Validation of a Risk Estimation of Tumor Recurrence After
Transplant (RETREAT) Score for Hepatocellular Carcinoma
Recurrence After Liver Transplant
Neil Mehta, MD, Julie Heimbach, MD, Denise M. Harnois, DO, Gonzalo Sapisochin, MD,
Jennifer L. Dodge, MPH, David Lee, MD, Justin M. Burns, MD, William Sanchez, MD, Paul D.
Greig, MD, David R. Grant, MD, John P. Roberts, MD, and Francis Y. Yao, MD
Division of Gastroenterology, Department of Medicine, University of California–San Francisco
(Mehta, Yao); Division of Transplantation, Department of Surgery, Mayo Clinic, Rochester,
Minnesota (Heimbach); Department of Transplantation, Mayo Clinic, Jacksonville, Florida
(Harnois, Lee, Burns); Multi-Organ Transplant Program, Division of General Surgery, Department
of Surgery, University of Toronto, Toronto, Ontario, Canada (Sapisochin, Greig, Grant); Division of
Transplant Surgery, Department of Surgery, University of California–San Francisco (Dodge,
Roberts, Yao); Division of Gastroenterology and Hepatology, Department of Medicine, Mayo
Clinic, Rochester, Minnesota (Sanchez)
Abstract
IMPORTANCE—Several factors are associated with increased hepatocellular carcinoma (HCC)
recurrence after liver transplantation (LT), but no reliable risk score has been established to
determine the individual risk for HCC recurrence.
OBJECTIVE—We aimed to develop and validate a Risk Estimation of Tumor Recurrence After
Transplant (RETREAT) score for patients with HCC meeting Milan criteria by imaging.
DESIGN, SETTING, AND PARTICIPANTS—Predictors of recurrence were tested in a
development cohort of 721 patients who underwent LT between 2002 and 2012 at 3 academic
transplant centers (University of California–San Francisco; Mayo Clinic, Rochester; and Mayo
Corresponding Author: Francis Yao, MD, Division of Gastroenterology, Department of Medicine, University of California–San
Francisco, 513 Parnassus Ave, Room S-357, San Francisco, CA 94143-0538, (francis.yao@ucsf.edu).
Conflict of Interest Disclosures: None reported.
Role of the Funder/Sponsor: The funder played a role in the collection, management, analysis, and interpretation of the data. The
funder had no role in the design and conduct of the study; preparation, review, or approval of the manuscript; and decision to submit
the manuscript for publication.
Previous Presentation: This research was presented at the American Association for the Study of Liver Diseases Liver Meeting;
November 13, 2016; Boston, Massachusetts.
Author Contributions: Drs Mehta and Yao had full access to all of the data in the study and take responsibility for the integrity of the
data and the accuracy of the data analysis.
Concept and design:
Mehta, Heimbach, Harnois, Sapisochin, Burns, Greig, Roberts, Yao.
Acquisition, analysis, or interpretation of data:
Mehta, Heimbach, Harnois, Sapisochin, Dodge, Lee, Sanchez, Grant, Yao.
Drafting of the manuscript:
Mehta, Harnois, Sapisochin, Dodge, Roberts, Yao.
Critical revision of the manuscript for important intellectual content:
Mehta, Heimbach, Harnois, Sapisochin, Dodge, Lee, Burns,
Sanchez, Greig, Grant, Yao.
Statistical analysis:
Mehta, Harnois, Dodge.
Administrative, technical, or material support:
Heimbach, Sapisochin, Sanchez, Grant.
Supervision:
Greig.
HHS Public Access
Author manuscript
JAMA Oncol
. Author manuscript; available in PMC 2018 April 01.
Published in final edited form as:
JAMA Oncol
. 2017 April 01; 3(4): 493–500. doi:10.1001/jamaoncol.2016.5116.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Clinic, Jacksonville) to create the RETREAT score. This was subsequently validated in a cohort of
341 patients also meeting Milan criteria by imaging who underwent LT at the University of
Toronto transplant center using the C concordance statistic and net reclassification index.
MAIN OUTCOMES AND MEASURES—Characteristics associated with post-LT HCC
recurrence.
RESULTS—A total of 1061 patients participated in the study; 77.8%(825) were men, and the
median (IQR) age was 58.2 (53.3–63.9) years in the development cohort and 56.4 (51.7–61.0)
years in the validation cohort (
P
< .001). In the development cohort of 721 patients (542 men),
median α-fetoprotein (AFP) level at the time of LT was 8.3 ng/mL; 9.4% had microvascular
invasion (n = 68), and 22.1% were beyond Milan criteria on explant (n = 159) owing to
understaging by pretransplantation imaging. Cumulative probabilities of HCC recurrence at 1 and
5 years were 5.7% and 12.8%, respectively. On multivariable Cox proportional hazards regression,
3 variables were independently associated with HCC recurrence: microvascular invasion, AFP at
time of LT, and the sum of the largest viable tumor diameter and number of viable tumors on
explant. The RETREAT score was created using these 3 variables, with scores ranging from 0 to 5
or higher that were highly predictive of HCC recurrence (C statistic, 0.77). RETREAT was able to
stratify 5-year post-LT recurrence risk ranging from less than 3%with a score of 0 to greater than
75% with a score of 5 or higher. The validation cohort (n = 340; 283 men) had significantly higher
microvascular invasion (23.8% [n = 81],
P
< .001), explant beyond Milan criteria (37.3% [n =
159],
P
< .001), and HCC recurrence at 5 years (17.9% [n = 159],
P
= .03). RETREAT showed
good model discrimination (C statistic, 0.82; 95%CI, 0.77–0.86) and superior recurrence risk
classification compared with explant Milan criteria (net reclassification index, 0.40;
P
= .001) in
the validation cohort.
CONCLUSIONS AND RELEVANCE—We have developed and validated a simple and novel
prognostic score that may improve post-LT HCC surveillance strategies and help identify patients
who may benefit from future adjuvant therapies.
For 2 decades, the Milan criteria (1 lesion of ≤5 cm, 2–3 lesions of ≤3 cm)
1
have been the
benchmark for the selection of candidates with hepatocellular carcinoma (HCC) for liver
transplantation (LT).
2,3
Treatment of HCC now accounts for more than 20% of all LTs
performed in the United States.
4
Despite physician adherence to the Milan criteria, HCC
recurrence still occurs in about 10% to 15% of patients,
5–7
with a median survival of only
about a year after HCC recurrence.
2,3,8
Only 10% to 30% of recurrent HCCs are eligible for
resection or ablation.
8,9
The impact of tumor size and number on HCC recurrence risk is best illustrated in the
“Metro-ticket forecast,”
10
which follows the paradigm of “the further the distance (from
conventional limits defined by Milan criteria), the higher the price (paid in terms of HCC
recurrence).” Microvascular invasion is a well-established predictor of HCC recurrence after
LT.
5,11–13
Other factors implicated in HCC recurrence include elevated α-fetoprotein (AFP)
levels
14–17
and possibly des-gamma-carboxyprothrombin,
18
poorly differentiated tumor
grade,
12
tumor progression despite locoregional therapy (LRT),
19,20
as well as short waiting
time before LT.
21,22
Mehta et al.
Page 2
JAMA Oncol
. Author manuscript; available in PMC 2018 April 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Despite known risk factors for HCC recurrence after LT, no validated risk score is available
to provide quantifiable and reliable measurements of an individual’s risk of post-LT HCC
recurrence. The lack of a reliable model to estimate the risk for HCC recurrence after LT
may explain why there is no standardized approach to HCC surveillance after LT
2,5
and
wide variation in this practice across LT centers.
5
In the present large multicenter study, we
aimed to develop and validate a recurrence risk score, the Risk Estimation of Tumor
Recurrence After Transplant (RETREAT), for patients with HCC who meet the Milan
criteria by imaging at the time of LT.
Methods
Study Design and Patient Population
This multicenter study was approved by the institutional review boards of all participating
institutions, and all boards waived patient written informed consent. The study included
adult patients (age ≥18 years) with HCC always meeting Milan criteria on imaging who
underwent LT with Model for End-Stage Liver Disease (MELD) score exception from June
2002 to December 2012. Patients requiring tumor downstaging to Milan criteria and those
with intrahepatic cholangiocarcinoma or mixed HCC–cholangiocarcinoma on explant were
excluded. Patients with incidental HCC were also excluded mainly because wait time to LT
was 1 of the variables evaluated as a predictor of post-LT HCC recurrence. The development
cohort consisted of 721 patients who underwent LT at 3 centers with different waiting times:
short (Mayo Clinic, Jacksonville), medium(Mayo Clinic, Rochester), and long (University of
California, San Francisco [UCSF]). The validation cohort consisted of 340 patients also
within Milan criteria on imaging who underwent LT with MELD exception over the same
time period at the University of Toronto.
The variables collected included age, sex, size and number of HCC tumors at time of
diagnosis, AFP at time of listing and LT, LRT, cause of the liver disease, MELD score, and
waiting time (defined as the time from HCC diagnosis to LT). All patients underwent
contrast-enhanced computed tomography or magnetic resonance imaging at a minimum of
once every 3 months after listing for LT.
Key Points
Question
What is an individual’s risk for recurrence of hepatocellular carcinoma (HCC) after liver
transplant based on their tumor characteristics?
Findings
Using a multicenter retrospective cohort study approach, we have developed and
validated a simple and novel score, the Risk Estimation of Tumor Recurrence After
Transplant (RETREAT), that incorporates 3 variables: explant tumor burden,
microvascular invasion, and α-fetoprotein level. The RETREAT score was highly
predictive of HCC recurrence risk after liver transplantation.
Meaning
Mehta et al.
Page 3
JAMA Oncol
. Author manuscript; available in PMC 2018 April 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
The RETREAT risk score will help identify patients who would potentially derive benefit
from future adjuvant therapies and also assist in determining posttransplant surveillance
strategies.
Explant pathology reports were reviewed to determine histologic grade based on the
modified Edmondson criteria,
23
presence of vascular invasion, and tumor stage. Explant
tumor staging was based on size and number of only viable tumors. The following categories
of tumor stage were evaluated: within vs outside Milan criteria, total tumor diameter,
number of viable lesions, and the sum of the largest diameter of viable tumor plus the
number of viable tumors on explant.
Statistical Analysis and Creation of the RETREAT Score
The study end points were 5-year post-LT HCC recurrence and survival. Recurrence and
survival probabilities were estimated by the Kaplan-Meier method and compared using the
log-rank test. Univariate and multivariable hazard ratios (HRs) for predictors of post-LT
HCC recurrence were determined by Coxproportional hazards regression models and
reported with 95% confidence intervals (CIs). Potential cutoffs were evaluated using the
Akaike information criterion (AIC), with lower AIC values indicating better model fit.
Predictors of HCC recurrence with a univariate
P
< .10 were included in the multivariable
analysis, and the final model was selected by backward step wise elimination (
P
> .05 for
removal). The RETREAT score was then created based on the final multivariable model
coefficients.
24
Model coefficients were scaled to the coefficient for AFP and rounded to the
nearest integer. This produced a simplified point scale reflecting the relative impact of model
covariables. The integer value for each model component was then summed to calculate the
RETREAT score.
In the validation cohort, the RETREAT score was tested and compared with Milan criteria
(based on explant pathology findings). The overall C statistic was used to assess model
discrimination, and net reclassification index
25
to evaluate improvement in model
performance by quantifying the proportion of correct risk reclassification for HCC
recurrence. Net reclassification improvement was estimated using a priori 1-and 5-year
recurrence risk groups (<5%, 5% to <10%, 10% to <20%, and ≥20%). Correct risk
reclassification was indicated by RETREAT score predicted probabilities that reclassified
patients with recurrence into higher risk groups and patients without recurrence into lower
risk groups compared with predicted probabilities estimated by explant Milan criteria.
Results
Patient Characteristics
The baseline characteristics of the 721 patients in the development cohort and the 340
patients in the validation cohort are summarized in Table 1. Compared with the development
cohort, the validation cohort was younger, had a higher percentage of men, and more
frequently had hepatitis B or alcohol abuse as the cause of the liver disease. White patients
made up 62.0% of the development cohort (n = 447); Asians, 19.1% (n = 138); Hispanics,
11.2%(n = 81);and African Americans 5.1% (n = 37). Patients in the validation cohort were
Mehta et al.
Page 4
JAMA Oncol
. Author manuscript; available in PMC 2018 April 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
more likely to have a single tumor and less likely to receive LRT prior to LT. Median wait
time from HCC diagnosis to LT ranged from 4.3 months in the center with the shortest wait
time to 12.9 months in the center with the longest wait time. Overall, wait time in the
development cohort was shorter than in the validation cohort (8.1 vs 10.5 months).
Compared with the developmental cohort, the validation cohort had significantly higher
proportions of microvascular invasion (23.8% [n = 81] vs 9.4% [n = 32]), moderately or
poorly differentiated tumor grade (76.7% [n = 260] vs 44.0% [n = 150]), and explant tumor
stage beyond Milan criteria (37.3% [n = 127] vs 22.1% [n = 75]). The median for the sum of
the largest viable tumor diameter (cm) plus the number of viable tumors on explant was 4.5
in the validation cohort vs 3.5 in the development cohort (
P
< .001).
Post-LT Outcomes—Development Cohort
Median post-LT follow-up was 4.5 years (interquartile range [IQR], 2.5–6.8 years) in the
development cohort. Follow-up included a combination of AFP and cross-sectional imaging
every 4 to 6 months for the first 2 years after LT. This was performed in all HCC LT
recipients at 2 centers and in patients deemed to be at high risk for HCC recurrence (AFP
>300, explant tumor burden beyond Milan criteria, and/or microvascular invasion) at a third
center. Overall post-LT survival was 93.1% (95% CI, 91.0%–94.8%) at 1 year and 77.0%
(95% CI, 73.4%–80.1%) at 5 years. Recurrence of HCC was found in 11.6%(84 of 721) at a
median of 13.0 months (IQR, 5.4–26.7 months) after LT. The most common sites of HCC
recurrence were the lung(44.0% [n = 37]), bone (29.8% [n = 25]), liver (26.2% [n = 22]),
and peritoneum (26.2% [n = 22]). The majority (75.0% [n = 63]) had a single site of HCC
recurrence; 20.2% [n = 17] had 2 sites; and 4.8%[n = 4] had 3 or more sites of recurrence at
diagnosis. Overall post-LT HCC recurrence at 1, 2, and 5 years was 5.7% (95% CI, 4.2%–
7.8%), 8.7%(95% CI, 6.8%–11.1%), and 12.8%(95% CI, 10.3%–15.7%), respectively
(Figure 1). There were no significant center-specific differences in rates of HCC recurrence.
Predictors of Post-LT HCC Recurrence and Creation of the RETREAT Score
Predictors of post-LT HCC recurrence in the development cohort on univariate analysis
included microvascular invasion, moderately or poorly differentiated tumor grade, AFP at
the time of LT as a continuous variable and at all tested cutoffs, and the sum of the largest
diameter of viable tumor (cm) plus the number of tumors (Table 2). The tumor sum criterion
had the lowest AIC compared with other assessments of tumor burden and therefore the best
model fit. Age, sex, race/ethnicity, MELD score, cause of the liver disease, number of
lesions at HCC diagnosis, and LRT(vs no LRT)were not predictive of HCC recurrence on
univariate analysis. On multivariable analysis, predictors of HCC included (1) microvascular
invasion, (2)sum of the largest diameter of viable tumor and number of viable tumors (at the
following cutoffs: 1.1–4.9, 5–9.9, and ≥10), and (3) AFP at LT (at the following cutoffs: 20–
99, 100–999, and ≥1000 ng/mL). The multivariable model coefficients for these 3 significant
variables were then used to calculate a simplified RETREAT score. An individual patient’s
RETREAT score is calculated by adding the individual points for each of the 3 variables
(Table 3). The actual statistical model and the coefficients are also summarized in Table 3.
Mehta et al.
Page 5
JAMA Oncol
. Author manuscript; available in PMC 2018 April 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript