This study enrolled 862 HCC patients who received ablation treatment at Beijing You’an Hospital, Capital Medical University, between January 1, 2015, and December 31, 2022. The diagnosis of HCC was according to the American Association for the Study of Liver Disease (AASLD) guideline. The inclusion criteria were as follows: (1) age from 18 to 80 years; (2) complete clinical and follow-up data; (3) BCLC stage 0, A or B; (4) Child-Pugh class A or B. Exclusion criteria: Patients were excluded if they met any one of the following criteria: (1) received other treatment before ablation; (2) distant metastasis of HCC; (3) secondary liver cancer; (4) coagulation function disorders or serious diseases of vital organs, such as the heart, brain, lung, and kidney.

Approval for this research was obtained from the Ethics Committee of Beijing You’an Hospital, Capital Medical University, ensuring adherence to the Declaration of Helsinki guidelines. The requirement for informed consent was waived due to the retrospective nature of this study.

Clinical characteristics were gathered before the local ablation, including: (1) demographic data: age, gender, history of smoking, drinking, antiviral therapy, hypertension, and diabetes; (2) tumor information: tumor number (TN), tumor size (TS) and AFP. (3) liver function indicators: Child-Pugh class, history of liver cirrhosis, alanine aminotransferase (ALT), aspartate transaminase (AST), total bilirubin (TBIL), direct bilirubin (DBIL), gamma-glutamyl transferase (GGT), albumin (ALB), alkaline phosphatase (ALP), prealbumin (Palb) and globulin (Glob). (4) other characteristics: prothrombin time (PT), thrombin time (TT), prothrombin time activity (PTA), activated partial thromboplastin time (APTT), red blood cell (RBC), white blood cell (WBC), and hemoglobin (Hb).

Tumor number and maximum tumor diameter were assessed based on preoperative multimodal imaging, including contrast-enhanced ultrasound, computed tomography (CT), and/or magnetic resonance imaging (MRI). These evaluations were performed by experienced radiologists who were blinded to the patients’ pathological outcomes and clinical prognoses, thereby minimizing interpretation bias and ensuring assessment consistency. Preoperative blood samples were collected and analyzed at our institutional central laboratory using standardized platforms and commercial assay kits to measure AFP, liver function parameters, and coagulation profiles. This centralized testing approach was implemented to reduce inter-assay variability and ensure data uniformity and reliability across all study participants.

In our study, the previously reported WATS score was used as the core variable to stratify patients into high-risk and low-risk groups according to its median value, for the prediction of OS in HCC patients undergoing local ablation. WATS = 0.73 × tumor number + 0.17 × tumor size + 0.10 × ln AFP (10).

Patients with HCC were recommended to undergo regular follow-ups in line with clinical guidelines after local ablation. Typically, these patients were followed up 3 to 6 months, which comprised clinical characteristics and adverse events. OS, as the primary endpoint of this study, was defined as the interval from local ablation to either the occurrence of death or the last follow-up.

All enrolled patients were treated with local ablation, which was performed by qualified hepatologists and interventional radiologists. The specific process includes 5 items: (1) Appropriate position for ablation was determined by CT or MRI. (2) The ablation needle was inserted in the marked skin, and followed by image scanning to track the ablation process. (3) For the purpose of attaining complete ablation, operators should expand the ablative range and contemplate multiple sites, overlapping, or repeated ablation. (4) In order to prevent tumor implantation and postoperative bleeding, the needle track required to be heated in the final stage. (5) Following the ablation, all patients underwent imaging examinations to assess treatment efficacy and possible complications.

Categorical variables were expressed as frequencies (percentages), whereas continuous variables were reported as means ± standard deviation or medians (quartiles). Comparisons between two groups were analyzed by Student’s t test, Chi-square test, or non-parametric test as appropriate. All statistical analyses were conducted using R software (version 4.5.0). All tests were two-tailed and statistical significance was set at P < 0.05.

Firstly, the Kaplan-Meier (KM) curve and log-rank test were used to assess the performance of WATS score in HCC OS. To address potential multicollinearity among variables, select the most predictive and parsimonious set of features, and mitigate the risk of overfitting, we employed Lasso regression analysis, incorporating 10-fold cross-validation. The optimal value of the regularization parameter (λ) was determined by minimizing the cross-validated partial likelihood deviance. Variables retaining non-zero coefficients after Lasso penalization were selected as candidate predictors. These Lasso-selected variables were subsequently incorporated into a multivariate Cox proportional hazards regression model. Variables that remained statistically significant (typically defined as P < 0.05) in this multivariate model were identified as independent prognostic factors for OS. Based on the results of the final multivariate Cox regression model, we constructed a nomogram using the rms package in R software. This nomogram provides a graphical representation of the multivariate Cox model. Each independent prognostic factor, including the WATS score as a key variable, is assigned a specific point value on the “Points” scale at the top of the nomogram. The points for each predictor are summed to yield a total point score. This total score is then projected downward to the lower scales of the nomogram to estimate the probability of OS at specific time points (3-year, 5-year, and 8-year). In order to determine the relative importance of the variables, we employed a statistical machine learning approach, Shapley Additive Explanations (SHAP). Furthermore, receiver operating characteristic curves (ROC) were plotted and the area under the curves (AUC) were calculated to evaluate the discrimination of the model. The calibration and decision curve analysis (DCA) were then used to validate the calibration performance and clinical utility. At last, patients were classified into low-risk and high-risk groups on the basis of the nomogram. KM curves were applied to compare the OS between two groups.

During the period from January 1, 2015, to December 31, 2022, a total of 862 HCC patients after local ablation were recruited in this study and randomized into two groups with a 7:3 ratio. There were 603 patients constituted the training set, while 259 formed the validation set ( Table 1 ). The result revealed no statistical differences between the two groups, except the history of diabetes. Our research concluded patients’ last follow-up on July 1, 2024, with a median follow-up duration of 52.7 months.

Among these patients, the average age was 56.60 years, with 685 (79.5%) males and 177 (20.5%) females. Notably, 377 (43.7%) patients had a history of smoking and 289 (33.5%) had a history of drinking. Furthermore, 222 (25.8%) individuals were diagnosed with hypertension and 180 (20.9%) with diabetes mellitus. The training and validation cohorts both had a majority of patients in Child-Pugh class A (76.3% and 75.3%, respectively).

According to the median value of WATS score, patients could be classified into low-risk group (n=431) and high-risk group (n=431). The comparison of baseline characteristics between two groups was shown in Table 2 . There were significant differences between two groups in terms of the WATS score.

The KM curve was plotted in Figure 1 for comparing the prognosis between two groups. The median OS was not reached in low-risk group and 92.0 months (95% CI: 85.1 months to not reached) in high-risk group, respectively. The cumulative OS rates for 3-, 5-, 8-year were 93.0%, 80.9% and 68.1% in the low-risk group, while 83.3%, 67.9% and 49.1% in the high-risk group (P < 0.0001).

Lasso regression, utilizing a loss function with L1 regularization to penalize model coefficients while minimizing the objective function, was employed to screen risk factors associated with OS ( Figure 2 ). The 10-fold cross-validation method was applied to select the optimal λ value, which was determined to be 0.0283 (Log λ = -1.548). Significant risk factors filtered by Lasso regression were age, history of drinking, Child-Pugh class, WBC, RBC, AST, Glob, GGT, Palb, PT, PTA, TT, and WATS. These variables were further incorporated into the multivariable Cox regression analysis, revealing that WATS (HR:1.334, 95% CI: 1.113-1.600), age (HR:1.031, 95% CI: 1.011-1.052), history of drinking (HR:1.449, 95% CI: 1.058-1.985) and Palb (HR:0.996, 95% CI: 0.992-0.999) as the independent prognostic factors for OS ( Table 3 ).

In order to assess the specific impact of each variable as a risk factor on OS, we employed the global interpretation method to construct SHAP. This approach quantifies the contribution of each feature to the model and identifies the prediction contribution of the model to a broader range of outcomes. The risk factors in Lasso-cox were further presented ( Figure 3A ), and the correlation heat map shows the correlation between factors ( Figure 3B ).

Based on these independent prognostic factors, we constructed a nomogram for predicting the OS of HCC patients who received local ablation ( Figure 4 ). Every risk factor corresponds to a specific score according to its value on the nomogram. It was necessary to sum the scores of factors and draw a vertical line at the corresponding total point. After these steps, the vertical line intersects with three lines representing mortality risk, which forecast the 3-, 5-, and 8-year OS.

In the training cohort, the time-dependent ROC curve was drawn and the C-index in the training set was 0.649 (95% CI: 0.606-0.692). It showed that AUCs of 3-, 5-, and 8-year were 0.679, 0.683, and 0.708, respectively ( Figure 5A ). These outcomes highlighted the advantageous discriminative ability. Additionally, a calibration curve ( Figure 6A ) and DCA curves ( Figures 7A–C ) were created, affirming that the nomogram demonstrated good calibration and clinical utility. According to the nomogram, patients were categorized into low-risk group (n=301) and high-risk group (n=302). The Kaplan-Meier curve was plotted, indicating that the median OS was 82.6 months for the high-risk group, while it was not reached in the low-risk group ( Figure 8A ). The cumulative OS rates for 3-, 5-, and 8-year were 83.7%, 64.3%, and 45.0% in the high-risk group, while 94.1%, 84.5%, and 73.2% in the low-risk group. There existed an obvious distinction in OS among the two groups (P < 0.0001).

In order to further validate the reliability of this nomogram, we performed internal validation in our study. The C-index in the validation cohort was 0.655 (95% CI: 0.542-0.676) and the AUCs for 3-, 5-, and 8-year were 0.701, 0.663, and 0.633, which suggested the favorable diagnostic value ( Figure 5B ). The calibration curve exhibited a good match ( Figure 6B ), and the DCA curves also had good clinical practicability ( Figures 7D–F ). Patients in the validation cohort were then also classified into two groups in light of the nomogram: low-risk group (n=129) and high-risk group (n=130) ( Figure 8B ). The median OS were not reached in both of the low-risk group and the high-risk group. The cumulative OS rates for 3-, 5-, and 8-year were 79.8%, 68.6%, and 50.6% in the high-risk group, while 93.5%, 84.5%, and 74.7% in the low-risk group. In concordance with the training cohort, there was also a statistically significant discrepancy in OS among the two groups (P = 0.00058).