Development and validation of nomogram to predict very early recurrence of combined hepatocellular-cholangiocarcinoma after hepatic resection: a multi-institutional study

In this retrospective multi-institutional study, a total of 221 patients received hepatic resection for cHCC between January 2011 and December 2015 were enrolled from 2 institutions, compromising Eastern Hepatobiliary Surgery Hospital of Second Military Medical University (Institution I) and Mengchao Hepatobiliary Hospital of Fujian Medical University (Institution II). This study was conducted in accordance with the ethical guideline of the 1975 Declaration of Helsinki and obtained approval from the Institutional Ethics Committee of the Mengchao Hepatobiliary Hospital of Fujian Medical University. Informed consent has been given to each participants enrolled in this study.

The inclusion criteria include (1) patients underwent curative hepatic resection for primary cHCC and pathologically confirmed as cHCC (only Allen and Lisa classification type C ); (2) Child-Pugh A or B liver function; (3) no previous anti-tumor treatment; (4) no extrahepatic metastasis; and (5) R0 resection, defined as complete removal of macroscopic tumor nodules with a clear margin. The exclusion criteria include (1) other malignant tumors; (2) incomplete clinical data; and (3) loss to follow-up within 12 months after the surgery.

One hundred thirty-one eligible patients form Institution I were assigned to development cohort for the construction of the predictive model and 90 eligible patients form Institution II were allocated to validation cohort for the verification of the models. The flow chart of this study was shown in Fig. 1.

All patients underwent hepatic resection, the type of which was determined by the tumor size, tumor location, preoperative diagnosis, and liver function. Hepatic resection without lymph node dissection was performed to the patients who were diagnosed as HCC preoperatively. Hepatic resection with lymph node dissection was performed to the patients who were diagnosed as ICC or cHCC preoperatively. Lymph node dissection was performed when an abnormal swollen lymph node around the hepatoduodenal ligament, the common hepatic artery, or behind the pancreas head was seen intraoperatively. Major resection was defined as resection of 3 or more Couinaud segments, while minor resection was defined as resection of fewer than 3 Couinaud segments [15].

The clinicopathological variables were listed as follows: gender, age, hepatitis B virus (HBV), hepatitis C virus (HBV), non-alcoholic fatty liver disease(NAFLD), red blood cell count, hemoglobin, platelet count, neutrophil count, lymphocyte count, neutrophils/lymphocytes ratio , aspartate aminotransferase, prothrombin time, total bilirubin, albumin, gamma-glutamyl transpeptidase, alkaline phosphatase, alpha-fetoprotein, carbohydrate antigen 19-9 level, carcinoembryonic antigen, decarboxylic prothrombin, liver cirrhosis, resection type, lymph node dissection, tumor number, maximum tumor size, tumor capsule, microvascular invasion (MiVI), macrovascular invasion(MaVI), satellite nodules (SN), Edmondson-Steiner classification [16], intraoperative blood loss, intraoperative blood transfusion, Child-Pugh classification, and AJCC 8th edition staging manual. We collected the serological examinations carried out within 2 weeks before the surgery. Two independent pathologists majored in hepatic tumors reviewed the resected specimens. MiVI was defined as the presence of tumor cell nests in the portal vein, the hepatic vein or the capsular vessel lined by endothelium visible only under microscopy [17], MaVI was defined as invasion of the first-and second- order branches of the portal veins, hepatic arteries or hepatic veins [18], SN was defined as tumor cell nests present on microscopy or tumors of which the maximum diameter is less than 2 cm resenting within 2 cm of the main tumor on macroscopy [19]. Tumor stage was defined in accordance with the AJCC 8th edition ICC staging manual [20].

After the surgery, patients were followed up every 3 months in the first 2 years and every 6–12 months thereafter. Serum tumor markers, contrast-enhanced magnetic resonance imaging (MRI) of abdomen were performed to diagnose the recurrence of cHCC. Patients with recurrence of cHCC received appropriate treatments including transarterial chemoembolization (TACE), radiofrequency ablation or hepatic re-resection. As the primary end point, recurrence-free survival (RFS) was defined as the interval from hepatic resection to recurrence. VER of cHCC was defined as recurrence within 6 months after surgery on the basis of previous studies [11‐13, 21]. We collected the survival information until March 31, 2020.

Categorical variables were presented as number (percentage) and compared using the chi-square test or Fisher exact test. Normally distributed continuous variables were presented as mean (standard deviation, IQR) and compared using Mann-Whitney U test. The P value less than 0.05 was set as statistical significance in this study. Statistical analysis was conducted by using R version 4.1.0 (http://​www.​r-project.​org/​) with R packages of Table 1, rms, pROC, ggplot2, shiny, survminer, and survival.

The optimal thresholds of continuous variables were determined by analyzing the ROC curves and Youden index. Clinicopathological variables showing potentially relevance (with p < 0.05 in the univariable logistic regression analysis) were employed for the multi-variable logistic regression analysis. The prognostic factors determined by using a stepwise selection method were integrated to construct the VER model. Nomogram and an online calculator are built to facilitate the clinical use of our model [22] (https://​chcrecurrence.​shinyapps.​io/​myCHCVER2). The VER risk scores of patients were calculated in accordance with the VER model, then the probabilities of VER were estimated by using the formula: VER probability= 1/{1+exp[−(VER risk score)]}. All patients were stratified into two risk groups depending on their VER probability by the cut-offs of 50th centile. Kaplan-Meier curves and log-rank test were employed to evaluate the difference in RFS between different subgroups.

To assess the predictive accuracy of our model, we drew the ROC curves and calculated the C-indexes of our model and AJCC staging for the development cohort and validation cohort with the bootstrapping resample method (n = 1000). We conducted the decision curve analysis(DCA) to estimate the clinical net benefits of our model and AJCC staging for both cohorts with the bootstrapping resample method (n = 1000) [23]. Calibration curves were employed to plot the predicted probabilities verses the actual outcomes.

A total of 363 cHCC patients underwent hepatic section in Eastern Hepatobiliary Surgery Hospital of Second Military Medical University (Institution I) and Mengchao Hepatobiliary Hospital of Fujian Medical University (Institution II) between 1 January 2011 and 31 December 2015 received careful reviews of their medical records, 221 patients were eligible for the study, 142 patients were excluded for receiving other anti-cancer treatment (n = 41), perioperative death (n = 4), history of liver cancer or other malignancies (n = 18), early lost to follow-up data (n = 19), lacking serological data within 2 weeks before surgery (n = 38), lacking other clinical data (n = 22). One hundred thirty-one eligible patients from Institution I were enrolled as the development cohort and 90 eligible patients from Institution II serve as the validation cohort. The flow chart of this study is shown in Fig. 1.

The clinicopathologic characteristics of patients in the development cohort and the validation cohort are summarized in Table 1. Variables including gender, cirrhosis, tumor capsule, PT, ALB, and Child-Pugh show differences between the two institutions. In the overall cohort, the development cohort and the validation cohort, respectively, 136 patients (61.5%), 82 patients (62.6%), and 54 patients (60%) had early recurrence (< 2 years); 96 patients (43.4%), 54 patients (41.2%), and 42 patients (46.7%) had 6-month VER; 121 patients (54.8%), 72 patients (60.0%), and 49 patients (54.4%) had 12-month recurrence.

The comparison of clinicopathologic characteristics between patients with and without VER is summarized in Table 2. We found that people developed VER tend to have these clinical factors: the performance of major resection, lymph node dissection, tumor size> = 5 cm, the presence of MiVI, MaVI, and SN, GGT> = 75 U/L, ALP > = 78 U/L, AFP > = 20 ng/mL, CA19-9 > = 25 U/mL, and DCP > = 200 mAU/mL.

The results of univariable logistic regression analysis and multi-variable logistic regression analysis of VER are listed in Table 3. The independent risk factors of VER including the presence of MiVI (odds ratio 3.73, 95% CI: 1.43–9.71, P = 0.007), the presence of MaVI (odds ratio 5.75, 95% CI: 1.73–19.04, P = 0.004), and CA19-9 > = 25 U/mL (odds ratio 2.48, 95% CI: 1–6.13, P = 0.049) were used to construct the VER model (Table 3). A nomogram integrating the independent risk factors described above was constructed to facilitate the use of our predicting model of the VER (Fig. 2).

The C-indexes of the prediction of the VER model in development cohort and validation cohort were 0.77(95% CI: 0.69–0.85) and 0.76 (95% CI: 0.66–0.86), while those of AJCC 8th staging was 0.64(95% CI: 0.57–0.72) and 0.64(95% CI: 0.54–0.74). P value less than 0.05 showed the significant difference, which signified the superiority of VER model than AJCC 8th staging (Fig. 3A, B). In addition, the VER model showed a better net benefit than AJCC 8th staging in both cohorts (Fig. 4A, B). Calibration analysis also displayed a favorable agreement between the prediction and actual outcome of VER in both cohorts (Fig. 5A, B). Using a 50th centile of the VER risk score, 3.96, in overall cohort, development cohort, and validation cohort, respectively, we stratified the patients into two risk groups with significant recurrence outcome (P < 0.001) (Fig. 6A–C).