. In the period of 2014-2019, 62% were transplanted, with 66% cases of discordance. No statistically significant differences were seen between these two periods (P = 0.25). Further analysis was performed to assess any modifying effect on the association of concordance/discordance and pre- and post-transplant variables during these two time periods. None of such modifying effects was seen for any of these variables except MC at diagnosis (P = 0.0294) and microvascular invasion (P = 0.0270). Patients outside of MC at diagnosis during the transplant period of 2008-2013 were noted to have higher discordance rates (27% vs. 3%, P = 0.013). No significant differences in concordance/discordance were seen for those within MC transplanted between 2014 - 2019 (P = 0.59) [Table 2]. Microvascular invasion during the transplant period of 2014-2019 was associated with discordance (26%) vs. concordance (4%) (P = 0.0055) [Table 5].

Post-transplant survival and recurrence

Median follow-up after transplant was 1507 days [min = 0 and max = 4622 (IQR: 822-2688)]. Among the 21% (n = 30) of patients who died, 13 % were in the discordant group and 8% were in the concordant group. Discordance was associated with an increased risk of death (P = 0.0136) [Figure 1]. While 10% (n = 14) had recurrence with 6% from the discordant group and 3% from the concordant group, no statistical significance was seen between recurrence-free survival and discordance (P = 0.07) [Figure 2]. Not all causes of death were HCC-related. Of the 30 deaths, 10 died from HCC recurrence, 4 from non-HCC, liver-related complications, and 16 from non-HCC, non-liver-related complications [Table 6]. A survival analysis considering only death as defined as HCC death (other causes of death and no death as censored) was performed, showing differences between the discordant and concordant groups [P = 0.0491] [Figure 3].

Figure 1. Comparison of survival distributions. Discordant (D): 19 deaths; Concordant (C): 11 deaths; P = 0.0136.

Figure 2. Comparison of recurrence-free survival distributions. Discordant (D): 9 recurrences; Concordant (C): 5 recurrences; P = 0.07.

Figure 3. Comparison of survival distributions -deaths due to HCC recurrence. Discordant (D): 7 deaths; Concordant (C): 3 deaths; P = 0.0491. HCC: hepatocellular carcinoma.

Pre-transplant characteristics

In our study, we found no association between concordance/discordance and those within or outside MC at diagnosis. Due to the small sample size, our study likely was underpowered to detect such differences. However, a sub-analysis between transplant periods revealed concordance was associated with MC in patients transplanted in the earlier era (2008-2013). We found no association between concordance/discordance for patients within UCSF criteria prior to listing, but a trend was noted between discordance and DS. While Mahmud found an association between DS and discordance, Ecker^[3] did not. One recent retrospective analysis of the UNOS database evaluated 3819 patients transplanted from 2012-2015^[6], and were classified as always within MC, meeting UNOS-DS inclusion criteria (within UCSF criteria) or “all-comers” (AC-DS) with initial HCC beyond UNOS-DS inclusion criteria. While 1/3 meeting UNOS- DS were found to be understaged with explant tumor beyond MC compared with < 15% within MC, HCC understaging increased by 10% for each 1- cm increase in total tumor diameter on last pre-LT imaging. The AC-DS group had inferior survival rates post- LT compared with UNOS-DS group and those within MC. Therefore, DS within MC should be the minimal requirement for LT.

We also evaluated baseline AFP at the time of HCC diagnosis and found no association between concordance/discordance when stratified according to increasing levels. However, it should be noted that AFP values for 37 patients were not available. We did not evaluate serial AFP values up to LT.

HCC characteristics

A novel finding in our study was the association of discordance with three or more HCC lesions at the time of diagnosis which was statistically significant. Multifocal HCC likely reflects aggressive disease, thus suggesting more aggressive tumor biology. These findings may impact clinical management in how these patients are approached when considering LT. Such considerations include more aggressive LRT with complete obliteration of tumor burden prior to LT and counseling patients on the high risk for understaging and recurrence after transplant. At the author’s institution, even if progression was seen on the last imaging before transplant, patients could still receive a transplant if the tumor remained within MC. This could be a factor contributing to higher rates of discordance. We found no association between concordance/discordance and location of the HCC lesion(s), tumor multifocality, new HCC lesion or increase in tumor size at last pre-LT imaging. The majority of our patients underwent MRI for HCC monitoring with no correlation between concordance and treatment response, whether partial or complete.

Pre-transplant treatments

TARE and SBRT have emerged as relatively newer LRT in HCC and may be effective in improving tumor control as a bridge to transplant, in DS and reducing post-LT recurrence compared to TACE^[6-13]. The use of TARE compared to TACE does have advantages including less abdominal pain, nausea, vomiting, elevated liver tests and fatigue. The advantage of SBRT includes very limited radiation doses to adjacent organs at risk, thus the ability to maintain liver function. Used both in combination with other LRT and as a single modality, SBRT is emerging as the therapeutic tool for small HCC lesions in “difficult to reach” locations or for “difficult to treat” lesions when TACE or TARE have failed. One study evaluated post-transplant outcomes in patients undergoing bridging LRT with TARE vs. TACE^[8]. A trend toward improved 3-year survival was seen in the TARE group. Microvascular invasion was seen in 3.6% of explants of the TARE group vs. 27% in the TACE group. Another study evaluated liver transplant outcomes over a 15-year period between 2004-2018 who received TARE as a bridge or DS^[9]. No differences in OS/recurrence-free survival were seen in those bridged or DS. These studies support the effectiveness of TARE for HCC in the setting of bridging and DS prior to LT.

Unlike Ecker and Mahmoud, we compared pre-LT LRT between newer generation TARE and/or SBRT vs. traditional TACE/ thermal ablation/ other. We evaluated combination vs. single LRT. Ecker evaluated only TACE or ablation^[3]. In Mahmoud’s study, patients received a single type of LRT prior to LT, with TACE being the most frequent at 63.5%, ablation at 33.3% and TARE at 2.7%^[5]. In our study, 17% received TARE, 6% received SBRT, 43% received ablation, and 71% received TACE. Under our “other” category, one received PEI, another proton therapy, and 3 received chemoinfusion. Ecker found no association between concordance/discordance with their LRT. Mahmud found LRT bridging/ DS status to be a predictor of understaging^[5]. In our study, no association between discordance was found between traditional LRT, newer generation LRT, a combination of both or multiple vs. single LRT. This lack of association may be attributed to smaller numbers in our study.

Explant characteristics

Similar to Ecker’s study, discordance was found in those with explant histology outside of MC, microvascular invasion, tumor multifocality, and pathologic response to LRT. The data on explant pathology will unlikely provide additional guidance pre-transplant in identifying high-risk understaged patients without reliable pre-transplant surrogates that can accurately predict histologic findings.

Transplant time periods

Our study is the first to evaluate concordance/discordance between two different time periods of transplantation: 2008-2013 vs. 2014-2019. Although we found no statistically significant differences, further analysis of any modifying effect on pre-transplant variables showed that MC at diagnosis during the transplant period of 2008-2013 had higher concordance rates compared to the period of 2014-2019. This could be explained by differences in practice patterns between these two time periods. In the earlier era, perhaps a more “conservative” approach to HCC diagnosis with strict adherence to MC criteria was followed, given the scarcity of donors.

Despite the newer generation LRT introduced over a decade ago, the advancement of radiologic technology including the advancement of MRI technology and the development of LI-RADS, based on our study, we have not yet improved our ability to predict discordance^[14,15]. To standardize the reporting and data collection of CT/ MRI imaging for HCC diagnosis, LI-RADS was first released by the ACR in 2011^[16]. The utility of LI-RADS has allowed the application of consistent terminology, reduction of imaging interpretation variability/errors, and has facilitated quality assurance and research.

Despite comparisons between the older and newer era of LT, advancement in our MRI technology appears not to have improved the ability to predict discordance. Variability in the interpretation of imaging results could be attributed to differences in imaging protocols, the level of radiologic expertise and HCC diagnostic criteria used. Another factor to consider is the criteria used to evaluate treatment response to newer LRT which are radiation-based. These differ from conventional LRT. Lesions may undergo coagulative necrosis or internal hemorrhage after TARE, making them more hyperintense on T1-weighted imaging or hyperdense on non-contrast CT. This makes interpretation of the response more challenging.

Limitations and strengths

Limitations to our study include the small number of patients and its retrospective nature. Small sample size could have contributed to type II error (i.e., low power). Although a single-center study, patients were referred from various VA hospitals along the east coast to our VA health system. This could contribute to radiologic interobserver variability in evaluating HCC treatment response and progression, as well as variability in imaging protocols and HCC diagnostic criteria. Additional limitations include the use of MRI as our center’s primary imaging modality in our HCC protocol, and multimodal treatments not evenly distributed. These could have impacted our results. Strengths of our study include evaluating patients under real-world conditions transplanted up to 2019 who underwent treatment with newer LRT with TARE and SBRT. We were able to evaluate two transplant time periods to further elucidate whether the improvement of our imaging technology, radiologic diagnostic criteria and newer treatment modalities would affect concordance/discordance.

In conclusion, our study supports the ongoing challenges of HCC stage discordance between pre-LT imaging and explant pathology. Despite the advent of more sophisticated radiologic technology, standardized LI-RADS, and novel LRT as a bridge to LT, discordance remains difficult to predict. In our study, the only variable associated with discordance was three or more HCC lesions detected at diagnosis. The only variable associated with concordance was those diagnosed within MC in the earlier era of LT. No other pre- or post-LT variables predicted discordance/concordance, including newer generation or older generation, single or combined LRT. Although we saw only a trend between DS and discordance, growing data supports this association. We also observed decreased survival with discordance. Emphasis on UNOS-mandated uniformity in selection criteria with adherence to UNOS-DS criteria in order to minimize center/regional variations, behavioral bias, and improve survival will likely play a critical role in selecting the ideal transplant candidate. Larger prospective studies incorporating newer LRT for HCC are needed to further elucidate the predictability of discordance to achieve optimal post-transplant outcomes.