Simple parameters predicting extrahepatic recurrence after curative hepatectomy for hepatocellular carcinoma

Extrahepatic recurrence (EHR) after curative hepatectomy for hepatocellular carcinoma (HCC) is associated with a poor prognosis. We investigated the features of EHR and identified its predictive factors. This retrospective study included 398 treatment-naive patients who underwent curative hepatectomy for HCC at two tertiary hospitals. Multivariate Cox-regression analysis was performed to identify the variables associated with EHR. EHR was diagnosed in 94 patients (23.6%) over a median follow-up period of 5.92 years, most commonly in the lungs (42.6%). The 5-/10-year cumulative rates of HCC recurrence and EHR were 63.0%/75.6% and 18.1%/35.0%, respectively. The median time to EHR was 2.06 years. Intrahepatic HCC recurrence was not observed in 38.3% of patients on EHR diagnosis. On multivariate analysis, pathologic modified Union for International Cancer Control stage (III, IVa), surgical margin involvement, tumor necrosis, sum of tumor size > 7 cm, and macrovascular invasion were predictive factors of EHR. Four risk levels and their respective EHR rates were defined as follows: very low risk, 1-/5-year, 3.1%/11.6%; low risk, 1-/5-year, 12.0%/27.7%; intermediate risk, 1-/5-year, 36.3%/60.9%; and high risk, 1-year, 100.0%. Our predictive model clarifies the clinical course of EHR and could improve the follow-up strategy to improve outcomes.


Comparison of surgical findings between patients with and without EHR. Microvascular and
serosal invasion were more prevalent among patients with EHR than in those without (16.9 versus 29.8% and 2.1 versus 6.5%, respectively) ( Table 2). Moreover, the presence of satellite nodules and tumor necrosis in resected specimens was more prominent in patients with EHR than in those without (13.0 versus 31.2% and 46.9 versus 68.8%, respectively).

Comparison of characteristics between patients with early and non-early EHR.
Both the radiologic and pathologic mUICC stages were more advanced in the early EHR than in the non-early EHR group (Supplementary Table S2). The early EHR group also had a markedly shorter recurrence-free-survival (RFS) and survival rates compared to the non-early EHR and non-EHR groups (Fig. 2). Moreover, the proportion of tumors with a mUICC stage ≥ III at the time of first recurrence was larger in the early EHR than in the non-early EHR group. Especially, in 54.8% of patients in the early EHR group, EHR was the first presenting recurrence after curative hepatectomy.

Prediction of EHR.
Based on our multivariate analyses, the following five variables were used to build a parametric model to predict EHR. Then, the risk of EHR was stratified into four levels based on the number of predictive factors present, as follows: very low risk, 0-1 risk factors; low risk, 2 risk factors; intermediate, 3 risk factors; and high, ≥ 4 risk factors. Then, cumulative rate of EHR was calculated using the Kaplan-Meier survival curve analysis for each risk level (Fig. 3a). The 1-, 3-, 5-, and 10-year cumulative rates of EHR were significantly related to the numbers of risk factors present: very low risk: 3.1%, 7.3%, 11.6%, and 25.1%, respectively; low risk: 12

Analysis of factors associated with the overall survival rates.
To analyze the association with EHR with the overall survival rate, we conducted Cox-regression analysis including many potential risk factors, such as EHR (

Discussion
Based on the data of many patients who underwent curative hepatectomy for HCC, we proposed a simple parametric model predicting the risk of EHR development. This model is straightforward and easy-to-use, and consists of five easy-to-obtain variables that constitute the essentials of pre-and post-operative clinical parameters and the postoperative pathologic findings. Because of the lack of current consensus on follow-up strategies for the detection of EHR after resection, our prediction models may aid in monitoring patients for individual risk and in appropriately assigning patients for participation in clinical trials for postoperative adjuvant therapy (e.g., patients would be categorized as intermediate or high risk, if they present a predicted 5-year EHR rate of 50.5% or 100%, respectively). Recent studies have reported improved prognosis for recurrent HCC based on the pattern of IHR 5,16 . However, these studies did not clarify the clinical features and pathological course of EHR, with the absence of models to predict EHR after curative treatment for HCC, thus, limiting the early detection of EHR. In our study, we www.nature.com/scientificreports/ identified the risk factors of EHR after curative hepatectomy for HCC and used these factors to stratify the risk for EHR into four levels. Our findings highlighted the necessity for the development of a predictive score based on risk stratification to inform optimal surveillance for prompt detection of EHR to improve patient outcomes. The current results of HCC recurrence and EHR development rates suggested some distinction from our previous study of HCC patients who underwent RFA 17 . In that study, the median times to first HCC recurrence and EHR after RFA were 1.75 and 2.68 years, respectively. Moreover, the 1-, 3-, 5-, 8-, and 10-year rates of EHR development were 1.0%, 2.9%, 8.1%, 15.7%, and 33.7%, respectively. These rates were comparably lower than those reported after curative hepatectomy.

Figure 2.
Comparison of the (a) intra-and/or extra-hepatic recurrence rate of hepatocellular carcinoma (HCC) and (b) overall survival rates between the early extrahepatic recurrence (EHR), non-early EHR, and non-EHR groups. The 6-month cumulative recurrence rates among the early EHR, non-early EHR, and non-EHR groups were 77.7%, 39.7%, and 21.9%, respectively. The 1-year, 5-year, and 10-year overall survival rates were 51.6%, 9.7%, and 0% in the early EHR group; 98.4%, 58.7%, and 4.8% in the non-early EHR group; and 96.4%, 63.2%, and 13.2% in the non-EHR group, respectively (p < 0.001). www.nature.com/scientificreports/ Regarding the pattern of HCC recurrence, the most common initial site of recurrence was within the intrahepatic area, which was consistent with a previous report 18 . Another study identified IHR as the most common initial site of recurrence, with EHR developing after several treatments for IHR 19 . Uchino et al. reported that 82.2% of patients with HCC with EHR presented with IHR, a finding comparable to those of our previous study 17 . Therefore, multiple IHRs may indicate EHR risk in patients with HCC 20 .
In our study cohort, the lungs were the most common site of EHR (42.6%). Thoracic metastases, which included the lungs and the vertebrae of the thoracic spine, were, thus, relatively common as previously reported 21 . Thoracic metastases reflect a systemic involvement with poor prognosis, as they are largely not curable. Of further concern is the fact that thoracic metastases may not be detected using conventional surveillance methods that rely on abdominal imaging. Therefore, the use of chest CT images should be included in the surveillance strategy for patients at risk for EHR after curative hepatectomy for HCC for the early detection of thoracic metastases. In addition, the rate of EHR at the time of the initial recurrence of HCC after hepatectomy was 34%, with 38.3% of these patients having no sign of IHR at the time of EHR diagnosis. Therefore, even if intrahepatic HCC lesions are stable, close surveillance for possible development of EHR may be necessary.
Regarding the risk factors for EHR, macrovascular invasion, pathologic mUICC stage (III, IVa), large tumor size (sum > 7 cm), surgical margin involvement, and necrotic HCC were associated with EHR after curative hepatectomy. Vascular invasion is a well-known prognostic indicator of HCC. Natsuizaka et al. showed that vascular invasion was more frequently observed in patients with EHR at the first diagnosis of HCC 1 . Senthilnathan et al. also reported a two-fold increase in EHR in the presence compared to the absence of portal vein invasion (24% versus 12%) 22 . Yang et al. reported that EHR was more common among patients with vascular invasion,  www.nature.com/scientificreports/ intrahepatic metastasis, and more advanced HCC stages 11 . A recent study revealed that the presence of tumor necrosis was associated with an advanced tumor stage, HCC recurrence, and patient survival after curative hepatectomy for HCC 23 . In agreement with these findings, we also identified that necrotic HCC was associated with a high rate of EHR. HCC tumors > 6 cm in size were also predictive of EHR after curative resection for HCC, exhibiting similar results to those of a previous study 24 . We identified involvement of the margin of resection as a significant risk factor of EHR (HR: 4.035, 95% CI: 1.28-12.725), which was consistent with the findings of a previous study 25 .
In addition to the aforementioned risk factors for EHR, the first recurrence free survival of < 12 months (HR: 5.748, 95% CI 3.787-8.722, p < 0.001) and the serum AFP level > 400 IU/mL at the time of first recurrence during follow-up after curative hepatectomy (HR: 3.127, 95% CI 1.935-5.057, p < 0.001) were significantly associated   26 . They suggested that aggressive tumor pathology was, therefore, a risk factor of early HCC recurrence. Recent studies have shown that high AFP levels were independent risk factors of HCC invasiveness [27][28][29][30] . Similarly, our previous study on EHR in RFA for HCC also demonstrated an association between the AFP level and HCC recurrence when the AFP level was > 400 IU/mL, in line with our findings 17 .
After performing multivariate analysis, we found that the presence of EHR was significantly associated with the overall survival rates. Furthermore, pathologic mUICC stage (III, IVa) and surgical margin involvement, which were risk factors of EHR, were also related to the overall survival rate. This finding suggested a close correlation of the EHR with the overall survival rates. Moreover, HCC beyond Milan criteria was associated with the risk of developing HER, as previously reported 31,32 .
Regarding the early EHR, apart from the mentioned risk factors (i.e., advanced pathologic mUICC stage, surgical margin involvement, and macrovascular invasion), the serum ALP levels and absence of fatty change were also significantly associated with early EHR. Although there is still controversy concerning the association between the ALP level and HCC prognosis, a recent study showed that the serum ALP levels increase in liver disease and may reflect liver injury 33 . A recent meta-analysis showed that high pre-treatment ALP levels were significantly associated with poor overall survival rate (HR: 1.14, 95% CI: 1.10-1.18) and RFS (HR: 1.78, 95% CI: 1.37-2.31) 34 . This finding was concordant with our present result presenting an association of higher ALP level with early EHR. The frequency of diffuse steatosis in early HCCs peaks at a diameter of approximately 1.5 cm, decreasing as a function of increasing tumor size and grade 35 . Thus, diffuse fatty change is uncommon in HCCs > 3 cm and with progressing HCC disease status, and is also not usually observed in patients with poorly differentiated HCC 35,36 . Therefore, the absence of fatty change in the liver with HCC is associated with the tumor aggressiveness. In our study, the proportion of patients with the absence of fatty change in the liver was higher for the early EHR than for the non-early EHR group.
We stratified the risk for EHR into four levels based on the number of predictive factors present for EHR. We demonstrated that the cumulative rates of EHR and the median time to her were correlated with these four risk levels. Moreover, we showed that the overall survival rates were correlated with risk levels representing dismal prognosis in patients with EHR. Further, to assess the calibration of our model, the Brier score was calculated (Supplementary Table S3) and showed fine correspondence. Our novel parametric model, albeit simple, could assist clinicians in identifying patients at high risk for EHR before surgery.
We finally noted that 54.8% of the participants in the early EHR group exhibited EHR at the time of first recurrence. This underlines the importance of identifying patients who are at high risk for early EHR, which may be informative concerning the best strategy for surveillance after surgery for the rapid identification of EHR development.
The limitations of our study should be noted in the interpretation of results. First, the diagnosis of EHR was mostly based on medical imaging and, thus, the possibility of other primary cancers from an origin other than the liver could not be completely ruled out. However, in circumstances when the origin of the tumor was not certain, pathologic confirmation was performed at the discretion of the treating physician. Second, this was a retrospective study based on medical records from patients enrolled from two tertiary hospitals. Therefore, the effect of bias on results cannot be denied. Moreover, as this was a retrospective study, there was no concrete treatment algorithm strategy and criteria for resection. Third, there was no uniform post-treatment or surveillance schedule, and the surveillance modality used for each patient was at the discretion of the treating physician. www.nature.com/scientificreports/ Moreover, patients underwent different treatment modalities for local HCC recurrence depending on the tumor and patient status. There may be diverse conditions concerning the tumor stage, liver reserve function, and patients' physical performance status. However, we tried to overcome these limitations by using a considerable number of patients with a long-term follow-up duration in multiple tertiary centers. Finally, it might have been insufficient to predict extrahepatic recurrence of HCC using only our model in all patients with HCC. There is a need to conduct a well-organized prospective study to validate more effective methods of EHR prediction and management.
In conclusion, we present a simple parametric model to predict EHR after curative hepatectomy for HCC. This tailored approach may be useful for the early detection of EHR and permit a more refined estimation of risk on an individual basis.

Methods
Patients. Between January 2004 and December 2013, 493 consecutive patients who underwent surgical resection for HCC at two tertiary hospitals were evaluated for study enrolment (Supplementary table S4). The inclusion criteria were as follows: age ≥ 18 years; and HCC diagnosis. Surgical resection of HCC was determined by physician's decision regarding tumor stage and patient's physical status. There had been no changes in the keynote of surgical resection during study period in both centers. Patients with hepatic tumors other than HCC and those with follow-up duration shorter than 90 days were excluded. After screening, 398 patients were enrolled ( Supplementary Fig. S1). Baseline clinical and tumor characteristics, resection method, pathological findings, status of recurrence, and RFS were assessed retrospectively.

Statement of ethics.
The design of our cohort study was approved by the Institutional Review Board of Chonnam National University Hospital (IRB No. CNUH-2019-203). Owing to the retrospective design of our study and the use of de-identified data, the requirement for informed consent was waived under approval of the Institutional Review Board of Chonnam National University Hospital. The study was performed in compliance with the tenets of the Helsinki Declaration of 1975.
Data collection. The following information was extracted from patients' medical records for analysis: age, sex, underlying diseases (including hepatitis infection status and alcohol use), blood chemistry [including the Child-Pugh and model for end stage liver disease scores], serum AFP level, pathological findings (tumor size, histological grade, micro-vascular invasion, and presence of satellite lesions), and abdominal imaging for tumor staging at the time of HCC diagnosis. Data on blood chemistry, serum tumor markers, and abdominal imaging obtained at each follow-up visit were also obtained for analysis. Measured data at the time of first recurrence were also evaluated. Survival was defined as the time interval between the date of HCC diagnosis and that of death or last follow-up examination.
Baseline HCC staging, curative hepatectomy, and follow-up. HCC was diagnosed according to the guidelines of the Korean Liver Cancer Study Group and the National Cancer Center 37 . HCC staging at the time of diagnosis was determined using the mUICC staging system 38 and the Barcelona Clinic Liver Cancer (BCLC) classification system 39 . Milan criteria were based on radiologic findings at initial diagnosis of HCC 40 . Abdominal CT or magnetic resonance imaging (MRI) and assessment of serological tumor markers were routinely performed at 1 month after surgical resection and at each 3-6-month follow-up visit.
The tumor size was measured by radiologic modalities with CT or MRI. The histological differentiation of HCC was graded according to the criteria of Edmondson and Steiner 41 . The presence of macrovascular invasion was detected on CT imaging or MRI, while microvascular invasion was defined as invasion of vascular structures on microscopic analysis of resected tumor specimens. Bile duct invasion and tumor necrosis were confirmed from pathological findings of resected tumor specimens.
Post-operative complications after surgical resection were analyzed (Supplementary Table S5) and treatment strategies after first intra-and/or extra-hepatic recurrence and extrahepatic recurrence were also assessed ( Supplementary Fig. S2). Overall survival days were defined as the time interval between the date diagnosis of HCC and the date of death or last follow-up examination. The cause of death of each patient is presented in Supplementary Table S6.

Diagnosis of EHR.
Most cases of EHR were diagnosed during routine follow-up studies, with few cases diagnosed during evaluation of new onset symptoms or based on significant elevation in AFP or serial AFP without definite intra-hepatic lesions. The diagnosis of EHR was confirmed by contrast-enhanced CT imaging or MRI, and by pathological examination in some patients. In addition, bone scintigraphy, positron emission tomography-CT, and brain MR or CT imaging were also performed at the discretion of the treating physician. Chest radiographs were obtained routinely at the time of admission and when pulmonary symptoms were present. Early EHR was defined as EHR developing within the 1st year after initial curative hepatectomy.
Statistical analysis. The data are presented as means ± standard deviations or as medians and ranges, as appropriate for the data type and distribution. Univariate analyses were performed using chi-squared test or student's t-test, as appropriate. Variables with a p-value < 0.05 on univariate analyses were included in a multivariate logistic regression analysis to identify factors predictive of EHR. The multivariable Cox regression model was built using stepwise backward selection of variables, with variables having a p-value < 0.05 retained as predictive factors. The risk levels of EHR were defined based on the number of risk factors present, with Kaplan-Meier www.nature.com/scientificreports/ survival curves constructed for each risk level. An equivalent of the concordance statistic for use with Cox regression was calculated by Harrell's C-index. All statistical analyses were performed using SPSS software (version 25.0, IBM Corp., Armonk, NY, USA).

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.