Development and Interpretability Analysis of a Machine Learning-Based Model for Predicting Early Liver Metastasis Risk After Pancreatic Cancer Surgery

Development and Interpretability Analysis of a Machine Learning-Based Model for Predicting Early Liver Metastasis Risk After Pancreatic Cancer Surgery | 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 Interpretability Analysis of a Machine Learning-Based Model for Predicting Early Liver Metastasis Risk After Pancreatic Cancer Surgery Chenhui Du, Shuo Zhang, gGuoyu Li, Xinling Cao, Tieying He This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7841940/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 15 You are reading this latest preprint version Abstract Background Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal malignancy. Despite undergoing radical surgical resection, patients are still at a high risk of recurrence and distant metastasis postoperatively. Among the organs prone to hematogenous metastasis, the liver is the most common site, and liver metastasis significantly shortens the survival period, becoming a key factor influencing prognosis. Objective This study aims to develop an interpretable machine learning model based on postoperative clinical variables to predict the risk of liver metastasis within one year after surgery in pancreatic cancer patients. Methods This study included data from 418 patients who underwent radical pancreatic cancer surgery at the Department of Gastrointestinal Surgery, First Affiliated Hospital of Xinjiang Medical University, between January 2015 and August 2024. The data were randomly divided into a training set and a test set in a 7:3 ratio. The performance of seven machine learning models was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC-ROC) and the area under the precision-recall curve (AP-PR). SHAP and LIME methods were used to determine feature importance and explain the best-performing model. Results After applying inclusion and exclusion criteria, 363 patients were included in the study. Among them, 118 patients (32.5%) developed liver metastasis within one year postoperatively. The final model incorporated 10 variables: chemotherapy status, tumor differentiation, vascular invasion (arterial/venous), hepatitis B infection, CA19-9 level, T stage, lymphocyte count, albumin level, alkaline phosphatase, and tumor size. Among the seven machine learning models, the Extra Trees (ET) model performed the best, achieving an AUC-ROC of 0.82 (95% CI: 0.73–0.90) and an average precision (AP-PR) of 0.77 in the test set. SHAP analysis revealed that postoperative chemotherapy, tumor differentiation, hepatic artery/portal vein invasion, and hepatitis B virus infection were the most influential predictors of liver metastasis. Conclusion An interpretable machine learning model was developed using postoperative clinical data, demonstrating good performance and interpretability. The model effectively predicts the risk of liver metastasis within one year after pancreatic cancer surgery. It holds promise as an auxiliary tool for postoperative follow-up and personalized interventions, providing clinicians with more precise decision-making support through feature contribution analysis. Machine Learning Pancreatic Cancer Liver Metastasis Predictive Model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Introduction Pancreatic cancer (PC) is one of the most challenging malignancies in clinical practice. Its high mortality rate and the lack of early symptoms make diagnosis and treatment extremely difficult. According to the 2022 global cancer statistics [ 1 ] , PC accounts for 2.6% of all cancers in age-standardized incidence rates, ranking 12th globally among malignant tumors. Its mortality rate, however, accounts for 4.8% of all cancer-related deaths, ranking it 6th, reflecting the extremely poor prognosis of pancreatic cancer, which is significantly worse than most other solid tumors. Due to the pancreas’s deep location in the abdomen, tumors are often difficult to detect early, resulting in most patients being diagnosed with locally advanced or metastatic disease [ 2 ] . Only about 10% of patients are eligible for radical surgery. Even among those who undergo radical resection, 90% of patients will experience tumor recurrence within 7 to 9 months postoperatively [ 3 ] . Liver metastasis [ 4 ] is one of the most common sites of distant metastasis in PC, accounting for 23.5% of initial recurrences, with most occurring within the first year after surgery. The appearance of liver metastasis usually signifies disease progression, and postoperative liver metastasis in pancreatic cancer not only significantly reduces survival but also affects subsequent treatment decisions. Studies have shown that early detection and treatment of postoperative liver metastasis can prolong survival. Therefore, predicting the occurrence of postoperative liver metastasis is of critical importance for clinicians in postoperative management, monitoring, and early intervention. Machine learning (ML), a branch of artificial intelligence(AI) [ 5 ] , has shown great potential in predicting clinical outcomes. Compared to traditional statistical methods, ML excels at handling complex interactions and nonlinear relationships, thereby reducing decision-making time [ 6 ] in clinical practice. Previous studies [ 7 – 9 ] have confirmed that ML models, trained on large amounts of clinical data, significantly improve the accuracy of risk prediction for various diseases, such as cancer and chronic diseases, and outperform traditional models in both prediction and diagnosis. Despite the great potential of ML in clinical practice, several challenges remain, with the "black box [ 10 ] " issue being one of the main obstacles. Most ML models are not transparent and difficult to interpret, making it hard for clinicians to understand and trust the decision-making process. Therefore, this study aims to develop an interpretable ML model using clinical data to predict high-risk populations for early liver metastasis after pancreatic cancer surgery, providing scientific evidence and decision support for the development of individualized treatment plans for postoperative patients. Materials and methods Subjects and study design Collection of Clinical Data from 418 Patients with PC Undergoing Radical Surgery at the Digestive Surgery Center of the First Affiliated Hospital of Xinjiang Medical University from January 2015 to August 2024. Clinical data from 418 patients who underwent radical surgery for PC at the Digestive Surgery Center of the First Affiliated Hospital of Xinjiang Medical University between January 2015 and August 2024 were collected. The inclusion criteria were as follows: (1)Pathological diagnosis of PC confirmed after surgery, with no history of other systemic tumors before surgery.(2)Patients who had not received neoadjuvant or conversion therapy [ 11 ] prior to surgery.(3)Preoperative imaging examinations, including enhanced CT, MRI, and PET-CT, showed no evidence of distant metastasis. Preoperative 3D vascular reconstruction did not show tumor invasion of blood vessels (such as the celiac trunk, common hepatic artery, or portal vein) or local vascular invasion (invasion ≤ 180°or complete removal of the arterial sheath is feasible).(4)Intraoperative frozen section and postoperative pathology both confirmed R0 resection.(5)All patients underwent close follow-up as per medical instructions postoperatively. The exclusion criteria were as follows:(1) History of other systemic malignant tumors.(2) Preoperative receipt of neoadjuvant or conversion therapy.(3) Lack of follow-up records or absence of relevant imaging data postoperatively.(4) Perioperative death: patients who died within 90 days after surgery. Based on the inclusion and exclusion criteria, 363 patients were ultimately included. These patients were randomly assigned to the training cohort and the testing cohort in a 7:3 ratio. Figure 1 presents a flowchart summarizing the study design. Variable collection This study retrospectively collected multimodal clinical data from patients, including:(1)Demographic Information: Age, gender, weight, medical history, smoking and alcohol consumption history, etc.(2) Preoperative Laboratory Data: Complete blood count, liver function tests, tumor markers, etc.(3) Preoperative Imaging Data: Abdominal enhanced CT, MRI reports, etc.(4) Intraoperative Data: Surgical method, operation time, blood loss, etc.(5) Postoperative Pathological Data: Maximum tumor diameter, lymph node metastasis, degree of differentiation, etc. The surgical methods in this study included pancreaticoduodenectomy and distal pancreatectomy with splenectomy (or adrenalectomy). Intraoperative vascular invasion was addressed by the lead surgeon, who selected either artificial vascular replacement or partial vessel wall resection based on the extent of vascular invasion. A drainage tube was routinely placed in the surgical area for all patients, with the number of drainage tubes determined by the lead surgeon according to the intraoperative situation. The surgeries were performed by four qualified surgeons, each holding a position of associate chief physician or higher, with over 10 years of surgical experience. Postoperative abdominal hemorrhage and pancreatic fistula were defined and graded according to the International Study Group of Pancreatic Surgery (ISGPS) [ 12 , 13 ] . In this study, the cases included were those with moderate to severe intra-abdominal hemorrhage and B or C grade pancreatic fistulas. The diagnostic criteria for abdominal infection were based on the "Revised Guidelines for the Management of Intra-abdominal Infections" by the Surgical Infection Society [ 14 ] . Postoperative follow-up Follow-up in this study was conducted through outpatient visits and telephone follow-ups. Patients were scheduled for follow-up every 3 months during the first year after surgery, and subsequently every 6–12 months. During each outpatient follow-up visit, patients underwent abdominal enhanced CT, complete blood count, blood biochemistry, tumor marker tests, and other relevant examinations. Telephone follow-up was conducted by inquiring or receiving the patient's local hospital examination results. When enhanced CT suggested suspected liver recurrence or metastasis, the patient was admitted for further confirmation of the lesion through various methods. The endpoint of the follow-up in this study was the early recurrence of tumors in the liver. There is currently no unified standard for defining early recurrence. Based on domestic and international research on early recurrence of PC after surgery, early liver recurrence [ 15 , 16 ] was defined as follows: during regular follow-up within one year after radical surgery for PC, liver lesions confirmed by imaging tests and/or laboratory tests and/or pathological examinations as new lesions originating from metastasis of the primary tumor. In cases where pathological diagnosis was missing or unclear, imaging diagnosis was prioritized, and tumor markers were referenced. The follow-up data was collected until August 31, 2025. Feature Selection Based on the occurrence of the follow-up endpoint event, patients were divided into two groups: the liver metastasis group and the non-metastasis group. The patients were randomly assigned to the training set and the testing set in a 7:3 ratio. Univariate analysis was first performed to preliminarily identify clinical variables with significant differences, with a significance threshold set at α < 0.05. Using the training set cohort, feature selection was conducted with LASSO regression (utilizing its L1 regularization feature to perform dimensionality reduction, with the optimal λ value determined by ten-fold cross-validation to avoid overfitting), the Boruta algorithm (an automated feature selection method based on random forests, which calculates the importance of both original and shadow features, where shadow features provide a random baseline and enhance generalization capability), and recursive feature elimination (RFE) based on XGBoost. The key features selected through these methods were then incorporated into the predictive model. Model construction and validation Based on the selected variables, seven machine learning methods were applied to construct predictive models in the training set cohort. These algorithms included Decision Tree (DT), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Logistic Regression (LR), and Extra Trees (ET). Subsequently, multiple metrics were used to evaluate the performance of each model, including the Area Under the Receiver Operating Characteristic Curve (AUC), Precision-Recall (PR) curve, accuracy, sensitivity, specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and F1 score. To assess the robustness and generalization ability of the models, hyperparameter grid search and 5-fold cross-validation were first conducted on the training set, followed by external validation on the testing set to evaluate generalization performance. The best model was ultimately selected. To further assess the clinical applicability of the model, Decision Curve Analysis (DCA) was performed, which calculates the net benefit at different risk thresholds to determine the clinical value of the best model. Statistical analysis All statistical analyses were performed using R 4.4.3 and Python 3.2 software. Data preprocessing was conducted first, analyzing the types of variables and missing data. Variables with a missing rate ≥ 30% were excluded, while variables with missing data < 30% were imputed using multiple imputation methods. For categorical data, comparisons between groups were made using the Chi-square (X²) test or Fisher’s exact test. For continuous data, normality tests were performed first. Normally distributed continuous data were expressed as mean ± standard deviation (x ± s), and comparisons between two groups were made using the t-test. For non-normally distributed continuous data, data were expressed as median (P25-P75), and the Mann-Whitney U test was used for comparisons between two groups. A two-sided P-value < 0.05 was considered statistically significant. Given the inherent "black-box" nature of machine learning models, we employed several interpretability techniques to explain the models. These included Shapley Additive Explanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Partial Dependence Plots (PDP)/Individual Conditional Expectation Plots (ICE). Results Baseline characteristics of patients A total of 363 patients were included in the study, of which 118 patients developed early liver metastasis, while 245 patients did not. The patients were randomly divided into two groups in a 7:3 ratio, with 254 patients in the training set and 109 patients in the testing set. In the training set, 83 patients (32.67%) developed early liver metastasis, and in the testing set, 35 patients (32.11%) developed early liver metastasis. There were no statistically significant differences in baseline characteristics between the two groups ( P > 0.05), as shown in Table 1 . Table 1 Comparison of baseline data features between the training and test sets. Train(n = 254) Test(n = 109) P Sex c 0.639 Female 122 (48) 56 (51) Male 132 (52) 53 (49) Chemotherapy c 0.946 No 28 (11) 13 (12) Yes 226 (89) 96 (88) Hypertension c 0.442 No 173 (68) 69 (63) Yes 81 (32) 40 (37) Diabetes2 c 0.525 No 186 (73) 84 (77) Yes 68 (27) 25 (23) Drinking history c 0.288 No 201 (79) 80 (73) Yes 53 (21) 29 (27) Smoking history c 0.512 No 194 (76) 79 (72) Yes 60 (24) 30 (28) PBD c 0.884 No 223 (88) 97 (89) Yes 31 (12) 12 (11) AV invasion c 1 No 194 (76) 83 (76) Yes 60 (24) 26 (24) Nerve invasion c 0.113 No 41 (16) 10 (9) Yes 213 (84) 99 (91) Vascular invasion c 0.9 No 148 (58) 65 (60) Yes 106 (42) 44 (40) Grade c 0.145 Moderately to well-differentiated 131 (52) 66 (61) Poorly differentiated 123 (48) 43 (39) Pathological type c 0.769 PDAC 224 (88) 98 (90) Other 30 (12) 11 (10) Tumor location c 0.889 pancreatic head 190 (75) 83 (76) Pancreatic body and tail 64 (25) 26 (24) POH c 0.567 No 242 (95) 106 (97) Yes 12 (5) 3 (3) POIAI c 0.965 No 184 (72) 78 (72) Yes 70 (28) 31 (28) POPF c 1 No 171 (67) 74 (68) Yes 83 (33) 35 (32) HBV c 0.318 No 211 (83) 85 (78) Yes 43 (17) 24 (22) Age b 62 (55, 69) 60 (56, 68) 0.686 Height(cm) b 167 (160, 174) 165 (160, 170) 0.156 Weight(kg) b 65 (58, 73) 64 (58, 71) 0.639 BMI(Kg/m 2 ) b 23.48 (20.87, 25.81) 23.44 (21.09, 25.71) 0.808 Surgical duration b 540 (450, 638.75) 545 (425, 655) 0.663 IBL b 500 (300, 787.5) 500 (300, 700) 0.826 WBC b 5.82 (4.87, 7.21) 5.92 (4.88, 7.38) 0.627 NE(10⁹/L) b 3.63 (2.96, 4.94) 3.74 (2.86, 4.83) 0.998 LYM(10⁹/L) b 1.41 (1.15, 1.84) 1.49 (1.15, 1.94) 0.398 PLT(10⁹/L) b 230.5 (188, 281) 232 (192, 286) 0.559 NLR b 2.56 (1.91, 3.74) 2.52 (1.66, 3.18) 0.213 PLR b 165.39 (120, 208.78) 155.48 (116.34, 204.41) 0.539 HB(g/L) b 132 (120, 142.75) 131 (120, 143) 0.983 TB (ummol/L) b 24.44 (14.7, 170.47) 23.81 (13.76, 141.85) 0.539 IB(ummol/L) b 16.98 (9.85, 35.75) 15.06 (9.19, 33.87) 0.469 ALB(g/L) a 39.55 ± 4.89 39.78 ± 4.95 0.681 A/G b 1.19 (1.05, 1.39) 1.23 (1.07, 1.4) 0.318 AST(U/L) b 43.72 (22.95, 132.19) 46 (25.3, 106.82) 0.986 ALT(U/L) b 50.37 (21.85, 213.4) 55.26 (21.33, 205.21) 0.601 AKP(U/L) b 134.43 (78.58, 334.62) 151.9 (78, 341.1) 0.842 CA19-9(U/mL) b 266.24 (37.84, 1200) 186.99 (41.76, 740.24) 0.427 CA50(U/mL) b 136.57 (34.32, 180) 78.17 (31.66, 180) 0.055 CEA(U/mL) b 3.05 (1.72, 5.79) 2.66 (1.82, 5.08) 0.305 Tumor size (cm) b 3.8 (3, 5) 3.5 (2.6, 4.5) 0.05 LNM b 0 (0, 1) 0 (0, 1) 0.482 T c 0.344 0 3 (1) 2 (2) 1 13 (5) 1 (1) 2 57 (22) 28 (26) 3 149 (59) 64 (59) 4 32 (13) 14 (13) N c 0.794 0 176 (69) 72 (66) 1 65 (26) 30 (28) 2 13 (5) 7 (6) a Continuous variables with a normal distribution were expressed as mean ± standard deviation, and inter-group comparisons were conducted using t-tests. b Continuous variables that did not follow a normal distribution were expressed as median [interquartile range], and inter-group comparisons were conducted using the Mann–Whitney U test. c Categorical variables were expressed as frequency (percentage), and inter-group comparisons were conducted using the X² test. PBD Preoperative Biliary Drainage, AV invasion Hepatic artery / portal vein invasion, POH Postoperative Hemorrhage, POIAI Postoperative Intra-abdominal Infection, POPF Postoperative Pancreatic Fistula, BMI Body Mass Index, IBL Intraoperative Blood Loss, NLR Neutrophil-lymphocyte ratio, PLR Platelet-to-Lymphocyte Ratio , LNM Lymph Node Metastasis. Feature Selection Based on the training set data, we performed feature selection using Lasso regression (Figs. 2 A, B), the Boruta algorithm (Figs. 2 C, D), and recursive feature elimination based on XGBoost (Fig. 2 F). According to the results of feature selection from these three methods (Fig. 2 G), we ultimately selected 10 variables for the development of the ML model. These 10 variables were: chemotherapy, hepatic artery/portal vein invasion, degree of differentiation, hepatitis B virus infection, CA19-9, T stage, lymphocyte count, albumin, tumor size, and alkaline phosphatase. Construction and Performance Comparison of Different ML Algorithms Based on the 10 selected features, this study used seven machine learning algorithms, including Decision Tree (DT), Support Vector Machine (SVM), XGBoost, Random Forest (RF), LightGBM, Logistic Regression (LR), and Extra Trees (ET), to construct the early liver metastasis model in the training set cohort. GridSearchCV was used for grid search to identify the optimal model. In the training set, all models, except for SVM, exhibited good performance. In the testing set, the ET and RF models showed excellent performance, with AUC values of 0.822 (95% CI: 0.726-0.900) and 0.814 (95% CI: 0.722–0.902), respectively (Figs. 3 A, B). Additionally, we calculated various performance metrics, including Accuracy, Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), F1 score, and Kappa score for model evaluation (Table 2 , Figs. 4 A, B). The confusion matrix (Fig. 5 ) and Precision-Recall (PR) curve (Fig. 6 A, B) for the testing set were also plotted. The ET model, which demonstrated a balanced performance across various metrics, was ultimately selected as the final model. Table 2 Performance of 7 ML algorithms on training and test sets Algorithms Accuracy Sensitivity Specificity PPV NPV F1 score Kappa score Train DT 0.913 0.831 0.953 0.896 0.831 0.863 0.799 ET 0.957 0.892 0.988 0.974 0.892 0.931 0.899 SVM 0.693 0.229 0.918 0.576 0.229 0.328 0.174 LightGBM 0.957 0.88 0.994 0.986 0.88 0.93 0.899 RF 0.976 0.94 0.994 0.987 0.94 0.963 0.946 LR 0.866 0.711 0.942 0.855 0.711 0.776 0.682 XGBoost 0.937 0.843 0.982 0.959 0.843 0.897 0.852 Test DT 0.761 0.629 0.824 0.629 0.629 0.629 0.453 ET 0.798 0.629 0.878 0.71 0.629 0.667 0.523 SVM 0.642 0.2 0.851 0.389 0.2 0.264 0.059 LightGBM 0.807 0.657 0.878 0.719 0.657 0.687 0.548 RF 0.798 0.629 0.878 0.71 0.629 0.667 0.523 LR 0.817 0.629 0.905 0.759 0.629 0.688 0.559 XGBoost 0.826 0.657 0.905 0.767 0.657 0.708 0.585 PPV positive predictive value, NPV negative predictive value Tuning and Performance of the Optimal Model To assess the robustness and generalization ability of the ET model, we plotted the learning curve and validation curve of the model (Figs. 7 A, B). The AUC for both the training set and cross-validation set increased with the sample size, and the gap between the two remained relatively stable, indicating that the model has good fitting ability and generalizability. To further optimize model performance and prevent overfitting, we adopted a two-stage hyperparameter tuning strategy. First, we used Randomized Search CV to perform a coarse search for the main hyperparameters, quickly identifying their effective ranges within a larger parameter space. Then, within the high-quality parameter regions identified by the randomized search, we conducted a more refined search using Grid Search CV to precisely select the optimal hyperparameter combination. The ROC curve, PR curve, and confusion matrix for the model under the best parameters were plotted (Figs. 8 A, B, C), showing that the performance of the optimal ET model improved across all aspects. Subsequently, we used Principal Component Analysis (PCA) for dimensionality reduction to visualize the data distribution in a two-dimensional space and observe the separation between different class samples. We performed PCA projection on both the training and testing datasets to evaluate the classification effect of the model and its generalization ability on different datasets (Figs. 9 A, B). In the training set, PC1 explained 80.05% of the variance, and the "non-metastatic" and "metastatic" categories showed clear separation in the two-dimensional space with almost no overlap, indicating that the model can effectively distinguish between different categories in the training set. In the testing set, PC1 explained 83.55% of the variance, and while there was still clear separation between "non-metastatic" and "metastatic" samples, some overlap was observed in certain areas. Interpretation Based on the Optimal ET Model This study provides global and local explanations of the model using SHAP, PDP, and ICE. The SHAP value plots visually demonstrate the direction and strength of the impact of each feature on the model's predictions. The heatmap reveals the positive or negative trend of the feature’s influence based on its value, while the bar and polar plots more clearly present the importance ranking of each feature in the global predictions (Fig. 10 A, B). To further explore the specific impact of each feature on the model's prediction, we plotted the SHAP dependence plots for each variable (Fig. 11 ). These plots show the relationship between the actual values of the features and their corresponding SHAP values, revealing how each feature influences the model's output in terms of direction and strength under different values. Additionally, we examined the nonlinear interactions between features and their effects on the model’s predictions. Using SHAP interaction values, we visualized the relationships between features. In the interaction between CA19-9 and tumor size, the distribution of interaction values was somewhat dispersed. As the largest diameter of the tumor increased, the SHAP interaction values for some high CA19-9 samples showed a slight upward trend, suggesting a potential joint effect between CA19-9 and tumor volume. In the interaction between CA19-9 and tumor differentiation, when the tumor was poorly differentiated, higher CA19-9 levels mainly resulted in negative SHAP interaction values, indicating that this combination was more likely to predict a positive outcome. In contrast, when the tumor was moderately to well-differentiated, a positive interaction was observed under low CA19-9 conditions, indicating a nonlinear combined effect of CA19-9 and tumor differentiation on the model’s prediction (Fig. 12 A、B). To explore the model's dependence on key features, we plotted partial dependence plots (PDPs) and individual conditional expectation plots (ICEs) for some important features. The PDPs illustrate the average effect of feature variation on the model's output, revealing global trends. Meanwhile, the ICE plots demonstrate the predictive response of individual samples to changes in feature values, highlighting heterogeneity among individual predictions (Fig. 13 A, B). Clinical identification of the optimal ET model To evaluate the contribution of features to individual patient predictions, we employed SHAP and LIME methods to explain the predictions for two sample patients. Two patients, A and B, were randomly selected from the true positive group, with predicted probabilities of 82% and 84%, respectively. Two patients, C and D, were randomly selected from the true negative group, with predicted probabilities of 13.7% and 24%, respectively (Fig. 14 ). Additionally, LIME was used to explain the prediction for patient A from the true positive group, where the predicted probability remained at 82% (Figs. 15 ). To further assess the clinical applicability of the model, we plotted the clinical decision curve (DCA). Decision curve analysis indicated that the machine learning model developed in this study provided a higher net benefit in predicting early postoperative liver metastasis, especially when the risk threshold ranged from 0.05 to 0.4. In this range, the model outperformed both the "all intervention" and "no intervention" strategies, suggesting that it holds clinical decision support potential, particularly for personalized management of high-risk patients (Fig. 16 ). Discussion Although curative surgery for resectable PC has become increasingly refined, the high postoperative recurrence rate and short survival time remain significant challenges for clinicians. Several studies [ 16 – 18 ] have shown that more than 50% of PC patients experience early recurrence after surgery, with the liver being the most common site of recurrence, often indicating poorer prognosis. This study retrospectively analyzed clinical data from a single-center cohort of PC patients who underwent curative surgery over a 10-year period, identifying predictive factors for early liver metastasis after surgery, which can aid in the formulation of appropriate treatment plans and monitoring strategies for patients with early liver metastasis. Postoperative adjuvant chemotherapy is a crucial component of treatment following curative resection of pancreatic cancer. In this study, SHAP analysis revealed that adjuvant chemotherapy is one of the most influential variables in predicting the risk of early liver metastasis after surgery, consistent with numerous domestic and international studies, further validating the clinical importance of adjuvant chemotherapy. Mechanistically, adjuvant chemotherapy can effectively eliminate microscopic residual disease that is undetectable preoperatively or intraoperatively, thereby delaying recurrence and suppressing metastasis. Several international multicenter randomized [ 19 ] controlled trials have also confirmed that both single-agent gemcitabine and combination chemotherapy regimens [ 20 ] can prolong overall survival and relapse-free survival in patients. However, in clinical practice, some patients fail to receive chemotherapy promptly due to slow postoperative recovery, concerns about toxicity, or poor adherence, leading to early recurrence and metastasis. Therefore, the timing of initiating adjuvant chemotherapy after surgery is critical. An analysis by Ma et al [ 21 ] . of 7,548 stage I–II pancreatic cancer patients from the National Cancer Database found that patients who started adjuvant chemotherapy 28 to 59 days postoperatively had the best overall survival. Domestic researchers also suggest that if there are no contraindications, adjuvant chemotherapy should be started as soon as possible after surgery, typically no later than 12 weeks postoperatively, and the treatment should continue for 6 months. This highlights the importance for clinicians to prioritize and promote the complete course of adjuvant chemotherapy in postoperative management. The liver, due to its unique dual blood supply system (portal vein and hepatic artery), makes it easier for tumor cells originating from the gastrointestinal tract and pancreas to become trapped in the circulation and colonize the liver parenchyma. The circulating tumor cell (CTC) hypothesis explains that once a tumor invades the portal venous system or celiac artery, cancer cells are easily disseminated to the liver through the bloodstream. In a meta-analysis involving 5242 PDAC patients, Song et al [ 22 ] . pointed out that those with portal vein/superior mesenteric vein (PV/SMV) invasion exhibited poorer biological behavior and had a higher risk of postoperative recurrence. Further research by Tong et al [ 23 ] . emphasized that the extent of portal vein invasion is an independent risk factor for poor postoperative prognosis. Although surgical resection may achieve R0, deeper vascular invasion implies higher risks of residual tumor and distant metastasis. For patients with locally advanced pancreatic tail cancer invading the celiac artery, although distal pancreatectomy with celiac artery resection [ 24 ] (DP-CAR) can provide surgical opportunities, it does not significantly improve the rates of curative resection or long-term survival. Compared with traditional distal pancreatectomy (DP), DP-CAR results in a lower R0 resection rate, higher intraoperative risk, and no significant difference in survival. Low-grade tumors are often more aggressive and metastatic. At the molecular level, poorly differentiated or undifferentiated PDAC cells exhibit higher activity in several key metastatic pathways [ 25 – 27 ] . Firstly, poorly differentiated tumors commonly activate epithelial-mesenchymal transition (EMT) pathways, which reduce cell adhesion and enhance migration capacity, significantly increasing the risk of liver metastasis. Secondly, low-grade PDAC is often accompanied by TP53 mutations and loss of SMAD4 expression, both of which are highly correlated with liver metastasis. Furthermore, poorly differentiated PDAC reshapes the liver microenvironment by releasing MIF-rich exosomes, establishing a pre-metastatic niche, and thereby gaining a metastatic advantage. Therefore, the degree of differentiation not only serves as a morphological pathological indicator but also represents the tumor's potential for systemic dissemination. This study focused on the potential of preoperative CA19-9 levels as a predictor for early liver metastasis risk after PC surgery. Previous studies have shown [ 28 ] that high preoperative CA19-9 levels are strongly associated with a high risk of distant metastasis. Guan et al [ 29 ] . found that elevated CA19-9 significantly shortened both disease-free survival (DFS) and overall survival (OS), making it an independent predictor for both DFS and OS. Further analysis by Raza et al [ 30 ] . revealed that the average preoperative CA19-9 level in metastatic patients was significantly higher than in non-metastatic patients, with an ROC cutoff point of 336 U/mL (sensitivity 90%, specificity 80%, AUC = 0.90). Tumor size has also been validated as a major risk factor for postoperative liver metastasis in multiple studies. Murakami et al [ 31 ] . discovered that a preoperative tumor size ≥ 22 mm is a significant predictor of hidden metastasis and is associated with poor overall prognosis. Ayoub et al [ 32 ] . compared the 1- and 2-year survival rates of patients with pancreatic head tumors ≤ 3 cm and > 3 cm, and found that patients with tumors > 3 cm had significantly lower 1-year and 2-year survival rates (50% vs. 79.1% and 19.2% vs. 40.3%, respectively). In this study, SHAP interaction analysis further revealed the synergistic effect between tumor size and CA19-9, particularly when CA19-9 levels exceeded 600 U/mL, where the interaction SHAP values of some samples increased significantly, suggesting that tumor size had a stronger positive contribution to the model’s output at these levels. Overall, larger tumor size and higher preoperative CA19-9 levels may reflect a larger tumor burden and the possible presence of hidden distant micro-metastasis, marking them as important risk indicators for early postoperative liver metastasis. Hepatitis B virus (HBV), a hepatotropic DNA virus, persistently infects the liver, triggering chronic inflammation. In this study, we found that patients with preoperative HBV infection had a significantly higher risk of early liver metastasis after surgery compared to those without HBV infection. We hypothesize that the chronic inflammatory environment in the liver may alter the liver microenvironment, promoting increased tumor invasiveness. This finding is consistent with the study by Wei et al [ 33 ] . However, a study by Chen [ 34 ] et al. in patients with unresectable PC reported that patients with chronic active HBV infection had a lower risk of postoperative liver metastasis, which contradicts our findings. These discrepancies may be due to differences in study populations (resectable vs. unresectable) and HBV infection status, among other factors. Overall, current evidence does not provide a consistent conclusion regarding whether HBV acts as a "risk factor" or "protective factor" for early postoperative liver metastasis in PC. In addition to traditional tumor burden indicators, this study also identified low albumin levels and lymphocyte count as predictors for early liver metastasis after PC surgery. Albumin, a sensitive marker reflecting nutritional status and chronic inflammation, was identified as a negative protective factor in our study, with lower levels being associated with an increased risk of early liver metastasis. Lymphocytes, as a key component of the immune surveillance system, are also one of the factors contributing to an increased risk of early liver metastasis postoperatively. A study by Gumus [ 35 ] et al. pointed out that low albumin levels and low HALP scores (a combination of hemoglobin, albumin, lymphocytes, and platelets) were significantly associated with shorter DFS and OS, making them independent prognostic factors for poor outcomes. Although our study provides valuable insights, several limitations must be acknowledged. First, as a single-center retrospective study, the model's generalizability needs to be further validated in multi-center, large-scale prospective studies. Second, the study primarily relied on conventional clinical and laboratory indicators and did not incorporate radiomics or pathological imaging data. Third, while chemotherapy was included as a variable in the model, the differences between various chemotherapy regimens were not specifically analyzed. Fourth, the sample size of this study was relatively small, limiting the full potential of the ML model. Additionally, patients who received neoadjuvant or conversion therapy before surgery were excluded, yet these treatment strategies have become increasingly widespread in recent years, meaning the study sample may not fully represent the broader patient population, thereby restricting the model's clinical applicability. Therefore, future research should focus on multi-center, large-scale data, integrating multi-modal data, and utilizing methods like deep learning to further enhance the model's accuracy. Conclusions This study developed a predictive model for early liver metastasis after PC surgery using ML methods, which demonstrated good validation results in the test set and strong clinical interpretability. The model provides robust support for the early identification and personalized management of high-risk patients post-surgery, assisting clinicians in formulating individualized treatment plans. Abbreviations PC Pancreatic cancer, ML Machine learning,AI artificial intelligence, LASSO Least absolute shrinkage and selection operator, AUC Area under the receiver operating characteristic, PPV Positive predictive value, NPV Negative predictive value, DT Decision Tree, SVM Support Vector Machine, XGBoost eXtreme Gradient Boosting, RF Random Forst, LightBGM Light Gradient Boosting Machine, LR Logistic Regression, ET Extra Trees. SHAP Shapley Additive Explanations, LIME Local Interpretable Model-agnostic Explanations, PDP Partial Dependence Plot, ICE Individual Conditional Expectation. ISGPS International Study Group of Pancreatic Surgery. RFE Recursive feature elimination. PBD Preoperative Biliary Drainage, AV invasion Hepatic artery / portal vein invasion, POH Postoperative Hemorrhage, POIAI Postoperative Intra-abdominal Infection, POPF Postoperative Pancreatic Fistula, BMI Body Mass Index, IBL Intraoperative Blood Loss, NLR Neutrophil-lymphocyte ratio , PLR Platelet-to-Lymphocyte Ratio ,LNM Lymph Node Metastasis. PCA Principal. AP Average Precision, PR Precision-Recall, DCA Decision curve analysis, CTC Circulating tumor cell, DP-CAR Distal pancreatectomy with celiac artery resection, DP Distal pancreatectomy, EMT Epithelial-mesenchymal transition, DFS Disease-free survival, OS overall survival, HBV Hepatitis B virus Declarations Ethics approval and consent to participate: This study was conducted in accordance with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University, and all participants provided written informed consent prior to participation in the study. Consent for publication: Written informed consent for publication of identifying information and/or images was obtained from the patient(s) included in the study. Availability of data and materials:The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Competing interests: The authors declare that they have no competing interests. Funding: Funded Project of the State Key Laboratory of Causes and Prevention of High Morbidity in Central Asia (No. SKL-HIDCA-2023-26), jointly built by the Ministry and the Province. Authors' contributions: Chenhui Du was responsible for manuscript writing and data analysis. Shuo Zhang contributed to data collection. Guoyu Li provided the research concept. Xinling Cao and Tieying He critically reviewed the manuscript. Acknowledgements:The authors would like to thank OpenAI's ChatGPT for its guidance and assistance in refining the code for the model construction in this study. References Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Cancer J Clin. 2024;74(3):229–63. Strobel O, Neoptolemos J, Jäger D, Büchler MW. Optimizing the outcomes of pancreatic cancer surgery. Nat Rev Clin Oncol. 2019;16(1):11–26. Rojas LA, Sethna Z, Soares KC, Olcese C, Pang N, Patterson E, Lihm J, Ceglia N, Guasp P, Chu A, Yu R, Chandra AK, Waters T, Ruan J, Amisaki M, Zebboudj A, Odgerel Z, Payne G, Derhovanessian E, Balachandran VP. Personalized RNA neoantigen vaccines stimulate T cells in pancreatic cancer. Nature. 2023;618(7963):144–50. Rompen IF, Levine J, Habib JR, Sereni E, Mughal N, Hewitt DB, Sacks GD, Welling TH, Simeone DM, Kaplan B, Berman RS, Cohen SM, Wolfgang CL, Javed AA. Progression of site-specific recurrence of pancreatic cancer and implications for treatment. Ann Surg. 2024;280(2):317–24. Cuocolo R, Caruso M, Perillo T, Ugga L, Petretta M. Machine learning in oncology: A clinical appraisal. Cancer Lett. 2020;481:55–62. Chung H, Ko Y, Lee I-S, Hur H, Huh J, Han S-U, Kim KW, Lee J. Prognostic artificial intelligence model to predict 5 year survival at 1 year after gastric cancer surgery based on nutrition and body morphometry. J Cachexia Sarcopenia Muscle. 2023;14(2):847–59. Gao Y, Li S, Jin Y, Zhou L, Sun S, Xu X, Li S, Yang H, Zhang Q, Wang Y. An assessment of the predictive performance of current machine learning-based breast cancer risk prediction models: Systematic review. JMIR Public Health Surveillance. 2022;8(12):e35750. Wang J, Yao X. Which approach better predicts diabetes: Traditional econometric methods or machine learning? Evidence from a cross-sectional study in south korea. Comput Biol Med. 2025;190:110035. Qian X, Pei J, Han C, Liang Z, Zhang G, Chen N, Zheng W, Meng F, Yu D, Chen Y, Sun Y, Zhang H, Qian W, Wang X, Er Z, Hu C, Zheng H, Shen D. A multimodal machine learning model for the stratification of breast cancer risk. Nat Biomedical Eng. 2025;9(3):356–70. Chaddad A, Peng J, Xu J, Bouridane A. Survey of explainable AI techniques in healthcare. Sensors. 2023;23(2):634. Conroy T, Pfeiffer P, Vilgrain V, Lamarca A, Seufferlein T, O’Reilly EM, Hackert T, Golan T, Prager G, Haustermans K, Vogel A, Ducreux M, ESMO Guidelines Committee. &. Electronic address: [email protected] . (2023). Pancreatic cancer: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. Annals of Oncology: Official Journal of the European Society for Medical Oncology , 34 (11), 987–1002. Wente MN, Veit JA, Bassi C, Dervenis C, Fingerhut A, Gouma DJ, Izbicki JR, Neoptolemos JP, Padbury RT, Sarr MG, Yeo CJ, Büchler MW. Postpancreatectomy hemorrhage (PPH): An international study group of pancreatic surgery (ISGPS) definition. Surgery. 2007;142(1):20–5. Bassi, C., Marchegiani, G., Dervenis, C., Sarr, M., Abu Hilal, M., Adham, M., Allen,P., Andersson, R., Asbun, H. J., Besselink, M. G., Conlon, K., Del Chiaro, M., Falconi,M., Fernandez-Cruz, L., Fernandez-Del Castillo, C., Fingerhut, A., Friess, H., Gouma,D. J., Hackert, T., … International Study Group on Pancreatic Surgery (ISGPS). (2017).The 2016 update of the international study group (ISGPS) definition and grading of postoperative pancreatic fistula: 11 years after. Surgery , 161 (3), 584–591. Mazuski JE, Tessier JM, May AK, Sawyer RG, Nadler EP, Rosengart MR, Chang PK, O’Neill PJ, Mollen KP, Huston JM, Diaz JJ, Prince JM. The surgical infection society revised guidelines on the management of intra-abdominal infection. Surg Infect. 2017;18(1):1–76. Zhang X-P, Xu S, Gao Y-X, Zhao Z-M, Zhao G-D, Hu M-G, Tan X-L, Lau WY, Liu R. Early and late recurrence patterns of pancreatic ductal adenocarcinoma after pancreaticoduodenectomy: A multicenter study. Int J Surg (London England). 2023;109(4):785–93. Daamen, L. A., Dorland, G., Brada, L. J. H., Groot, V. P., van Oosten, A. F., Besselink,M. G., Bosscha, K., Bonsing, B. A., Busch, O. R., Cirkel, G. A., van Dam, R. M., Festen,S., Groot Koerkamp, B., Haj Mohammad, N., van der Harst, E., de Hingh, I. H. J. T.,Intven, M. P. W., Kazemier, G., Los, M., … Dutch Pancreatic Cancer Group. (2022).Preoperative predictors for early and very early disease recurrence in patients undergoing resection of pancreatic ductal adenocarcinoma. HPB: The Official Journal of the International Hepato Pancreato Biliary Association , 24 (4), 535–546. Groot VP, Gemenetzis G, Blair AB, Ding D, Javed AA, Burkhart RA, Yu J, Borel Rinkes IH, Molenaar IQ, Cameron JL, Fishman EK, Hruban RH, Weiss MJ, Wolfgang CL, He J. Implications of the Pattern of Disease Recurrence on Survival Following Pancreatectomy for Pancreatic Ductal Adenocarcinoma. Ann Surg Oncol. 2018;25(8):2475–83. Groot VP, Rezaee N, Wu W, Cameron JL, Fishman EK, Hruban RH, Weiss MJ, Zheng L, Wolfgang CL, He J. Patterns, timing, and predictors of recurrence following pancreatectomy for pancreatic ductal adenocarcinoma. Ann Surg. 2018;267(5):936–45. Klaiber U, Hackert T, Neoptolemos JP. Adjuvant treatment for pancreatic cancer. Translational Gastroenterol Hepatol. 2019;4:27. Oettle H, Post S, Neuhaus P, Gellert K, Langrehr J, Ridwelski K, Schramm H, Fahlke J, Zuelke C, Burkart C, Gutberlet K, Kettner E, Schmalenberg H, Weigang-Koehler K, Bechstein W-O, Niedergethmann M, Schmidt-Wolf I, Roll L, Doerken B, Riess H. Adjuvant chemotherapy with gemcitabine vs observation in patients undergoing curative-intent resection of pancreatic cancer: A randomized controlled trial. JAMA. 2007;297(3):267–77. Ma SJ, Oladeru OT, Miccio JA, Iovoli AJ, Hermann GM, Singh AK. Association of timing of adjuvant therapy with survival in patients with resected stage I to II pancreatic cancer. JAMA Netw Open. 2019;2(8):e199126. Song A, Liu F, Wu L, Si X, Zhou Y. Histopathologic tumor invasion of superior mesenteric vein/ portal vein is a poor prognostic indicator in patients with pancreatic ductal adenocarcinoma: Results from a systematic review and meta-analysis. Oncotarget. 2017;8(20):32600–7. Tong J, Jiang W, Mao S, Wu S, Lu C. Development and validation of a nomogram to predict liver metastasis for pancreatic ductal adenocarcinoma after radical resection. Front Oncol. 2022;12:1040411. Feng Q, Xin Z, Du Y, Mao F, Li L, Zhai H, Yao J. Distal pancreatectomy with en bloc celiac axis resection does not improve the R0 rate or median survival time of patients with locally advanced pancreatic cancer: A systematic review and meta-analysis. Translational Cancer Res. 2020;9(11):7205–13. Hoshino, A., Costa-Silva, B., Shen, T.-L., Rodrigues, G., Hashimoto, A., Tesic Mark,M., Molina, H., Kohsaka, S., Di Giannatale, A., Ceder, S., Singh, S., Williams, C.,Soplop, N., Uryu, K., Pharmer, L., King, T., Bojmar, L., Davies, A. E., Ararso, Y.,… Lyden, D. (2015). Tumour exosome integrins determine organotropic metastasis. Nature , 527 (7578), 329–335. Borrelli, C., Roberts, M., Eletto, D., Hussherr, M.-D., Fazilaty, H., Valenta, T.,Lafzi, A., Kretz, J. A., Guido Vinzoni, E., Karakatsani, A., Adivarahan, S., Mannhart,A., Kimura, S., Meijs, A., Baccouche Mhamedi, F., Acar, I. E., Handler, K., Ficht,X., Platt, R. J., … Moor, A. E. (2024). In vivo interaction screening reveals liver-derived constraints to metastasis. Nature , 632 (8024), 411–418. Masugi, Y., Takamatsu, M., Tanaka, M., Hara, K., Inoue, Y., Hamada, T., Suzuki, T.,Arita, J., Hirose, Y., Kawaguchi, Y., Nakai, Y., Oba, A., Sasahira, N., Shimane, G.,Takeda, T., Tateishi, K., Uemura, S., Fujishiro, M., Hasegawa, K., … GTK Pancreatic Cancer Study Group in Japan. (2023). Post-operative mortality and recurrence patterns in pancreatic cancer according to KRAS mutation and CDKN2A, p53, and SMAD4 expression. The Journal of Pathology. Clinical Research , 9 (5), 339–353. Kim TH, Han S-S, Park S-J, Lee WJ, Woo SM, Yoo T, Moon SH, Kim SH, Hong EK, Kim DY, Park J-W. CA 19 – 9 level as indicator of early distant metastasis and therapeutic selection in resected pancreatic cancer. Int J Radiat Oncol Biol Phys. 2011;81(5):e743–748. Guan S, Chen Y, Han Q, Deng G, Wang Y, Shi Y, Dai G. Preoperative CA19-9 levels predict disease-free survival and overall survival in pancreatic adenocarcinoma patients after resection. Translational Cancer Res. 2019;8(3):811–20. Raza SS, Khan H, Hajibandeh S, Hajibandeh S, Bartlett D, Chatzizacharias N, Roberts K, Marudanayagam R, Sutcliffe RP. Can preoperative carbohydrate antigen 19 – 9 predict metastatic pancreatic cancer? Results of a systematic review and meta-analysis. HPB: Official J Int Hepato Pancreato Biliary Association. 2024;26(5):630–8. Murakami T, Kimura Y, Imamura M, Nagayama M, Kato T, Kukita K, Yoshida M, Masaki Y, Nakase H, Takemasa I. Predictors of occult metastases in potentially resectable pancreatic ductal adenocarcinoma. Surg Open Sci. 2024;20:222–9. Ayoub II, Talab TAE, Omar H, Akoud SAB, Shoreem HAM, El-Deen EMS, Marwan IK, Macshut M. Impact of pancreatic head tumor size on the outcome of surgical management. Egypt J Surg. 2024;43(1):258. Wei X, Qiu M, Chen W, Jin Y, Ren C, Wang F, Luo H, Wang Z, Zhang D, Wang F, Li Y, Xu R. The status of HBV infection influences metastatic pattern and survival in Chinese patients with pancreatic cancer. J Translational Med. 2013;11:249. Chen, Q., Ning, Z., Wang, L., Ying, H., Dong, S., Zhang, C., Shen, X., Guo, Y., Chen,H., Zhu, X., Shen, Y., Shi, W., Hua, Y., Wang, K., Lin, J., Xu, L., Chen, L., Feng,L., Zhang, X., … Meng, Z. (2016). Is chronic hepatitis B infection a protective factor for the progression of advanced pancreatic ductal adenocarcinoma? An analysis from a large multicenter cohort study. Oncotarget , 7 (51), 85603–85612. Gumus T, Umman V, Cetin B, Uguz A. Utilizing albumin value, HALP score and LCR value for predicting survival in patients with pancreatic adenocarcinoma. Med (Kaunas Lithuania). 2025;61(4):639. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 12 Nov, 2025 Reviews received at journal 11 Nov, 2025 Reviews received at journal 10 Nov, 2025 Reviewers agreed at journal 09 Nov, 2025 Reviews received at journal 05 Nov, 2025 Reviewers agreed at journal 04 Nov, 2025 Reviewers agreed at journal 04 Nov, 2025 Reviewers agreed at journal 04 Nov, 2025 Reviewers agreed at journal 03 Nov, 2025 Reviews received at journal 03 Nov, 2025 Reviewers agreed at journal 03 Nov, 2025 Reviewers invited by journal 03 Nov, 2025 Editor assigned by journal 21 Oct, 2025 Submission checks completed at journal 20 Oct, 2025 First submitted to journal 20 Oct, 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-7841940","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":543803422,"identity":"f236812a-74d5-4ac8-b676-71f796b29c32","order_by":0,"name":"Chenhui Du","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvElEQVRIiWNgGAWjYDACCQYGZiDJ2MDe2PjwA2laeA43G0uQoIWBsUEivU2Ahxgd/LObjz0ubLOQ3XDzYRtQv52cbgMhS+4cSzee2SZhvOF2YtuDAoZkY7MDBLQYSOSYSfO2SSQCtbQbSDAcSNxGWEv+N4iWmwfbJHiI05LDBtFyg5FILRI30sykec5JGM88kwgMZAMi/MI/I/mZNE9ZnWzf8eMPH36osJMjqAXdnaQpHwWjYBSMglGAAwAAJH9AVcp3sv8AAAAASUVORK5CYII=","orcid":"","institution":"The First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":true,"prefix":"","firstName":"Chenhui","middleName":"","lastName":"Du","suffix":""},{"id":543803423,"identity":"9ed931dc-2a1b-4dc8-8f9a-e2f79ffe6101","order_by":1,"name":"Shuo Zhang","email":"","orcid":"","institution":"The First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shuo","middleName":"","lastName":"Zhang","suffix":""},{"id":543803424,"identity":"419784ec-7d43-44d0-b5c0-4d42b5510683","order_by":2,"name":"gGuoyu Li","email":"","orcid":"","institution":"Jining Hospital, China Academy of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"gGuoyu","middleName":"","lastName":"Li","suffix":""},{"id":543803425,"identity":"3e3c4acf-217e-45c9-968d-0530bbc7438e","order_by":3,"name":"Xinling Cao","email":"","orcid":"","institution":"The First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xinling","middleName":"","lastName":"Cao","suffix":""},{"id":543803426,"identity":"72b9e4bb-e316-4936-a7c4-a82aea6baced","order_by":4,"name":"Tieying He","email":"","orcid":"","institution":"The First Affiliated Hospital of Xinjiang Medical University","correspondingAuthor":false,"prefix":"","firstName":"Tieying","middleName":"","lastName":"He","suffix":""}],"badges":[],"createdAt":"2025-10-12 16:23:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7841940/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7841940/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96240878,"identity":"4b2534b4-924f-46bd-a260-510bfa4e0019","added_by":"auto","created_at":"2025-11-19 07:09:39","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2942109,"visible":true,"origin":"","legend":"","description":"","filename":"DevelopmentandInterpretabilityAnalysisofaMachineLearningBasedModelforPredictingEarlyLiverMetastasisRiskAfterPancreaticCancerSurgery.docx","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/f620d5314bac3fdc90b05b95.docx"},{"id":95845403,"identity":"8593f4e9-ef1c-4f2c-bd50-7d9b7e65ea24","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7588,"visible":true,"origin":"","legend":"","description":"","filename":"68d68afd74ec4072ba14447ed4a6e23b.json","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/84b05500ad9198dab0df785c.json"},{"id":95845412,"identity":"642bc467-b051-4662-80bc-504307e6be11","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":171608,"visible":true,"origin":"","legend":"","description":"","filename":"68d68afd74ec4072ba14447ed4a6e23b1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/4ded30f84246fde1b419077c.xml"},{"id":95845410,"identity":"212e6b73-51c3-4d6e-a999-c13b8cf6241e","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"jpeg","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":377509,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/eb0efe1e3c255770d2651015.jpeg"},{"id":96239903,"identity":"ba43b462-fc4a-41e3-9ae1-36c98a8f8e32","added_by":"auto","created_at":"2025-11-19 07:07:56","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":263213,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/eb6b7cc6bc22bbd4953633a8.jpeg"},{"id":96239911,"identity":"0a099aca-2446-4cd9-9842-b1f39d8acc32","added_by":"auto","created_at":"2025-11-19 07:07:58","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":336210,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/c55b999d9c4a3ff9fc3589f2.jpeg"},{"id":96240235,"identity":"33e588a6-334b-44d0-97fc-80cf3e250fa8","added_by":"auto","created_at":"2025-11-19 07:08:37","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":199014,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage12.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/946e8b21845e4733b0c2a181.jpeg"},{"id":96240707,"identity":"d20be636-e57e-4ae0-80ea-9d5b40e2c9e4","added_by":"auto","created_at":"2025-11-19 07:09:25","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":312526,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/420da5d5d032e42ae6ecdf76.jpeg"},{"id":96240834,"identity":"2966b346-f208-4a1c-ab63-1722223c6e43","added_by":"auto","created_at":"2025-11-19 07:09:37","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":423668,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage14.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/ae65e4eb3bd4785bad6bad35.jpeg"},{"id":96240093,"identity":"e76e21ea-8170-4dda-bc3d-6bd1f25d8d87","added_by":"auto","created_at":"2025-11-19 07:08:23","extension":"jpeg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":225601,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage15.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/8782c68d9383b64c25fd1a00.jpeg"},{"id":95845418,"identity":"caa245d5-daec-4864-bc1b-8640a7521030","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":177101,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage16.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/15e78c2a6c5c4bc6c14c3549.jpeg"},{"id":95845426,"identity":"28671f54-3295-4c51-a6b1-62be0e0d56fc","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":232739,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage17.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/a36a887d044d83f8feaabb10.jpeg"},{"id":95845431,"identity":"e60dbddb-2f74-4faf-95a4-c144c90850d2","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":715597,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/3432dd89b92f614e13e1a846.jpeg"},{"id":96240306,"identity":"75b19638-8827-478a-aea1-de438fbdc43b","added_by":"auto","created_at":"2025-11-19 07:08:46","extension":"jpeg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":277967,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/baf991eb7f4c630de9a302f0.jpeg"},{"id":95845428,"identity":"97dfd61b-8c69-4b09-8282-66d00f26e345","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"jpeg","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":290913,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/a8313f4685e4e8b5604cb58c.jpeg"},{"id":95845433,"identity":"d8c5dd5d-da76-4d53-b0f9-a030a9a05952","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"jpeg","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":341158,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/625a7b94c1a71352fd204a0f.jpeg"},{"id":95845442,"identity":"7cecaa15-5d55-4414-89df-1a44a3034c77","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"jpeg","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":281480,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/3a7cddd20842195e764456b7.jpeg"},{"id":96240753,"identity":"1092ded9-62ea-48b3-bc0a-fc9fff46ef2f","added_by":"auto","created_at":"2025-11-19 07:09:30","extension":"jpeg","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":196623,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/2b34d73cd855cc40ad8fb137.jpeg"},{"id":96240152,"identity":"df95bd6a-3977-48dd-b7e0-e1214400dc14","added_by":"auto","created_at":"2025-11-19 07:08:31","extension":"jpeg","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":187983,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/d290f1cfee3d6b60c8d65b54.jpeg"},{"id":95845434,"identity":"0d5606e8-d0d2-42e2-80ae-bf3855942ad1","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"jpeg","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":239485,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/12cbfdc2c6f9a36aa4bf725f.jpeg"},{"id":95845440,"identity":"c2809eda-fa06-4e28-9646-f6fd884465fc","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":92514,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/07446ebc9ca51d8211eb785b.png"},{"id":95845453,"identity":"95a8dba3-6097-42ec-84fa-d063a1866f12","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":44281,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/893a102b086e6819a86bc336.png"},{"id":96240383,"identity":"5a82002d-82e7-47b2-a83e-6fd40d97bdbf","added_by":"auto","created_at":"2025-11-19 07:08:52","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":66426,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/3c5a82139d661fa68ff62c04.png"},{"id":95845439,"identity":"116e94df-f63a-402c-9f69-ebc3e7f4c5bc","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40027,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/b7db232d4013edfabfe50cc7.png"},{"id":96240423,"identity":"c18dd394-64b0-4a28-a467-466a581626dd","added_by":"auto","created_at":"2025-11-19 07:08:54","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":85983,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/ca84c039a2f532e5b7b6e07e.png"},{"id":95845432,"identity":"d0e2de9d-750d-4b29-bbce-0f4e8bd97122","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":60037,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/459c763b609707babc7b0ebd.png"},{"id":96240873,"identity":"b45e9883-8e60-4ed2-90c1-d5bad1f3f875","added_by":"auto","created_at":"2025-11-19 07:09:39","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":43208,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/53f10be479deef488f63a7a4.png"},{"id":96240881,"identity":"ecbb5a4f-8626-40e4-aec3-2a61f2f5831b","added_by":"auto","created_at":"2025-11-19 07:09:39","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":30988,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/be906e17e05d32e52f5ac8c5.png"},{"id":96241571,"identity":"27893067-f679-41a6-b2c0-1552a23cd3ca","added_by":"auto","created_at":"2025-11-19 07:10:57","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":61682,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/71abee9796b347fb0a34cb55.png"},{"id":96240400,"identity":"77570d53-7d6c-47e3-9c7f-630f067833f0","added_by":"auto","created_at":"2025-11-19 07:08:53","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":130539,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/9ffce4ab60b53853e7d1e73f.png"},{"id":96240298,"identity":"49883f29-0963-4ffb-8258-2babf2984317","added_by":"auto","created_at":"2025-11-19 07:08:45","extension":"png","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45146,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/fcba63dca9e47d24b44e72ee.png"},{"id":96240246,"identity":"b5a500ca-d020-4611-af60-62b83701527c","added_by":"auto","created_at":"2025-11-19 07:08:38","extension":"png","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57022,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/413dc6afea2ed7f5c825b020.png"},{"id":96240318,"identity":"7ad0fe11-a0df-49fd-808a-a6012adc5ca2","added_by":"auto","created_at":"2025-11-19 07:08:48","extension":"png","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":77003,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/d3ef64b3c2c01e846b86b80d.png"},{"id":95845451,"identity":"c5b6b1ca-88fb-4746-b1b8-6923917270a7","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"png","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":50390,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/062f143d6b263de5c5d76bf7.png"},{"id":96240868,"identity":"7b349438-55b9-496a-ad63-36bb5403d42f","added_by":"auto","created_at":"2025-11-19 07:09:38","extension":"png","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40858,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/a5ea8bd3136b327bce6d88a6.png"},{"id":96240524,"identity":"a4983585-81b1-436f-a440-6cea2c6044d2","added_by":"auto","created_at":"2025-11-19 07:09:02","extension":"png","order_by":35,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31825,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/cc29c4844676c9d0f6fa172c.png"},{"id":95845457,"identity":"c126c517-860e-4a22-a435-bfe44a394f52","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"png","order_by":36,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":64509,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/300cd9ca4bdfbc8a52f93fa6.png"},{"id":95845448,"identity":"0ced33fb-d02c-4153-a52b-ad22c67e7e6b","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"xml","order_by":37,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":169832,"visible":true,"origin":"","legend":"","description":"","filename":"68d68afd74ec4072ba14447ed4a6e23b1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/0a41a621509facbd5e458203.xml"},{"id":95845455,"identity":"a9d46043-6b3a-430b-b538-fec0541eaa0f","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"html","order_by":38,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":178510,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/f9b81da93c8b5483717f95af.html"},{"id":96240151,"identity":"35834ea4-889f-427b-bce8-3fe6760127c9","added_by":"auto","created_at":"2025-11-19 07:08:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":192152,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the study. ML machine learning, LASSO Least absolute shrinkage and selection operator, AUC area under the receiver operating characteristic, PPV positive predictive value, NPV negative predictive value, DT Decision Tree,SVM Support Vector Machine,XGBoost eXtreme Gradient Boosting, RF Random Forst, LightBGM Light Gradient Boosting Machine, LR Logistic Regression,ET Extra Trees.SHAP Shapley Additive Explanations,LIME Local Interpretable Model-agnostic Explanations,PDP Partial Dependence Plot,ICE Individual Conditional Expectation.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/632040652144232ed0029904.png"},{"id":96240866,"identity":"0b7cf23e-8a9b-4ca3-8de9-9d68e5a1dcd7","added_by":"auto","created_at":"2025-11-19 07:09:38","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":116861,"visible":true,"origin":"","legend":"\u003cp\u003eFeature Selection. A: Lasso regression cross-validation error plot. The x-axis represents the regularization strength, and the y-axis shows the corresponding Mean Squared Error (MSE). The black dashed line (λ_min) represents the λ value with the minimum MSE, and the blue dashed line (λ_1se) represents the λ value within one standard error; B: Lasso regression path plot. This plot shows the change trajectory of each feature coefficient at different λ values. The x-axis represents log(λ), and the y-axis represents the regression coefficients; C: Boruta feature importance ranking boxplot. The more important the feature, the higher its ranking;D: Boruta feature importance trajectory plot.This plot shows the change in feature importance over iterations.The x-axis represents the number of iterations, and the y-axis represents the importance value of each feature in each round. Lines of different colors represent the performance of different features during 50 random forest trainings;E: Feature contribution and AUC performance change plot.This plot displays the importance of each feature (bar chart) and the change in model performance (AUC) as each feature is sequentially introduced (line chart); F: Venn diagram of the intersection of three feature selection algorithms.The Venn diagram shows the intersection and differences of the selected variables from the Boruta (red circle), Lasso (green circle), and XGBoost (blue circle) feature selection methods.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/a3da17198ba25b545a74fa95.png"},{"id":95845406,"identity":"f45740b5-099e-4779-8f73-f56cad6a974d","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":127593,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of 7 ML models on training and test sets. A Training Set ROC curves;B Test Set ROC curves\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/a3fa2b53b0e2264022a3eb84.png"},{"id":96240203,"identity":"dc041af9-0978-4207-abce-24b873fb2a59","added_by":"auto","created_at":"2025-11-19 07:08:33","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":141510,"visible":true,"origin":"","legend":"\u003cp\u003eLine plots of model evaluation metrics for 7 ML algorithms on the training and test sets.A Training Set;B Test\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/6d776e0858a08d31bc8d729b.png"},{"id":95845458,"identity":"0c96c372-f6a8-4239-9b63-b27e2bef8ec9","added_by":"auto","created_at":"2025-11-13 14:49:25","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":70922,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix of 7 ML models on the test set .The x-axis represents the predicted values, and the y-axis represents the actual values. The color intensity highlights the regions with strong classification accuracy (darker colors indicate more accurate predictions). Taking the ET model as an example, 65 cases (87.8%) were predicted correctly, with non-metastatic patients predicted as non-metastatic; 9 cases (12.2%) were predicted incorrectly, with non-metastatic patients predicted as metastatic; 13 cases (37.1%) were predicted incorrectly, with metastatic patients predicted as non-metastatic; and 22 cases (62.9%) were predicted correctly, with metastatic patients predicted as metastatic.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/358ff208072a88f6ece1aa50.png"},{"id":96240192,"identity":"62a0cce5-f759-41b9-bb60-7876b763915f","added_by":"auto","created_at":"2025-11-19 07:08:33","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":128456,"visible":true,"origin":"","legend":"\u003cp\u003ePR curves of 7 ML models on training and test sets. A:In the training set, most models exhibited precision and recall values close to 1, with Average Precision (AP) values nearly reaching 1, indicating excellent performance; B: In the testing set, the curves became less smooth, showing “step-like” patterns and fluctuations, suggesting the presence of overfitting. Among all models, the ET model demonstrated the most robust performance, with an AP of 0.992 in the training set and 0.768 in the testing set, both indicating strong predictive ability.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/51f1285cd76cf4d9ea8dd783.png"},{"id":96240496,"identity":"6ad0608e-dfd0-451b-b075-a3bd28c23715","added_by":"auto","created_at":"2025-11-19 07:08:58","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":79639,"visible":true,"origin":"","legend":"\u003cp\u003eLearning curves and validation curves for ET models. A: The x-axis represents the number of training samples, and the y-axis represents the AUC score. The blue line represents the AUC on the training set, and the orange line represents the AUC on the cross-validation set. As the sample size increases, the model tends to stabilize;B: The x-axis represents the minimum number of samples per leaf node (min_samples_leaf), which controls the model's complexity, and the y-axis represents the AUC score. The blue line represents the AUC on the training set, and the orange line represents the AUC on the cross-validation set. min_samples_leaf = 2 is the optimal choice, as it balances fitting and generalization ability.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/8ee44039f65e9a389e0c2c45.png"},{"id":95845414,"identity":"04920e88-a986-4dc3-9ea6-9fb7d966a75b","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":85494,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves, PR curves and confusion matrices for optimal ET models. A: The AUC values for the optimized ET model were 0.974 for the training set and 0.827 for the testing set. Compared to the previous results, the model still performs well on the training set, while the performance on the testing set is more stable;B: The optimized ET model showed improved performance in predicting \"non-metastatic\" cases (89.2% accuracy) and \"metastatic\" cases (65.7% accuracy);C: The optimized ET model demonstrated an improvement in precision when handling positive samples, and the Average Precision (AP) also increased. This further proves the stability of the optimized ET model across multiple metrics.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/99af0b356b34d9e240463305.png"},{"id":96240220,"identity":"3c50ebf5-8bb8-4bbe-baee-c8b2b2216bcb","added_by":"auto","created_at":"2025-11-19 07:08:36","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":134060,"visible":true,"origin":"","legend":"\u003cp\u003eClassification effects of optimal ET models based on PCA downscaling with model prediction probabilities. The green dots represent samples without early liver metastasis, and the purple dots represent samples with early liver metastasis.\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/f4d56534f758012e602ff8b4.png"},{"id":96240888,"identity":"093ba563-7b19-45b9-a00a-260088b585e2","added_by":"auto","created_at":"2025-11-19 07:09:39","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":111586,"visible":true,"origin":"","legend":"\u003cp\u003eModel interpretation plot for SHAP values. A: The x-axis represents the degree of influence of a particular feature on the prediction results. Each row corresponds to a feature, with higher rows indicating greater importance of the feature. Each point represents a sample’s value for that feature and its corresponding SHAP value. The color gradient, from blue to red, reflects the range of feature values from low to high. B: The bar chart displays each feature with a horizontal bar, where the x-axis represents the absolute mean SHAP value of the feature. The rose chart illustrates the proportion of the contribution of each feature to the overall model prediction.\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/0ead1032f16949fb67b87b3a.png"},{"id":95845421,"identity":"44b35229-805a-49ce-8948-30e55dace335","added_by":"auto","created_at":"2025-11-13 14:49:23","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":151070,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP Dependence Plot. Each subplot corresponds to the SHAP dependence plot of a specific feature. The x-axis represents the actual value of the feature, while the y-axis denotes the corresponding SHAP value. The color of each point indicates either the magnitude of the feature value or the level of another interacting feature, thereby reflecting the strength of its influence on the prediction.\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/466fc59750476acd487641a3.png"},{"id":96241883,"identity":"053bcfce-60c4-46fa-a4ef-b51ba2867456","added_by":"auto","created_at":"2025-11-19 07:11:34","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":109914,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP Interaction Diagram.A: When the tumor size was large (yellowish color) , the interaction SHAP value of some samples with high CA19-9 levels showed a slight upward trend, suggesting that there may be a synergistic effect between CA19-9 and tumor size.B: For low-grade (poorly differentiated) tumors (yellow points), higher CA19-9 levels lead to predominantly negative SHAP interaction values, indicating that this combination favors a positive prediction in the model\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/0775068fb2d878ad08fd2a60.png"},{"id":96240335,"identity":"befab5d2-f507-44aa-ba98-9239d1261396","added_by":"auto","created_at":"2025-11-19 07:08:49","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":258262,"visible":true,"origin":"","legend":"\u003cp\u003ePartial Dependency Graphs and Individual Conditional Expectation Graphs. A: Each subplot illustrates the average effect of a feature on the model's prediction. The x-axis represents the feature value, and the y-axis represents the model’s prediction output. B: Each line represents the prediction curve of an individual sample, showing the effect of feature variation on the prediction for that specific sample.\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/88c7197dc5b83844f173a9c1.png"},{"id":95845438,"identity":"0004fd9a-a1d9-4b38-9ed1-4686e3de4a2d","added_by":"auto","created_at":"2025-11-13 14:49:24","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":60847,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP Explanatory Model Individual Predictions.\u003c/p\u003e","description":"","filename":"14.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/1ac638b1f3a1324a2847d85c.png"},{"id":96241753,"identity":"a42178f9-d39f-45ab-a673-619b43bf6657","added_by":"auto","created_at":"2025-11-19 07:11:19","extension":"png","order_by":15,"title":"Figure 15","display":"","copyAsset":false,"role":"figure","size":66418,"visible":true,"origin":"","legend":"\u003cp\u003eLIME Explanatory Model Individual Predictions.\u003c/p\u003e","description":"","filename":"15.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/187197814d2cec47d95bbece.png"},{"id":96240286,"identity":"2ecbc604-5a24-4b1e-a4ae-b1d598a2b06a","added_by":"auto","created_at":"2025-11-19 07:08:44","extension":"png","order_by":16,"title":"Figure 16","display":"","copyAsset":false,"role":"figure","size":76076,"visible":true,"origin":"","legend":"\u003cp\u003eDCA curve for the test set.The x-axis represents the risk threshold, and the y-axis represents the net benefit. The red curve shows the net benefit of the machine learning prediction model at different risk thresholds, the solid black line represents the \"all intervention\" strategy, and the dashed line represents the \"no intervention\" strategy.\u003c/p\u003e","description":"","filename":"16.png","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/3f3d6c38ad3e75e3ad757f65.png"},{"id":96452745,"identity":"254cbb32-2c8f-461d-af1c-b3e64461d460","added_by":"auto","created_at":"2025-11-21 09:41:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2728582,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7841940/v1/1056e72c-3399-4159-8f26-ed262dd0350b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and Interpretability Analysis of a Machine Learning-Based Model for Predicting Early Liver Metastasis Risk After Pancreatic Cancer Surgery","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePancreatic cancer (PC) is one of the most challenging malignancies in clinical practice. Its high mortality rate and the lack of early symptoms make diagnosis and treatment extremely difficult. According to the 2022 global cancer statistics\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e, PC accounts for 2.6% of all cancers in age-standardized incidence rates, ranking 12th globally among malignant tumors. Its mortality rate, however, accounts for 4.8% of all cancer-related deaths, ranking it 6th, reflecting the extremely poor prognosis of pancreatic cancer, which is significantly worse than most other solid tumors. Due to the pancreas\u0026rsquo;s deep location in the abdomen, tumors are often difficult to detect early, resulting in most patients being diagnosed with locally advanced or metastatic disease\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. Only about 10% of patients are eligible for radical surgery. Even among those who undergo radical resection, 90% of patients will experience tumor recurrence within 7 to 9 months postoperatively\u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. Liver metastasis\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e is one of the most common sites of distant metastasis in PC, accounting for 23.5% of initial recurrences, with most occurring within the first year after surgery. The appearance of liver metastasis usually signifies disease progression, and postoperative liver metastasis in pancreatic cancer not only significantly reduces survival but also affects subsequent treatment decisions. Studies have shown that early detection and treatment of postoperative liver metastasis can prolong survival. Therefore, predicting the occurrence of postoperative liver metastasis is of critical importance for clinicians in postoperative management, monitoring, and early intervention.\u003c/p\u003e\u003cp\u003eMachine learning (ML), a branch of artificial intelligence(AI)\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e, has shown great potential in predicting clinical outcomes. Compared to traditional statistical methods, ML excels at handling complex interactions and nonlinear relationships, thereby reducing decision-making time\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e in clinical practice. Previous studies\u003csup\u003e[\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e have confirmed that ML models, trained on large amounts of clinical data, significantly improve the accuracy of risk prediction for various diseases, such as cancer and chronic diseases, and outperform traditional models in both prediction and diagnosis. Despite the great potential of ML in clinical practice, several challenges remain, with the \"black box\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e\" issue being one of the main obstacles. Most ML models are not transparent and difficult to interpret, making it hard for clinicians to understand and trust the decision-making process. Therefore, this study aims to develop an interpretable ML model using clinical data to predict high-risk populations for early liver metastasis after pancreatic cancer surgery, providing scientific evidence and decision support for the development of individualized treatment plans for postoperative patients.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e\u003ch2\u003eSubjects and study design\u003c/h2\u003e\u003cp\u003eCollection of Clinical Data from 418 Patients with PC Undergoing Radical Surgery at the Digestive Surgery Center of the First Affiliated Hospital of Xinjiang Medical University from January 2015 to August 2024. Clinical data from 418 patients who underwent radical surgery for PC at the Digestive Surgery Center of the First Affiliated Hospital of Xinjiang Medical University between January 2015 and August 2024 were collected. The inclusion criteria were as follows: (1)Pathological diagnosis of PC confirmed after surgery, with no history of other systemic tumors before surgery.(2)Patients who had not received neoadjuvant or conversion therapy\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e prior to surgery.(3)Preoperative imaging examinations, including enhanced CT, MRI, and PET-CT, showed no evidence of distant metastasis. Preoperative 3D vascular reconstruction did not show tumor invasion of blood vessels (such as the celiac trunk, common hepatic artery, or portal vein) or local vascular invasion (invasion\u0026thinsp;\u0026le;\u0026thinsp;180\u0026deg;or complete removal of the arterial sheath is feasible).(4)Intraoperative frozen section and postoperative pathology both confirmed R0 resection.(5)All patients underwent close follow-up as per medical instructions postoperatively. The exclusion criteria were as follows:(1) History of other systemic malignant tumors.(2) Preoperative receipt of neoadjuvant or conversion therapy.(3) Lack of follow-up records or absence of relevant imaging data postoperatively.(4) Perioperative death: patients who died within 90 days after surgery. Based on the inclusion and exclusion criteria, 363 patients were ultimately included. These patients were randomly assigned to the training cohort and the testing cohort in a 7:3 ratio. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents a flowchart summarizing the study design.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eVariable collection\u003c/h2\u003e\u003cp\u003eThis study retrospectively collected multimodal clinical data from patients, including:(1)Demographic Information: Age, gender, weight, medical history, smoking and alcohol consumption history, etc.(2) Preoperative Laboratory Data: Complete blood count, liver function tests, tumor markers, etc.(3) Preoperative Imaging Data: Abdominal enhanced CT, MRI reports, etc.(4) Intraoperative Data: Surgical method, operation time, blood loss, etc.(5) Postoperative Pathological Data: Maximum tumor diameter, lymph node metastasis, degree of differentiation, etc. The surgical methods in this study included pancreaticoduodenectomy and distal pancreatectomy with splenectomy (or adrenalectomy). Intraoperative vascular invasion was addressed by the lead surgeon, who selected either artificial vascular replacement or partial vessel wall resection based on the extent of vascular invasion. A drainage tube was routinely placed in the surgical area for all patients, with the number of drainage tubes determined by the lead surgeon according to the intraoperative situation. The surgeries were performed by four qualified surgeons, each holding a position of associate chief physician or higher, with over 10 years of surgical experience. Postoperative abdominal hemorrhage and pancreatic fistula were defined and graded according to the International Study Group of Pancreatic Surgery (ISGPS)\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e. In this study, the cases included were those with moderate to severe intra-abdominal hemorrhage and B or C grade pancreatic fistulas. The diagnostic criteria for abdominal infection were based on the \"Revised Guidelines for the Management of Intra-abdominal Infections\" by the Surgical Infection Society\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003ePostoperative follow-up\u003c/h3\u003e\n\u003cp\u003eFollow-up in this study was conducted through outpatient visits and telephone follow-ups. Patients were scheduled for follow-up every 3 months during the first year after surgery, and subsequently every 6\u0026ndash;12 months. During each outpatient follow-up visit, patients underwent abdominal enhanced CT, complete blood count, blood biochemistry, tumor marker tests, and other relevant examinations. Telephone follow-up was conducted by inquiring or receiving the patient's local hospital examination results. When enhanced CT suggested suspected liver recurrence or metastasis, the patient was admitted for further confirmation of the lesion through various methods. The endpoint of the follow-up in this study was the early recurrence of tumors in the liver. There is currently no unified standard for defining early recurrence. Based on domestic and international research on early recurrence of PC after surgery, early liver recurrence\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e was defined as follows: during regular follow-up within one year after radical surgery for PC, liver lesions confirmed by imaging tests and/or laboratory tests and/or pathological examinations as new lesions originating from metastasis of the primary tumor. In cases where pathological diagnosis was missing or unclear, imaging diagnosis was prioritized, and tumor markers were referenced. The follow-up data was collected until August 31, 2025.\u003c/p\u003e\n\u003ch3\u003eFeature Selection\u003c/h3\u003e\n\u003cp\u003eBased on the occurrence of the follow-up endpoint event, patients were divided into two groups: the liver metastasis group and the non-metastasis group. The patients were randomly assigned to the training set and the testing set in a 7:3 ratio. Univariate analysis was first performed to preliminarily identify clinical variables with significant differences, with a significance threshold set at α\u0026thinsp;\u0026lt;\u0026thinsp;0.05. Using the training set cohort, feature selection was conducted with LASSO regression (utilizing its L1 regularization feature to perform dimensionality reduction, with the optimal λ value determined by ten-fold cross-validation to avoid overfitting), the Boruta algorithm (an automated feature selection method based on random forests, which calculates the importance of both original and shadow features, where shadow features provide a random baseline and enhance generalization capability), and recursive feature elimination (RFE) based on XGBoost. The key features selected through these methods were then incorporated into the predictive model.\u003c/p\u003e\n\u003ch3\u003eModel construction and validation\u003c/h3\u003e\n\u003cp\u003eBased on the selected variables, seven machine learning methods were applied to construct predictive models in the training set cohort. These algorithms included Decision Tree (DT), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Logistic Regression (LR), and Extra Trees (ET). Subsequently, multiple metrics were used to evaluate the performance of each model, including the Area Under the Receiver Operating Characteristic Curve (AUC), Precision-Recall (PR) curve, accuracy, sensitivity, specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and F1 score. To assess the robustness and generalization ability of the models, hyperparameter grid search and 5-fold cross-validation were first conducted on the training set, followed by external validation on the testing set to evaluate generalization performance. The best model was ultimately selected. To further assess the clinical applicability of the model, Decision Curve Analysis (DCA) was performed, which calculates the net benefit at different risk thresholds to determine the clinical value of the best model.\u003c/p\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eAll statistical analyses were performed using R 4.4.3 and Python 3.2 software. Data preprocessing was conducted first, analyzing the types of variables and missing data. Variables with a missing rate\u0026thinsp;\u0026ge;\u0026thinsp;30% were excluded, while variables with missing data\u0026thinsp;\u0026lt;\u0026thinsp;30% were imputed using multiple imputation methods. For categorical data, comparisons between groups were made using the Chi-square (X\u0026sup2;) test or Fisher\u0026rsquo;s exact test. For continuous data, normality tests were performed first. Normally distributed continuous data were expressed as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (x\u0026thinsp;\u0026plusmn;\u0026thinsp;s), and comparisons between two groups were made using the t-test. For non-normally distributed continuous data, data were expressed as median (P25-P75), and the Mann-Whitney U test was used for comparisons between two groups. A two-sided P-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant. Given the inherent \"black-box\" nature of machine learning models, we employed several interpretability techniques to explain the models. These included Shapley Additive Explanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Partial Dependence Plots (PDP)/Individual Conditional Expectation Plots (ICE).\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003eBaseline characteristics of patients\u003c/h2\u003e\u003cp\u003eA total of 363 patients were included in the study, of which 118 patients developed early liver metastasis, while 245 patients did not. The patients were randomly divided into two groups in a 7:3 ratio, with 254 patients in the training set and 109 patients in the testing set. In the training set, 83 patients (32.67%) developed early liver metastasis, and in the testing set, 35 patients (32.11%) developed early liver metastasis. There were no statistically significant differences in baseline characteristics between the two groups (\u003cb\u003eP\u003c/b\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05), as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eComparison of baseline data features between the training and test sets.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTrain(n\u0026thinsp;=\u0026thinsp;254)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTest(n\u0026thinsp;=\u0026thinsp;109)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSex\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.639\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e122 (48)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e56 (51)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e132 (52)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e53 (49)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eChemotherapy\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.946\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e28 (11)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e13 (12)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e226 (89)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e96 (88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHypertension\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.442\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e173 (68)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e69 (63)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e81 (32)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e40 (37)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDiabetes2\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.525\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e186 (73)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e84 (77)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e68 (27)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e25 (23)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDrinking history\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.288\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e201 (79)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e80 (73)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e53 (21)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e29 (27)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSmoking history\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.512\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e194 (76)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e79 (72)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e60 (24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e30 (28)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePBD\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.884\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e223 (88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e97 (89)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e31 (12)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e12 (11)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAV invasion\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e194 (76)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e83 (76)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e60 (24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e26 (24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNerve invasion\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.113\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e41 (16)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e10 (9)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e213 (84)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e99 (91)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVascular invasion\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e148 (58)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e65 (60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e106 (42)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e44 (40)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGrade\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.145\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModerately to well-differentiated\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e131 (52)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e66 (61)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePoorly differentiated\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e123 (48)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e43 (39)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePathological type\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.769\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePDAC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e224 (88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e98 (90)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOther\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e30 (12)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e11 (10)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTumor location\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.889\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003epancreatic head\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e190 (75)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e83 (76)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePancreatic body and tail\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e64 (25)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e26 (24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePOH\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.567\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e242 (95)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e106 (97)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e12 (5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3 (3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePOIAI\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.965\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e184 (72)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e78 (72)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e70 (28)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e31 (28)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePOPF\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e171 (67)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e74 (68)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e83 (33)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e35 (32)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHBV\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.318\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e211 (83)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e85 (78)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e43 (17)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e24 (22)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e62 (55, 69)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e60 (56, 68)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.686\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHeight(cm)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e167 (160, 174)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e165 (160, 170)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.156\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWeight(kg)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e65 (58, 73)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e64 (58, 71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.639\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBMI(Kg/m\u003csup\u003e2\u003c/sup\u003e)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e23.48 (20.87, 25.81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e23.44 (21.09, 25.71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.808\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSurgical duration\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e540 (450, 638.75)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e545 (425, 655)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.663\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIBL\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e500 (300, 787.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e500 (300, 700)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.826\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWBC\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5.82 (4.87, 7.21)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e5.92 (4.88, 7.38)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.627\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNE(10⁹/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.63 (2.96, 4.94)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.74 (2.86, 4.83)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.998\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLYM(10⁹/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.41 (1.15, 1.84)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.49 (1.15, 1.94)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.398\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePLT(10⁹/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e230.5 (188, 281)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e232 (192, 286)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.559\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNLR\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2.56 (1.91, 3.74)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.52 (1.66, 3.18)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.213\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePLR\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e165.39 (120, 208.78)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e155.48 (116.34, 204.41)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.539\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHB(g/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e132 (120, 142.75)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e131 (120, 143)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.983\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTB (ummol/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e24.44 (14.7, 170.47)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e23.81 (13.76, 141.85)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.539\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIB(ummol/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e16.98 (9.85, 35.75)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15.06 (9.19, 33.87)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.469\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eALB(g/L)\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e39.55\u0026thinsp;\u0026plusmn;\u0026thinsp;4.89\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e39.78\u0026thinsp;\u0026plusmn;\u0026thinsp;4.95\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.681\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eA/G\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.19 (1.05, 1.39)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.23 (1.07, 1.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.318\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAST(U/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e43.72 (22.95, 132.19)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e46 (25.3, 106.82)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.986\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eALT(U/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e50.37 (21.85, 213.4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e55.26 (21.33, 205.21)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.601\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAKP(U/L)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e134.43 (78.58, 334.62)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e151.9 (78, 341.1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.842\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCA19-9(U/mL)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e266.24 (37.84, 1200)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e186.99 (41.76, 740.24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.427\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCA50(U/mL)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e136.57 (34.32, 180)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e78.17 (31.66, 180)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.055\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCEA(U/mL)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.05 (1.72, 5.79)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.66 (1.82, 5.08)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.305\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTumor size (cm)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.8 (3, 5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.5 (2.6, 4.5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.05\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLNM\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0 (0, 1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0 (0, 1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.482\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eT\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.344\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3 (1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2 (2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e13 (5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1 (1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e57 (22)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e28 (26)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e149 (59)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e64 (59)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e32 (13)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e14 (13)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eN\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.794\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e176 (69)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e72 (66)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e65 (26)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e30 (28)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e13 (5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7 (6)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003csup\u003ea\u003c/sup\u003eContinuous variables with a normal distribution were expressed as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation, and inter-group comparisons were conducted using t-tests.\u003c/p\u003e\u003cp\u003e\u003csup\u003eb\u003c/sup\u003eContinuous variables that did not follow a normal distribution were expressed as median [interquartile range], and inter-group comparisons were conducted using the Mann\u0026ndash;Whitney U test.\u003c/p\u003e\u003cp\u003e\u003csup\u003ec\u003c/sup\u003eCategorical variables were expressed as frequency (percentage), and inter-group comparisons were conducted using the X\u0026sup2; test.\u003c/p\u003e\u003cp\u003e\u003cb\u003ePBD\u003c/b\u003e Preoperative Biliary Drainage, \u003cb\u003eAV invasion\u003c/b\u003e Hepatic artery / portal vein invasion,\u003cb\u003ePOH\u003c/b\u003e Postoperative Hemorrhage,\u003cb\u003ePOIAI\u003c/b\u003e Postoperative Intra-abdominal Infection, \u003cb\u003ePOPF\u003c/b\u003e Postoperative Pancreatic Fistula, \u003cb\u003eBMI\u003c/b\u003e Body Mass Index,\u003cb\u003eIBL\u003c/b\u003e Intraoperative Blood Loss, \u003cb\u003eNLR\u003c/b\u003e Neutrophil-lymphocyte ratio, \u003cb\u003ePLR\u003c/b\u003e Platelet-to-Lymphocyte Ratio ,\u003cb\u003eLNM\u003c/b\u003e Lymph Node Metastasis.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eFeature Selection\u003c/h3\u003e\n\u003cp\u003eBased on the training set data, we performed feature selection using Lasso regression (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA, B), the Boruta algorithm (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC, D), and recursive feature elimination based on XGBoost (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eF). According to the results of feature selection from these three methods (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eG), we ultimately selected 10 variables for the development of the ML model. These 10 variables were: chemotherapy, hepatic artery/portal vein invasion, degree of differentiation, hepatitis B virus infection, CA19-9, T stage, lymphocyte count, albumin, tumor size, and alkaline phosphatase.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eConstruction and Performance Comparison of Different ML Algorithms\u003c/h2\u003e\u003cp\u003eBased on the 10 selected features, this study used seven machine learning algorithms, including Decision Tree (DT), Support Vector Machine (SVM), XGBoost, Random Forest (RF), LightGBM, Logistic Regression (LR), and Extra Trees (ET), to construct the early liver metastasis model in the training set cohort. GridSearchCV was used for grid search to identify the optimal model. In the training set, all models, except for SVM, exhibited good performance. In the testing set, the ET and RF models showed excellent performance, with AUC values of 0.822 (95% CI: 0.726-0.900) and 0.814 (95% CI: 0.722\u0026ndash;0.902), respectively (Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA, B). Additionally, we calculated various performance metrics, including Accuracy, Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), F1 score, and Kappa score for model evaluation (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA, B). The confusion matrix (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e) and Precision-Recall (PR) curve (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA, B) for the testing set were also plotted. The ET model, which demonstrated a balanced performance across various metrics, was ultimately selected as the final model.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance of 7 ML algorithms on training and test sets\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"9\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAlgorithms\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003ePPV\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eNPV\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e\u003cp\u003eF1 score\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c9\"\u003e\u003cp\u003eKappa score\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTrain\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.913\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.831\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.953\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.896\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.831\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.863\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.799\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eET\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.957\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.892\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.988\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.974\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.892\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.931\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.899\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.693\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.229\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.918\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.576\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.229\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.328\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.174\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLightGBM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.957\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.994\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.986\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.93\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.899\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.976\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.94\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.994\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.987\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.94\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.963\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.946\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.866\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.711\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.942\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.855\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.711\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.776\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.682\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.937\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.843\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.982\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.959\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.843\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.897\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.852\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.761\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.824\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.453\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eET\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.798\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.878\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.667\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.523\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.642\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.851\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.389\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.264\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.059\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLightGBM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.807\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.657\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.878\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.719\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.657\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.687\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.548\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.798\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.878\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.667\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.523\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.817\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.905\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.759\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.688\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.559\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.826\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.657\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.905\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.767\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.657\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0.708\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e0.585\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003ePPV positive predictive value, NPV negative predictive value\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eTuning and Performance of the Optimal Model\u003c/h2\u003e\u003cp\u003eTo assess the robustness and generalization ability of the ET model, we plotted the learning curve and validation curve of the model (Figs.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA, B). The AUC for both the training set and cross-validation set increased with the sample size, and the gap between the two remained relatively stable, indicating that the model has good fitting ability and generalizability. To further optimize model performance and prevent overfitting, we adopted a two-stage hyperparameter tuning strategy. First, we used Randomized Search CV to perform a coarse search for the main hyperparameters, quickly identifying their effective ranges within a larger parameter space. Then, within the high-quality parameter regions identified by the randomized search, we conducted a more refined search using Grid Search CV to precisely select the optimal hyperparameter combination. The ROC curve, PR curve, and confusion matrix for the model under the best parameters were plotted (Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA, B, C), showing that the performance of the optimal ET model improved across all aspects. Subsequently, we used Principal Component Analysis (PCA) for dimensionality reduction to visualize the data distribution in a two-dimensional space and observe the separation between different class samples. We performed PCA projection on both the training and testing datasets to evaluate the classification effect of the model and its generalization ability on different datasets (Figs.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003eA, B). In the training set, PC1 explained 80.05% of the variance, and the \"non-metastatic\" and \"metastatic\" categories showed clear separation in the two-dimensional space with almost no overlap, indicating that the model can effectively distinguish between different categories in the training set. In the testing set, PC1 explained 83.55% of the variance, and while there was still clear separation between \"non-metastatic\" and \"metastatic\" samples, some overlap was observed in certain areas.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eInterpretation Based on the Optimal ET Model\u003c/h2\u003e\u003cp\u003eThis study provides global and local explanations of the model using SHAP, PDP, and ICE. The SHAP value plots visually demonstrate the direction and strength of the impact of each feature on the model's predictions. The heatmap reveals the positive or negative trend of the feature\u0026rsquo;s influence based on its value, while the bar and polar plots more clearly present the importance ranking of each feature in the global predictions (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003eA, B). To further explore the specific impact of each feature on the model's prediction, we plotted the SHAP dependence plots for each variable (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e). These plots show the relationship between the actual values of the features and their corresponding SHAP values, revealing how each feature influences the model's output in terms of direction and strength under different values. Additionally, we examined the nonlinear interactions between features and their effects on the model\u0026rsquo;s predictions. Using SHAP interaction values, we visualized the relationships between features. In the interaction between CA19-9 and tumor size, the distribution of interaction values was somewhat dispersed. As the largest diameter of the tumor increased, the SHAP interaction values for some high CA19-9 samples showed a slight upward trend, suggesting a potential joint effect between CA19-9 and tumor volume. In the interaction between CA19-9 and tumor differentiation, when the tumor was poorly differentiated, higher CA19-9 levels mainly resulted in negative SHAP interaction values, indicating that this combination was more likely to predict a positive outcome. In contrast, when the tumor was moderately to well-differentiated, a positive interaction was observed under low CA19-9 conditions, indicating a nonlinear combined effect of CA19-9 and tumor differentiation on the model\u0026rsquo;s prediction (Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003eA、B). To explore the model's dependence on key features, we plotted partial dependence plots (PDPs) and individual conditional expectation plots (ICEs) for some important features. The PDPs illustrate the average effect of feature variation on the model's output, revealing global trends. Meanwhile, the ICE plots demonstrate the predictive response of individual samples to changes in feature values, highlighting heterogeneity among individual predictions (Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e13\u003c/span\u003eA, B).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eClinical identification of the optimal ET model\u003c/h2\u003e\u003cp\u003eTo evaluate the contribution of features to individual patient predictions, we employed SHAP and LIME methods to explain the predictions for two sample patients. Two patients, A and B, were randomly selected from the true positive group, with predicted probabilities of 82% and 84%, respectively. Two patients, C and D, were randomly selected from the true negative group, with predicted probabilities of 13.7% and 24%, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e14\u003c/span\u003e). Additionally, LIME was used to explain the prediction for patient A from the true positive group, where the predicted probability remained at 82% (Figs.\u0026nbsp;\u003cspan refid=\"Fig15\" class=\"InternalRef\"\u003e15\u003c/span\u003e). To further assess the clinical applicability of the model, we plotted the clinical decision curve (DCA). Decision curve analysis indicated that the machine learning model developed in this study provided a higher net benefit in predicting early postoperative liver metastasis, especially when the risk threshold ranged from 0.05 to 0.4. In this range, the model outperformed both the \"all intervention\" and \"no intervention\" strategies, suggesting that it holds clinical decision support potential, particularly for personalized management of high-risk patients (Fig.\u0026nbsp;\u003cspan refid=\"Fig16\" class=\"InternalRef\"\u003e16\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eAlthough curative surgery for resectable PC has become increasingly refined, the high postoperative recurrence rate and short survival time remain significant challenges for clinicians. Several studies\u003csup\u003e[\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e have shown that more than 50% of PC patients experience early recurrence after surgery, with the liver being the most common site of recurrence, often indicating poorer prognosis. This study retrospectively analyzed clinical data from a single-center cohort of PC patients who underwent curative surgery over a 10-year period, identifying predictive factors for early liver metastasis after surgery, which can aid in the formulation of appropriate treatment plans and monitoring strategies for patients with early liver metastasis.\u003c/p\u003e\u003cp\u003ePostoperative adjuvant chemotherapy is a crucial component of treatment following curative resection of pancreatic cancer. In this study, SHAP analysis revealed that adjuvant chemotherapy is one of the most influential variables in predicting the risk of early liver metastasis after surgery, consistent with numerous domestic and international studies, further validating the clinical importance of adjuvant chemotherapy. Mechanistically, adjuvant chemotherapy can effectively eliminate microscopic residual disease that is undetectable preoperatively or intraoperatively, thereby delaying recurrence and suppressing metastasis. Several international multicenter randomized\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e controlled trials have also confirmed that both single-agent gemcitabine and combination chemotherapy regimens\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e can prolong overall survival and relapse-free survival in patients. However, in clinical practice, some patients fail to receive chemotherapy promptly due to slow postoperative recovery, concerns about toxicity, or poor adherence, leading to early recurrence and metastasis. Therefore, the timing of initiating adjuvant chemotherapy after surgery is critical. An analysis by Ma et al\u003csup\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e. of 7,548 stage I\u0026ndash;II pancreatic cancer patients from the National Cancer Database found that patients who started adjuvant chemotherapy 28 to 59 days postoperatively had the best overall survival. Domestic researchers also suggest that if there are no contraindications, adjuvant chemotherapy should be started as soon as possible after surgery, typically no later than 12 weeks postoperatively, and the treatment should continue for 6 months. This highlights the importance for clinicians to prioritize and promote the complete course of adjuvant chemotherapy in postoperative management.\u003c/p\u003e\u003cp\u003eThe liver, due to its unique dual blood supply system (portal vein and hepatic artery), makes it easier for tumor cells originating from the gastrointestinal tract and pancreas to become trapped in the circulation and colonize the liver parenchyma. The circulating tumor cell (CTC) hypothesis explains that once a tumor invades the portal venous system or celiac artery, cancer cells are easily disseminated to the liver through the bloodstream. In a meta-analysis involving 5242 PDAC patients, Song et al\u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e. pointed out that those with portal vein/superior mesenteric vein (PV/SMV) invasion exhibited poorer biological behavior and had a higher risk of postoperative recurrence. Further research by Tong et al\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e. emphasized that the extent of portal vein invasion is an independent risk factor for poor postoperative prognosis. Although surgical resection may achieve R0, deeper vascular invasion implies higher risks of residual tumor and distant metastasis. For patients with locally advanced pancreatic tail cancer invading the celiac artery, although distal pancreatectomy with celiac artery resection\u003csup\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e (DP-CAR) can provide surgical opportunities, it does not significantly improve the rates of curative resection or long-term survival. Compared with traditional distal pancreatectomy (DP), DP-CAR results in a lower R0 resection rate, higher intraoperative risk, and no significant difference in survival. Low-grade tumors are often more aggressive and metastatic. At the molecular level, poorly differentiated or undifferentiated PDAC cells exhibit higher activity in several key metastatic pathways\u003csup\u003e[\u003cspan additionalcitationids=\"CR26\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e. Firstly, poorly differentiated tumors commonly activate epithelial-mesenchymal transition (EMT) pathways, which reduce cell adhesion and enhance migration capacity, significantly increasing the risk of liver metastasis. Secondly, low-grade PDAC is often accompanied by TP53 mutations and loss of SMAD4 expression, both of which are highly correlated with liver metastasis. Furthermore, poorly differentiated PDAC reshapes the liver microenvironment by releasing MIF-rich exosomes, establishing a pre-metastatic niche, and thereby gaining a metastatic advantage. Therefore, the degree of differentiation not only serves as a morphological pathological indicator but also represents the tumor's potential for systemic dissemination.\u003c/p\u003e\u003cp\u003eThis study focused on the potential of preoperative CA19-9 levels as a predictor for early liver metastasis risk after PC surgery. Previous studies have shown\u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e that high preoperative CA19-9 levels are strongly associated with a high risk of distant metastasis. Guan et al\u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e. found that elevated CA19-9 significantly shortened both disease-free survival (DFS) and overall survival (OS), making it an independent predictor for both DFS and OS. Further analysis by Raza et al\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e. revealed that the average preoperative CA19-9 level in metastatic patients was significantly higher than in non-metastatic patients, with an ROC cutoff point of 336 U/mL (sensitivity 90%, specificity 80%, AUC\u0026thinsp;=\u0026thinsp;0.90). Tumor size has also been validated as a major risk factor for postoperative liver metastasis in multiple studies. Murakami et al\u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e. discovered that a preoperative tumor size\u0026thinsp;\u0026ge;\u0026thinsp;22 mm is a significant predictor of hidden metastasis and is associated with poor overall prognosis. Ayoub et al\u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. compared the 1- and 2-year survival rates of patients with pancreatic head tumors\u0026thinsp;\u0026le;\u0026thinsp;3 cm and \u0026gt;\u0026thinsp;3 cm, and found that patients with tumors\u0026thinsp;\u0026gt;\u0026thinsp;3 cm had significantly lower 1-year and 2-year survival rates (50% vs. 79.1% and 19.2% vs. 40.3%, respectively). In this study, SHAP interaction analysis further revealed the synergistic effect between tumor size and CA19-9, particularly when CA19-9 levels exceeded 600 U/mL, where the interaction SHAP values of some samples increased significantly, suggesting that tumor size had a stronger positive contribution to the model\u0026rsquo;s output at these levels. Overall, larger tumor size and higher preoperative CA19-9 levels may reflect a larger tumor burden and the possible presence of hidden distant micro-metastasis, marking them as important risk indicators for early postoperative liver metastasis.\u003c/p\u003e\u003cp\u003eHepatitis B virus (HBV), a hepatotropic DNA virus, persistently infects the liver, triggering chronic inflammation. In this study, we found that patients with preoperative HBV infection had a significantly higher risk of early liver metastasis after surgery compared to those without HBV infection. We hypothesize that the chronic inflammatory environment in the liver may alter the liver microenvironment, promoting increased tumor invasiveness. This finding is consistent with the study by Wei et al\u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. However, a study by Chen\u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e et al. in patients with unresectable PC reported that patients with chronic active HBV infection had a lower risk of postoperative liver metastasis, which contradicts our findings. These discrepancies may be due to differences in study populations (resectable vs. unresectable) and HBV infection status, among other factors. Overall, current evidence does not provide a consistent conclusion regarding whether HBV acts as a \"risk factor\" or \"protective factor\" for early postoperative liver metastasis in PC. In addition to traditional tumor burden indicators, this study also identified low albumin levels and lymphocyte count as predictors for early liver metastasis after PC surgery. Albumin, a sensitive marker reflecting nutritional status and chronic inflammation, was identified as a negative protective factor in our study, with lower levels being associated with an increased risk of early liver metastasis. Lymphocytes, as a key component of the immune surveillance system, are also one of the factors contributing to an increased risk of early liver metastasis postoperatively. A study by Gumus\u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e et al. pointed out that low albumin levels and low HALP scores (a combination of hemoglobin, albumin, lymphocytes, and platelets) were significantly associated with shorter DFS and OS, making them independent prognostic factors for poor outcomes.\u003c/p\u003e\u003cp\u003eAlthough our study provides valuable insights, several limitations must be acknowledged. First, as a single-center retrospective study, the model's generalizability needs to be further validated in multi-center, large-scale prospective studies. Second, the study primarily relied on conventional clinical and laboratory indicators and did not incorporate radiomics or pathological imaging data. Third, while chemotherapy was included as a variable in the model, the differences between various chemotherapy regimens were not specifically analyzed. Fourth, the sample size of this study was relatively small, limiting the full potential of the ML model. Additionally, patients who received neoadjuvant or conversion therapy before surgery were excluded, yet these treatment strategies have become increasingly widespread in recent years, meaning the study sample may not fully represent the broader patient population, thereby restricting the model's clinical applicability. Therefore, future research should focus on multi-center, large-scale data, integrating multi-modal data, and utilizing methods like deep learning to further enhance the model's accuracy.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study developed a predictive model for early liver metastasis after PC surgery using ML methods, which demonstrated good validation results in the test set and strong clinical interpretability. The model provides robust support for the early identification and personalized management of high-risk patients post-surgery, assisting clinicians in formulating individualized treatment plans.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003e\u003cstrong\u003ePC\u003c/strong\u003e Pancreatic cancer,\u003cstrong\u003eML\u003c/strong\u003e Machine learning,AI artificial intelligence,\u003cstrong\u003eLASSO\u0026nbsp;\u003c/strong\u003eLeast absolute shrinkage and selection operator, \u003cstrong\u003eAUC\u003c/strong\u003e Area under the receiver operating characteristic, \u003cstrong\u003ePPV\u003c/strong\u003e Positive predictive value, \u003cstrong\u003eNPV\u003c/strong\u003e Negative predictive value, \u003cstrong\u003eDT\u003c/strong\u003e Decision Tree,\u003cstrong\u003eSVM\u003c/strong\u003e Support Vector Machine,\u003cstrong\u003eXGBoost\u003c/strong\u003e eXtreme Gradient Boosting, \u003cstrong\u003eRF\u003c/strong\u003e Random Forst, \u003cstrong\u003eLightBGM\u003c/strong\u003e Light Gradient Boosting Machine, \u003cstrong\u003eLR\u003c/strong\u003e Logistic Regression,\u003cstrong\u003eET\u003c/strong\u003e Extra Trees.\u003cstrong\u003eSHAP\u003c/strong\u003e Shapley Additive Explanations,\u003cstrong\u003eLIME\u003c/strong\u003e Local Interpretable Model-agnostic Explanations,\u003cstrong\u003ePDP\u003c/strong\u003e\u0026nbsp; \u0026nbsp;Partial Dependence Plot,\u003cstrong\u003eICE\u0026nbsp;\u003c/strong\u003eIndividual Conditional Expectation.\u003cstrong\u003eISGPS\u003c/strong\u003e International Study Group of Pancreatic Surgery. \u003cstrong\u003eRFE\u003c/strong\u003e Recursive feature elimination.\u003cstrong\u003e\u0026nbsp;PBD\u003c/strong\u003e Preoperative Biliary Drainage, \u003cstrong\u003eAV invasion\u003c/strong\u003e Hepatic artery / portal vein invasion,\u003cstrong\u003ePOH\u0026nbsp;\u003c/strong\u003ePostoperative Hemorrhage,\u003cstrong\u003ePOIAI\u003c/strong\u003e Postoperative Intra-abdominal Infection, \u003cstrong\u003ePOPF\u003c/strong\u003e Postoperative Pancreatic Fistula, \u003cstrong\u003eBMI\u003c/strong\u003e Body Mass Index,\u003cstrong\u003eIBL\u003c/strong\u003e Intraoperative Blood Loss, \u003cstrong\u003eNLR\u003c/strong\u003e Neutrophil-lymphocyte ratio , \u003cstrong\u003ePLR\u0026nbsp;\u003c/strong\u003ePlatelet-to-Lymphocyte Ratio\u003cstrong\u003e\u0026nbsp;,LNM\u003c/strong\u003e Lymph Node Metastasis. \u003cstrong\u003ePCA\u003c/strong\u003e Principal. \u003cstrong\u003eAP\u003c/strong\u003e Average Precision, \u003cstrong\u003ePR\u003c/strong\u003e Precision-Recall,\u003cstrong\u003eDCA\u003c/strong\u003e Decision curve analysis,\u003cstrong\u003eCTC\u003c/strong\u003e Circulating tumor cell, \u003cstrong\u003eDP-CAR\u003c/strong\u003e Distal pancreatectomy with celiac artery resection, \u003cstrong\u003eDP\u003c/strong\u003e Distal pancreatectomy, \u003cstrong\u003eEMT\u003c/strong\u003e Epithelial-mesenchymal transition,\u003cstrong\u003eDFS\u003c/strong\u003e Disease-free survival,\u003cstrong\u003eOS\u003c/strong\u003e overall survival, \u003cstrong\u003eHBV\u0026nbsp;\u003c/strong\u003eHepatitis B virus\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate: This study was conducted in accordance with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University, and all participants provided written informed consent prior to participation in the study.\u003c/p\u003e\n\u003cp\u003eConsent for publication: Written informed consent for publication of identifying information and/or images was obtained from the patient(s) included in the study.\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials:The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003eCompeting interests: The authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eFunding:\u0026nbsp;Funded Project of the State Key Laboratory of Causes and Prevention of High Morbidity in Central Asia (No. SKL-HIDCA-2023-26), jointly built by the Ministry and the Province.\u003c/p\u003e\n\u003cp\u003eAuthors\u0026apos; contributions: Chenhui Du was responsible for manuscript writing and data analysis. \u0026nbsp;Shuo Zhang contributed to data collection. Guoyu Li provided the research concept. Xinling Cao and Tieying He critically reviewed the manuscript.\u003c/p\u003e\n\u003cp\u003eAcknowledgements:The authors would like to thank OpenAI\u0026apos;s ChatGPT for its guidance and assistance in refining the code for the model construction in this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Cancer J Clin. 2024;74(3):229\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStrobel O, Neoptolemos J, J\u0026auml;ger D, B\u0026uuml;chler MW. Optimizing the outcomes of pancreatic cancer surgery. Nat Rev Clin Oncol. 2019;16(1):11\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRojas LA, Sethna Z, Soares KC, Olcese C, Pang N, Patterson E, Lihm J, Ceglia N, Guasp P, Chu A, Yu R, Chandra AK, Waters T, Ruan J, Amisaki M, Zebboudj A, Odgerel Z, Payne G, Derhovanessian E, Balachandran VP. Personalized RNA neoantigen vaccines stimulate T cells in pancreatic cancer. Nature. 2023;618(7963):144\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRompen IF, Levine J, Habib JR, Sereni E, Mughal N, Hewitt DB, Sacks GD, Welling TH, Simeone DM, Kaplan B, Berman RS, Cohen SM, Wolfgang CL, Javed AA. Progression of site-specific recurrence of pancreatic cancer and implications for treatment. Ann Surg. 2024;280(2):317\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCuocolo R, Caruso M, Perillo T, Ugga L, Petretta M. Machine learning in oncology: A clinical appraisal. Cancer Lett. 2020;481:55\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChung H, Ko Y, Lee I-S, Hur H, Huh J, Han S-U, Kim KW, Lee J. Prognostic artificial intelligence model to predict 5 year survival at 1 year after gastric cancer surgery based on nutrition and body morphometry. J Cachexia Sarcopenia Muscle. 2023;14(2):847\u0026ndash;59.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGao Y, Li S, Jin Y, Zhou L, Sun S, Xu X, Li S, Yang H, Zhang Q, Wang Y. An assessment of the predictive performance of current machine learning-based breast cancer risk prediction models: Systematic review. JMIR Public Health Surveillance. 2022;8(12):e35750.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang J, Yao X. Which approach better predicts diabetes: Traditional econometric methods or machine learning? Evidence from a cross-sectional study in south korea. Comput Biol Med. 2025;190:110035.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQian X, Pei J, Han C, Liang Z, Zhang G, Chen N, Zheng W, Meng F, Yu D, Chen Y, Sun Y, Zhang H, Qian W, Wang X, Er Z, Hu C, Zheng H, Shen D. A multimodal machine learning model for the stratification of breast cancer risk. Nat Biomedical Eng. 2025;9(3):356\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChaddad A, Peng J, Xu J, Bouridane A. Survey of explainable AI techniques in healthcare. Sensors. 2023;23(2):634.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eConroy T, Pfeiffer P, Vilgrain V, Lamarca A, Seufferlein T, O\u0026rsquo;Reilly EM, Hackert T, Golan T, Prager G, Haustermans K, Vogel A, Ducreux M, ESMO Guidelines Committee. \u0026amp;. Electronic address: [email protected]. (2023). Pancreatic cancer: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. \u003cem\u003eAnnals of Oncology: Official Journal of the European Society for Medical Oncology\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(11), 987\u0026ndash;1002.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWente MN, Veit JA, Bassi C, Dervenis C, Fingerhut A, Gouma DJ, Izbicki JR, Neoptolemos JP, Padbury RT, Sarr MG, Yeo CJ, B\u0026uuml;chler MW. Postpancreatectomy hemorrhage (PPH): An international study group of pancreatic surgery (ISGPS) definition. Surgery. 2007;142(1):20\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBassi, C., Marchegiani, G., Dervenis, C., Sarr, M., Abu Hilal, M., Adham, M., Allen,P., Andersson, R., Asbun, H. J., Besselink, M. G., Conlon, K., Del Chiaro, M., Falconi,M., Fernandez-Cruz, L., Fernandez-Del Castillo, C., Fingerhut, A., Friess, H., Gouma,D. J., Hackert, T., \u0026hellip; International Study Group on Pancreatic Surgery (ISGPS). (2017).The 2016 update of the international study group (ISGPS) definition and grading of postoperative pancreatic fistula: 11 years after. \u003cem\u003eSurgery\u003c/em\u003e, \u003cem\u003e161\u003c/em\u003e(3), 584\u0026ndash;591.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMazuski JE, Tessier JM, May AK, Sawyer RG, Nadler EP, Rosengart MR, Chang PK, O\u0026rsquo;Neill PJ, Mollen KP, Huston JM, Diaz JJ, Prince JM. The surgical infection society revised guidelines on the management of intra-abdominal infection. Surg Infect. 2017;18(1):1\u0026ndash;76.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang X-P, Xu S, Gao Y-X, Zhao Z-M, Zhao G-D, Hu M-G, Tan X-L, Lau WY, Liu R. Early and late recurrence patterns of pancreatic ductal adenocarcinoma after pancreaticoduodenectomy: A multicenter study. Int J Surg (London England). 2023;109(4):785\u0026ndash;93.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDaamen, L. A., Dorland, G., Brada, L. J. H., Groot, V. P., van Oosten, A. F., Besselink,M. G., Bosscha, K., Bonsing, B. A., Busch, O. R., Cirkel, G. A., van Dam, R. M., Festen,S., Groot Koerkamp, B., Haj Mohammad, N., van der Harst, E., de Hingh, I. H. J. T.,Intven, M. P. W., Kazemier, G., Los, M., \u0026hellip; Dutch Pancreatic Cancer Group. (2022).Preoperative predictors for early and very early disease recurrence in patients undergoing resection of pancreatic ductal adenocarcinoma. \u003cem\u003eHPB: The Official Journal of the International Hepato Pancreato Biliary Association\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(4), 535\u0026ndash;546.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGroot VP, Gemenetzis G, Blair AB, Ding D, Javed AA, Burkhart RA, Yu J, Borel Rinkes IH, Molenaar IQ, Cameron JL, Fishman EK, Hruban RH, Weiss MJ, Wolfgang CL, He J. Implications of the Pattern of Disease Recurrence on Survival Following Pancreatectomy for Pancreatic Ductal Adenocarcinoma. Ann Surg Oncol. 2018;25(8):2475\u0026ndash;83.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGroot VP, Rezaee N, Wu W, Cameron JL, Fishman EK, Hruban RH, Weiss MJ, Zheng L, Wolfgang CL, He J. Patterns, timing, and predictors of recurrence following pancreatectomy for pancreatic ductal adenocarcinoma. Ann Surg. 2018;267(5):936\u0026ndash;45.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKlaiber U, Hackert T, Neoptolemos JP. Adjuvant treatment for pancreatic cancer. Translational Gastroenterol Hepatol. 2019;4:27.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOettle H, Post S, Neuhaus P, Gellert K, Langrehr J, Ridwelski K, Schramm H, Fahlke J, Zuelke C, Burkart C, Gutberlet K, Kettner E, Schmalenberg H, Weigang-Koehler K, Bechstein W-O, Niedergethmann M, Schmidt-Wolf I, Roll L, Doerken B, Riess H. Adjuvant chemotherapy with gemcitabine vs observation in patients undergoing curative-intent resection of pancreatic cancer: A randomized controlled trial. JAMA. 2007;297(3):267\u0026ndash;77.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa SJ, Oladeru OT, Miccio JA, Iovoli AJ, Hermann GM, Singh AK. Association of timing of adjuvant therapy with survival in patients with resected stage I to II pancreatic cancer. JAMA Netw Open. 2019;2(8):e199126.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSong A, Liu F, Wu L, Si X, Zhou Y. Histopathologic tumor invasion of superior mesenteric vein/ portal vein is a poor prognostic indicator in patients with pancreatic ductal adenocarcinoma: Results from a systematic review and meta-analysis. Oncotarget. 2017;8(20):32600\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTong J, Jiang W, Mao S, Wu S, Lu C. Development and validation of a nomogram to predict liver metastasis for pancreatic ductal adenocarcinoma after radical resection. Front Oncol. 2022;12:1040411.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFeng Q, Xin Z, Du Y, Mao F, Li L, Zhai H, Yao J. Distal pancreatectomy with en bloc celiac axis resection does not improve the R0 rate or median survival time of patients with locally advanced pancreatic cancer: A systematic review and meta-analysis. Translational Cancer Res. 2020;9(11):7205\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHoshino, A., Costa-Silva, B., Shen, T.-L., Rodrigues, G., Hashimoto, A., Tesic Mark,M., Molina, H., Kohsaka, S., Di Giannatale, A., Ceder, S., Singh, S., Williams, C.,Soplop, N., Uryu, K., Pharmer, L., King, T., Bojmar, L., Davies, A. E., Ararso, Y.,\u0026hellip; Lyden, D. (2015). Tumour exosome integrins determine organotropic metastasis. \u003cem\u003eNature\u003c/em\u003e, \u003cem\u003e527\u003c/em\u003e(7578), 329\u0026ndash;335.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBorrelli, C., Roberts, M., Eletto, D., Hussherr, M.-D., Fazilaty, H., Valenta, T.,Lafzi, A., Kretz, J. A., Guido Vinzoni, E., Karakatsani, A., Adivarahan, S., Mannhart,A., Kimura, S., Meijs, A., Baccouche Mhamedi, F., Acar, I. E., Handler, K., Ficht,X., Platt, R. J., \u0026hellip; Moor, A. E. (2024). In vivo interaction screening reveals liver-derived constraints to metastasis. \u003cem\u003eNature\u003c/em\u003e, \u003cem\u003e632\u003c/em\u003e(8024), 411\u0026ndash;418.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMasugi, Y., Takamatsu, M., Tanaka, M., Hara, K., Inoue, Y., Hamada, T., Suzuki, T.,Arita, J., Hirose, Y., Kawaguchi, Y., Nakai, Y., Oba, A., Sasahira, N., Shimane, G.,Takeda, T., Tateishi, K., Uemura, S., Fujishiro, M., Hasegawa, K., \u0026hellip; GTK Pancreatic Cancer Study Group in Japan. (2023). Post-operative mortality and recurrence patterns in pancreatic cancer according to KRAS mutation and CDKN2A, p53, and SMAD4 expression.\u003cem\u003eThe Journal of Pathology. Clinical Research\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(5), 339\u0026ndash;353.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim TH, Han S-S, Park S-J, Lee WJ, Woo SM, Yoo T, Moon SH, Kim SH, Hong EK, Kim DY, Park J-W. CA 19\u0026thinsp;\u0026ndash;\u0026thinsp;9 level as indicator of early distant metastasis and therapeutic selection in resected pancreatic cancer. Int J Radiat Oncol Biol Phys. 2011;81(5):e743\u0026ndash;748.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuan S, Chen Y, Han Q, Deng G, Wang Y, Shi Y, Dai G. Preoperative CA19-9 levels predict disease-free survival and overall survival in pancreatic adenocarcinoma patients after resection. Translational Cancer Res. 2019;8(3):811\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRaza SS, Khan H, Hajibandeh S, Hajibandeh S, Bartlett D, Chatzizacharias N, Roberts K, Marudanayagam R, Sutcliffe RP. Can preoperative carbohydrate antigen 19\u0026thinsp;\u0026ndash;\u0026thinsp;9 predict metastatic pancreatic cancer? Results of a systematic review and meta-analysis. HPB: Official J Int Hepato Pancreato Biliary Association. 2024;26(5):630\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMurakami T, Kimura Y, Imamura M, Nagayama M, Kato T, Kukita K, Yoshida M, Masaki Y, Nakase H, Takemasa I. Predictors of occult metastases in potentially resectable pancreatic ductal adenocarcinoma. Surg Open Sci. 2024;20:222\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAyoub II, Talab TAE, Omar H, Akoud SAB, Shoreem HAM, El-Deen EMS, Marwan IK, Macshut M. Impact of pancreatic head tumor size on the outcome of surgical management. Egypt J Surg. 2024;43(1):258.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWei X, Qiu M, Chen W, Jin Y, Ren C, Wang F, Luo H, Wang Z, Zhang D, Wang F, Li Y, Xu R. The status of HBV infection influences metastatic pattern and survival in Chinese patients with pancreatic cancer. J Translational Med. 2013;11:249.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen, Q., Ning, Z., Wang, L., Ying, H., Dong, S., Zhang, C., Shen, X., Guo, Y., Chen,H., Zhu, X., Shen, Y., Shi, W., Hua, Y., Wang, K., Lin, J., Xu, L., Chen, L., Feng,L., Zhang, X., \u0026hellip; Meng, Z. (2016). Is chronic hepatitis B infection a protective factor for the progression of advanced pancreatic ductal adenocarcinoma? An analysis from a large multicenter cohort study. \u003cem\u003eOncotarget\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(51), 85603\u0026ndash;85612.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGumus T, Umman V, Cetin B, Uguz A. Utilizing albumin value, HALP score and LCR value for predicting survival in patients with pancreatic adenocarcinoma. Med (Kaunas Lithuania). 2025;61(4):639.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"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":"Machine Learning, Pancreatic Cancer, Liver Metastasis, Predictive Model","lastPublishedDoi":"10.21203/rs.3.rs-7841940/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7841940/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal malignancy. Despite undergoing radical surgical resection, patients are still at a high risk of recurrence and distant metastasis postoperatively. Among the organs prone to hematogenous metastasis, the liver is the most common site, and liver metastasis significantly shortens the survival period, becoming a key factor influencing prognosis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eObjective \u003c/strong\u003eThis study aims to develop an interpretable machine learning model based on postoperative clinical variables to predict the risk of liver metastasis within one year after surgery in pancreatic cancer patients.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e This study included data from 418 patients who underwent radical pancreatic cancer surgery at the Department of Gastrointestinal Surgery, First Affiliated Hospital of Xinjiang Medical University, between January 2015 and August 2024. The data were randomly divided into a training set and a test set in a 7:3 ratio. The performance of seven machine learning models was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC-ROC) and the area under the precision-recall curve (AP-PR). SHAP and LIME methods were used to determine feature importance and explain the best-performing model.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e After applying inclusion and exclusion criteria, 363 patients were included in the study. Among them, 118 patients (32.5%) developed liver metastasis within one year postoperatively. The final model incorporated 10 variables: chemotherapy status, tumor differentiation, vascular invasion (arterial/venous), hepatitis B infection, CA19-9 level, T stage, lymphocyte count, albumin level, alkaline phosphatase, and tumor size. Among the seven machine learning models, the Extra Trees (ET) model performed the best, achieving an AUC-ROC of 0.82 (95% CI: 0.73–0.90) and an average precision (AP-PR) of 0.77 in the test set. SHAP analysis revealed that postoperative chemotherapy, tumor differentiation, hepatic artery/portal vein invasion, and hepatitis B virus infection were the most influential predictors of liver metastasis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion \u003c/strong\u003eAn interpretable machine learning model was developed using postoperative clinical data, demonstrating good performance and interpretability. The model effectively predicts the risk of liver metastasis within one year after pancreatic cancer surgery. It holds promise as an auxiliary tool for postoperative follow-up and personalized interventions, providing clinicians with more precise decision-making support through feature contribution analysis.\u003c/p\u003e","manuscriptTitle":"Development and Interpretability Analysis of a Machine Learning-Based Model for Predicting Early Liver Metastasis Risk After Pancreatic Cancer Surgery","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-13 14:49:18","doi":"10.21203/rs.3.rs-7841940/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-12T09:00:03+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-11T08:55:44+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-10T05:47:37+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"73905346536319465253411805901855269046","date":"2025-11-10T04:53:38+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-05T07:36:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"23628476035394115144193166799770241737","date":"2025-11-05T01:59:01+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"333169220657139458419366572888467409392","date":"2025-11-04T17:01:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"112952047725942090465725416803131457734","date":"2025-11-04T15:09:58+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"106957201624932275517767074334209414446","date":"2025-11-04T01:21:10+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-03T21:59:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"200658800545413983073867245211237116140","date":"2025-11-03T21:42:23+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-03T20:33:00+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-21T11:05:00+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-20T16:57:30+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cancer","date":"2025-10-20T16:53:36+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":"d0223298-b4a4-44eb-a8f9-9625fea6d610","owner":[],"postedDate":"November 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-12-04T08:38:21+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-13 14:49:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7841940","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7841940","identity":"rs-7841940","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}