Integrated nomograms to predict overall survival and recurrence-free survival in patients with combined hepatocellular cholangiocarcinoma (cHCC) after liver resection

Clinical characteristics of the study patients

After careful reviews of the medical records, a total of 212 patients with cHCC confirmed by pathology who underwent curative liver resection were eligible for this study. All cHCC patients were followed up after initial treatment until December 2018. Among the entire set, the median OS was 16.5 months (range: 4.9–84.6 months). The 1-, 2- and 4-year OS rate were 79.7%, 27.4% and 8.5%, respectively. The demographic and clinical characteristics of the patients in the training and validation sets are listed in Table 1. The baseline demographic and clinical characteristics of the patients in the training and validation sets were similar (P>0.05). There was no significant difference in the cHCC patients RFS and OS between two groups (Supplementary Figure 1).

Independent factors significantly associated with OS and RFS

The results of univariate and multivariate Cox regression analyses of RFS and OS after curative resection of cHCC are listed in Table 2 and Table 3. In the univariate analysis of OS in the training set, sex, maximum tumor size, multiple tumors, MVI, macroscopic vascular invasion, SAT, LN metastasis were associated with shorter OS (all P < 0.05, Table 2), Parameters with P < 0.05 in univariate analyses were subjected to the multivariate Cox regression model for stepwise variable selection to determine the prognostic factors. Multivariate Cox regression analysis identified that tumor number (multiple vs. single, HR,1.851; 95% CI, 1.188-2.883; P =0.006), MVI (Yes vs. No, HR,1.938; 95% CI, 1.273-2.950; P=0.002), SAT (Yes vs. No, HR, 2.152; 95% CI,1.336-3.466; P=0.002), and LN metastasis (Yes vs. No, HR, 2.397; 95% CI,1.391-4.131; P=0.002) were independent predictors of OS in patients after curative resection of cHCC, as listed in Table 2. The univariate analysis of RFS in the training set showed that age, maximum tumor size, tumor number, MVI, macroscopic vascular invasion, SAT, LN metastasis were associated with an increased tumor recurrence rate (all P< 0.05). Multivariate Cox regression analysis identified that maximum tumor size (>5 vs. ≤5 cm, HR,1.507; 95% CI, 1.030-2.204; P=0.035), tumor number (multiple vs. single, HR,1.972; 95% CI, 1.290-3.016; P =0.002), MVI (Yes vs. No, HR, 1.528; 95% CI,1.024-2.279; P=0.038), SAT (Yes vs. No, HR, 1.953; 95% CI,1.259-3.029; P=0.003), and LN metastasis (Yes vs. No, HR, 2.059; 95% CI,1.228-3.453; P=0.006) were independent predictors of RFS in patients after curative resection of cHCC, as listed in Table 3. Based on the results of the multivariate analysis, a nomogram integrating all significant independent factors was constructed to predict OS and RFS for cHCC patients, as shown in Figure 1. The C-index of prediction of OS and RFS in training set were 0.685 (95% CI, 0.638 to 0.732) and 0.685 (95% CI, 0.639 to 0.731), respectively. The C-index of prediction of OS and RFS in validation set were 0.654 (95% CI, 0.567 to 0.741) and 0.669 (95% CI, 0.582 to 0.756), respectively.

As shown in Supplementary Figure 2, the BCLC stage, the 8^th edition HCC AJCC TNM stage, the 8^th edition ICC AJCC TNM stage, and CNLC stage have had good prognostic stratification ability for patients between stage I and later stages in the training group. However, the current stage systems did not perform well in the prognostic stratification of advanced cHCC with later stages in the training group.

The C-index of the nomogram predicting OS in the training set was significantly higher than that of the BCLC staging system(0.601, 95% CI: 0.547 to 0.655, P<0.001), 8^th edition AJCC HCC TNM staging system(0.625, 95% CI: 0.574 to 0.676, P= 0.013), 8^th edition AJCC ICC TNM staging system(0.593, 95% CI: 0.542 to 0.644, P<0.001), and CNLC staging system(0.604, 95% CI: 0.550 to 0.658, P<0.001). The calibration plot for the probability of OS and RFS at 1, 2 or 4-years after surgery showed an optimal agreement between the nomogram prediction and actual observation in the training set and validation set, as shown in Supplementary Figure 3. Furthermore, we found that our nomogram had a better net benefit across a wider scale of threshold probabilities for predicting 1-, 2- and 4-year overall survival than the BCLC staging system, the 8^th edition AJCC staging system (HCC and ICC), CNLC staging system in the DCA, as shown in Supplementary Figure 4.

Independent factors significantly associated with the presence of SAT and MVI

Furthermore, in order to predict SAT and MVI well before surgery and guide clinical decision making, univariate logistic regression analysis was performed to estimate the impacts of clinical and imaging features on the presence of SAT and MVI in the training set. According to univariable logistic regression analysis, maximum tumor size, tumor encapsulation, and the Ishak fibrosis score were associated with SAT. Additionally, univariate logistic regression analysis showed that portal hypertension, MCI, tumor encapsulation, and the Ishak fibrosis score were associated with MVI. Stepwise multivariate logistic regression analysis was further performed to identify significant independent risk factors. The multivariate analyses revealed that maximum tumor size (>5 vs. ≤5 cm, OR, 2.484, 95%CI, 1.011-6.107, P =0.047), tumor encapsulation (OR, 2.914, 95%CI, 1.239-5.725, P =0.014), Ishak fibrosis score (OR, 2.421, 95% CI, 1.024-5.725, P=0.044) were identified as independent risk factors for (Table 4). In addition, portal hypertension (OR, 2.477, 95%CI, 1.073-5.718, P =0.034), MCI(OR, 2.406, 95%CI, 1.067-5.423, P =0.034), hepatic encapsulation (OR, 2.563, 95%CI, 1.171-5.611, P =0.019) were independently associated with MVI (Table 5).. To elucidate and synthetically estimate the significant values for the blood indexes, the LASSO logistic regression algorithm was used to select the candidate blood index, Using the coefficients derived from the LASSO logistic regression models in the training set, we then constructed a formula to calculate for each patient. The LASSO coefficient profiles of the selected blood features are shown in Figure 2. The blood signature score is based on their personalized levels of the 25 blood features, where the blood-satellite score=0.007× activated partial thromboplastin time (APTT) -0.001× platelet (PLT) +0.523×international normalized ratio (INR)-0.257×total bilirubin (TBIL)-0.003×aspartate transaminase(AST)+0.031×globulin (GLB)-0.081 × albumin (ALB) + 0.001 × CA199 + 1.053 × HBV-DNA +0.427, and the blood-MVI score = 0.037×prothrombin time (PT)+0.032×TBIL-0.016×ALB-1.105, as listed in Supplement Table 1. Using the ROC curve, we classified patients into a type A^SAT group and type B^SAT group with a blood signature score of -1.228 as the cut-off value. We further classified patients into a type A^MVI group and type B^MVI group with a blood signature score of -0.605 as the cut-off value, as shown in Supplementary Figure 5. Based on the results of the blood signatures and multivariate logistic regression, nomogram for predicting SAT and MVI was established (Figure 3). The model had good predictive ability for SAT and MVI. The C-index for the nomogram for the prediction of SAT was 0.826 (95% CI, 0.743 to 0.909) for the training set and 0.778 (95% CI, 0.630 to 0.926) for the validation set, while the C-index for the nomogram for the prediction of MVI was 0.771 (95% CI, 0.688 to 0.854) for the training set and 0.702 (95% CI, 0.572 to 0.832) for the validation set. The calibration curve for nomograms showed good agreement between predictions and observations for SAT prediction in the training set and validation set, as well as for MVI prediction in the training set and validation set (Figure 4). The decision curves of the nomograms for predicting the presence of SAT and MVI are presented for the training set and validation set. (Figure 5).

Patients and study design

A total of 212 consecutive patients (174 men, 38 women) who underwent curative liver resection for cHCC between January 2006 and December 2017 at West China Hospital were enrolled in this study. The inclusion criteria were as follows: (1) liver resection, with tumor tissues pathologically confirmed as cHCC mixed cancer, not double cancer type or collision cancer; (2) Child-Pugh A or B7 (score ≤7 [less than or equal to]) liver function; (3) curative liver resection, defined as the complete removal of all macroscopic nodules with a clear margin (R0 resection); and (4) available detailed clinical characteristics. The exclusion criteria were (1) a history of extrahepatic malignancies and (2) poor clinical data integrity. Eligible patients(n=212) who underwent surgery were assigned to the training and validation sets at a ratio of 7:3 according to the scanning date: the early data before the 70 percent scanning date were allocated to the training set(n=148) for the development of the nomogram, whereas the other patients were allocated to the validation set(n=64) for the verification of the nomogram. The flowchart of this present study selection is shown in Figure 6 and the clinicopathologic characteristics of patients in the training and validation sets are listed in Table 1. This study obtained ethics approval from the ethics committee of Sichuan University and was performed in accordance with the 1975 Declaration of Helsinki. Written informed consent was obtained from each participant in the study.

Data collection

The clinical medical data of cHCC patients who underwent curative liver resection were retrospectively collected from our hospital and included demographics, comorbid illnesses, portal hypertension, liver and renal function tests, hepatitis B and C immunology, HBV-DNA load, preoperative α-fetoprotein (AFP) level, preoperative serum carbohydrate antigen 19-9 (CA19-9) level, preoperative serum carcinoembryonic antigen (CEA) level, imaging data of tumors (including the maximum tumor size, tumor number, tumor location, and encapsulation status), pathological results of cHCC (including the differentiation grade, microvascular invasion (MVI), SAT, LN metastasis and Ishak fibrosis score), and surgery-related factors including the extent of liver resection (major or minor), intraoperative blood transfusion (yes or no), Barcelona Clinic Liver Cancer (BCLC) stage, 8^th American Joint Committee on Cancer (AJCC) TNM clinical stage(HCC and ICC) and China liver cancer (CNLC) stage [41–43]. Comorbid illnesses included diabetes mellitus, hypertension, and hyperlipidemia. Portal hypertension was defined by the presence of either esophageal varices or splenomegaly with a decreased platelet count (100 × 10⁹/L or less). MVI was defined as the presence of tumor in a portal vein, hepatic vein, or a large capsular vessel of the surrounding hepatic tissue lined by endothelium that was visible only on microscopy [44]. Macroscopic vascular invasion included major hepatic vessel invasion, defined as invasion of the first-and second-order branches of the portal veins or hepatic arteries, or as invasion of one or more of the three hepatic veins. 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 [45]. The Ishak scoring system uses a 0-6 scale; F0 is defined as a fibrosis score 0-4 (no to moderate fibrosis), and F1 is defined as a fibrosis score 5 -6 (severe fibrosis or cirrhosis);

The extent of liver resection was determined according to the location of the tumor, tumor diameter, liver function and indocyanine green retention rate at 15 minutes (ICG-R15). The resection of the liver parenchyma was performed with an ultrasonic scalpel, CUSA, monopolar electrocoagulation, LigaSure, Endo-GIA and clips. Intermittent Pringle manipulation or a selective vascular clamp was used if necessary. Regional lymph nodes were dissected if metastasis was was suspected or diagnosed preoperatively or found intraoperatively. Intraoperative ultrasonography was undertaken routinely to identify lesion(s) and the relationship to surrounding vascular and biliary structures and determine whether there were still additional lesions that could not be seen by preoperative imaging findings in the remnant liver.

Follow-up and recurrence treatment

In general, all patients who received curative liver resection were prospectively followed up through outpatient clinic visits or phone calls at intervals of 2-3 months during the first year after operation and 3-6 months thereafter. Chest CT examination and bone scintigraphy were performed when extrahepatic cHCC recurrence was suspected. Recurrent cHCC was treated with postoperative adjuvant therapy, repeated liver resection, radiofrequency ablation, and liver transplantation, depending on the status of the cHCC and liver function at the time of recurrence. In addition, for patients with high-risk factors for tumor recurrence, we recommended patients to undergo adjuvant therapy after surgery. Postoperative adjuvant therapy included systemic chemotherapy (mainly 5-fluorouracil) and transarterial chemoembolization (TACE). Survival information, including OS and RFS, was collected until December 31, 2018. OS was defined as the interval between resection and death, or the period up to the observation point. RFS was identified as the interval between resection and the recurrence of the primary tumors detected by dynamic radiological findings including intrahepatic recurrence and extrahepatic metastasis. The OS and RFS were measured in months.

Risk factors for OS and RFS

Univariate and multivariate Cox regression analyses were used to identify the independent risk factors of cHCC. Parameters with P < 0.05 in univariate analyses were included in the multivariate Cox regression model for stepwise variable selection to determine the prognostic factors. A nomogram was constructed based on the results of multivariate analysis. A calibration curve was used to describe the consistency of the nomogram predictions of 1-year, 2-year, and 4-year OS and RFS with the actual values. Harrell's concordance index (C-index) was used to quantify the performance of this nomogram. Bootstraps with 1,000 resamples were used for these analyses. In addition, we used clinical data from the validation set for validation. The total points for each patient in the validation set were calculated with the established nomogram, and the C-index and calibration curve were derived based on the regression analysis. Decision curve analysis (DCA) was used to determine the clinical application value of the nomogram models by calculating the net benefits at each risk threshold.

Risk factors for presence of satellite nodules and MVI

The least absolute shrinkage and selection operator (LASSO) logistic regression model was used to build a prognostic classifier for SAT and MVI in the training set, which integrated all types of serological variables. Using the coefficients derived from the LASSO logistic regression models, we then constructed a formula to calculate a score for each patient. Formula=expressionindex1× βindex1+…+ ex-pressionindexn × βindexn (where β is the regression coefficient derived from LASSO regression). We use the receiver operating characteristic (ROC) curve with calculations of the area under the curve (AUC) to determine the optimal cut-off value of the blood signature index. Logistic regression analysis was used to evaluate the outcomes based on the identified variables and other clinically relevant variables (odds ratio [OR], 95% confidence interval [CI]). After univariate analysis, selected variables with a P value < 0.10 were considered for inclusion in multivariate regression analysis to investigate the factors related to SAT and MVI. In the multivariate regression model, the P value was set at 0.05. In addition, in the multivariable logistic model, the Hosmer-Lemeshow goodness of fit test was also applied. The nomogram was drawn using the results of the multivariable logistic regression model for SAT and MVI. The predictive accuracy of the models was measured using the C-index, quantifying the level of agreement between the predicted probabilities and the actual possibility of having the event of interest, and the bootstrap estimate of slope shrinkage [46]. The Bootstrap resampling method was chosen for the internal validation of the predictive models’ selecting 1000 repetitions. DCA was performed to determine the clinical application value of the nomogram models by calculating the net benefits at each risk threshold probability [47].

Categorization of patients with different conventional staging Systems

Eligible patients were categorized according to four conventional staging systems (the BCLC staging system, the 8^th edition of the AJCC TNM classification system (HCC and ICC) and the CNLC staging system). We conducted a group-stratified analysis to compare the discriminative ability of the nomogram with that those of the other staging systems in the training and the validation sets and were evaluated by the C-index. The larger the C-index was, the more accurate was the prognostic prediction.

Statistical analysis

The Mann-Whitney U test was used to compare continuous variables between two patient groups. The chi-squared test and two-tailed Fisher's exact test were used for the comparison of categorical variables data between two groups. Continuous variables are expressed as medians and ranges, and categorical variables data are expressed as numbers and percentages. The OS and RFS were calculated with the Kaplan-Meier and Log-rank method using GraphPad Prism 8.0 software. R version 3.6.1 (http://www.r-project.org/) was used for ROC curve analysis, LASSO logistic regression, nomogram generation, C-index assessment, calibration plot generation, DCA, and clinical impact curve analysis. The rest of the analyses were conducted using SPSS statistical software version 24.0 (IBM Corporation, Armonk, NY). In all analyses, P < 0.05 was considered to indicate statistical significance.

Synopsis

There is little information on the surgical outcomes of and prognostic factors for cHCC from previous research. Using the clinical data obtained at West China Hospital, the authors discovered prognostic factors of this malignancy for cHCC. Moreover, a nomogram was established by combining clinical risk factors using least absolute shrinkage and selection operator (LASSO) regression that can objectively and accurately predict the preoperative risks of individualized satellite nodules and microvascular invasion.