Development and validation of a small-sample machine learning model to predict 5-year overall survival in patients with hepatocellular carcinoma

Development and validation of a small-sample machine learning model to predict 5-year overall survival in patients with hepatocellular carcinoma | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development and validation of a small-sample machine learning model to predict 5-year overall survival in patients with hepatocellular carcinoma Tingting Jiang, Xingyu Liu, Wencan He, Hepei Li, Xiang Yan, Qian Yu, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6176116/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 01 Jul, 2025 Read the published version in BMC Cancer → Version 1 posted 10 You are reading this latest preprint version Abstract Background: Early-onset hepatocellular carcinoma (HCC) is insidious, with characteristics of easy metastasis, high recurrence rate, and significant mortality. To address the substantial time and resource demands associated with HCC prognostic prediction, we extract meaningful insights from limited small-sample data to develop and validate a prediction model for HCC 5‑year overall survival (OS) by machine learning (ML). Methods ：76 newly diagnosed patients with HCC were eventually enrolled between September 2018 and July 2019. The follow-up time was 1-67 months. Patients who survived for 5 years after the first surgery, were divided into a surviving group (n=34) and a nonsurviving group (n=42). The pathological data and related survival factors of patients were collected before treatment. The final subset of features was filtered. Prediction models for 5-year OS in patients with HCC were established by logistic regression (LR), support vector machine (SVM), decision tree classification (DTC), random forests (RF), and extreme gradient Boosting (XGBoost), respectively. Additionally, the optimal model was established after rigorous validation. The models were evaluated by values of specificity, F1 score, recall, accuracy and area under the receiver operating characteristic curve (AUC-ROC). Results: The significant variable set, which included 22 variables, was screened. Ranking the importance of variables, the top 22 characteristic variables were as follows: the maximum diameter, presence or absence of distant metastasis, CNLC stage, ALB, age, RBC, the large sizeCTC, total bilirubin (TBIL), PD-L1 (-) CTC, ≥ Pentaploid CTC, AFP, vascular cancer thrombus and satellite nodules, WBC, CTC, BCLC stage, multiple nodules, AST, PD-L1 (-) CTC-WBC cluster, Triploid CTC, LYM, PD-L1 (-) CEC-WBC cluster and degree of cirrhosis. The AUC-ROC values for predicting the 5-year OS rate of HCC patients by the logistic regression, SVM, DTC, RF, and XGBoost models were 0.737, 0.971, 0.657, 0.741, and 0.703, respectively. Among them, the SVM model had the best performance (Accuracy=0.987, F1 score=0.988, Recall value =1.000). Conclusion: The SVM model could predict the 5-year OS in HCC with good recognition ability and achieves significantly greater accuracy compared to traditional models. Diagnosis and treatment could be utilized to intervene in the risk factors in this model, thereby improving patient prognosis. HCC OS Small-Sample ML Clinicopathological Characteristic Hematological Biomarker CTC CEC Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Hepatocellular carcinoma (HCC) is the main pathological subtype of primary liver cancer (PLC). HCC is a malignant tumor that originates from liver cells and bile duct epithelial cells. In 2023, HCC ranks fifth in terms of cancer prevalence but is the third leading cause of mortality among malignancies [ 1 ] . In 2022, there were 367700 new cases of HCC in China, ranking fourth among new cases of malignant tumors [ 2 ] . The number of deaths from HCC reached 316500, making it the second leading cause of cancer-related deaths in China [ 3 ] . Although advancements have been made in both fundamental research and clinical diagnosis and treatment in recent years, the OS of HCC patients remain poor. The recurrence rate within 5 years is nearly 70%, while the 5-year survival rate is only 5%. Currently, surgical resection and liver transplantation are available interventions for HCC, but the prognostic methodologies for HCC lack precise physical and chemical indicators to timely determine its potential recurrence risk, leading patients to miss the potential treatment opportunities [ 4 – 5 ] . Therefore, it is imperative to develop novel biomarker-based models for evaluating the recurrence risk of HCC. OS analysis is frequently employed in clinical research to identify potential risk factors and predict various outcomes using statistical methods. LR and Cox regression are commonly utilized to assess the impact of specific features on particular events. However, this linear assumption oversimplifies the complex relationship between predictive factors and prognosis. In recent years, with the continuous advancements of artificial intelligence (AI), machine learning (ML) methods have been able to aid physicians in disease diagnosis, disease staging [ 6 ] , risk factor prediction [ 7 ] , treatment planning [ 8 ] , and prognosis analysis [ 9 – 10 ] . Serum tumor markers are typically generated and secreted into circulation by neoplastic tissues. CTCs have emerged as a pivotal tumor marker in 'liquid biopsy', enabling the detection of CTCs in peripheral blood and paving the way for precise patient treatment. CTCs are rare tumor cells that detach from primary or metastatic solid tumor lesion and enter the peripheral circulatory system. When two or more CTCs are clumped together, they form circulating tumor thrombi (CTM) [ 11 – 12 ] . Within the peripheral blood system, in addition to blood cells and CTCs, there are endothelial cells (CECs), which are single endothelial cells derived from blood vessels. The coalescence of two or more CECs is referred to as CEC cluster. Both CTCs and CECs aren’t only associated with the development of tumors, but also intricately linked to tumor recurrence, metastasis and survival. The quantifications of CTCs and CECs levels are widely employed for early screening, diagnostic staging, treatment efficacy assessment, prognostic evaluation, and personalized therapy in various solid tumors [ 13 – 15 ] . Unveiling the occurrence, development and prognosis of HCC requires a considerable amount of time and resources, including large sample sizes, long-term follow-ups and costly experiments. To address these issues, small-sample learning has emerged as a potentially effective approach, particularly in scenarios where sample recruitment is difficult or analysis costs are high. The crucial to improving the survival rate and prognosis of HCC patients in this study lies in the selection of appropriate epidemiological methods. In this study, based on real-world data, the 5-year survival status of HCC patients was predicted for the first time using a combination of patient pathological, hematological, CTCs, and CECs variables. We established a 5-year OS model for newly diagnosed HCC patients and identified a more sensitive set of variables. This model had the potential to improve prognostic accuracy and provide a theoretical basis for treatment planning in HCC patients. Materials and Methods Data collection The baseline data of the patients were collected, including demographic characteristics, admission status, medical history, personal history of smoking and drinking, and degree of liver cirrhosis.This retrospective observational study included 76 patients diagnosed with HCC who underwent surgical treatment at West China Hospital of Sichuan University between September 2018 and July 2019. The cohort comprised 64 males (84.21%) and 12 females (15.79%), a median age of 55.0 years and a mean age of 55.48±12.02 years. Pathological staging and grading were conducted according to the Chinese Liver Cancer Staging (CNLC) system. The inclusion criteria were as follows: ①HCC diagnosis was confirmed through pathological examination. ②The patient had no previous history of other tumors or exposure to radiotherapy or chemotherapy. ③There was no vascular diseases, severe heart disease, kidney disease, abdominal infection, hepatic failure or other complications. ④The patient’s WBC count was within the normal reference range of (4.0-10.0)×10^9/L. The exclusion criteria were as follows: ①Patients who have previously received anti-tumor treatment. ②History of other tumors. ③History of autoimmune diseases. Treatment plans Individualized standard treatment plans were developed based on PCL guidelines, patients’overall health status, and their personal preferences. Among patients undergoing radical resection for PCL, 35 cases (46.05%) received conservative medical management, 27 cases (35.53%) underwent interventional therapy combined with local ablation, 5 cases (6.58%) had simple surgical resection, 4 cases (5.26%) received a combination of surgical resection and interventional therapy, 3 cases (3.95%) underwent liver transplantation, and 2 cases (2.63%) were treated with interventional therapy combined with systemic anti-tumor treatment. All surgeries were performed by the same group of surgeons. Post surgery, routine antibiotic prophylaxis was administered to prevent infection. Follow-up content and outcome The commencement of follow-up was determined as the date of the patient's initial surgery (surgery day), and data were obtained from the electronic medical record system, outpatient records, and telephone follow-ups with patients or their family members at West China Hospital of Sichuan University. The minimum duration for follow-up was 5 years. Follow-up assessments included postoperative radiotherapy and chemotherapy, including frequency and treatment plans, recurrence status, time and treatment following recurrence, mortality events with time and cause of death. Patients who died were followed up until the time of death (OS), while other patients' actual follow-up times were recorded. OS was defined as the time interval from the completion of surgery to either from the completion of surgery to either the last follow-up or the patient's death. The deadline for all follow-ups was March 2024. Detection of serum and plasma biomarkers The level of PIVKA-Ⅱwas detected by G1200 chemiluminescence enzyme immunoassay system (LUMIPULSE, Japan). The level of AFP was measured by an I2000 immunochemiluminescence detection system (ABBOTT, USA). The level of CEA was determined with a fully automatic Cobas e801 measurement analyzer (Roche, Germany). The level of HBV-DNA was assessed by a Lepgen-96 real-time fluorescence PCR instrument (LUEPU, China). Additionally, the RBC, WBC, PLT, LYM and MONO levels were analyzed by XN-9000/9001(SYSMEX, Japan). Furthermore, AST, ALT, ALB and TBIL levels were quantified by a hematology analyzer (Hitachi, Japan) according to the instructions provided by the reagent kit. The detailed normal reference ranges for the serum and plasma biomarkers could be found in Appendix Table 1. Detection of CTCs and CECs CTCs and CECs were isolated by differential enrichment tumor marker-immunofluorescence staining-fluorescence in situ hybridization (SE-iFISH) technology according to the manufacturer’s instructions. A volume of 6 mL of venous blood from the elbow was collected from the patient within one week before surgery into an anticoagulant vacuum tube containing ACD (Becton, Dickinson and Company, NJ, USA). The samples were storaged at a low temperature and shielded from light. CTCs and CECs were conducted strictly within 24 hours following the instructions of the reagent kit (Cytointelligen, San Diego, CA, USA). The SE kit was utilized for the synchronous separation and enrichment of CTCs and CECs in peripheral blood, while the iFISH kit was used for the identification and classification of CTCs and CECs. Upon completion of the aforementioned experimental procedures, automatic scanning was performed on an automated scanning platform. The interpretation standard for CTCs was as follows: PD-L1±/CD31-/CD45-/DAPI+/CEP8≥2. The interpretation criterion for CTM was a cluster of CTCs with a CTC count more than 2. The interpretation criterion for CECs was PD-L1±/CD31+/CD45-/DAPI+/CEP8≥2. The interpretation criteria for CEC clusters indicated that the number of CECs should exceed 2 [16] (Figure 1). Screening Features of ML The study cohort comprised 76 treatment-naive HCC patients, who were stratified based on their 5-year post surgery survival status. The dataset included a total of 56 independent variables, comprising 14 categorical and 42 continuous variables. The outcome variable was related to the probability of reaching the 5-year survival milestone. The 56 independent variables were primarily categorized into: 10 personal medical history features, 6 pathological features, 14 hematology features, and 26 CTC and CEC subtype features (Figure 2). The personal medical history features were as follows: gender, age, genetic history, smoking history, drinking history, other underlying diseases (excluding viral hepatitis), gallbladder Murphy sign, presence of distant metastasis, presence of liver cirrhosis, and degree of liver cirrhosis. The pathological features were as follows: the number of intrahepatic nodules, CNLC stage, BCLC stage, the maximum diameter of nodules, presence of vascular cancer thrombi and satellite nodules, and presence of vascular invasion. The hematological features were as follows: PIVKA-Ⅱ, AFP, CEA, HBV-DNA, TBIL, AST, ALT, ALB, RBC, WBC, PLT, LYM, MONO, and NEU. The subtype characteristics of CTCs and CECs were as follows: CTC, CTM, CEC, CEC cluster; haploid, amphiploid, triploid, tetraploid, ≥ pentaploid CTC; PD-L1(+) CTC, PD-L1(-)CTC, large size CTC, small size CTC; haploid, amphiploid, triploid, tetraploid, ≥pentaploid CEC; PD-L1(+) CEC, PD-L1(-) CEC; the large size CEC, the small size CEC; PD-L1(+) CTC-WBC cluster, PD-L1(-) CTC-WBC cluster, PD-L1(+) CEC-WBC cluster, PD-L1(-) CEC-WBC cluster. Model construction and validation The R language was utilized for data processing, and univariate hypothesis testing was conducted on all independent variables. The categorical variables were analyzed by the χ 2 test, and the continuous variables were analyzed by the t test. A statistically significant subset was identified based on P<0.05 . The SVM-RFE algorithm had been extensively utilized in a wide range of practical applications and had been demonstrated to be an effective feature selection method [19] . Compared to other conventional feature selection techniques, the SVM-RFE algorithm leveraged the kernel method and decision boundaries of SVM to better capture complex data structures. This was particularly advantageous in high-dimensional datasets, where SVM-RFE effectively mitigated the risk of overfitting by identifying the most informative features, thereby enhancing the model’s accuracy and generalizability. Subsequently, the data of these variables were standardized and screened using the SVM-RFE algorithm to obtain an optimal subset of features. The accuracy measure included a line graph depicting changes in the root mean square error (RMSE) obtained through cross-validation to identify the optimal feature subset. The significant variable set and optimal feature subset obtained from the aforementioned methods were separately screened, and then merged to form the final significant feature set. The selected variables were input into 5 different ML models for prediction and evaluation. Logistic regression, SVM, DTC, RF, and XGBoost were used to predict the mortality risk after 5 years. The ML model parameter settings were shown in Table 1. Considering the small-sample size in this study, leave-one-out-cross-validation (LOOCV) was selected to evaluate model performance. This approach maximized data utilization, reduced evaluation bias, and helped prevent overfitting. In LOOCV, one sample was excluded as the test set in each iteration, while the remaining samples were used for training [17] . Finally, the values of specificity, F1 score, recall and AUC-ROC were selected as the performance evaluation parameters for each model. Table 1 The main parameters of machine learning models Machine learning models Parameter Value SVM kernel ‘radial’ cost 1 gamma 0.1 DTC width Console_width() trim FALSE RF mtry 15 ntree 1500 XGBoost eval_metric ‘auc’ eta 0.01 gamma 0.005 max_depth 3 subsample 0.5 colsample_bytree 0.4 nrounds 500 verbose 1 print_every_n 100 early_stopping_rounds 200 Statistical analysis The R programming language (version 4.1.2, R Development Core Team, Vienna, Austria) was used for the computation of the statistical data and generation of the visual charts. The data obtained from 76 participants were described by examples, and intergroup comparisons were conducted by the chi-square test. Normally distributed data were presented as the mean ± standard deviation (M±SD). Additionally, independent sample t-tests were performed for intergroup comparisons. For nonnormally distributed data, the median (P25, P75) was utilized, and nonparametric tests were applied for group comparisons. Univariate logistic regression models and other ML algorithms leveraged clinical data. The diagnostic value of 5-year survival biomarkers was assessed with AUC-ROC analysis. A two-tailed P<0.05 was considered to indicate statistical significance [18-22] . Results Characteristics and Kaplan-Meier analysis of the study population. All 76 HCC patients completed follow-up with a duration ranging from 1 to 67 months. The shortest and longest OS times recorded were 2 months and 66 months, respectively. At the end of the follow-up period, 34 patients (44.7%) survived for more than 5 years, while 42 patients (55.3%) had died from the disease. Additionally, 43 patients experienced tumor recurrence or metastasis (56.6%), and terminal tumor recurrence reached 6.6% (n=5). The majority of the patients had no family history of inherited diseases (n=71,93.4%). Smoking history was present in 46 patients (60.5%), alcohol consumption history in 31 patients (40.8%), HBV infection history in 40 patients (52.6%), liver cirrhosis in 71 patients (93.4%), distant metastasis in 19 patients (25.0%), tumors with a diameter of ≥5 cm in 54 patients (71.1%), multiple liver lesions in 50 patients (65.8%), vascular tumor thrombi and satellite nodules in 50 patients (65.8%), and vascular invasion in 33 patients (43.4%). The Kaplan-Meier survival curve demonstrated a progressive decline in the survival rate of HCC patients over time, with significantly lower survival rates observed in patients at advanced stages during the 5-year follow-up period (Fig.3). Among the 3 patients with CNLC stage I disease, the 5-year OS rate was 66.7% (n=2). Among the 11 patients with CNLC stage II disease, the 5-year OS rate was 63.6% (n=7). Among the 49 patients with CNLC stage III disease, the 5-year OS rate was 51.0% (n=25). Among the 13 patients with CNLC stage IV disease, the 5-year OS rate was 0% (n=0). Furthermore, patients who exhibited PD-L1 (-) CTC-WBC cluster numbers ≥1 had more lower survival rate than did those with PD-L1 (-) CTC-WBC cluster numbers <1 (HR:1.87, CI:0.95-3.72, P =0.038). Similarly, patients who displayed PD-L1 (-) CEC-WBC cluster numbers ≥1 had more lower survival rate than those with numbers <1 (HR:2.19, CI:0.84-5.75, P =0 .029). Analysis results of HCC patient 5-year OS prognostic variables based on the chi-square test, Student’s test, and SVM-RFE This study involved 56 variables, which were primarily categorized into four groups as follows: personal medical history characteristics (Number of variables = 10), pathological characteristics (Number of variables = 6), hematological characteristics (Number of variables = 14), and CTC and CEC subtype characteristics (Number of variables = 26). Univariate hypothesis tests were conducted on continuous and discrete variables, identifying 11 features associated with survival rate (Table 2). The significant variables set included: the maximum tumor diameter ( P =0.0006), serum ALB concentration ( P =0.015), age ( P =0.03), RBC count ( P =0.0223), TBIL level ( P =0.0204), PD-L1(-) CTC-WBC cluster count ( P =0.0123), PD-L1(-) CEC-WBC cluster count ( P =0.0295), distant metastasis status( P =0.0014), CNLC stage ( P =0.037), BCLC stage ( P =0.908), and degree of liver fibrosis ( P =0.154). The remaining 45 variables didn’t show statistical significance, and these included gender, genetic history, smoking history, drinking history, other underlying diseases, gallbladder Murphy sign presence, liver cirrhosis status, number of intrahepatic nodules, vessel cancer thrombi and satellite nodules, presence or absence of vascular invasion, abnormal prothrombin levels, AFP levels, CEA levels, HBV-DNA, AST levels, ALT levels, WBC, PLT, LYM, MONO, NEU, CTC, CTM, CEC, CEC cluster, haploid CTC, amphiploid CTC, triploid CTC, tetraploid CTC, ≥ 5 pentaploid CTC, PD-L1(+)CTC、PD-L1(-)CTC, large size CTC, small size CTC, haploid CEC, amphiploid CEC, triploid CEC, tetraploid CEC, ≥ 5 pentaploid CEC, PD-L1(+)CEC、PD-L1(-)CEC, large CEC, haploid CEC, PD-L1(+)CTC-WBC cluster, PD-L1(+)CEC-WBC cluster (Appendix table 2). Table 2 Comparison of selected significant baseline characteristics between the 5-year survival group and the non-survival group in HCC patents. Characteristics Overall (N=76) Statistics P-value Continuous variables Mean value (SD) Age 56.21(12.02) -2.21 0.030 * The largest diameter of the tumor tissues 5.40(3.38) -3.63 0.0006 * TBIL 30.41（45.25） -2.40 0.0204 * ALB 42.50(7.68) 2.50 0.015 * AST 53.07（42.41） -1.09 0.2786 AFP 928.96(3300.49) -1.63 0.1113 WBC 6.25（2.91） -0.90 0.3706 PD-L1(-) CTC 7.87（6.28） 0.12 0.9020 The number of ≥Pentaploid CTC 3.47（3.45） 0.52 0.6037 The large size CTC 6.75（5.27） 0.24 0.8132 The number of CTC 9.16（6.88） -0.01 0.9903 RBC 4.60（0.71） 2.34 0.0223 * PD-L1(-) CTC-WBC clusters 0.47（0.90） -2.58 0.0123 * Triploid CTC 3.13（3.55） -0.97 0.3354 LYM 2.46（5.55） 1.55 0.1308 PD-L1(-) CEC-WBC clusters 0.29（0.91） -2.25 0.0295 * Discrete variables Number value Whether distant metastasis or not Yes 19 10.22 0.0014 * No 57 The CNLC staging I 3 13.48 0.0037 * II 11 III 49 IV 13 Vascular cancer thrombus and satellite nodules Yes 50 5.60 0.0179 * No 26 The BCLC staging 0 2 6.47 0.0908 A 2 B 19 C 53 The number of intrahepatic nodules The single nodule 26 3.54 0.06 The multiple nodules 50 The Child-pugh Classing in Cirrhosis No 10 10.41 0.0154 * A 20 B 27 C 19 * indicated that it was statistically significant. The data were processed in accordance with the principle of the maximum interval in the SVM-RFE algorithm, the variables were standardized, and the data were then input into the SVM-RFE for filtering. The change in the root mean square error (RMSE) was obtained through cross validation, and gradually increased with the number of variables (Fig. 4). When the number of variables was 20, the RMSE reached its lowest point. Thus, the optimal feature subset included a total of 20 features (Table 3). Considering both the comprehensiveness and predictive ability of the model, the 11 significant variables were integrated with the optimal feature subset to form the final set of significant variables (22 variables). The significant variables set were as follows: the maximum diameter, presence or absence of distant metastasis, CNLC stage, ALB, age, RBC, large size CTC,TBIL, PD-L1(-)CTC, ≥ pentaploid CTC, AFP, vascular cancer thrombus and satellite nodules, WBC, CTC, BCLC stage, multiple nodules, AST, PD-L1(-)CTC-WBC cluster, triploid CTC, LYM, PD-L1(-)CEC-WBC cluster, and degree of cirrhosis. Table 3 The importance ranking of screening variables set by the SVM-RFE algorithm Variable The importance ranking The maximum diameter of nodules 1 Presence of distant metastasis 2 CNLC stage 3 ALB 4 Age 5 RBC 6 The number of the large CTC 7 TBIL 8 The number of PD-L1(-) CTC 9 The number of ≥ pentaploid CTC 10 AFP 11 Vascular cancer thrombi and satellite nodules 12 WBC 13 The number of CTC 14 BCLC stage 15 The number of intrahepatic nodules 16 AST 17 The number of PD-L1(-) CTC-WBC cluster 18 The number of triploid CTC 19 LYM 20 Comparison and evaluation of five ML algorithms for establishing multi-indicator joint detection models The 5-year OS risk factor dataset were trained and predicted by logistic regression, SVM, DTC, FR, and XGBoost algorithms. We employed the LOOCV algorithm to optimize the hyperparameters for maximizing the area under the ROC curve. Time-dependent ROC analysis was utilized to assess the predictive performance of the curve in estimating patients’5-year OS rate. Table 4 presented the evaluation metrics for five ML algorithms. Among them, SVM demonstrated the best performance across all metrics: its specificity (0.971) and recall (1.000) reached the highest values, while its F1 Score (0.988), accuracy (0.987), and AUC-ROC (0.971) were significantly superior to those of the other models. Notably, the recall of SVM achieved the theoretical maximum, indicating that the model was able to fully capture positive samples, while its specificity remained at a high level of 0.971, demonstrating excellent suppression of false positive samples. LR exhibited high specificity, second only to SVM, but its recall and F1 Score were significantly lower, reflecting limitations in identifying positive samples. XGBoost showed a good balance, with accuracy comparable to LR, though its AUC-ROC was slightly lower. Both DTC and RF showed relatively poor overall performance, with these models significantly lagging behind others in key metrics such as specificity, accuracy, and AUC-ROC. According to the ROC curves of the five algorithms, the SVM algorithm exhibited notably near-perfect discriminatory ability (Fig. 5a). Dynamic channel allocation (DCA) represented a novel approach for assessing alternative predictive method, demonstrating superior clinical value evaluation to the AUC-ROC. When the risk threshold fell within the range of 0-0.90, the clinical net benefits of the five models were all positive. Furthermore, the SVM algorithm exhibited the best positive clinical net benefit (Fig. 5b). To further validate the robustness of our model, particularly given the relatively small-sample size in this study, we performed both internal and external validation. For internal validation, we collected an additional independent internal dataset as the test set. Subsequently, all models were trained using the full cohort of 76 patients from our study, and their predictive performance was evaluated on this independent test set. For external validation, we utilized a large-scale dataset to conduct model training and prediction. The results from both validations demonstrated the superiority and robustness of our model, with the SVM consistently outperforming the other methods. Detailed results are presented in Supplementary Information B and C. In addition, given the commonly cited limitation of ML models regarding interpretability in clinical settings, we conducted a SHAP (SHapley Additive exPlanations) analysis on the optimal model (SVM), using all 22 input features. The resulting feature importance chart and beeswarm plot provided valuable insights into the contribution of each feature to the model’s predictions, thereby enhancing the model’s transparency and clinical interpretability. The visualizations were provided in Supplementary Information A. Table 4 Evaluation of the classification performance on five ML models Parameter Logistic SVM DTC RF XGBoost Specificity 0.912 0.971 0.765 0.765 0.735 F1 Score 0.696 0.988 0.649 0.667 0.741 Recall Value 0.571 1.000 0.571 0.595 0.714 Accuracy 0.724 0.9867 0.658 0.671 0.724 AUC-ROC 0.737 0.971 0.657 0.738 0.727 Discussion The delayed discovery and a low rate of early diagnosis in patients with HCC are significant contributing factors to the poor 5-year OS rate [ 23 – 25 ] . The urgent need exists for the development of a sensitive and highly specific prognostic model for HCC. In recent years, extensive researches have focused on the application of AI technology in HCC-related screening, diagnosis, treatment efficacy evaluation, and outcome prediction, with promising initial results [ 26 – 28 ] . When employing conventional epidemiological approaches to analyze large-scale database information on HCC progression, researchers frequently encounter challenges related to high-dimensional data and analytical precision. Integrating smaller datasets with extensive data sources to provide contextual background is essential for advancing HCC studies. Recently, techniques that combine large and small datasets have emerged as the focus across various domains and have demonstrated significant value in the field of HCC [ 29 – 30 ] . In contrast to conventional approaches that rely on cutoff values as the criteria, ML algorithms have demonstrated greater potential for accurately identifying disease indicators. Particularly, the disease diagnosis and prognosis assessment when dealing with high values fall below these thresholds. Frid-Adar et al. proposed a deep learning framework based on CNNs, which effectively differentiates cysts, metastases, and vascular tumors [ 31 ] . Moreover, Morshid et al. developed a predictive model that integrated imaging and genomic features extracted from CT scans with blood biochemical data to anticipate patient responses to interventional treatments [ 32 ] . Compared to using only BCLC staging results, the combined prediction accuracy of utilizing BCLC staging and model results predictions reached 74.2% [ 32 ] . Previous studies haven ’ t established an individualized prediction model for forecasting the 5-year survival prognosis of patients with HCC based on small-sample data. Based on prior research, our team developed an optimized ML model tailored for the small-sample database [ 17 ] . On this basis, we further collect appropriate samples and verified the effectiveness and accuracy of the model through practical cases. This study is the first to categorize features influencing OS into four groups and employ chi-square test, Student’s test, and SVM-RFE algorithms for independent screening of prognostic indicators. The final selected variable set (22 variables) served as an independent prognostic factor. The levels of CTCs and CTM can be utilized for dynamic monitoring of HCC, assessment of therapeutic effects, and guidance for prognosis determination. They can facilitate the formulation of precision treatment plans based on detection results and prompt early clinical interventions accordingly. The SVM algorithm demonstrated superior discriminative ability for 5-year survival and OS. Its excellent performance can be attributed to the synergistic effect of several factors. Firstly, the SVM algorithm optimizes the classification boundary by employing the margin maximization principle. By integrating kernel methods, it maps the data into a higher-dimensional space to enhance linear separability, thereby effectively improving the model’s generalization capability in small-sample scenarios. Secondly, the SVM-RFE method, through recursive feature elimination, selects the features with the least contribution to the classification hyperplane, significantly reducing noise interference while retaining highly discriminative features, thereby further improving the model’s robustness and classification efficiency. In contrast, LR is limited by its linear assumption and sensitivity to feature collinearity, making it difficult to capture complex nonlinear relationships. On the other hand, tree-based models (DTC, RF) and XGBoost, may experience significant performance degradation in small-sample data due to unstable calculation of information gain for node splits and insufficient diversity of base learners. Conventional blood tumor markers and HBV-DNA quantitative detection provide reference indicators for the preoperative assessment of liver cancer patients in clinical practice [ 33 ] . AFP, CEA, and CA199 were commonly used as tumor markers for the detection and monitoring of liver cancer. [ 34 ] . The next-generation SE-iFISH technology not only enabled simultaneous detection, classification, and enumeration of both CTCs and CECs but also preserved the protein profile and genetic information of intact cells, thereby laying a solid foundation for further in-depth research [ 35 – 36 ] . Our previous studies demonstrated that different subtypes of CTCs had distinct clinical significance for nasopharyngeal carcinoma patients [ 37 ] . In our investigation, patients with elevated CTCs and CECs levels exhibited significantly less survival times and more higher recurrence rates. The levels of CTCs and CECs exerted a statistically significant impact on outcomes and were consequently integrated into the model we developed. It has been validated that the variable set encompassing four aspects yields heightened higher specificity in model construction. An investigation revealed that the interaction between CTCs and immune cells leading to the formation of CTC-WBC clusters significantly augmented the propensity for metastasis in CTCs, resulting in increasing metastatic capacity and elevating probability of tumor recurrence and spread [ 38 ] . In a fundamental inquiry into HCC, it was demonstrated that neutrophils could adhere to CTCs via intercellular adhesion molecule-1, thereby enhancing the cytotoxicity of natural killer (NK) cells against these CTCs while also facilitating their evasion from immune surveillance [ 39 ] . Our findings indicated that individuals with elevated levels of PD-L1(-) CTC-WBC clusters or PD-L1(-) CEC-WBC clusters had significantly decreased OS. Nevertheless, the further investigations are warranted to elucidate the underlying molecular mechanisms governing how interactions among CTCs/CECs and WBC contribute to HCC recurrence and metastasis. Our study has several limitations. Firstly, the impact of treatment plan on prognosis isn ’ t analyzed in this study due to the sample size. Secondly, our findings lack external data validation. Thirdly, in addition to pursuing accuracy when choosing a model algorithm, it is also necessary to comprehensively consider the balance among the model's complexity, robustness, generalization ability and interpretability. Fourthly, we anticipate that our research findings can be integrated with radiogenomic or metabolomic markers in future studies to enhance their predictive capacity. Finally, we plan to conduct multicenter clinical trials and develop novel algorithms to rigorously validate the model effectiveness. Conclusion In conclusion, the 5-year OS of HCC patients is influenced by a combination of tumor-related pathological factors and nontumor-related factors. This study utilizes the SVM algorithm to construct a predictive model, which demonstrates superior performance compared to traditional single tumor markers and has potential for assisting in clinical prognosis assessment. Abbreviations HCC: Hepatocellular carcinoma; PLC: Primary liver cancer; ML: Machine learning; CTCs: Circulating tumor cells; CECs: Circulating endothelial cells; OS: Overall survival; CTM: Circulating tumor thrombi; SVM-RFE: Support vector machine recursive feature elimination feature; LR: Logistic regression; DTC: Decision tree classification; RF: Random forests; XGBoost: Extreme gradient Boosting; RUC-ROC: Area under the receiver operating characteristic curve; LOOCV: Leave-one-out-cross-validation; DCA: Dynamic channel allocation; AI: Artificial intelligence; CNLC: Chinese Liver Cancer Staging; BCLC: Barcelona Clinic Liver Cancer; PIVKA_Ⅱ: Abnormal Prothrombin; AFP: Alpha-fetoprotein; CEA: Carcinoembryonic antigen; HBV-DNA: Deoxyribonucleic acid of Hepatitis B virus; TBIL: Total bilirubin; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; ALB: Albumin; RBC: Erythrocyte; WBC: Leukocyte; PLT: Thrombocyte; LYM: Lymphocyte; MONO: Monocyte. Declarations Acknowledgments The authors thank all individuals who contributed to this study for providing advice. Author contributions TJ analyzed and interpreted the data, and was a major contributor in writing the manuscript. TJ, XL and SM analyzed the data and wrote the manuscript. WH, HL and XY contributed in study concept. QY and SM contributed in funding and supervision. All authors read and approved the final manuscript. Funding This work was supported by the National Natural Science Foundations of China (82201339, 12401381), the Natural Science Foundations of Hunan Province (2025JJ60043), the Excellent Young Scholar Project of Hunan Provincial Education Department (23B0033) and the Natural Science Foundation of Changsha (kq2402048). Data availability Availability of data and materials Data are available upon reasonable request. Corresponding author: [email protected] . Ethics approval and consent to participate The study followed the guidelines for the Quality Management of Drug Clinical Trials (2003), the Regulations for Clinical Trials of Medical Devices (2004), the International Ethical Guidelines for Biomedical Research on Humans, the Declaration of Helsinki and the Ethical Review of Biomedical Research on Human Designers (2007). This study was approved by the ethics committee of West China Hospital of Sichuan University (No. HXYY-2015-141). The enrolled participants voluntarily signed informed consent forms, Sichuan Province, China. Ethical review and approval were waived for this study because the data were fully deidentified and no interventions were performed. Consent for Publication Not applicable. Competing interests The authors declare no potential conflicts of interest. References Lynette M Sequeira , N Begum Ozturk , Leandro Sierra, et al. Hepatocellular Carcinoma and the Role of Liver Transplantation: An Update and Review. J Clin Transl Hepatol,2025,13(4): 327-338. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin, 2024,74(3):229-263. Han B, Zheng R, Zeng H, et al. Cancer incidence and mortality in China,2022. J Natl Cancer Cent,2024, (1):47-53. Zhang SW, Sun KX, Zheng RS, et al. Cancer incidence and mortality in China, J Nat Cancer Cent,2021,1(1):2-11. Amar M, Niraj K, Amol S, et al. Transarterial Chemoembolization: A Consistent and Continuously Evolving Therapy for Hepatocellular Carcinoma. J Clin Exp Hepatol,2025,15(4):102538. JOHRI A M, SINGH K V, MANTELLA L E, et al. Deep learning artificial intelligence framework for multiclass coronary artery disease prediction using combination of conventional risk factors, carotid ultrasound, and intraplaque neovascularization. Com put Biol Med, 2022, 150:106018. Xu Z, Chi C, Yan W, et al. Recurrence risk prediction models for hepatocellular carcinoma after liver transplantation. J Gastroenterol Hepatol, 2024,39(11):2272-2280. Z G Yuan, S L Ye. Systemic therapeutic strategies for hepatocellular carcinoma: current status and prospects. Zhonghua Gan Zang Bing Za Zhi, 2024,32(6):565-571. WANG H, LIU Y, XUN, et al. Development and validation of a deep learning model for survival prognosis of transcatheter arterial chemoembolization in patients with intermediate-stage Liver cancer. Eur J Radiol, 2022, 156:110527. HU G, HU X, YANG K, et al. Radiomics-based machine learning to predict recurrence in glioma patients using magnetic resonance imaging. J Comput Assist Tomogr. 2023,47(1):129-135. Geoffroy P, Joséphine M, Valerie T. Liquid biopsy: general concepts. Acta Cytol, 2019, 63(6): 449-455. Serafina M, Demi W, Eleonora L, et al. Liquid biopsy: An innovative tool in oncology. Where do we stand? Semin Oncol,2025,52(2):152343. Lisanne M,Lisanne F,Jaco K,et al. Generating human prostate cancer organoids from leukapheresis enriched circulating tumor cells. Eur J Cancer,2021,150:179-189. Carolina R, Eleonora N, Surbhi S,et al. Unveiling the impact of circulating tumor cells: Two decades of discovery and clinical advancements in solid tumors. Crit Rev Oncol Hematol, 2024,203:104483. Léa S, William J, Ludovic G,et al. Programmed Cell Death Ligand 1-Expressing Circulating Tumor Cells: A New Prognostic Biomarker in Non-Small Cell Lung Cancer,2021,131. Liu X, Li J, Cadilha B, et al. Epithelial-type systemic breast carcinoma cells with a restricted mesenchymal transition are a major source of metastasis. Sci Adv, 2019, 5(6): 4275. Shanjun Mao, Xiaodan Fan, Jie Hu. Correlation for tree-shaped datasets and its Bayesian estimation. Computational Statistics and Data Analysis,2021,164(2021): 107307. Zheyu Z, Tianze C, Yuexia H, et al. CirclizePlus: using ggplot2 feature to write readable R code for circular visualization. Front Genet,2025,16:1535368. Richhariya, Bharat, et al. Diagnosis of Alzheimer's disease using universum support vector machine based recursive feature elimination (USVM-RFE). Biomedical Signal Processing and Control, 2020, (59): 101903. R. Kolde, pheatmap: Pretty Heatmaps. R Package Version 1.0.12, https://cran.r project.org/web/packags/pheatmap/index.html, 2019. P. S.-C. Patrick J. Heagerty, survivalROC: Time-dependent ROC curve estimation from censored survival data. R Package Version project.org/web/packages/survivalROC/index.html, 2015. T. T, A Package for Survival Analysis in R. R package version 3.2-7, https://CRAN.R project.org/package=survival, 2020. Ahmed S , Ahmed M E , Mohamed M, et al. Impact of tumor size on the outcomes of hepatic resection for hepatocellular carcinoma: a retrospective study. BMC Surg,2024,24(1):7. L Mocan. Multimodal therapy for hepatocellular carcinoma: the role of surgery. Eur Rev Med Pharmacol Sci,2021,25(13):4470-4477. Zachary J, Diamantis I, Samantha M , et al. Management of Hepatocellular Carcinoma: A Review. JAMA Surg,2023,158(4):410-420. Lee YT, Wang JJ, Luu M, et al. The mortality and overall survival of primary liver cancer in the United States. J Natl Cancer Inst, 2021, 113(11):1531-1541. QIAO M, LIU C, LI Z, et al. Breast tumor classification based on MRI-US images by disentangling modality features. IEEE J Biomed Health Inform, 2022,26(7):3059-3067. SEKHAR A, BISWAS S, HAZRA R, et al. Brain tumor classification using fine-tuned GoogLeNet features and machine learning algorithms: IoMT enabled CAD system. IEEE J Biomed Health Inform, 2022, 26(3):983-991. Johnston KG, Grieco SF, Nie Q, et al. Small data methods in omics: the power of one. Nat Methods. 2024, 21(9):1597-1602. Tang, X., Guo, R., Mo, Z. et al. Causality-driven candidate identification for reliable DNA methylation biomarker discovery. Nat Commun. 2025 ,16(1), 680. FRID-ADAR M, DIAMANT I, KLANG E, et al. GAN-based synthetic medical image augmentation for increased CNN performance in liver lesion classification. Neurocomputing, 2018,321, 321-331. MORSHID A, ELSAYES K M, KHALAF A M, et al. A machine learning model to predict Liver cancer response to transcatheter arterial chemoembolization. Radiol Artif Intell, 2019,1(5): e180021. You W, Sheng N, Yan L, et al. The difference in prognosis of stage II and III colorectal cancer based on preoperative serum tumor markers. J Cancer, 2019,10(16): 3757-3766. Luo P, Wu S, Yu Y, et al. Current Status and Perspective Biomarkers in AFP Negative HCC: Towards Screening for and Diagnosing Liver cancer at an Earlier Stage. Pathol Oncol Res,2020,26(2):599-603. Lin PP. Aneuploid CTC and CTEC. Diagnostics (Basel),2018,8(2):26. Eva O, Christiane A, Eva S, et al. Molecular Characterization of Circulating Tumor Cells Enriched by A Microfluidic Platform in Patients with Small-Cell Lung Cancer. Cells, 2019,8:880. Zhang J, Shi H, Jiang T, et al. Circulating tumor cells with karyotyping as a novel biomarker for diagnosis and treatment of nasopharyngeal carcinoma. BMC Cancer, 2018,18(1):1133-1145. Qiu Y, Zhang X, Deng X, et al. Circulating tumor cell-associated white blood cell cluster is associated with poor survival of patients with gastric cancer following radical gastrectomy. Eur J Surg Oncol,2022,48(5):1039-1045. Qi LN, Xiang BD, Wu FX, et al. Circulating tumor cells undergoing EMT provide a metric for diagnosis and prognosis of patients with hepatocellular carcinoma. Cancer Res, 2018, 78(16): 4731‑4744. Additional Declarations No competing interests reported. Supplementary Files SupplementaryInformationrevised.docx Cite Share Download PDF Status: Published Journal Publication published 01 Jul, 2025 Read the published version in BMC Cancer → Version 1 posted Editorial decision: Revision requested 15 May, 2025 Reviews received at journal 11 May, 2025 Reviews received at journal 09 May, 2025 Reviewers agreed at journal 27 Apr, 2025 Reviewers agreed at journal 24 Apr, 2025 Reviews received at journal 22 Apr, 2025 Reviewers agreed at journal 22 Apr, 2025 Reviewers invited by journal 21 Apr, 2025 Submission checks completed at journal 21 Apr, 2025 First submitted to journal 21 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6176116","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":446143774,"identity":"e295572b-96d2-4acf-ae45-bdac077af70d","order_by":0,"name":"Tingting Jiang","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Tingting","middleName":"","lastName":"Jiang","suffix":""},{"id":446143776,"identity":"26b8f97d-815c-4cad-a4b8-801ba06655ea","order_by":1,"name":"Xingyu Liu","email":"","orcid":"","institution":"Hunan University","correspondingAuthor":false,"prefix":"","firstName":"Xingyu","middleName":"","lastName":"Liu","suffix":""},{"id":446143778,"identity":"f60e2e12-c0a5-4bfb-8221-5a4f0f91f92c","order_by":2,"name":"Wencan He","email":"","orcid":"","institution":"Central South University","correspondingAuthor":false,"prefix":"","firstName":"Wencan","middleName":"","lastName":"He","suffix":""},{"id":446143779,"identity":"3e4ea54a-3f3a-49d5-883f-a21944e93660","order_by":3,"name":"Hepei Li","email":"","orcid":"","institution":"Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Hepei","middleName":"","lastName":"Li","suffix":""},{"id":446143784,"identity":"a9d82b8a-3a85-4cff-946a-09fab5d83638","order_by":4,"name":"Xiang Yan","email":"","orcid":"","institution":"IMN-CNM, CSIC (CEI UAM + CSIC)","correspondingAuthor":false,"prefix":"","firstName":"Xiang","middleName":"","lastName":"Yan","suffix":""},{"id":446143785,"identity":"b42d8d27-eab9-4230-82b0-aba69c054e69","order_by":5,"name":"Qian Yu","email":"","orcid":"","institution":"Sun Yat-sen University","correspondingAuthor":false,"prefix":"","firstName":"Qian","middleName":"","lastName":"Yu","suffix":""},{"id":446143786,"identity":"d2fe3b65-febc-41c5-bac0-f8c44b35396b","order_by":6,"name":"Shanjun Mao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAwklEQVRIiWNgGAWjYDACCYYEIGmTAObwkKAljTQtIHCYBC3ysxuebvhRcT7PXCKB8cHbNgZ5c0JaDO4cSLvZc+Z2seWMBGbDuW0MhjsbCGmRSEi7wdt2O3HDjQQ2ad42hgSDA4QcNiMh7ebftnMgLey/idLCcCMh7TZv2wGwLcxEaTEAaZE5k5y44czDZsk55yQMNxB2WE7azTcVdokbjicf/PCmzEaesMMYeBKgDMYGBlg0EQDshE0dBaNgFIyCEQ4AClpGe5+VjxkAAAAASUVORK5CYII=","orcid":"","institution":"Hunan University","correspondingAuthor":true,"prefix":"","firstName":"Shanjun","middleName":"","lastName":"Mao","suffix":""}],"badges":[],"createdAt":"2025-03-07 07:53:26","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6176116/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6176116/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12885-025-14425-0","type":"published","date":"2025-07-01T15:57:01+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":81694056,"identity":"6cbbb2fe-e5a6-4be2-a1b3-938a245a969f","added_by":"auto","created_at":"2025-04-30 11:51:25","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2745828,"visible":true,"origin":"","legend":"\u003cp\u003eCTC/CEC subtypes in the peripheral blood of the patient (× 100).\u003c/p\u003e\n\u003cp\u003e1: PD-L1+(green); 2: CD45+(red); 3: CEP8+(orange); 4: CD31+(yellow); 5: DAPI+ (blue); 6: The merge images of five kinds of fluorescence.\u003c/p\u003e\n\u003cp\u003ea6: CTC-WBC clusters formed by three white blood cells (PD-L1-/CD45+/CD31-/CEP8=4/DAPI+) next to CTC cells (PD-L1+/ CD45-/CD31-/DAPI+).\u003c/p\u003e\n\u003cp\u003eb6: The CTM formed by an amphiploid CTC cell and a tetraploid CTC cell (PD-L1+/CD45-/CD31-/CEP8=4 and CEP8=2/DAPI+).\u003c/p\u003e\n\u003cp\u003ec6: ≥Pentaploid CEC (PD-L1+/CD45-/CD31+/CEP8≥5/DAPI+).\u003c/p\u003e\n\u003cp\u003ed6: CEC clusters formed by many CEC cells(PD-L1+/CD45-/CD31+/CEP8+/ DAPI+).\u003c/p\u003e","description":"","filename":"Fig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-6176116/v1/412692e572819463c033c5a1.png"},{"id":81692593,"identity":"f1ce9b40-4042-44d3-ac2f-e7918587dd4f","added_by":"auto","created_at":"2025-04-30 11:43:25","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1597029,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the study.\u003c/p\u003e","description":"","filename":"Fig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-6176116/v1/29b3b9c3217155d1c5d32682.png"},{"id":81692586,"identity":"1661222b-2064-440e-aa28-c78bff41b49d","added_by":"auto","created_at":"2025-04-30 11:43:25","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":263858,"visible":true,"origin":"","legend":"\u003cp\u003eKaplan-Meier survival curves to evaluate the influence of the significant variables.\u003c/p\u003e","description":"","filename":"Fig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-6176116/v1/d95b15c496d5838e81b52e50.png"},{"id":81694053,"identity":"9e1997f1-4ef1-4264-89b8-157570367224","added_by":"auto","created_at":"2025-04-30 11:51:25","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1468309,"visible":true,"origin":"","legend":"\u003cp\u003eAnalysis of 5-year OS prognostic variables for HCC patients.\u003c/p\u003e\n\u003cp\u003ea The line graph of RMSE changing with increasing number of variables.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eb Framework of screening of the final subset by chi-square test, student’s test and SVM-RFE algorithm.\u003c/p\u003e","description":"","filename":"Fig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-6176116/v1/60d0e3dfc2abc8b3d77f54f2.png"},{"id":81692595,"identity":"b67c8bcc-0a3f-4ddc-adac-e149ffcb9157","added_by":"auto","created_at":"2025-04-30 11:43:25","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":2908205,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation of the classification performance to five ML models\u003c/p\u003e\n\u003cp\u003ea The ROC curves of five ML algorithms.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eb The DCA curves of five ML algorithms.\u003c/p\u003e","description":"","filename":"Fig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-6176116/v1/f20da0008d0e416aaf76c6f0.png"},{"id":86178878,"identity":"b152fbc9-8bca-47dd-9d31-b473123b2935","added_by":"auto","created_at":"2025-07-07 16:06:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":11202062,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6176116/v1/f95b195e-cd09-49ed-81b9-c5958cf488af.pdf"},{"id":81694059,"identity":"95bc5a37-eed8-479b-a43d-810d16725357","added_by":"auto","created_at":"2025-04-30 11:51:25","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":783412,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformationrevised.docx","url":"https://assets-eu.researchsquare.com/files/rs-6176116/v1/18ad2a4d0bb350fa20b8f9a0.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and validation of a small-sample machine learning model to predict 5-year overall survival in patients with hepatocellular carcinoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHepatocellular carcinoma (HCC) is the main pathological subtype of primary liver cancer (PLC). HCC is a malignant tumor that originates from liver cells and bile duct epithelial cells. In 2023, HCC ranks fifth in terms of cancer prevalence but is the third leading cause of mortality among malignancies \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. In 2022, there were 367700 new cases of HCC in China, ranking fourth among new cases of malignant tumors \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. The number of deaths from HCC reached 316500, making it the second leading cause of cancer-related deaths in China \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. Although advancements have been made in both fundamental research and clinical diagnosis and treatment in recent years, the OS of HCC patients remain poor. The recurrence rate within 5 years is nearly 70%, while the 5-year survival rate is only 5%. Currently, surgical resection and liver transplantation are available interventions for HCC, but the prognostic methodologies for HCC lack precise physical and chemical indicators to timely determine its potential recurrence risk, leading patients to miss the potential treatment opportunities \u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e. Therefore, it is imperative to develop novel biomarker-based models for evaluating the recurrence risk of HCC.\u003c/p\u003e \u003cp\u003eOS analysis is frequently employed in clinical research to identify potential risk factors and predict various outcomes using statistical methods. LR and Cox regression are commonly utilized to assess the impact of specific features on particular events. However, this linear assumption oversimplifies the complex relationship between predictive factors and prognosis. In recent years, with the continuous advancements of artificial intelligence (AI), machine learning (ML) methods have been able to aid physicians in disease diagnosis, disease staging \u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e, risk factor prediction \u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e, treatment planning \u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e, and prognosis analysis \u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eSerum tumor markers are typically generated and secreted into circulation by neoplastic tissues. CTCs have emerged as a pivotal tumor marker in 'liquid biopsy', enabling the detection of CTCs in peripheral blood and paving the way for precise patient treatment. CTCs are rare tumor cells that detach from primary or metastatic solid tumor lesion and enter the peripheral circulatory system. When two or more CTCs are clumped together, they form circulating tumor thrombi (CTM) \u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. Within the peripheral blood system, in addition to blood cells and CTCs, there are endothelial cells (CECs), which are single endothelial cells derived from blood vessels. The coalescence of two or more CECs is referred to as CEC cluster. Both CTCs and CECs aren\u0026rsquo;t only associated with the development of tumors, but also intricately linked to tumor recurrence, metastasis and survival. The quantifications of CTCs and CECs levels are widely employed for early screening, diagnostic staging, treatment efficacy assessment, prognostic evaluation, and personalized therapy in various solid tumors \u003csup\u003e[\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eUnveiling the occurrence, development and prognosis of HCC requires a considerable amount of time and resources, including large sample sizes, long-term follow-ups and costly experiments. To address these issues, small-sample learning has emerged as a potentially effective approach, particularly in scenarios where sample recruitment is difficult or analysis costs are high. The crucial to improving the survival rate and prognosis of HCC patients in this study lies in the selection of appropriate epidemiological methods. In this study, based on real-world data, the 5-year survival status of HCC patients was predicted for the first time using a combination of patient pathological, hematological, CTCs, and CECs variables. We established a 5-year OS model for newly diagnosed HCC patients and identified a more sensitive set of variables. This model had the potential to improve prognostic accuracy and provide a theoretical basis for treatment planning in HCC patients.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e\u003cstrong\u003eData collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe baseline data of the patients were collected, including demographic characteristics, admission status, medical history, personal history of smoking and drinking, and degree of liver cirrhosis.This retrospective observational study included 76 patients diagnosed with HCC who underwent surgical treatment at West China Hospital of Sichuan University between September 2018 and July 2019. The cohort comprised 64 males (84.21%) and 12 females (15.79%), a median age of 55.0 years and a mean age of 55.48\u0026plusmn;12.02 years. Pathological staging and grading were conducted according to the Chinese Liver Cancer Staging (CNLC) system.\u003c/p\u003e\n\u003cp\u003eThe inclusion criteria were as follows: ①HCC diagnosis was confirmed through pathological examination. ②The patient had no previous history of other tumors or exposure to radiotherapy or chemotherapy. ③There was no vascular diseases, severe heart disease, kidney disease, abdominal infection, hepatic failure or other complications. ④The patient\u0026rsquo;s WBC count was within the normal reference range of (4.0-10.0)\u0026times;10^9/L. The exclusion criteria were as follows: ①Patients who have previously received anti-tumor treatment. ②History of other tumors. ③History of autoimmune diseases.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTreatment plans\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIndividualized standard treatment plans were developed based on PCL guidelines, patients\u0026rsquo;overall health status, and their personal preferences. Among patients undergoing radical resection for PCL, 35 cases (46.05%) received conservative medical management, 27 cases (35.53%) underwent interventional therapy combined with local ablation, 5 cases (6.58%) had simple surgical resection, 4 cases (5.26%) received a combination of surgical resection and interventional therapy, 3 cases (3.95%) underwent liver transplantation, and 2 cases (2.63%) were treated with interventional therapy combined with systemic anti-tumor treatment. All surgeries were performed by the same group of surgeons. Post surgery, routine antibiotic prophylaxis was administered to prevent infection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFollow-up content and outcome\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe commencement of follow-up was determined as the date of the patient\u0026apos;s initial surgery (surgery day), and data were obtained from the electronic medical record system, outpatient records, and telephone follow-ups with patients or their family members at West China Hospital of Sichuan University. The minimum duration for follow-up was 5 years. Follow-up assessments included postoperative radiotherapy and chemotherapy, including frequency and treatment plans, recurrence status, time and treatment following recurrence, mortality events with time and cause of death. Patients who died were followed up until the time of death (OS), while other patients\u0026apos; actual follow-up times were recorded. OS was defined as the time interval from the completion of surgery to either from the completion of surgery to either the last follow-up or the patient\u0026apos;s death. The deadline for all follow-ups was March 2024.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDetection of serum and plasma biomarkers\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe level of PIVKA-Ⅱwas detected by G1200 chemiluminescence enzyme immunoassay system (LUMIPULSE, Japan). The level of AFP was measured by an I2000 immunochemiluminescence detection system (ABBOTT, USA). The level of CEA was determined with a fully automatic Cobas e801 measurement analyzer (Roche, Germany). The level of HBV-DNA was assessed by a Lepgen-96 real-time fluorescence PCR instrument (LUEPU, China). Additionally, the RBC, WBC, PLT, LYM and MONO levels were analyzed by XN-9000/9001(SYSMEX, Japan).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFurthermore, AST, ALT, ALB and TBIL levels were quantified by a hematology analyzer (Hitachi, Japan) according to the instructions provided by the reagent kit. The detailed normal reference ranges for the serum and plasma biomarkers could be found in Appendix Table 1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDetection of CTCs and CECs\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCTCs and CECs were isolated by differential enrichment tumor marker-immunofluorescence staining-fluorescence in situ hybridization (SE-iFISH) technology according to the manufacturer\u0026rsquo;s instructions. A volume of 6 mL of venous blood from the elbow was collected from the patient within one week before surgery into an anticoagulant vacuum tube containing ACD (Becton, Dickinson and Company, NJ, USA). The samples were storaged at a low temperature and shielded from light. CTCs and CECs were conducted strictly within 24 hours following the instructions of the reagent kit (Cytointelligen, San Diego, CA, USA). The SE kit was utilized for the synchronous separation and enrichment of CTCs and CECs in peripheral blood, while the iFISH kit was used for the identification and classification of CTCs and CECs. Upon completion of the aforementioned experimental procedures, automatic scanning was performed on an automated scanning platform.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe interpretation standard for CTCs was as follows: PD-L1\u0026plusmn;/CD31-/CD45-/DAPI+/CEP8\u0026ge;2. The interpretation criterion for CTM was a cluster of CTCs with a CTC count more than 2. The interpretation criterion for CECs was PD-L1\u0026plusmn;/CD31+/CD45-/DAPI+/CEP8\u0026ge;2. The interpretation criteria for CEC clusters indicated that the number of CECs should exceed 2\u003csup\u003e\u0026nbsp;[16]\u003c/sup\u003e (Figure 1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eScreening Features of ML\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study cohort comprised 76 treatment-naive HCC patients, who were stratified based on their 5-year post surgery survival status. The dataset included a total of 56 independent variables, comprising 14 categorical and 42 continuous variables. The outcome variable was related to the probability of reaching the 5-year\u0026nbsp;survival milestone. The 56 independent variables were primarily categorized into: 10 personal medical history features, 6 pathological features, 14 hematology features, and 26 CTC and CEC subtype features (Figure 2).\u003c/p\u003e\n\u003cp\u003eThe personal medical history features were as follows: gender, age, genetic history, smoking history, drinking history, other underlying diseases (excluding viral hepatitis), gallbladder Murphy sign, presence of distant metastasis, presence of liver cirrhosis, and degree of liver cirrhosis.\u003c/p\u003e\n\u003cp\u003eThe pathological features were as follows: the number of intrahepatic nodules, CNLC stage, BCLC stage, the maximum diameter of nodules, presence of vascular cancer thrombi and satellite nodules, and presence of vascular invasion.\u003c/p\u003e\n\u003cp\u003eThe hematological features were as follows: PIVKA-Ⅱ, AFP, CEA, HBV-DNA, TBIL, AST, ALT, ALB, RBC, WBC, PLT, LYM, MONO, and NEU.\u003c/p\u003e\n\u003cp\u003eThe subtype characteristics of CTCs and CECs were as follows: CTC, CTM, CEC, CEC cluster; haploid, amphiploid, triploid, tetraploid,\u0026nbsp;\u0026ge;\u0026nbsp;pentaploid CTC; PD-L1(+) CTC, PD-L1(-)CTC, large size CTC, small size CTC; haploid, amphiploid, triploid, tetraploid,\u0026nbsp;\u0026ge;pentaploid CEC; PD-L1(+) CEC, PD-L1(-) CEC; the large size CEC, the small size CEC; PD-L1(+) CTC-WBC cluster, PD-L1(-) CTC-WBC cluster, PD-L1(+) CEC-WBC cluster, PD-L1(-) CEC-WBC cluster.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel construction and validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe R language was utilized for data processing, and univariate hypothesis testing was conducted on all independent variables. The categorical variables were analyzed by the \u0026chi;\u003csup\u003e2\u003c/sup\u003e test, and the continuous variables were analyzed by the \u003cem\u003et\u003c/em\u003e test. A statistically significant subset was identified based on \u003cem\u003eP\u0026lt;0.05\u003c/em\u003e. The SVM-RFE algorithm had been extensively utilized in a wide range of practical applications and had been demonstrated to be an effective feature selection method \u003csup\u003e[19]\u003c/sup\u003e. Compared to other conventional feature selection techniques, the SVM-RFE algorithm leveraged the kernel method and decision boundaries of SVM to better capture complex data structures. This was particularly advantageous in high-dimensional datasets, where SVM-RFE effectively mitigated the risk of overfitting by identifying the most informative features, thereby enhancing the model\u0026rsquo;s accuracy and generalizability. Subsequently, the data of these variables were standardized and screened using the SVM-RFE algorithm to obtain an optimal subset of features. The accuracy measure included a line graph depicting changes in the root mean square error (RMSE) obtained through cross-validation to identify the optimal feature subset. The significant variable set and optimal feature subset obtained from the aforementioned methods were separately screened, and then merged to form the final significant feature set.\u003c/p\u003e\n\u003cp\u003eThe selected variables were input into 5 different ML models for prediction and evaluation. Logistic regression, SVM, DTC, RF, and XGBoost were used to predict the mortality risk after 5 years. The ML model parameter settings were shown in Table 1. Considering the small-sample size in this study, leave-one-out-cross-validation (LOOCV) was selected to evaluate model performance. This approach maximized data utilization, reduced evaluation bias, and helped prevent overfitting. In LOOCV, one sample was excluded as the test set in each iteration, while the remaining samples were used for training \u003csup\u003e[17]\u003c/sup\u003e. Finally, the values of specificity, F1 score, recall and AUC-ROC were selected as the performance evaluation parameters for each model.\u003c/p\u003e\n\u003cp\u003eTable 1 The main parameters of machine learning models\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"75%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003eMachine learning models\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003eParameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003eValue\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003ekernel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026lsquo;radial\u0026rsquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003ecost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003egamma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003eDTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003ewidth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003eConsole_width()\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003etrim\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003eFALSE\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003emtry\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003entree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e1500\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003eeval_metric\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026lsquo;auc\u0026rsquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003eeta\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003egamma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e0.005\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003emax_depth\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003esubsample\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003ecolsample_bytree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003enrounds\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e500\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003everbose\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003eprint_every_n\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 40px;\"\u003e\n \u003cp\u003eearly_stopping_rounds\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e200\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe R programming language (version 4.1.2, R Development Core Team, Vienna, Austria) was used for the computation of the statistical data and generation of the visual charts. The data obtained from 76 participants were described by examples, and intergroup comparisons were conducted by the chi-square test. Normally distributed data were presented as the mean \u0026plusmn; standard deviation (M\u0026plusmn;SD). Additionally, independent sample t-tests were performed for intergroup comparisons. For nonnormally distributed data, the median (P25, P75) was utilized, and nonparametric tests were applied for group comparisons. Univariate logistic regression models and other ML algorithms leveraged clinical data. The diagnostic value of 5-year survival biomarkers was assessed with AUC-ROC analysis. A two-tailed \u003cem\u003eP\u0026lt;0.05\u003c/em\u003e was considered to indicate statistical significance \u003csup\u003e[18-22]\u003c/sup\u003e.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eCharacteristics and Kaplan-Meier analysis of the study population.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll 76 HCC patients completed follow-up with a duration ranging from 1 to 67 months. The shortest and longest OS times recorded were 2 months and 66 months, respectively. At the end of the follow-up period, 34 patients (44.7%) survived for more than 5 years, while 42 patients (55.3%) had died from the disease. Additionally, 43 patients experienced tumor recurrence or metastasis (56.6%), and terminal tumor recurrence reached 6.6% (n=5). The majority of the patients had no family history of inherited diseases (n=71,93.4%). Smoking history was present in 46 patients (60.5%), alcohol consumption history in 31 patients (40.8%), HBV infection history in 40 patients (52.6%), liver cirrhosis in 71 patients (93.4%), distant metastasis in 19 patients (25.0%), tumors with a diameter of \u0026ge;5 cm in 54 patients (71.1%), multiple liver lesions in 50 patients (65.8%), vascular tumor thrombi and satellite nodules in 50 patients (65.8%), and vascular invasion in 33 patients (43.4%).\u003c/p\u003e\n\u003cp\u003eThe Kaplan-Meier survival curve demonstrated a progressive decline in the survival rate of HCC patients over time, with significantly lower survival rates observed in patients at advanced stages during the 5-year follow-up period (Fig.3). Among the 3 patients with CNLC stage I disease, the 5-year OS rate was 66.7% (n=2). Among the 11 patients with CNLC stage II disease, the 5-year OS rate was 63.6% (n=7). Among the 49 patients with CNLC stage III disease, the 5-year OS rate was 51.0% (n=25). Among the 13 patients with CNLC stage IV disease, the 5-year OS rate was 0% (n=0). Furthermore, patients who exhibited PD-L1 (-) CTC-WBC cluster numbers \u0026ge;1 had more lower survival rate than did those with PD-L1 (-) CTC-WBC cluster numbers \u0026lt;1 (HR:1.87, CI:0.95-3.72, \u003cem\u003eP\u003c/em\u003e=0.038). Similarly, patients who displayed PD-L1 (-) CEC-WBC cluster numbers \u0026ge;1 had more lower survival rate than those with numbers \u0026lt;1 (HR:2.19, CI:0.84-5.75, \u003cem\u003eP\u003c/em\u003e=0 .029).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis results of HCC patient 5-year OS prognostic variables based on the chi-square test, Student\u0026rsquo;s test, and SVM-RFE\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study involved 56 variables, which were primarily categorized into four groups as follows: personal medical history characteristics (Number of variables = 10), pathological characteristics (Number of variables = 6), hematological characteristics (Number of variables = 14), and CTC and CEC subtype characteristics (Number of variables = 26). Univariate hypothesis tests were conducted on continuous and discrete variables, identifying 11 features associated with survival rate (Table 2). The significant variables set included: the maximum tumor diameter (\u003cem\u003eP\u003c/em\u003e=0.0006), serum ALB concentration (\u003cem\u003eP\u003c/em\u003e=0.015), age (\u003cem\u003eP\u003c/em\u003e=0.03), RBC count (\u003cem\u003eP\u003c/em\u003e=0.0223), TBIL level (\u003cem\u003eP\u003c/em\u003e=0.0204), PD-L1(-) CTC-WBC cluster count (\u003cem\u003eP\u003c/em\u003e=0.0123), PD-L1(-) CEC-WBC cluster count (\u003cem\u003eP\u003c/em\u003e=0.0295), distant metastasis status(\u003cem\u003eP\u003c/em\u003e=0.0014), CNLC stage (\u003cem\u003eP\u003c/em\u003e=0.037), BCLC stage (\u003cem\u003eP\u003c/em\u003e=0.908), and degree of liver fibrosis (\u003cem\u003eP\u003c/em\u003e=0.154). The remaining 45 variables didn\u0026rsquo;t show statistical significance, and these included gender, genetic history, smoking history, drinking history, other underlying diseases, gallbladder Murphy sign presence, liver cirrhosis status, number of intrahepatic nodules, vessel cancer thrombi and satellite nodules, presence or absence of vascular invasion, abnormal prothrombin levels, AFP levels, CEA levels, HBV-DNA, AST levels, ALT levels, WBC, PLT, LYM, MONO, NEU, CTC, CTM, CEC, CEC cluster, haploid CTC, amphiploid CTC, triploid CTC, tetraploid CTC,\u0026nbsp;\u0026ge;\u0026nbsp;5 pentaploid CTC, PD-L1(+)CTC、PD-L1(-)CTC, large size CTC, small size CTC, haploid CEC, amphiploid CEC, triploid CEC, tetraploid CEC,\u0026nbsp;\u0026ge;\u0026nbsp;5 pentaploid CEC, PD-L1(+)CEC、PD-L1(-)CEC, large CEC, haploid CEC, PD-L1(+)CTC-WBC cluster, PD-L1(+)CEC-WBC cluster (Appendix table 2).\u003c/p\u003e\n\u003cp\u003eTable 2 Comparison of selected significant baseline characteristics between the 5-year survival group and the non-survival group in HCC patents.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"106%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eCharacteristics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003eOverall (N=76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003eStatistics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003eP-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eContinuous variables\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003eMean value (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e56.21(12.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-2.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.030\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe largest diameter of the tumor tissues\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e5.40(3.38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-3.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0006\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eTBIL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e30.41（45.25）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-2.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0204\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eALB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e42.50(7.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e2.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.015 \u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eAST\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e53.07（42.41）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-1.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.2786\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eAFP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e928.96(3300.49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-1.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.1113\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e6.25（2.91）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.3706\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003ePD-L1(-) CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e7.87（6.28）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.9020\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe number of\u0026nbsp;\u0026ge;Pentaploid CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e3.47（3.45）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.6037\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe large size CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e6.75（5.27）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.8132\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe number of CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e9.16（6.88）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.9903\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eRBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e4.60（0.71）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e2.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0223\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003ePD-L1(-) CTC-WBC clusters\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e0.47（0.90）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-2.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0123\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eTriploid CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e3.13（3.55）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.3354\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eLYM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e2.46（5.55）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e1.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.1308\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003ePD-L1(-) CEC-WBC clusters\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e0.29（0.91）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e-2.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0295\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eDiscrete variables\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003eNumber value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eWhether distant metastasis or not\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e10.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0014\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe CNLC staging\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e13.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0037\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eII\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eIII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eIV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eVascular cancer thrombus and satellite nodules\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e5.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0179\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe BCLC staging\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e6.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe number of intrahepatic nodules\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe single nodule\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e3.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe multiple nodules\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eThe Child-pugh Classing in Cirrhosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e10.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0.0154\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 41px;\"\u003e\n \u003cp\u003eC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e* indicated that it was statistically significant.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe data were processed in accordance with the principle of the maximum interval in the SVM-RFE algorithm, the variables were standardized, and the data were then input into the SVM-RFE for filtering. The change in the root mean square error (RMSE) was obtained through cross validation, and gradually increased with the number of variables (Fig. 4). When the number of variables was 20, the RMSE reached its lowest point. Thus, the optimal feature subset included a total of 20 features (Table 3). Considering both the comprehensiveness and predictive ability of the model, the 11 significant variables were integrated with the optimal feature subset to form the final set of significant variables (22 variables). The significant variables set were as follows: the maximum diameter, presence or absence of distant metastasis, CNLC stage, ALB, age, RBC, large size CTC,TBIL, PD-L1(-)CTC, \u0026ge; pentaploid CTC, AFP, vascular cancer thrombus and satellite nodules, WBC, CTC, BCLC stage, multiple nodules, AST, PD-L1(-)CTC-WBC cluster, triploid CTC, LYM, PD-L1(-)CEC-WBC cluster, and degree of cirrhosis.\u003c/p\u003e\n\u003cp\u003eTable 3 The importance ranking of screening variables set by the SVM-RFE algorithm\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"83%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 39px;\"\u003e\n \u003cp\u003eThe importance ranking\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe maximum diameter of nodules\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003ePresence of distant metastasis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eCNLC stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eALB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eRBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe number of the large CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eTBIL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe number of PD-L1(-) CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe number of\u0026nbsp;\u0026ge;\u0026nbsp;pentaploid CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eAFP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eVascular cancer thrombi and satellite nodules\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe number of CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eBCLC stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe number of intrahepatic nodules\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eAST\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe number of PD-L1(-) CTC-WBC cluster\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eThe number of triploid CTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003eLYM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 32px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 6px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eComparison and evaluation of five ML algorithms for establishing multi-indicator joint detection models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe 5-year OS risk factor dataset were trained and predicted by logistic regression, SVM, DTC, FR, and XGBoost algorithms. We employed the LOOCV algorithm to optimize the hyperparameters for maximizing the area under the ROC curve. Time-dependent ROC analysis was utilized to assess the predictive performance of the curve in estimating patients\u0026rsquo;5-year OS rate. Table 4 presented the evaluation metrics for five ML algorithms. Among them, SVM demonstrated the best performance across all metrics: its specificity (0.971) and recall (1.000) reached the highest values, while its F1 Score (0.988), accuracy (0.987), and AUC-ROC (0.971) were significantly superior to those of the other models. Notably, the recall of SVM achieved the theoretical maximum, indicating that the model was able to fully capture positive samples, while its specificity remained at a high level of 0.971, demonstrating excellent suppression of false positive samples. LR exhibited high specificity, second only to SVM, but its recall and F1 Score were significantly lower, reflecting limitations in identifying positive samples. XGBoost showed a good balance, with accuracy comparable to LR, though its AUC-ROC was slightly lower. Both DTC and RF showed relatively poor overall performance, with these models significantly lagging behind others in key metrics such as specificity, accuracy, and AUC-ROC. According to the ROC curves of the five algorithms, the SVM algorithm exhibited notably near-perfect discriminatory ability (Fig. 5a).\u003c/p\u003e\n\u003cp\u003eDynamic channel allocation (DCA) represented a novel approach for assessing alternative predictive method, demonstrating superior clinical value evaluation to the AUC-ROC. When the risk threshold fell within the range of 0-0.90, the clinical net benefits of the five models were all positive. Furthermore, the SVM algorithm exhibited the best positive clinical net benefit (Fig. 5b).\u003c/p\u003e\n\u003cp\u003eTo further validate the robustness of our model, particularly given the relatively small-sample size in this study, we performed both internal and external validation. For internal validation, we collected an additional independent internal dataset as the test set. Subsequently, all models were trained using the full cohort of 76 patients from our study, and their predictive performance was evaluated on this independent test set. For external validation, we utilized a large-scale dataset to conduct model training and prediction. The results from both validations demonstrated the superiority and robustness of our model, with the SVM consistently outperforming the other methods. Detailed results are presented in Supplementary Information B and C.\u003c/p\u003e\n\u003cp\u003eIn addition, given the commonly cited limitation of ML models regarding interpretability in clinical settings, we conducted a SHAP (SHapley Additive exPlanations) analysis on the optimal model (SVM), using all 22 input features. The resulting feature importance chart and beeswarm plot provided valuable insights into the contribution of each feature to the model\u0026rsquo;s predictions, thereby enhancing the model\u0026rsquo;s transparency and clinical interpretability. The visualizations were provided in Supplementary Information A.\u003c/p\u003e\n\u003cp\u003eTable 4 Evaluation of the classification performance on five ML models\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eParameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003eLogistic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003eDTC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.912\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.971\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.765\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.765\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.735\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eF1 Score\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.696\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.988\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.649\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.667\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.741\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eRecall Value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e1.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.595\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.724\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.9867\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.658\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.671\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.724\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003eAUC-ROC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.737\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.971\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e0.657\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e0.738\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15px;\"\u003e\n \u003cp\u003e0.727\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe delayed discovery and a low rate of early diagnosis in patients with HCC are significant contributing factors to the poor 5-year OS rate \u003csup\u003e[\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e. The urgent need exists for the development of a sensitive and highly specific prognostic model for HCC. In recent years, extensive researches have focused on the application of AI technology in HCC-related screening, diagnosis, treatment efficacy evaluation, and outcome prediction, with promising initial results \u003csup\u003e[\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e. When employing conventional epidemiological approaches to analyze large-scale database information on HCC progression, researchers frequently encounter challenges related to high-dimensional data and analytical precision. Integrating smaller datasets with extensive data sources to provide contextual background is essential for advancing HCC studies. Recently, techniques that combine large and small datasets have emerged as the focus across various domains and have demonstrated significant value in the field of HCC \u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn contrast to conventional approaches that rely on cutoff values as the criteria, ML algorithms have demonstrated greater potential for accurately identifying disease indicators. Particularly, the disease diagnosis and prognosis assessment when dealing with high values fall below these thresholds. Frid-Adar et al. proposed a deep learning framework based on CNNs, which effectively differentiates cysts, metastases, and vascular tumors \u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e. Moreover, Morshid et al. developed a predictive model that integrated imaging and genomic features extracted from CT scans with blood biochemical data to anticipate patient responses to interventional treatments \u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. Compared to using only BCLC staging results, the combined prediction accuracy of utilizing BCLC staging and model results predictions reached 74.2% \u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. Previous studies haven\u003cb\u003e\u0026rsquo;\u003c/b\u003et established an individualized prediction model for forecasting the 5-year survival prognosis of patients with HCC based on small-sample data. Based on prior research, our team developed an optimized ML model tailored for the small-sample database \u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e. On this basis, we further collect appropriate samples and verified the effectiveness and accuracy of the model through practical cases. This study is the first to categorize features influencing OS into four groups and employ chi-square test, Student\u0026rsquo;s test, and SVM-RFE algorithms for independent screening of prognostic indicators. The final selected variable set (22 variables) served as an independent prognostic factor. The levels of CTCs and CTM can be utilized for dynamic monitoring of HCC, assessment of therapeutic effects, and guidance for prognosis determination. They can facilitate the formulation of precision treatment plans based on detection results and prompt early clinical interventions accordingly. The SVM algorithm demonstrated superior discriminative ability for 5-year survival and OS. Its excellent performance can be attributed to the synergistic effect of several factors. Firstly, the SVM algorithm optimizes the classification boundary by employing the margin maximization principle. By integrating kernel methods, it maps the data into a higher-dimensional space to enhance linear separability, thereby effectively improving the model\u0026rsquo;s generalization capability in small-sample scenarios. Secondly, the SVM-RFE method, through recursive feature elimination, selects the features with the least contribution to the classification hyperplane, significantly reducing noise interference while retaining highly discriminative features, thereby further improving the model\u0026rsquo;s robustness and classification efficiency. In contrast, LR is limited by its linear assumption and sensitivity to feature collinearity, making it difficult to capture complex nonlinear relationships. On the other hand, tree-based models (DTC, RF) and XGBoost, may experience significant performance degradation in small-sample data due to unstable calculation of information gain for node splits and insufficient diversity of base learners.\u003c/p\u003e \u003cp\u003eConventional blood tumor markers and HBV-DNA quantitative detection provide reference indicators for the preoperative assessment of liver cancer patients in clinical practice \u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. AFP, CEA, and CA199 were commonly used as tumor markers for the detection and monitoring of liver cancer. \u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. The next-generation SE-iFISH technology not only enabled simultaneous detection, classification, and enumeration of both CTCs and CECs but also preserved the protein profile and genetic information of intact cells, thereby laying a solid foundation for further in-depth research \u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/sup\u003e. Our previous studies demonstrated that different subtypes of CTCs had distinct clinical significance for nasopharyngeal carcinoma patients \u003csup\u003e[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/sup\u003e. In our investigation, patients with elevated CTCs and CECs levels exhibited significantly less survival times and more higher recurrence rates. The levels of CTCs and CECs exerted a statistically significant impact on outcomes and were consequently integrated into the model we developed. It has been validated that the variable set encompassing four aspects yields heightened higher specificity in model construction.\u003c/p\u003e \u003cp\u003eAn investigation revealed that the interaction between CTCs and immune cells leading to the formation of CTC-WBC clusters significantly augmented the propensity for metastasis in CTCs, resulting in increasing metastatic capacity and elevating probability of tumor recurrence and spread \u003csup\u003e[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e. In a fundamental inquiry into HCC, it was demonstrated that neutrophils could adhere to CTCs via intercellular adhesion molecule-1, thereby enhancing the cytotoxicity of natural killer (NK) cells against these CTCs while also facilitating their evasion from immune surveillance \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. Our findings indicated that individuals with elevated levels of PD-L1(-) CTC-WBC clusters or PD-L1(-) CEC-WBC clusters had significantly decreased OS. Nevertheless, the further investigations are warranted to elucidate the underlying molecular mechanisms governing how interactions among CTCs/CECs and WBC contribute to HCC recurrence and metastasis.\u003c/p\u003e \u003cp\u003eOur study has several limitations. Firstly, the impact of treatment plan on prognosis isn\u003cb\u003e\u0026rsquo;\u003c/b\u003et analyzed in this study due to the sample size. Secondly, our findings lack external data validation. Thirdly, in addition to pursuing accuracy when choosing a model algorithm, it is also necessary to comprehensively consider the balance among the model's complexity, robustness, generalization ability and interpretability. Fourthly, we anticipate that our research findings can be integrated with radiogenomic or metabolomic markers in future studies to enhance their predictive capacity. Finally, we plan to conduct multicenter clinical trials and develop novel algorithms to rigorously validate the model effectiveness.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, the 5-year OS of HCC patients is influenced by a combination of tumor-related pathological factors and nontumor-related factors. This study utilizes the SVM algorithm to construct a predictive model, which demonstrates superior performance compared to traditional single tumor markers and has potential for assisting in clinical prognosis assessment.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eHCC: Hepatocellular carcinoma; PLC: Primary liver cancer; ML: Machine learning; CTCs: Circulating tumor cells; CECs: Circulating endothelial cells; OS: Overall survival; CTM: Circulating tumor thrombi; SVM-RFE: Support vector machine recursive feature elimination feature; LR: Logistic regression; DTC: Decision tree classification; RF: Random forests; XGBoost: Extreme gradient Boosting; RUC-ROC: Area under the receiver operating characteristic curve; LOOCV: Leave-one-out-cross-validation; DCA: Dynamic channel allocation; AI: Artificial intelligence; CNLC: Chinese Liver Cancer Staging; BCLC: Barcelona Clinic Liver Cancer; PIVKA_Ⅱ: Abnormal Prothrombin; AFP: Alpha-fetoprotein; CEA: Carcinoembryonic antigen; HBV-DNA: Deoxyribonucleic acid of Hepatitis B virus; TBIL: Total bilirubin; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; ALB: Albumin; RBC: Erythrocyte; WBC: Leukocyte; PLT: Thrombocyte; LYM: Lymphocyte; MONO: Monocyte.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank all individuals who contributed to this study for providing advice.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTJ analyzed and interpreted the data, and was a major contributor in writing the manuscript. TJ, XL and SM analyzed the data and wrote the manuscript. WH, HL and XY contributed in study concept. QY and SM contributed in funding and supervision. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the National Natural Science Foundations of China (82201339, 12401381), the Natural Science Foundations of Hunan Province (2025JJ60043), the Excellent Young Scholar Project of Hunan Provincial Education Department (23B0033) and the Natural Science Foundation of Changsha (kq2402048).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials Data are available upon reasonable request. Corresponding author: [email protected].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study followed the guidelines for the Quality Management of Drug Clinical Trials (2003), the Regulations for Clinical Trials of Medical Devices (2004), the International Ethical Guidelines for Biomedical Research on Humans, the Declaration of Helsinki and the Ethical Review of Biomedical Research on Human Designers (2007). This study was approved by the ethics committee of West China Hospital of Sichuan University (No. HXYY-2015-141). The enrolled participants voluntarily signed informed consent forms, Sichuan Province, China.\u0026nbsp;Ethical review and approval were waived for this study because the data were fully deidentified and no interventions were performed.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for Publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no potential conflicts of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eLynette M Sequeira , N Begum Ozturk , Leandro Sierra, et al. Hepatocellular Carcinoma and the Role of Liver Transplantation: An Update and Review. J Clin Transl Hepatol,2025,13(4): 327-338.\u003c/li\u003e\n\u003cli\u003eBray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin, 2024,74(3):229-263.\u003c/li\u003e\n\u003cli\u003eHan B, Zheng R, Zeng H, et al. Cancer incidence and mortality in China,2022. J Natl Cancer Cent,2024, (1):47-53.\u003c/li\u003e\n\u003cli\u003eZhang SW, Sun KX, Zheng RS, et al. Cancer incidence and mortality in China, J Nat Cancer Cent,2021,1(1):2-11.\u003c/li\u003e\n\u003cli\u003eAmar M, Niraj K, Amol S, et al. Transarterial Chemoembolization: A Consistent and Continuously Evolving Therapy for Hepatocellular Carcinoma. J Clin Exp Hepatol,2025,15(4):102538. \u003c/li\u003e\n\u003cli\u003eJOHRI A M, SINGH K V, MANTELLA L E, et al. Deep learning artificial intelligence framework for multiclass coronary artery disease prediction using combination of conventional risk factors, carotid ultrasound, and intraplaque neovascularization. Com put Biol Med, 2022, 150:106018.\u003c/li\u003e\n\u003cli\u003eXu Z, Chi C, Yan W, et al. Recurrence risk prediction models for hepatocellular carcinoma after liver transplantation. J Gastroenterol Hepatol, 2024,39(11):2272-2280.\u003c/li\u003e\n\u003cli\u003eZ G Yuan, S L Ye. Systemic therapeutic strategies for hepatocellular carcinoma: current status and prospects. Zhonghua Gan Zang Bing Za Zhi, 2024,32(6):565-571.\u003c/li\u003e\n\u003cli\u003eWANG H, LIU Y, XUN, et al. Development and validation of a deep learning model for survival prognosis of transcatheter arterial chemoembolization in patients with intermediate-stage Liver cancer. Eur J Radiol, 2022, 156:110527.\u003c/li\u003e\n\u003cli\u003eHU G, HU X, YANG K, et al. Radiomics-based machine learning to predict recurrence in glioma patients using magnetic resonance imaging. J Comput Assist Tomogr. 2023,47(1):129-135.\u003c/li\u003e\n\u003cli\u003eGeoffroy P, Jos\u0026eacute;phine M, Valerie T. Liquid biopsy: general concepts. Acta Cytol, 2019, 63(6): 449-455.\u003c/li\u003e\n\u003cli\u003eSerafina M, Demi W, Eleonora L, et al. Liquid biopsy: An innovative tool in oncology. Where do we stand? Semin Oncol,2025,52(2):152343.\u003c/li\u003e\n\u003cli\u003eLisanne M,Lisanne F,Jaco K,et al. Generating human prostate cancer organoids from leukapheresis enriched circulating tumor cells. Eur J Cancer,2021,150:179-189.\u003c/li\u003e\n\u003cli\u003eCarolina R, Eleonora N, Surbhi S,et al. Unveiling the impact of circulating tumor cells: Two decades of discovery and clinical advancements in solid tumors. Crit Rev Oncol Hematol, 2024,203:104483.\u003c/li\u003e\n\u003cli\u003eL\u0026eacute;a S, William J, Ludovic G,et al. Programmed Cell Death Ligand 1-Expressing Circulating Tumor Cells: A New Prognostic Biomarker in Non-Small Cell Lung Cancer,2021,131.\u003c/li\u003e\n\u003cli\u003eLiu X, Li J, Cadilha B, et al. Epithelial-type systemic breast carcinoma cells with a restricted mesenchymal transition are a major source of metastasis. Sci Adv, 2019, 5(6): 4275.\u003c/li\u003e\n\u003cli\u003eShanjun Mao, Xiaodan Fan, Jie Hu. Correlation for tree-shaped datasets and its Bayesian estimation. Computational Statistics and Data Analysis,2021,164(2021): 107307.\u003c/li\u003e\n\u003cli\u003eZheyu Z, Tianze C, Yuexia H, et al. CirclizePlus: using ggplot2 feature to write readable R code for circular visualization. Front Genet,2025,16:1535368.\u003c/li\u003e\n\u003cli\u003eRichhariya, Bharat, et al. Diagnosis of Alzheimer\u0026apos;s disease using universum support vector machine based recursive feature elimination (USVM-RFE). Biomedical Signal Processing and Control, 2020, (59): 101903.\u003c/li\u003e\n\u003cli\u003eR. Kolde, pheatmap: Pretty Heatmaps. R Package Version 1.0.12, https://cran.r project.org/web/packags/pheatmap/index.html, 2019.\u003c/li\u003e\n\u003cli\u003eP. S.-C. Patrick J. Heagerty, survivalROC: Time-dependent ROC curve estimation from censored survival data. R Package Version project.org/web/packages/survivalROC/index.html, 2015.\u003c/li\u003e\n\u003cli\u003eT. T, A Package for Survival Analysis in R. R package version 3.2-7, https://CRAN.R project.org/package=survival, 2020.\u003c/li\u003e\n\u003cli\u003eAhmed S , Ahmed M E , Mohamed M, et al. Impact of tumor size on the outcomes of hepatic resection for hepatocellular carcinoma: a retrospective study. BMC Surg,2024,24(1):7.\u003c/li\u003e\n\u003cli\u003eL Mocan. Multimodal therapy for hepatocellular carcinoma: the role of surgery. Eur Rev Med Pharmacol Sci,2021,25(13):4470-4477.\u003c/li\u003e\n\u003cli\u003eZachary J, Diamantis I, Samantha M , et al. Management of Hepatocellular Carcinoma: A Review. JAMA Surg,2023,158(4):410-420.\u003c/li\u003e\n\u003cli\u003eLee YT, Wang JJ, Luu M, et al. The mortality and overall survival of primary liver cancer in the United States. J Natl Cancer Inst, 2021, 113(11):1531-1541.\u003c/li\u003e\n\u003cli\u003eQIAO M, LIU C, LI Z, et al. Breast tumor classification based on MRI-US images by disentangling modality features. IEEE J Biomed Health Inform, 2022,26(7):3059-3067. \u003c/li\u003e\n\u003cli\u003eSEKHAR A, BISWAS S, HAZRA R, et al. Brain tumor classification using fine-tuned GoogLeNet features and machine learning algorithms: IoMT enabled CAD system. IEEE J Biomed Health Inform, 2022, 26(3):983-991.\u003c/li\u003e\n\u003cli\u003eJohnston KG, Grieco SF, Nie Q, et al. Small data methods in omics: the power of one. Nat Methods. 2024, 21(9):1597-1602.\u003c/li\u003e\n\u003cli\u003eTang, X., Guo, R., Mo, Z. et al. Causality-driven candidate identification for reliable DNA methylation biomarker discovery. Nat Commun. 2025 ,16(1), 680.\u003c/li\u003e\n\u003cli\u003eFRID-ADAR M, DIAMANT I, KLANG E, et al. GAN-based synthetic medical image augmentation for increased CNN performance in liver lesion classification. Neurocomputing, 2018,321, 321-331. \u003c/li\u003e\n\u003cli\u003eMORSHID A, ELSAYES K M, KHALAF A M, et al. A machine learning model to predict Liver cancer response to transcatheter arterial chemoembolization. Radiol Artif Intell, 2019,1(5): e180021.\u003c/li\u003e\n\u003cli\u003eYou W, Sheng N, Yan L, et al. The difference in prognosis of stage II and III colorectal cancer based on preoperative serum tumor markers. J Cancer, 2019,10(16): 3757-3766.\u003c/li\u003e\n\u003cli\u003eLuo P, Wu S, Yu Y, et al. Current Status and Perspective Biomarkers in AFP Negative HCC: Towards Screening for and Diagnosing Liver cancer at an Earlier Stage. Pathol Oncol Res,2020,26(2):599-603.\u003c/li\u003e\n\u003cli\u003eLin PP. Aneuploid CTC and CTEC. Diagnostics (Basel),2018,8(2):26.\u003c/li\u003e\n\u003cli\u003eEva O, Christiane A, Eva S, et al. Molecular Characterization of Circulating Tumor Cells Enriched by A Microfluidic Platform in Patients with Small-Cell Lung Cancer. Cells, 2019,8:880.\u003c/li\u003e\n\u003cli\u003eZhang J, Shi H, Jiang T, et al. Circulating tumor cells with karyotyping as a novel biomarker for diagnosis and treatment of nasopharyngeal carcinoma. BMC Cancer, 2018,18(1):1133-1145.\u003c/li\u003e\n\u003cli\u003eQiu Y, Zhang X, Deng X, et al. Circulating tumor cell-associated white blood cell cluster is associated with poor survival of patients with gastric cancer following radical gastrectomy. Eur J Surg Oncol,2022,48(5):1039-1045. \u003c/li\u003e\n\u003cli\u003eQi LN, Xiang BD, Wu FX, et al. Circulating tumor cells undergoing EMT provide a metric for diagnosis and prognosis of patients with hepatocellular carcinoma. Cancer Res, 2018, 78(16): 4731‑4744.\u003c/li\u003e\n\u003c/ol\u003e\n"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"HCC, OS, Small-Sample ML, Clinicopathological Characteristic, Hematological Biomarker, CTC, CEC","lastPublishedDoi":"10.21203/rs.3.rs-6176116/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6176116/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e Early-onset hepatocellular carcinoma (HCC) is insidious, with characteristics of easy metastasis, high recurrence rate, and significant mortality. To address the substantial time and resource demands associated with HCC prognostic prediction, we extract meaningful insights from limited small-sample data to develop and validate a prediction model for HCC 5‑year overall survival (OS) by machine learning (ML).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e：76 newly diagnosed patients with HCC were eventually enrolled between September 2018 and July 2019. The follow-up time was 1-67 months. Patients who survived for 5 years after the first surgery, were divided into a surviving group (n=34) and a nonsurviving group (n=42). The pathological data and related survival factors of patients were collected before treatment. The final subset of features was filtered. Prediction models for 5-year OS in patients with HCC were established by logistic regression (LR), support vector machine (SVM), decision tree classification (DTC), random forests (RF), and extreme gradient Boosting (XGBoost), respectively. Additionally, the optimal model was established after rigorous validation. The models were evaluated by values of specificity, F1 score, recall, accuracy and area under the receiver operating characteristic curve (AUC-ROC).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e The significant variable set, which included 22 variables, was screened. Ranking the importance of variables, the top 22 characteristic variables were as follows: the maximum diameter, presence or absence of distant metastasis, CNLC stage, ALB, age, RBC, the large sizeCTC, total bilirubin (TBIL), PD-L1 (-) CTC, ≥ Pentaploid CTC, AFP, vascular cancer thrombus and satellite nodules, WBC, CTC, BCLC stage, multiple nodules, AST, PD-L1 (-) CTC-WBC cluster, Triploid CTC, LYM, PD-L1 (-) CEC-WBC cluster and degree of cirrhosis. The AUC-ROC values for predicting the 5-year OS rate of HCC patients by the logistic regression, SVM, DTC, RF, and XGBoost models were 0.737, 0.971, 0.657, 0.741, and 0.703, respectively. Among them, the SVM model had the best performance (Accuracy=0.987, F1 score=0.988, Recall value =1.000).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003e The SVM model could predict the 5-year OS in HCC with good recognition ability and achieves significantly greater accuracy compared to traditional models. Diagnosis and treatment could be utilized to intervene in the risk factors in this model, thereby improving patient prognosis.\u003c/p\u003e","manuscriptTitle":"Development and validation of a small-sample machine learning model to predict 5-year overall survival in patients with hepatocellular carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-30 11:43:20","doi":"10.21203/rs.3.rs-6176116/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-05-15T21:12:30+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-11T15:32:17+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-09T06:43:51+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"295785281081836993740434037432997517865","date":"2025-04-27T14:48:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"134980981514886944143478066291891954316","date":"2025-04-24T14:08:30+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-22T07:17:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"156080612925855893813359678353302883732","date":"2025-04-22T07:09:46+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-22T02:15:41+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-21T22:37:18+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cancer","date":"2025-04-21T15:37:50+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"13cc8b04-d728-4355-8465-5e744c206b00","owner":[],"postedDate":"April 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-07-07T15:59:37+00:00","versionOfRecord":{"articleIdentity":"rs-6176116","link":"https://doi.org/10.1186/s12885-025-14425-0","journal":{"identity":"bmc-cancer","isVorOnly":false,"title":"BMC Cancer"},"publishedOn":"2025-07-01 15:57:01","publishedOnDateReadable":"July 1st, 2025"},"versionCreatedAt":"2025-04-30 11:43:20","video":"","vorDoi":"10.1186/s12885-025-14425-0","vorDoiUrl":"https://doi.org/10.1186/s12885-025-14425-0","workflowStages":[]},"version":"v1","identity":"rs-6176116","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6176116","identity":"rs-6176116","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}