Parastomal hernia (PSH) is one of the most frequent long-term complications following abdominoperineal resection (APR) for rectal cancer. The optimal colostomy route to minimize PSH remains controversial. This study aimed to compare PSH risk between extraperitoneal colostomy (EPC) and transperitoneal colostomy (TPC) after laparoscopic APR. A retrospective cohort study was conducted including patients who underwent laparoscopic APR for rectal cancer between 2014 and 2017. Patients were categorized according to colostomy route (EPC vs. TPC). The primary endpoint was PSH, and secondary endpoints included other short- and long-term stoma-related complications and perioperative outcomes. Propensity score matching (1:3) was applied to balance baseline characteristics. Risk factors for PSH were further analyzed using logistic regression. A total of 464 patients were included. After matching, 102 patients in the EPC group and 243 in the TPC group were analyzed. Perioperative outcomes and overall stoma-related complication rates were comparable between groups. However, PSH occurred less frequently in the EPC group than in the TPC group (10/102 [9.8

Liver metastasis constitutes the principal cause of mortality in colorectal cancer, with about 50% of patients developing liver metastasis during disease progression. While surgical resection remains the cornerstone of curative-intent treatment, thermal ablation is rapidly reshaping the landscape of local tumor control in well-selected individuals. This comprehensive review analyzes the evolving role of thermal ablation in the management of resectable, unresectable, and recurrent disease, with the aim of establishing contemporary best practices and identifying critical frontiers for future research.

Objective.Current automatic segmentation models in radiotherapy, which are predominantly unimodal and image-based, have limited generalizability due to boundary ambiguity and the lack of guideline integration. This study proposes a text-guided segmentation network, termed (TG-SegNet), for the automatic delineation of clinical target volumes (CTVs) in rectal cancer radiotherapy.Approach.Data from 567 preoperative patients with rectal cancer were retrospectively collected. Text prompts contained (i) patient case information (age, sex, tumor stage, tumor location, position) and (ii) guideline-derived descriptions indicating which CTV subsites should be included. TG-SegNet integrates computed tomography-derived visual features with structured clinical text prompts encoded by PubMedBERT, fused via cross-attention and fine-grained fusion. The model was trained on 452 patients and tested on 115. Its performance was compared with that of nnU-Net and two ablated variants (TG-SegNet without text prompts and TG-SegNet with simplified fusion). The evaluation comprised quantitative metrics, including dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), mean surface distance (MSD), surface DSC (S-DSC), and average path length (APL), along with blinded expert scoring and an efficiency analysis. In additional analyses, we conducted text-prompt and module ablations.Main results.TG-SegNet achieved the best performance across all quantitative metrics: DSC 0.927 ± 0.022, HD95 7.01 ± 6.05 mm, MSD 1.94 ± 1.08 mm, S-DSC 0.799 ± 0.074, and APL 7372 ± 4452 (allp< 0.01). In clinical evaluation, TG-SegNet significantly improved target coverage, guideline adherence, and overall clinical acceptability compared with nnU-Net and ablations (p< 0.05), with boundary appropriateness comparable to nnU-Net. TG-SegNet had the shortest correction time (3.39 ± 1.10 min), corresponding to 82.1% time savings versus manual delineation. Text-prompt ablations suggested that the CTV-subsite prompt component contributed more to performance. Module ablations showed that both cross-attention and fine-grained fusion were beneficial.Significance.By integrating clinical semantics with imaging, TG-SegNet demonstrated superior accuracy, efficiency, and clinical acceptability over nnU-Net and ablated models, highlighting its potential for clinical translation.

Background: Conventional laparoscopic-assisted surgery (CLS) for sigmoid and upper rectal cancer requires an abdominal extraction incision, linked to pain, surgical site infection and poor cosmesis. Natural orifice specimen extraction surgery (NOSES) removes tumours via the anus without abdominal wounds, yet high-quality randomised controlled trial evidence on long-term oncological safety remains scarce. This multicentre trial aimed to verify whether NOSES is non-inferior to CLS regarding 3-year disease-free survival (DFS), alongside evaluating short-term recovery and complications. Methods: This open-label, parallel-group, non-inferiority randomised trial enrolled patients with cT1-3N0-2M0 sigmoid/upper rectal adenocarcinoma across 13 Chinese tertiary centres between Aug 30, 2020, and Nov 5, 2023. Participants were 1:1 allocated via centre-stratified web randomisation; outcome assessors at discharge were masked to group assignment. The prespecified non-inferiority margin for 3-year DFS was 10%. Primary analysis used the modified intention-to-treat (mITT) population; per-protocol (PP) data served for sensitivity analysis. Trial registration: ChiCTR2000036314. Findings: 516 patients were randomised (258 per group). The mITT survival cohort included 205 CLS and 208 NOSES participants, with median follow-up of 36.9 months. 3-year DFS was 89.7% (95% CI 85.1-94.6) for CLS and 94.7% (91.1-98.4) for NOSES (absolute difference 5.0%, 95% CI -3.4 to 13.3, meeting non-inferiority; log-rank p=0.088). 3-year overall survival and local recurrence rates were similar between groups. NOSES patients had earlier first flatus (p<0.001), lower postoperative NRS pain scores, and less rescue analgesic use (17.5% vs 37.4%, p<0.001). Overall 30-day complication rate was numerically lower in NOSES (12.0% vs 18.3%, p=0.056); all incisional surgical site infections occurred only in the CLS group (6.0%). Hospital costs were higher with NOSES (mean difference 6525 CNY, p<0.001). No 30-day deaths occurred in either arm. Interpretation: For selected patients with cT1-3N0-2M0 sigmoid or upper rectal cancer, NOSES performed by experienced surgeons delivers non-inferior long-term oncological outcomes versus CLS, with meaningful improvements in postoperative pain, bowel recovery and surgical site infection risk. NOSES represents a patient-friendly minimally invasive option for suitable candidates.

Early colorectal cancer is often treated through endoscopic procedures to remove tumors. However, when initial removal is incomplete or shows high-risk features, further major surgery is required to ensure no cancer remains. This consensus provides a standard for surgeons to decide when and how to perform this additional surgery. A multidisciplinary group of Chinese medical experts reviewed global research and clinical evidence published between 2010 and 2024. The medical experts formulated key clinical recommendations, which were then discussed and finalized through expert voting to ensure high agreement. The consensus identifies specific high-risk factors necessitating additional surgery. The consensus recommends that these follow-up surgeries should ideally take place approximately 4 weeks after the first procedure. Furthermore, the consensus provides detailed protocols for marking the tumor location and selecting the best surgical approach. These consensuses offer a practical framework to improve the safety and effectiveness of treating early colorectal cancer. By standardizing surgical decisions, the consensus aims to help patients achieve better long-term recovery and quality of life.

LnCeVar 2.0 (available at http://bio-bigdata.hrbmu.edu.cn/LnCeVar or http://www.bio-bigdata.net/LnCeVar) is an updated database investigating genomic variations that disrupt competing endogenous RNA (ceRNA) networks via single-cell and spatial transcriptomics. Enhancements include expanded data and improved features: (i) 16 937 experimentally supported cancer biomarkers as well as 5785 validated ceRNA interactions and single nucleotide variant (SNV)-ceRNA events, manually curated and linked to key cancer pathogenic processes; (ii) 812 single-cell RNA sequencing/spatial transcriptomics RNA sequencing datasets covering 102 diseases, clinical treatments (e.g. chemotherapy, immunotherapy), and normal tissues; (iii) 5 218 062 single-cell- and spatial-specific SNV-ceRNA events across 2 673 603 cells/spots, with cellular functional perturbation networks; (iv) 5 comprehensive and 12 mini tools for multilevel cross talk analysis and 3D visualization; and (v) novel inference of SNV effects on cell types, states, and functions at single-cell and spatial levels. LnCeVar 2.0 features a user-friendly interface for searching, browsing, and analyzing data. For instance, the CeVarState interface illustrates how SNV-ceRNA events influence cell states during developmental processes, revealing interactions that determine cell fate. The CeVarSC3D and CeVarST3D tools perform multilevel cross talk analyses of SNVs, ceRNA networks, and cell states in disease pathology, providing interactive 3D visualizations. Overall, we anticipate that the updated database will facilitate the high-resolution investigation of SNV-ceRNA networks and advance our understanding of the regulatory mechanisms in complex disease ecosystems.

Abstract The tumor microenvironment (TME) plays a critical role in cancer progression and therapeutic response, with cancer-associated fibroblasts (CAFs) being a key stromal component. Conventional tumor organoid models lack TME elements, and existing co-culture systems have limitations in recapitulating dynamic, multidimensional interactions. Here, we developed a novel dynamic co-culture chip (BAC) to better model the TME and investigate CAF-tumor cell crosstalk. Using this platform, we established a non-contact co-culture model of patient-derived colorectal cancer cells and CAFs. The resulting model closely recapitulated the morphological and molecular characteristics of the original patient tumors. Compared with tumor organoids cultured alone, the co-culture system exhibited significantly higher resistance to the clinically common chemotherapeutics 5-fluorouracil and oxaliplatin. Moreover, the presence of CAFs promoted tumor recurrence. Notably, drug responses in the co-culture model showed superior concordance with clinical outcomes relative to both organoid-only and animal models. Transcriptomic profiling under different culture conditions provided further insights into the mechanisms driving CAF-mediated interactions. These findings demonstrate that the BAC-based tumor organoid-CAF co-culture model serves as a more accurate platform for predicting drug responses, investigating TME-dependent mechanisms, and guiding personalized cancer therapy.

Background: Current evidence suggests that neoadjuvant chemoradiotherapy (nCRT) followed by total mesorectal excision (TME) alone is insufficient for magnetic resonance imaging (MRI)–suspected lateral lymph node metastasis (LLNM) in rectal cancer. However, whether upfront TME with lateral lymph node dissection (LLND) is adequate, and whether adding nCRT before planned LLND confers additional benefit, remains controversial. Methods: Between May 2021 and September 2022, a total of 342 patients from 20 Chinese centers were enrolled, of whom 293 were included in the final analysis and received either long-course nCRT plus TME with LLND or upfront TME+LLND. Groups were balanced by propensity score matching. The primary endpoint was 3-year recurrence-free survival (RFS). Findings: After matching, the nCRT group had significantly better 3-year RFS (HR 0.54; 95% CI. 0.32–0.92; P=0.023) and locoregional recurrence-free survival (HR 0.37; 95% CI, 0.19–0.74; P=0.005); distant metastasis-free survival did not differ (HR 0.74; 95% CI, 0.37–1.47; P=0.385). In patients with pretreatment positive lateral nodes (n=141), nCRT also improved RFS (HR 0.53; 95% CI, 0.28–0.99; P=0.049) and local control (HR 0.35; 95% CI, 0.15–0.77; P=0.010). Pathologic complete response in lateral nodes after nCRT was 44.7% (34/76). Overall postoperative complication rates were similar between groups (19.8% vs 16.0%; P=0.474), as were severe complications (grade III–V, 9.4% vs 8.5%; P=0.811). Interpretation: In MRI-suspected LLNM, adding nCRT to TME with LLND significantly improves RFS and local control without increasing morbidity. Upfront surgery alone is insufficient. These findings support a combined treatment paradigm and confirm the necessity of nCRT in these patients.

Background: Current evidence suggests that neoadjuvant chemoradiotherapy (nCRT) followed by total mesorectal excision (TME) alone is insufficient for magnetic resonance imaging (MRI)–suspected lateral lymph node metastasis (LLNM) in rectal cancer. However, whether upfront TME with lateral lymph node dissection (LLND) is adequate, and whether adding nCRT before planned LLND confers additional benefit, remains controversial. Methods: Between May 2021 and September 2022, a total of 342 patients from 20 Chinese centers were enrolled, of whom 293 were included in the final analysis and received either long-course nCRT plus TME with LLND or upfront TME+LLND. Groups were balanced by propensity score matching. The primary endpoint was 3-year recurrence-free survival (RFS). Findings: After matching, the nCRT group had significantly better 3-year RFS (HR 0.54; 95% CI. 0.32–0.92; P=0.023) and locoregional recurrence-free survival (HR 0.37; 95% CI, 0.19–0.74; P=0.005); distant metastasis-free survival did not differ (HR 0.74; 95% CI, 0.37–1.47; P=0.385). In patients with pretreatment positive lateral nodes (n=141), nCRT also improved RFS (HR 0.53; 95% CI, 0.28–0.99; P=0.049) and local control (HR 0.35; 95% CI, 0.15–0.77; P=0.010). Pathologic complete response in lateral nodes after nCRT was 44.7% (34/76). Overall postoperative complication rates were similar between groups (19.8% vs 16.0%; P=0.474), as were severe complications (grade III–V, 9.4% vs 8.5%; P=0.811). Interpretation: In MRI-suspected LLNM, adding nCRT to TME with LLND significantly improves RFS and local control without increasing morbidity. Upfront surgery alone is insufficient. These findings support a combined treatment paradigm and confirm the necessity of nCRT in these patients.

Background:No previous study has assessed the relationship between macronutrient quality and colorectal cancer (CRC) incidence and mortality. Thus, to further explore the associations between macronutrient quality and CRC risk, we conducted a large prospective cohort study involving 101,709 people in the United States from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. Methods:Our study population was derived from 154,887 adults aged 55 to 74 years who were recruited from 10 screening centers in the United States. The macronutrient quality index (MQI) was calculated based on dietary history questionnaire (DHQ). Cox regression analysis was utilized to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs) of the associations between MQI and CRC incidence and mortality. We used subgroup analyses to identify potential effect modifiers. Sensitivity analysis was performed to ensure the study findings were robust. Results:During the study period, 1,100 colorectal cancer (CRC) diagnoses and 314 CRC-related deaths were recorded. Higher adherence to the MQI was significantly associated with reduced CRC risk, demonstrating a 22% lower incidence (HR Q4 vs. Q1: 0.78; 95% CI: 0.65-0.93; p = 0.006 for trend) and 38% lower mortality (HR Q4 vs. Q1: 0.62; 95% CI: 0.44-0.86; p = 0.001 for trend) in the highest vs. lowest quartiles. These associations were robust across sensitivity analyses. Subsite-specific analyses revealed pronounced protective effects for distal colon cancer incidence (36% reduction; HR: 0.64; 95% CI: 0.43-0.96; p = 0.010 for trend) and mortality (56% reduction; HR: 0.44; 95% CI: 0.19-1.01; p = 0.037 for trend), with significant mortality reductions also observed for proximal colon cancer (34%; HR: 0.66; 95% CI: 0.44-1.00; p = 0.031 for trend). Conclusion:Our findings suggest focusing on higher quality of macronutrient consumption may be an effective approach to reduce the risk of CRC in the US population.

Multiple primary colorectal cancer (MPCRC) is uncommon but clinically challenging, and differences between synchronous MPCRC (SMPCRC) and metachronous MPCRC (MMPCRC) remain incompletely defined. We compared clinicopathological and surgical features between SMPCRC and MMPCRC and explored prognostic factors for overall survival (OS) in MPCRC. This retrospective cohort study consecutively included patients with pathologically confirmed multiple primary colorectal adenocarcinoma who underwent curative-intent resection at our hospital. SMPCRC was defined as tumors identified within 6 months and MMPCRC as a subsequent primary diagnosed after 6 months. Clinicopathological and perioperative variables were extracted from medical records and pathology reports. Mismatch repair protein expression was assessed as an exploratory pathological variable. OS was analyzed using Kaplan–Meier methods and Cox proportional hazards regression. A total of 165 patients were included (120 SMPCRC and 45 MMPCRC) with follow-up until December 2024. Baseline characteristics were broadly comparable between groups. SMPCRC more frequently underwent laparoscopic surgery (97.5

Overall survival (OS) of colorectal cancer (CRC) patients remains suboptimal, especially in advanced disease. This study aimed to construct and validate explainable machine learning (ML) models using routine blood indices for accurate CRC prognosis across multicenter cohorts. The training cohort included 850 CRC patients (demographic and routine blood data) from Union; validation cohorts were 403 patients (Hefei) and 217 (Shihezi). Seven time-to-event models and SHapley Additive exPlanation (SHAP) (for interpretation) were used. Among the evaluated models, the random survival forest (RSF) algorithm demonstrated superior predictive performance. RSF algorithm demonstrated high discriminatory performance in the Union test cohort with AUCs of 0.768, 0.775, and 0.731 for 1-year, 2-year and 3-year OS, which was sustained in external validation cohorts: Hefei (0.820, 0.805, 0.775) and Shihezi (0.651, 0.706, 0.747). SHAP analysis identified CEA, CA125, age, MPV, CA19-9, INR and monocyte that contributed to the accurate prediction of RSF model. This study provides an innovative strategy for the convenient and accurate prediction of survival outcome of CRC individuals based on routine blood laboratory indices. RSF model helps oncologists to early identify CRC patients with high risk of death and provides a basis for personalized treatment.

Background: Colorectal cancer lung metastases (CRCLM) significantly influence treatment planning and prognosis in colorectal cancer (CRC). This study aimed to develop and validate machine learning-based models to support individualized risk stratification for chest computed tomography (CT) utilization during baseline evaluation by predicting synchronous CRCLM at diagnosis. Methods: Patients with primary CRC diagnosed between 2010 and 2015 were identified from the Surveillance, Epidemiology, and End Results (SEER) database using International Classification of Diseases for Oncology, 3rd edition (ICD-O-3) codes. Synchronous CRCLM was defined by the variable "CS Mets at DX-Lung". Predictors included age, sex, race, primary tumor site, grade, histologic type, tumor stage (T stage), node stage (N stage), tumor size, carcinoembryonic antigen (CEA) level, tumor deposits, and perineural invasion. The cohort was randomly divided into training (70%) and validation (30%) sets. eXtreme gradient boosting (XGB), random forest (RF), decision tree (DT), and logistic regression (LR) models were developed and evaluated mainly by receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA). Model interpretability was assessed using SHapley Additive exPlanation (SHAP). Results: Among 51,553 patients, 1,329 (2.6%) had synchronous CRCLM. In the validation cohort, the area under the curve was 0.81 for XGB, 0.81 for RF, 0.79 for DT, and 0.73 for LR after hyperparameter optimization. Calibration curves indicated high consistency between predictions and observations. DCA revealed substantial clinical utility for all models. SHAP analysis highlighted CEA and N stage as the strongest predictors in the RF model, while CEA and T stage were most influential in the XGB model. Conclusions: Machine learning models, particularly XGB and RF, demonstrated robust performance in predicting synchronous CRCLM. CEA was consistently identified as the most important risk factor, supporting personalized chest CT utilization during initial CRC staging.

Purpose: This study aimed to determine the optimal temporal threshold for distinguishing "early" from "late" liver metastasis in patients who developed liver metastasis after colorectal cancer (CRC) surgery, and to evaluate whether KRAS and BRAFV600E mutations, along with other clinicopathological factors, are associated with the timing of liver metastasis. Methods: This retrospective study utilized clinical and pathological data from patients who developed liver metastasis after radical CRC surgery at two centers from 2019 to 2023. X-tile software was used to identify the optimal temporal threshold. Logistic regression analysis was applied to determine if KRAS/BRAFV600E mutations and other potential factors are independently associated with the time to onset of liver metastasis. Results: X-tile analysis identified 11 months post-surgery as the optimal cutoff for distinguishing early metachronous liver metastasis (EMLM) from late metachronous liver metastasis (LMLM), classifying 114 cases into the EMLM group and 72 into the LMLM group. Comparative analysis indicated statistically significant differences between the two groups in lymphovascular tumor emboli, perineural invasion, and postoperative adjuvant therapy (p < 0.05). Logistic regression analysis revealed that neither KRAS mutation (OR, 1.185; 95% CI: 0.641-2.190; p = 0.587) nor BRAFV600E mutation (OR, 2.836; 95% CI: 0.302-26.642; p = 0.363) was independently associated with the timing of liver metastasis. In contrast, postoperative adjuvant therapy showed a statistical association with a likelihood of LMLM (OR, 0.253; 95% CI: 0.105-0.611; p = 0.002). Conclusions: This study identified 11 months post-CRC surgery as the optimal cutoff for differentiating EMLM versus LMLM. In this cohort, no statistically significant association was observed between KRAS/BRAFV600E mutations and the timing of liver metastasis, whereas postoperative adjuvant therapy was statistically correlated with the likelihood of LMLM. This stratification may guide personalized surveillance strategies and provide valuable insights for future mechanistic investigations into the temporal heterogeneity of post-surgical liver metastasis. However, the interpretation and generalization of the findings require external validation in prospective cohorts.

Background:Colorectal cancer (CRC) remains a leading global malignancy with a rising obesity-attributable burden. Emerging evidence highlights concerning trends in early-onset CRC and marked regional disparities, underscoring the need for comprehensive epidemiological assessments to inform targeted prevention strategies. Methods:Using Global Burden of Disease 2023 data, we analysed high body mass index (BMI)-related CRC deaths and disability-adjusted life years (DALYs) among adults (>40 years) from 1990-2023. We analysed both absolute counts and age-standardised rates, stratifying by sex, age, region, and sociodemographic index (SDI) categories. Decomposition analysis quantified the contributions of ageing, population growth, and epidemiological factors. We used Bayesian age-period-cohort analysis to project future trends. Results:From 1990 to 2023, the global number of high BMI-related CRC deaths increased more than 2-fold, accompanied by a corresponding marked increase in DALYs. Western Europe had the highest burden, while South Asia had the most rapid growth in deaths, as measured by the estimated annual percentage change. Generally, as SDI decreased, the corresponding numbers of deaths and DALYs decreased. Cluster analysis based on the estimated annual percentage changes in age-standardised rates of high BMI-related CRC deaths and DALYs identified distinct regional patterns, with significant decreases in these rates in Western Europe and high-income North America, contrasted by significant increases in South Asia and Central Sub-Saharan Africa. Decomposition analysis indicated that population growth was the primary driver of the rise in mortality, followed by population ageing, and these were partially offset by improvements in epidemiological risk. Projections suggest a continuing increase in the age-standardised death rates for both males and females by 2038. Conclusions:High BMI has become a key driver of CRC mortality and incidence worldwide. Reducing this burden requires efforts in healthy lifestyles, policy reforms, and international scientific cooperation.

One of the most important changes in the transformation of normal cells into tumor cells is metabolism. In order to satisfy the more active proliferation, migration and metastasis of cancer cells, abnormal changes occur in various pathways and molecules involved in metabolism, which eventually lead to metabolic reprogramming of tumor cells. This process involves the uptake of nutrients and changes in major metabolic forms. As an important part of post-transcriptional epigenetics, RNA methylation modifications can regulate RNA processing and metabolism, while dynamically and reversibly influencing the expression of specific molecules, thereby ultimately affecting diverse biological processes and cellular phenotypes. In this review, various types of RNA methylation modifications involved in cancer are summarized. Subsequently, we systematically elucidate the mechanism of RNA modification for metabolic reprogramming in cancer, including glucose, lipid, amino acid and mitochondrial metabolism. Most importantly, we discuss in depth the clinical significance of RNA modification in metabolic targeted therapy and immunotherapy from mechanism to therapeutic application.

BACKGROUND:The critical role of long non-coding RNAs (lncRNAs) in tumor immunity has garnered increasing attention, and immune pathways are key regulatory factors in tumor immune mechanisms. However, current research on the association between lncRNAs and immune pathways remains limited. Therefore, quantifying this association may provide new insights for predicting prognosis and immunotherapy response. METHODS:A method was developed to classify candidate immune-related lncRNAs into complex (CIlncRNAs), moderate (MIlncRNAs), and specific (SIlncRNAs) types based on the breadth of their associations with immune pathway activity. The features of these lncRNA types were then compared from multiple aspects. RESULTS:The genomic variations and correlations with DNA methylation of three types of immune-related lncRNAs were significantly different in multiple cancers. CIlncRNAs were more strongly associated with immune infiltration than MIlncRNAs and SIlncRNAs. Lung adenocarcinoma (LUAD) patients were classified into two subtypes based on CIlncRNAs, with one subtype exhibiting better prognosis and higher immune cell infiltration. High expression of CIlncRNAs in immune cells and their role in CD8+ T cell differentiation in LUAD were validated using single-cell data. Furthermore, a CIlncRNA signature (CIsig) was established for each cancer, and significant differences in overall survival (OS) were observed across risk stratifications in most cancers. Lastly, CIlncRNA was validated in independent datasets for its ability to predict immunotherapy response and enhance the predictive performance of established biomarkers. CONCLUSIONS:Overall, our analysis innovatively explores the breadth of associations between lncRNAs and immune pathway activity. We identified CIlncRNAs as potential novel biomarkers for predicting patient prognosis and immunotherapy response, offering valuable insights for clinical practice.

PURPOSE:Internal iliac and obturator lymph nodes are common sites of metastasis in rectal cancer. This study developed a machine learning (ML) model using clinical data to predict lymph node metastasis and applied the Shapley Additive explanations (SHAP) method for interpretation. MATERIALS AND METHODS:Retrospectively, data from patients with rectal cancer at four Chinese centers-who underwent total mesorectal excision and lateral pelvic lymph node dissection without neoadjuvant therapy-were collected. Two centers provided training/test sets (3:1 ratio) and two centers supplied external validation. Lymph node enlargement was determined by imaging and confirmed by pathology. Five ML models were evaluated by AUC, accuracy, and F1 score. Key features included demographics, tumor stage, tumor-to-anal verge distance, imaging measurements, tumor histological differentiation, preoperative carcinoembryonic antigen, and carbohydrate antigen 19-9. SHAP was used to assess feature importance. RESULTS:Of the 411 cases (174 positives) in the training/test sets and 109 cases (43 positives) in external validation, the random forest (RF) model ranked second in terms of AUC and accuracy in the training set (0.999, 0.995), whereas it achieved the highest AUC and accuracy (0.877 and 0.788) in the test set. In the external validation, the RF model outperformed all other ML models (AUC of 0.899, accuracy of 0.827). Overall, the RF model demonstrates the superior overall performance. According to the SHAP analysis, the most important predictors of internal iliac and obturator lymph node metastasis were, in descending order, the short-axis diameter of enlarged lymph nodes, regional lymph node metastasis, and tumor-to-anal verge distance. At the individual patient level, SHAP force plots provided explanations of the RF model predictions for internal iliac and obturator lymph node metastasis. CONCLUSION:An interpretable ML model was developed that accurately predicts internal iliac and obturator lymph node metastasis using clinical data. SHAP analysis enhances understanding of feature contributions, supporting personalized treatment planning.