A total of 748 patients were included in the study with a mean age of 51.6±8.6 and 89.6% were male. HBV infection was the most common cause (96%) and cirrhosis was found in 84 %. The 486 patients of the training cohort had similar characteristics to the 262 patients of validation cohort without any significant differences as summarized in (Table1). For the training and validation cohorts, the median post-LT follow-up was 338 days, (IQR: 205-673 days) and 416.5 days, (IQR:205-672 days), respectively. The 2-year OS was 82.6 % (95%CI:0.77-0.87) vs 81.9% (95%CI: 0.74-0.87), p=0.870(Fig.1A). Recurrence of HCC was observed in 12.8 % (62 of 486) vs 16.8% (44 of 262) at a median of 11.3months (IQR: 5.9-21.2, months) vs 12.0 months IQR:6.2-20.5 months) after LT. The 2-year overall HCC recurrence was 17.1 % (95%CI:0.13-0.22) vs 25.9% (95%CI: 0.19-0.36), p=0.180 (Fig.1B).
Factors affecting post-LT HCC recurrence
Factors of post-LT HCC recurrence in the training cohort were identified by univariable Cox regression analysis (Table2). The preoperative factors that associated with post-LT HCC recurrence included pre-LT TACE, pre-LT AFP, total tumor diameter (cm), the largest tumor diameter (cm) and the number of nodules at all tested cut-off values. The postoperative factors included; vascular invasion, poorly differentiated tumor grade. Other factors including recipient age, gender, MELD score, number of HCC nodules, use of other neoadjuvant therapies (i.e. hepatectomy, RFA) did not associate with post-LT HCC recurrence. We emphasize that the tumor diameters and the number of nodules were obtained from the last imaging before transplantation by which Milan and Hangzhou criteria were calculated. On multivariable analysis, only pre-LT AFP, largest tumor size and tumor number were the preoperative predictors found to be associated with increasing the risk of post-LT HCC recurrence. While the postoperative predictors were the only presence of vascular invasion and poorly differentiated tumor grade.
Development of the 5-8 Score
Preoperative factors with p-value <0.05 on univariable analysis were then used in the multivariable model. A Cox regression analysis was then performed with backward selection to conduct a multivariable analysis of clinicopathologic factors that associated significantly with post-LT HCC recurrence. Preoperative independent factors of post-LT HCC recurrence were utilized to construct the 5-8 score including (1) pre-LT AFP (at the following cut-off values: 10-200, 201-1000, and > 1000 ng/mL), (2) the largest diameter of tumor (at the following cut-off values: 4-6,6.1-8 and >8 ), and (3) number of nodules (single vs multiple). It is important to note that the model with the lowest Akaike information criterion (AIC), was chosen as the final model for the risk score. The multivariable HR of these factors derived from the Cox regression model were rounded to the nearest integer, then used to calculate a simplified 5-8 score. To calculate the score for each patient, the individual points for each of the three variables can be added together giving a minimum point of 0 and a maximum point of 24(Table3).
Prediction of HCC recurrence risk by the 5-8 score
Based on the 5-8 score, patients were then accurately stratified according to their risk of recurrence into three categories; the low-risk group had a score of 0 to 5, the medium-risk group had a score of 6–8, the high-risk group had a score of >8(Table3). The most common group being low risk group [n=253(52.1%)] then medium-risk group [n=129(26.5%)] and high-risk group (n=104(21.4%)]. The risk of 2-year HCC recurrence was increased significantly from low to high-risk group as shown by KM curves and log-rank test. The 2-year HCC recurrence rate was 4.5 % (95%CI:0.02-0.09), 20.0% (95%CI:0.12-0.34) and 51.4% (95%CI:0.36-0.73) in the low, medium and high-risk group respectively (overall log-rank p < 0.001) )(Fig.2A). The corresponding 2- year OS was 92.5% (95%CI: 0.86-0.96), 82.9%(95%CI:0.72-0.90) and 57.2% (95%CI: 0.42-0.70), respectively (overall log-rank p<0.001) (Fig.2B). According to the 5-8 score, patients before transplantation with AFP level of ≤10 ng/mL and single nodule with the largest tumor diameter of <4cm on radiological assessment would be categorized in the low-risk group. There was no deviation from the proportional hazard assumption according to the visual inspection of log-log survival curves and the Schoenfeld test (p=0.708). Also, for prediction HCC recurrence in the training cohort, visual assessment of the KM curves showed good discrimination between the three 5-8 model prognostic subgroups also the Harrell’s C and Somers’ D of 5-8score were 79%(95%CI:0.73-0.86) and 59%(95%CI:0.40-0.72) (Table4). Based on the competing risk analysis, the 2-year cumulative incidence of mortality as estimated by the competing risk analysis, while controlling for the risk of HCC-related death, was 5.2%(95%CI: 0.02-0.11), 12.9%(95%CI: 0.06-0.22)and 35.0%(95%CI: 0.22-0.49, overall p <0.001) in patients with low, medium and high 5-8score(Fig.2C). Furthermore, 2- year cumulative incidence of HCC-unrelated death which not related to HCC recurrence were, 2.4%(95%CI: 0.01-0.05) ,4.2% (95%CI: 0.01-0.10) and 7.9%(95%CI: 0.03-0.16), (overall p = 0.120) in patients with low , medium and high 5-8score (Fig.2D).
Comparison of the 5-8 Model with Milan and Hangzhou Criteria
The 2-years recurrence rate for patients meeting and exceeding Milan criteria, was 4.8% and 33.8% respectively (p<0.001) (Fig.3A), while it was 9.1% vs 57.5% in patients meeting and exceeding Hangzhou criteria respectively, (p<0.001) (Fig.3D). We further analyzed the risk-stratified patients of 5-8 score for the patients who were fulfilling and exceeding either Milan or Hangzhou criteria. Among 259 patients who were fulfilling Milan criteria (53.2%) , the risk of 2-year HCC recurrence according to 5-8 score was 2.6% (95%CI: 0.01-0.07) in the low risk group, 8.5% (95%CI: 0.03-0.29) in medium- risk group and 23.6% (95%CI: 0.06-0.95) in the high-risk group (overall p=0.009) (Fig.3B).While for 227 patients who are exceeding Milan criteria (46.7%), the risk of 2-year HCC recurrence was 11.5% (95%CI: 0.05-0.28),29.0% (95% CI:0.16-0.51) and 55.0%(95%CI: 0.38-0.79), respectively (overall p<0.001) (Fig.3C). Likewise, for the 390 patients within Hangzhou criteria (80.3%), the risk of 2-year HCC recurrence according to 5-8 score was 3.0% (95%CI:0.01-0.07) in the low risk group, 16.3% (95%CI:0.09-0.31) in medium- risk group and 24.6% (95%CI: 0.12-0.49) in the high-risk group (overall p<0.009) (Fig.3E).While for 96 patients who are exceeding Milan criteria (19.8%), the risk of 2-year HCC recurrence was 26.2% (95%CI: 0.08-0.82), 37.2%(95%CI: 0.16-0.87) and 79.3% (95% CI:0.52-1.21), respectively (overall p<0.001) (Fig.3F). For prediction HCC recurrence, Harrell’s C and Somers’ D of 5-8score were 79%(95%CI:0.73-0.86) and 59%(95%CI:0.40-0.72) in the training cohort compared with 72%(95%CI:0.67-0.76) and 43%(95%CI:0.35-0.58)for Milan criteria and 72%(95%CI:0.65-0.78) and 43%(95%CI:0.31-0.61)for Hangzhou criteria (Table4).
Comparison of the 5-8 Model with AFP model
For patients exceeding and fulfilling the AFP model, the 2-year rates of HCC recurrence were 35.0% and 7.9%, respectively (Fig.4A). Further analysis for the risk-stratified patients of 5-8 score for the patients who were within and outside the AFP model showed that among 301 patients who were within AFP model (62%), the risk of 2-year HCC recurrence was 4.0% (95%CI: 0.02-0.09) in the low-risk group, 19.6% (95%CI: 0.09-0.43) in medium- risk group (p=0.006) (Fig.4B).While for 185 patients who are exceeding AFP model (38.1%), the risk of 2-year HCC recurrence was 9.8% (95% CI: 0.02-0.39) ,20.8% (95%CI: 0.11-0.41), and 51.4%(95%CI: 0.36-0.73), respectively (overall p=0.006) (Fig.4C).
Validation of the 5-8 Model
As mentioned above, there were no significant differences in baseline characteristics among the training and validation cohort. In the validation cohort, the median post-LT follow-up was 416.5 days [IQR:205-672].The 5-8 model could also accurately stratify patients into low, medium and high-risk prognostic subgroups with 2-year HCC recurrence rates of 10.8% (95%CI: 0.06-0.21), 31.7% (95%CI: 0.17- 0.59) and 62.4% (95%CI: 0.39-1.00) respectively, (overall log-rank p < 0.001) (Fig.4D). The 5-8 score also achieved a good performance in post-LT recurrence risk prediction in the validation cohort according to the visual assessment of the resulting KM curves and the Harrell’s C and Somers’ D of 74%(95%CI:0.66-0.82) and 49%(95%CI:0.32-0.74) compared with 65%(95%CI:0.59-0.71) and 30%(95%CI: 0.17-0.45) for Milan criteria and 61%(95%CI:0.54-0.69), 23%(95%CI:0.07-0.40)for Hangzhou criteria (Table 4 ).