Abstract
Background
The role of hilar lymph node metastasis (LNM) in intrahepatic cholangiocarcinoma (ICC) was investigated and a simple postoperative risk-stratification system combining LNM and preoperative anemia was established to predict the outcome of patients with ICC after curative resection.
Methods
A retrospective analysis was conducted on 384 patients who underwent hepatectomy for ICC. Both postoperative clinical outcomes and long-term oncological outcomes were systematically evaluated. Subsequently, predictive models were developed to assess the risks of tumor recurrence, overall survival (OS) and recurrence-free survival (RFS).
Results
The primary cohort demonstrated 65.2% recurrence and 13.1-month median RFS (1-/3-/5-year rates: 51.9%/37.6%/25.0%), with median OS of 31.1 months (1-/3-/5-year rates: 81.8%/63.3%/54.1%). The external validation cohort showed 79.4% recurrence, 11.7-month median RFS (1-/3-/5-year rates: 49.5%/11.8%/5.9%), and 16.2-month median OS (1-/3-/5-year rates: 98.1%/10.4%/6.4%). We developed nomograms for recurrence, OS, and RFS incorporating LNM and preoperative anemia, yielding C-indices of 0.881, 0.802, and 0.758, respectively. Risk-stratified analysis revealed significant prognostic discrimination: high-risk patients demonstrated substantially inferior 5-year OS compared to low-risk patients (19.6% vs. 86.6%, P < 0.001), with median OS of 16.7 months versus not reached. Similarly, median RFS declined from not reached in low-risk patients to 6.1 months in high-risk patients. These results were confirmed in the external validation cohort, where high-risk status corresponded to a median OS of 14.4 months and median RFS of 8.1 months (both P < 0.001).
Conclusions
The integration of LNM status and preoperative anemia emerges as a novel and clinically valuable prognostic indicator, effectively predicting outcomes in patients with ICC following curative resection, thereby providing tailored clinical guidance for postoperative adjuvant therapy selection and intensive surveillance planning.
Graphical Abstract
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12885-026-16604-z.
Keywords: Intrahepatic cholangiocarcinoma, Lymph node metastasis, Anemia, Hepatectomy, Prognosis, Nomogram
Background
Intrahepatic cholangiocarcinoma (ICC) represents the second most prevalent primary hepatic malignancy, arising from the biliary epithelium of intrahepatic ducts [1, 2]. Its characteristically aggressive phenotype, marked by highly aggressive biological behavior and early metastatic propensity, confers a 5-year survival rate below 20% even with contemporary therapies [2, 3]. While R0 resection remains the only potentially curative intervention, refractory recurrence rates exceeding 80% post-resection necessitate molecular stratification paradigms [4, 5]. Such stratification may identify discrete biological subgroups warranting tailored surgical intensification or novel adjuvant approaches, thereby addressing a critical precision oncology imperative.
A clinically actionable staging system should deliver precise survival prognostication and evidence-based adjuvant therapy recommendations. Current ICC prognostication relies predominantly on conventional clinicopathological parameters, notably TNM staging and nodal status, yet their discriminative capacity remains constrained [4, 6]. Consequently, constructing an integrated prognostication framework that synthesizes conventional clinicopathological determinants with validated biological signatures becomes imperative to enable risk-adapted therapeutic strategies and enhance prognostic precision. Such a novel model holds transformative potential for optimizing postoperative management in the precision oncology era.
While accurate assessment of hilar lymph node metastasis (LNM) status is a critical determinant for adjuvant therapy selection in ICC, relying solely on preoperative radiographic evaluation often introduces staging discrepancies [7]. Although imaging modalities form the initial clinical cornerstone, definitive histopathological confirmation remains the gold standard for prognosis. Thus, incorporating pathologically verified nodal status into a comprehensive risk-assessment framework is vital for optimizing subsequent clinical paths. Emerging liquid biopsy strategies propose circulating biomarkers for recurrence risk stratification, yet technical complexity and prohibitive costs impede clinical translation [8, 9]. Contemporary evidence establishes perisurgical immunological dynamics, systemic inflammation, and anemia severity as critical drivers of oncogenesis [10, 11]. In ICC, which is characterized by a prominent desmoplastic stroma, this intense localized and systemic inflammatory cascade directly shapes the hematological profile [12]. Specifically, pro-inflammatory cytokines hypersecreted within the tumor microenvironment, notably interleukin-6 and tumor necrosis factor-alpha, signal the hepatic upregulation of hepcidin [13, 14]. This master iron regulatory hormone traps iron within macrophages and impairs dietary iron absorption, culminating in iron-restricted erythropoiesis, fundamentally known as anemia of chronic disease [15, 16]. Consequently, preoperative anemia in ICC represents far more than a simple hematological deficit. It serves as a functional, liquid biopsy-like surrogate reflecting an aggressive systemic inflammatory state and a heavy tumor burden. These hematological indicators demonstrate proven prognostic utility across malignancies, but biological redundancy limits single-marker fidelity [17, 18]. The combinatorial prognostic value of integrated biomarker panels remains underexplored, a critical knowledge gap in precision risk stratification.
Therefore, we developed and validated a multidimensional postoperative prognostic framework that couples baseline host physiological depletion with definitive surgical and histopathological characteristics. This composite model specifically targets the optimization of postoperative surveillance intensity and adjuvant treatment planning by providing individualized, post-surgical risk quantification.
Methods
Patients and study design
This multi-institutional retrospective analysis encompassed a primary cohort of 384 patients diagnosed with ICC who underwent curative-intent hepatectomy across 6 specialized hepatobiliary centers between January 2006 and December 2023. Eligibility for study inclusion was strictly defined by histopathologically confirmed ICC, preserved hepatic function classified as Child-Pugh grade A or B, and an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1, indicating adequate physiological reserve for major hepatic resection. Comprehensive perioperative data availability within institutional registries constituted a mandatory requirement for enrollment. Conversely, the exclusion criteria eliminated candidates subject to palliative surgical interventions, those presenting with synchronous extrahepatic malignancies, or patients lacking definitive surgical management.
To evaluate model generalizability, an independent external validation cohort was assembled from separate institutions using identical inclusion and exclusion protocols. Clinical and pathological variables were systematically curated from baseline assessments through the immediate postoperative interval, with the study design and reporting rigor fully aligned with the TRIPOD guidelines [19]. During the preparation of this work, the authors used large language models (Deepseek/Gemini) in order to improve the language quality, readability, and narrative flow of the manuscript.
The training cohort comprised 161 patients from the Department of Liver Surgery, Sun Yat-sen University Cancer Center (Guangzhou, Guangdong province, China). For external validation, 223 patients were enrolled from five independent institutions: the Department of General Surgery, The Seventh Affiliated Hospital, Sun Yat-sen University (Shenzhen, Guangdong province; n = 25); Department of General Surgery, The First Affiliated Hospital of Shantou University Medical College (Shantou, Guangdong province; n = 31); Department of Surgery, Shenzhen Guangming District People’s Hospital (Shenzhen, Guangdong province; n = 3); Department of Hepatobiliary Surgery, Affiliated Hospital of Guangdong Medical University (Zhanjiang, Guangdong province; n = 5); and Department of General Surgery, Nanjing Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University (Nanjing, Jiangsu province; n = 159).
Evaluation and follow-up
A standardized preoperative workup was mandated for all candidates to assess surgical resectability and physiological reserve. This assessment encompassed comprehensive hepatic, renal, and coagulation profiles alongside serological tumor marker quantification. Anatomical staging was thoroughly established via multimodality cross-sectional imaging, primarily contrast-enhanced abdominal computed tomography (CT) and/or contrast-enhanced magnetic resonance imaging (MRI), paired with non-contrast chest CT for pulmonary metastatic screening. To ensure maximum diagnostic objectivity and reproducibility, all preoperative baseline imaging data were independently consensus-reviewed by two senior abdominal radiologists (each possessing over 10 years of dedicated subspecialty experience) who remained strictly blinded to the subsequent post-resection histopathological diagnoses. Inter-observer discrepancies regarding nodal status were systematically resolved through a joint reevaluation to reach a final consensus. Radiological LNM was strictly defined based on standardized criteria, requiring the presence of at least one regional lymph node demonstrating a short-axis diameter of 10 mm or greater, a distinctly rounded contour, irregular or ill-defined borders, or a prominent heterogeneous/ring-like contrast enhancement pattern on portal venous phase imaging. In our cohort, the concordance rate between preoperative radiological suspicion of LNM and pathologically confirmed LNM was 78.7%. Physiological operability was further stratified using ECOG performance scoring and cardiopulmonary functional testing. Postoperative surveillance adhered to a strict institutional protocol involving outpatient reviews every two months for the initial two years, extending to three-to-six-month intervals thereafter. These visits entailed serial serum tumor marker analysis, liver function panels, and cross-sectional abdominal imaging. Tumor recurrence was diagnosed based on the appearance of new intrahepatic lesions exhibiting characteristic ICC features on two independent imaging modalities or the identification of de novo extrahepatic metastases not present at initial staging. Survival endpoints included overall survival (OS), calculated from the date of resection to death from any cause or last follow-up, and recurrence-free survival (RFS), defined as the interval from surgery to the first radiological or pathological evidence of relapse.
Statistical analysis
Continuous variables were expressed as medians with interquartile ranges (IQR) and compared using the Mann-Whitney U test or independent t-test based on distribution normality, while categorical data were summarized as frequencies and proportions, subject to Chi-square or Kruskal-Wallis analysis as indicated. To mitigate potential bias from missing baseline values, multiple imputation was performed using fully conditional specification, applying discriminant functions for categorical inputs and regression modeling for continuous parameters. Survival probabilities were estimated via the Kaplan-Meier method and compared using the log-rank test. A stepwise variable selection strategy was adopted where factors achieving a screening threshold of P < 0.2 in univariate analysis were incorporated into multivariable Cox proportional hazards models for OS and RFS, or logistic regression models for binary recurrence outcomes. Independent prognostic factors derived from these multivariable analyses served as the foundation for constructing the prognostic nomogram. Model performance was rigorously quantified using the concordance index (C-index) for discrimination and calibration plots assessed via Hosmer-Lemeshow goodness-of-fit testing. Internal validation was achieved through 1000 bootstrap resamples to minimize overfitting. Statistical significance was defined as a two-tailed P value < 0.05. Data analyses were executed using IBM SPSS Statistics (Version 26.0; IBM Corp., Armonk, NY, USA) and R software (Version 4.4.3; R Foundation for Statistical Computing, Vienna, Austria), leveraging the rms package for nomogram development.
Results
Patient characteristics
The final analytic cohort consisted of 384 patients undergoing curative-intent resection, characterized by a male predominance (60.2%) and a median age of 58 years (range: 26–87). Long-term surveillance, spanning a median follow-up of 31.1 months (IQR: 20.9–46.3), revealed a substantial recurrence burden, with 73.4% of the population experiencing disease recurrence. Consequently, the median OS and RFS were 18.3 months (95% CI: 16.1–20.5) and 12.5 months (95% CI: 11.4–13.7), respectively. The OS rates at 1, 3, and 5 years were 91.2%, 33.9%, and 28.1%, while the corresponding RFS rates were 50.7%, 24.5%, and 16.8%. Four patients (2.5%) in the primary cohort and 27 patients (12.1%) in the validation cohort underwent concurrent extrahepatic bile duct resection and reconstruction. Baseline clinicopathologic demographics are comprehensively detailed in Table 1, and the study inclusion flowchart is illustrated in Fig. 1.
Table 1.
Participant Characteristics
| Characteristics | Training cohort (n = 161) |
External validation cohort (n = 223) |
P-value |
|---|---|---|---|
| Age, mean ± SD | 56.77 ± 9.5944 | 59.928 ± 10.674 | 0.003 |
| Gender, n (%) | 0.808 | ||
| Male | 98 (60.9%) | 133 (59.6%) | |
| Female | 63 (39.1%) | 90 (40.4%) | |
| Child-Pugh grade, n (%) | 0.342 | ||
| A | 159 (98.8%) | 223 (100%) | |
| B | 2 (1.2%) | 0 (0) | |
| ECOG performance status, n (%) | 0.022 | ||
| 0 | 133 (82.6%) | 162 (72.6%) | |
| 1 | 28 (17.4%) | 61 (27.4%) | |
| Etiology (HBV/HCV infection), n (%) | < 0.001 | ||
| Yes | 112 (69.6%) | 113 (50.7%) | |
| No | 49 (30.4%) | 110 (49.3%) | |
| Biliary tract diseases history, n (%) | 0.511 | ||
| Present | 33 (20.5%) | 52 (23.3%) | |
| Absent | 128 (79.5%) | 171 (76.7%) | |
| Tumor size, median (IQR) | 5.1 (3.8, 7.3) | 5 (3.6, 7) | 0.625 |
| Tumor number, n (%) | 0.251 | ||
| Solitary | 119 (73.9%) | 176 (78.9%) | |
| Multiple | 42 (26.1%) | 47 (21.1%) | |
| Vascular invasion, n (%) | 0.057 | ||
| Present | 28 (17.4%) | 57 (25.6%) | |
| Absent | 133 (82.6%) | 166 (74.4%) | |
| MVI, n (%) | 0.872 | ||
| Present | 45 (28%) | 64 (28.7%) | |
| Absent | 116 (72%) | 159 (71.3%) | |
| Intraoperative blood transfusion, n (%) | < 0.001 | ||
| Yes | 15 (9.3%) | 73 (32.7%) | |
| No | 146 (90.7%) | 150 (67.3%) | |
| Resection extent of hepatic segments, n (%) | 0.230 | ||
| 1 | 26 (16.1%) | 34 (8.9%) | |
| 2 | 51 (31.7%) | 51 (22.93.3%) | |
| 3 | 52 (32.3%) | 85 (22.1%) | |
| ≥ 4 | 32 (19.9%) | 53 (13.8%) | |
| Extrahepatic bile duct resection and reconstruction | 4 (2.5%) | 27 (12.1%) | 0.001 |
| AJCC stage, n (%) | |||
| Ⅰ | 37 (23%) | 98 (43.9%) | |
| Ⅱ | 21 (13%) | 51 (22.9%) | |
| Ⅲ | 101 (62.7%) | 71 (31.8%) | |
| Ⅳ | 2 (1.2%) | 3 (1.3%) | |
| Preoperative hilar lymph node metastasis, n (%) | < 0.001 | ||
| Present | 35 (21.7%) | 87 (39%) | |
| Absent | 126 (78.3%) | 136 (61%) | |
| Tumor differentiation, n (%) | < 0.001 | ||
| Well | 7 (4.3%) | 14 (6.3%) | |
| Moderate | 42 (26.1%) | 98 (43.9%) | |
| Moderate-Poor | 88 (54.7%) | 64 (28.7%) | |
| Poor | 24 (14.9%) | 47 (21.1%) | |
| Perineural invasion, n (%) | < 0.001 | ||
| Present | 31 (19.3%) | 124 (55.6%) | |
| Absent | 130 (80.7%) | 99 (44.4%) | |
| Preoperative anemia (Hgb < 120), n (%) | < 0.001 | ||
| Present | 14 (8.7%) | 61 (27.4%) | |
| Absent | 147 (91.3%) | 162 (72.6%) | |
| CRP, n (%) | < 0.001 | ||
| < 10 mg/L | 137 (85.1%) | 141 (63.2%) | |
| ≥ 10 mg/L | 24 (14.9%) | 82 (36.8%) | |
| CA19-9, n (%) | 0.030 | ||
| ≤ 35 U/ml | 91 (56.5%) | 101 (45.3%) | |
| > 35 U/ml | 70 (43.5%) | 122 (54.7%) | |
| CEA, n (%) | 0.023 | ||
| ≤ 5 ng/ml | 126 (32.8%) | 151 (39.3%) | |
| > 5 ng/ml | 35 (9.1%) | 72 (18.8%) | |
| TBIL, n (%) | |||
| ≤ 20.5 µmol/L | 149 (92.5%) | ||
| > 20.5 µmol/L | 12 (7.5%) | 46 (20.6%) | |
| ALB, n (%) | 0.004 | ||
| < 35 g/L | 159 (41.4%) | 206 (53.6%) | |
| ≥ 35 g/L | 2 (0.5%) | 17 (4.4%) | |
| NLR, n (%) | 0.011 | ||
| < 5 | 150 (39.1%) | 189 (49.2%) | |
| ≥ 5 | 11 (2.9%) | 34 (8.9%) | |
| PLR, n (%) | < 0.001 | ||
| ≤ 78.24 | 15 (3.9%) | 97 (25.3%) | |
| > 78.24 | 146 (38%) | 126 (32.8%) | |
| TBS, n (%) | 0.782 | ||
| Low | 81 (21.1%) | 109 (28.4%) | |
| High | 80 (20.8%) | 114 (29.7%) | |
Abbreviations: AJCC the American joint committee on cancer, ALB Albumin, HBV Hepatitis B virus, HCV Hepatitis B virus, CA19-9 Carbohydrate antigen 19 − 9, CEA Carcinoembryonic antigen, CRP C-reactive protein, ECOGPS Eastern cooperative oncology group performance score, HGB hemoglobin, MVI Microvascular invasion, NLR Neutrophil-to-lymphocyte ratio, PLR Platelet-to-lymphocyte ratio, SD Standard deviation, TBIL Total bilirubin, TBS Tumor burden score
Fig. 1.
Flowchart of the patient enrollment process for the primary and validation cohorts. ICC, intrahepatic cholangiocarcinoma. ICC, intrahepatic cholangiocarcinoma
Regression analysis in the training cohort
Multivariable analyses stratified distinct prognostic profiles for recurrence risk and long-term survival. In the logistic regression model targeting binary recurrence outcomes, preoperative anemia (odds ratio (OR) 15.21, 95% CI: 1.57–147.51) and radiologic evidence of hilar LNM (OR 9.62, 95% CI: 1.17–79.28) emerged as the most potent independent predictors. Other significant factors contributing to recurrence included male sex, poor tumor differentiation, presence of microvascular invasion (MVI), intraoperative blood transfusion, and elevated preoperative C-reactive protein (CRP) levels.
Regarding survival endpoints, Cox proportional hazards modeling identified a convergence of risk factors alongside distinct outcome-specific determinants. For OS, preoperative anemia (hazard ratio (HR) 40.81) and hilar LNM (HR 2.27) remained strongly prognostic. Notably, the extent of hepatectomy exhibited a stepwise association with mortality risk, with HRs increasing from 1.96 for 2-segment resections to 5.00 for resections involving ≥ 4 segments compared to single-segment controls. Additional independent predictors of diminished OS included higher tumor burden score (TBS), postoperative CRP elevation, perineural invasion, poor differentiation, and intraoperative blood loss. Parallel analysis of RFS corroborated the prognostic significance of hilar LNM (HR 1.97), preoperative anemia (HR 7.43), and tumor differentiation. Uniquely, the RFS model highlighted the impact of systemic inflammatory and tumor markers, identifying preoperative platelet-to-lymphocyte ratio (PLR) (HR 2.64), neutrophil-to-lymphocyte ratio (NLR) (HR 3.21), and carcinoembryonic antigen (CEA) levels as specific drivers of disease-free interval reduction, alongside TBS and intraoperative blood loss. The predictors are listed in Tables 2 and 3, and Table 4, respectively.
Table 2.
Univariate and multivariate logistic regression analysis of prognostic factors for recurrence probability in the training cohort
| Characteristics | Univariate analysis | Multivariate analysis | ||
|---|---|---|---|---|
| Odds Ratio (95% CI) | P value | Odds Ratio (95% CI) | P-value | |
| Gender, Male vs. Female | 2.243 (1.154–4.360) | 0.017 | 3.841 (1.070-13.787) | 0.039 |
| Tumor size, ≥5 cm vs. <5 cm | 2.615 (1.343–5.094) | 0.005 | ||
| Tumor number, Multiple vs. Solitary | 4.348 (1.702–11.106) | 0.002 | ||
| Vascular invasion, Present vs. Absent | 3.852 (1.264–11.739) | 0.018 | ||
| LNM, Present vs. Absent | 7.744 (2.252–26.632) | 0.001 | 9.621 (1.168–79.275) | 0.035 |
| Tumor differentiation | ||||
| Moderate vs. Well | 1.125 (1.251–5.352) | 0.003 | 1.419 (1.075–6.751) | 0.011 |
| Moderate-Poor vs. Well | 6.823 (2.157–10.894) | 0.001 | 8.562 (1.508–14.285) | 0.007 |
| Poor vs. Well | 11.472 (3.127–18.552) | < 0.001 | 12.069 (1.629–23.680) | < 0.001 |
| MVI, Present vs. Absent | 2.725 (1.201–6.181) | 0.016 | 4.785 (1.148–19.947) | 0.032 |
| Local invasion, Present vs. Absent | 3.176 (0.884–11.413) | 0.077 | ||
| Intraoperative blood loss | ||||
| 200-400 ml vs. ≤200 ml | 3.638 (1.686–7.848) | 0.001 | ||
| >400 ml vs. ≤200 ml | 4.941 (1.645–14.839) | 0.004 | ||
| Intraoperative blood transfusion, Yes vs. No | 8.462 (1.083–66.136) | 0.042 | 12.962 (1.093-340.924) | 0.025 |
| Preoperative anemia (Hgb < 120), Present vs. Absent | 3.830 (1.216–12.057) | 0.022 | 15.210 (1.568-147.508) | 0.019 |
| Lymph node dissection, Yes vs. No | 2.481 (1.194–5.152) | 0.015 | ||
| Resection extent of hepatic segments | ||||
| 2 vs. 1 | 1.048 (0.402–2.729) | 0.024 | ||
| 3 vs. 1 | 1.385 (0.528–3.637) | 0.058 | ||
| ≥ 4 vs. 1 | 3.178 (0.976–10.346) | 0.005 | ||
| Margin, ≥1 cm vs. <1 cm | 1.859 (0.935–3.697) | 0.077 | ||
| CRP, ≥ 10 mg/L vs. <10 mg/L | 6.857 (2.957–15.899) | < 0.001 | 3.449 (1.084–13.460) | 0.045 |
| CA19-9, > 35 U/ml vs. ≤35 U/ml | 2.647 (1.321–5.302) | 0.006 | ||
| CEA, > 5 ng/ml vs. ≤5 ng/ml | 3.180 (1.231–8.212) | 0.017 | ||
| ALP, ≥ 135 U/L vs. <135 U/L | 3.187 (1.147–8.861) | 0.026 | ||
| γ-GT, ≥ 45 U/L vs. <45 U/L | 2.705 (1.386–5.281) | 0.004 | ||
| PLR, > 78.24 vs. ≤78.24 | 3.160 (1.063–9.395) | 0.039 | ||
| GPR, > 0.262 vs. ≤0.262 | 2.495 (1.277–4.877) | 0.007 | ||
| TBS, High vs. Low | 2.700 (1.374–5.305) | 0.004 | ||
Abbreviations: ALP, Alkaline phosphatase, CA19-9 Carbohydrate antigen 19 − 9, CEA Carcinoembryonic antigen, CI Confidence interval, CRP C-reactive protein, γ-GT Gamma-glutamyl transferase, GPR Gamma-glutamyl transferase-to-platelet ratio, HGB Hemoglobin, LNM Lymph node metastasis, MVI Microvascular invasion, PLR Platelet-to-lymphocyte ratio, TBS Tumor burden score
Table 3.
Univariate and multivariate Cox regression analysis of prognostic factors for overall survival in the training cohort
| Characteristics | Univariate analysis | Multivariate analysis | ||
|---|---|---|---|---|
| Hazard ratio (95% CI) | P value | Hazard ratio (95% CI) | P-value | |
| Tumor size, ≥5 cm vs. <5 cm | 2.360 (1.517–3.671) | < 0.001 | ||
| Tumor number, Multiple vs. Solitary | 2.238 (1.348–3.714) | 0.002 | ||
| Vascular invasion, Present vs. Absent | 2.569 (1.500-4.397) | 0.001 | ||
| MVI, Present vs. Absent | 1.651 (0.995–2.740) | 0.052 | ||
| Capsular invasion, Present vs. Absent | 2.525 (1.204–5.295) | 0.014 | ||
| Margin, ≥1 cm vs. <1 cm | 1.303 (0.799–2.127) | 0.128 | ||
| LNM, Present vs. Absent | 2.952 (1.791–4.864) | < 0.001 | 2.401 (1.855–6.742) | 0.006 |
| Perineural invasion, Present vs. Absent | 2.453 (1.432–4.203) | 0.001 | 3.630 (1.385–9.512) | 0.009 |
| Local invasion, Present vs. Absent | 2.071 (1.079–3.976) | 0.029 | ||
| Tumor differentiation | ||||
| Moderate vs. Well | 1.475 (1.059–2.054) | 0.022 | 1.146 (1.002–2.982) | 0.047 |
| Moderate-Poor vs. Well | 2.963 (1.258–4.740) | 0.019 | 3.305 (1.365–5.603) | 0.008 |
| Poor vs. Well | 5.500 (1.739–10.156) | 0.002 | 6.312 (1.845–9.856) | 0.001 |
| Resection extent of hepatic segments | ||||
| 2 vs. 1 | 1.407 (1.092–1.814) | 0.036 | 1.958 (1.178–5.338) | 0.018 |
| 3 vs. 1 | 2.004 (1.895–4.490) | 0.019 | 4.356 (1.476–12.856) | 0.008 |
| ≥ 4 vs. 1 | 2.190 (1.940–5.105) | 0.006 | 4.997 (1.244–20.075) | 0.023 |
| Intraoperative blood loss | ||||
| 200-400 ml vs. ≤200 ml | 1.135 (1.206–2.481) | 0.001 | 1.998 (1.995–2.301) | 0.037 |
| >400 ml vs. ≤200 ml | 1.938 (1.338–2.807) | < 0.001 | 11.125 (3.326–37.214) | < 0.001 |
| Intraoperative blood transfusion, Yes vs. No | 3.167 (1.668–6.011) | < 0.001 | ||
| Lymph node dissection, Yes vs. No | 2.285 (1.403–3.721) | 0.001 | ||
| Preoperative anemia (Hgb < 120), Present vs. Absent | 2.711 (1.038–8.768) | 0.036 | 40.810 (5.656-294.477) | < 0.001 |
| TBS, High vs. Low | 2.621 (1.563–4.397) | < 0.001 | 3.440 (1.958–12.349) | 0.008 |
| LNM number, Multiple vs. Solitary | 1.767 (1.462–2.135) | < 0.001 | ||
| ECOG score, 1 vs. 0 | 2.535 (1.481–4.339) | 0.001 | ||
| ALBI grade, Grade 2 vs. Grade 1 | 2.793 (1.416–5.510) | 0.003 | ||
| CA19-9, > 35 U/ml vs. ≤35 U/ml | 1.934 (1.185–3.156) | 0.008 | ||
| CEA, > 5 ng/ml vs. ≤5 ng/ml | 2.395 (1.419–4.041) | 0.001 | ||
| WBC, > 10 × 10^9 vs. ≤10 × 10^9 | 3.860 (2.036–7.317) | < 0.001 | ||
| CRP, ≥ 10 mg/L vs. <10 mg/L | 2.957 (1.794–4.873) | < 0.001 | ||
| ALP, ≥ 135 U/L vs. <135 U/L | 2.579 (1.530–4.346) | < 0.001 | ||
| NLR, ≥ 5 vs. <5 | 2.124 (0.910–4.960) | 0.082 | ||
| LMR, ≥ 3 vs. <3 | 2.959 (1.670–5.245) | < 0.001 | ||
| PLR, > 78.24 vs. ≤78.24 | 8.418 (1.165–60.814) | 0.035 | ||
| SIRI, ≥ 1.26 vs. <1.26 | 2.092 (1.270–3.444) | 0.004 | ||
| PNI, < 45 vs. ≥45 | 3.803 (1.620–8.923) | 0.002 | ||
| GPR, > 0.262 vs. ≤0.262 | 1.747 (1.051–2.903) | 0.031 | ||
| Postoperative ALBI grade, Grade 2 vs. Grade 1 | 3.250 (1.860–5.678) | < 0.001 | ||
| Postoperative CA19-9, > 35 U/ml vs. ≤35 U/ml | 1.876 (1.135-3.100) | 0.014 | ||
| Postoperative CRP, ≥ 10 mg/L vs. <10 mg/L | 2.151 (1.304–3.549) | 0.003 | 2.896 (1.381–6.074) | 0.005 |
| Postoperative ALP, ≥ 135 U/L vs. <135 U/L | 2.308 (1.322–4.027) | 0.003 | ||
| Postoperative NLR, ≥ 5 vs. <5 | 1.483 (0.642–4.764) | 0.058 | ||
| Postoperative LMR, ≥ 3 vs. <3 | 2.380 (1.258–4.504) | 0.008 | ||
| Postoperative SIRI, ≥ 1.26 vs. <1.26 | 2.139 (1.170–3.910) | 0.013 | ||
| Postoperative PNI, < 45 vs. ≥45 | 2.433 (1.190–4.975) | 0.015 | ||
| Postoperative GPR, > 0.262 vs. ≤0.262 | 1.711 (1.021–2.868) | 0.041 | ||
Abbreviations: ALBI albumin-bilirubin, ALPAlkaline phosphatase, CA19-9 Carbohydrate antigen 19 − 9, CEA, Carcinoembryonic antigen, CI Confidence interval, CRP C-reactive protein, GPR Gamma-glutamyl transferase-to-platelet ratio, ECOGPS Eastern cooperative oncology group performance score, HGB Hemoglobin, LMR Lymphocyte-to-monocyte ratio, LNM Lymph node metastasis, MVI Microvascular invasion, NLR Neutrophil-to-lymphocyte ratio PLR Platelet-to-lymphocyte ratio, PNI Prognostic nutritional index, SIRI Systemic inflammation response index, TBS Tumor burden score, WBC white blood cell
Table 4.
Univariate and multivariate Cox regression analysis of prognostic factors for recurrence-free survival in the training cohort
| Characteristics | Univariate analysis | Multivariate analysis | ||
|---|---|---|---|---|
| Hazard ratio (95% CI) | P value | Hazard ratio (95% CI) | P value | |
| Gender, Male vs. Female | 1.618 (1.074–2.436) | 0.021 | ||
| Tumor size, ≥5 cm vs. <5 cm | 2.057 (1.379–3.068) | < 0.001 | ||
| Tumor number, Multiple vs. Solitary | 2.270 (1.503–3.427) | < 0.001 | ||
| Vascular invasion, Present vs. Absent | 2.434 (1.531–3.870) | < 0.001 | ||
| Capsular invasion, Present vs. Absent | 1.693 (1.028–2.788) | 0.038 | ||
| MVI, Present vs. Absent | 1.671 (1.115–2.505) | 0.013 | ||
| LNM, Present vs. Absent | 2.527 (1.660–3.848) | < 0.001 | 1.965 (1.086–4.485) | 0.018 |
| Tumor differentiation | ||||
| Moderate vs. Well | 5.336 (1.716–39.739) | 0.007 | 5.015 (4.803–21.106) | 0.032 |
| Moderate-Poor vs. Well | 9.905 (1.370-71.628) | 0.023 | 9.843 (2.953–60.157) | 0.002 |
| Poor vs. Well | 11.675 (1.548–88.032) | 0.017 | 13.845 (3.737–80.142) | < 0.001 |
| Abdominal adhesions, Present vs. Absent | 1.517 (1.010–2.278) | 0.045 | ||
| Metastatic seeding, Present vs. Absent | 3.276 (1.030-10.419) | 0.044 | ||
| Lymph node dissection, Yes vs. No | 1.751 (1.186–2.585) | 0.005 | ||
| Local invasion, Present vs. Absent | 2.028 (1.184–3.472) | 0.010 | ||
| Resection extent of hepatic segments | ||||
| 2 vs. 1 | 1.001 (0.538–1.862) | 0.08 | ||
| 3 vs. 1 | 1.375 (1.047–2.528) | 0.036 | ||
| ≥ 4 vs. 1 | 1.898 (1.000-3.601) | 0.05 | ||
| Margin, ≥1 cm vs. <1 cm | 1.410 (0.959–2.073) | 0.080 | ||
| Intraoperative blood loss | ||||
| 200-400 ml vs. ≤200 ml | 2.479 (1.452–4.231) | 0.001 | 2.583 (2.410–4.952) | 0.044 |
| >400 ml vs. ≤200 ml | 2.663 (1.403–5.058) | 0.003 | 3.136 (1.502–6.741) | 0.005 |
| Intraoperative blood transfusion, Yes vs. No | 2.823 (1.585–5.028) | < 0.001 | ||
| TBS, High vs. Low | 2.186 (1.476–3.238) | < 0.001 | 1.484 (1.156–3.582) | 0.038 |
| ECOG score, 1 vs. 0 | 1.824 (1.146–2.904) | 0.011 | ||
| ALBI grade, Grade 2 vs. Grade 1 | 2.029 (1.107–3.719) | 0.022 | ||
| Preoperative anemia (Hgb < 120), Present vs. Absent | 2.794 (1.129–6.912) | 0.026 | 7.426 (2.005–27.506) | 0.003 |
| CA19-9, > 35 U/ml vs. ≤35 U/ml | 1.858 (1.264–2.730) | 0.002 | ||
| CEA, > 5 ng/ml vs. ≤5 ng/ml | 2.077 (1.350–3.196) | 0.001 | 1.961 (1.075–3.580) | 0.028 |
| WBC, > 10 × 10^9 vs. ≤10 × 10^9 | 2.729 (1.486–5.009) | 0.001 | ||
| CRP, ≥ 10 mg/L vs. <10 mg/L | 2.787 (1.893–4.104) | < 0.001 | ||
| ALP, ≥ 135 U/L vs. <135 U/L | 1.802 (1.144–2.836) | 0.011 | ||
| γ-GT, > 45 U/L vs. ≤45 U/L | 1.797 (1.188–2.717) | 0.006 | ||
| LDH, > 250 U/L vs. ≤250 U/L | 2.159 (1.152–4.046) | 0.016 | ||
| INR, > 1.2 vs. ≤1.2 | 3.574 (1.299–9.830) | 0.014 | ||
| NLR, ≥ 5 vs. <5 | 2.663 (1.336–5.305) | 0.005 | 3.207 (1.032–9.964) | 0.044 |
| LMR, ≥ 3 vs. <3 | 2.211 (1.341–3.645) | 0.002 | ||
| PLR, > 78.24 vs. ≤78.24 | 2.345 (1.027–5.350) | 0.043 | 2.641 (1.559–7.301) | 0.016 |
| SIRI, ≥ 1.26 vs. <1.26 | 1.747 (1.166–2.618) | 0.007 | ||
| PNI, < 45 vs. ≥45 | 2.630 (1.141–6.060) | 0.023 | ||
| GPR, > 0.262 vs. ≤0.262 | 1.645 (1.115–2.427) | 0.012 | ||
| Postoperative ALBI, Grade 2 vs. Grade 1 | 2.481 (1.531–4.023) | < 0.001 | ||
| Postoperative CA19-9, > 35 U/ml vs. ≤35 U/ml | 1.578 (1.047–2.378) | 0.029 | ||
| postoperative ALP, ≥ 135 U/L vs. <135 U/L | 1.717 (1.070–2.755) | 0.025 | ||
| Postoperative CRP, ≥ 10 mg/L vs. <10 mg/L | 1.651 (1.121–2.430) | 0.011 | ||
| Postoperative SII, > 330 vs. ≤330 | 1.572 (1.072–2.306) | 0.021 | ||
| Postoperative SIRI, ≥ 1.26 vs. <1.26 | 1.934 (1.156–3.234) | 0.012 | ||
| Postoperative GPR, > 0.262 vs. ≤0.262 | 1.516 (1.022–2.247) | 0.039 | ||
Abbreviations: ALBI Albumin-bilirubin, ALP Alkaline phosphatase, CA19-9 Carbohydrate antigen 19-9, CEA Carcinoembryonic antigen, CI Confidence interval, CRP C-reactive protein, γ-GT Gamma-glutamyl transferase, GPR Gamma-glutamyl transferase-to-platelet ratio, ECOGPS eastern cooperative oncology group performance score, HGB Hemoglobin, INR International normalized ratio, LMR Lymphocyte-to-monocyte ratio, LDH Lactate dehydrogenase, LNM Lymph node metastasis, MVI Microvascular invasion, NLR Neutrophil-to-lymphocyte ratio, PLR Platelet-to-lymphocyte ratio, PNI Prognostic nutritional index, SII Systemic immune inflammatory index, SIRI Systemic inflammation response index, TBS Tumor burden score, WBC white blood cell
Development of Prognostic Nomograms
To translate the identified independent risk factors into clinically actionable tools, we constructed individualized nomograms for stratifying the probability of tumor recurrence (Fig. 2A), OS (Fig. 3A), and RFS (Fig. 4A) in patients undergoing radical resection. Across these models, radiological evidence of LNM, preoperative anemia, and tumor differentiation grade emerged as consistent core determinants. The recurrence-specific nomogram further incorporated male sex, MVI, intraoperative blood transfusion requirement, and elevated preoperative CRP levels. Conversely, the OS prediction model extended beyond these baseline pathological features to include surgical and tumor burden metrics, specifically the extent of hepatectomy, intraoperative blood loss, TBS, perineural invasion, and postoperative CRP elevation. Similarly, the RFS nomogram synthesized the core predictors with inflammatory and tumor markers, encompassing preoperative NLR ≥ 5, PLR, CEA levels, TBS, and intraoperative blood transfusion volume.
Fig. 2.
Development and performance of the nomogram for predicting tumor recurrence in ICC patients following liver resection. A The nomogram was constructed using clinicopathological factors to estimate the probability of tumor recurrence. B, C Calibration plots demonstrate the agreement between predicted and actual recurrence probabilities in both the primary cohort and the external validation cohort. D, E ROC curves illustrate the discriminative ability of the nomogram for predicting recurrence in the primary and external validation cohorts, respectively. F, G Comparative ROC analyses depict the predictive performance of the nomogram alongside other established models in the primary and external validation cohorts, highlighting its relative discriminatory power. ICC, intrahepatic cholangiocarcinoma; ROC, receiver operating characteristic
Fig. 3.
Development and performance of the nomogram for predicting OS in ICC patients following liver resection. A The nomogram was developed based on clinicopathological features to predict OS rates. B, C Calibration plots assess the agreement between predicted and observed 1-, 3-, and 5-year OS probabilities in the primary cohort and external validation cohort. D, E Time-dependent ROC curves evaluate the discriminative ability of the nomogram and other prognostic systems for OS prediction in the primary and external validation cohorts, respectively. F, G Kaplan-Meier survival analyses compare OS outcomes between high-, intermediate-, and low-risk groups stratified by the nomogram in the primary and external validation cohorts, demonstrating its clinical utility for risk stratification. ICC, intrahepatic cholangiocarcinoma; OS, overall survival; ROC, receiver operating characteristic
Fig. 4.
Development and performance of the nomogram for predicting RFS in ICC patients following liver resection. A The nomogram was developed based on clinicopathological features to predict RFS rates. B, C Calibration plots demonstrate the nomogram's predictive accuracy by comparing predicted versus actual RFS outcomes at 1-, 3-, and 5-years in both the primary and external validation cohorts. D, E Time-dependent ROC analyses quantify the discriminative performance of the nomogram and comparator prognostic models across different follow-up durations in the primary and validation cohorts, respectively. F, G Risk-stratified Kaplan-Meier survival curves illustrate significant differences in RFS outcomes among patients categorized into high-, intermediate-, and low-risk groups by the nomogram in both cohorts, thereby validating its clinical utility for personalized prognosis assessment.ICC, intrahepatic cholangiocarcinoma; RFS, recurrence-free survival; ROC, receiver operating characteristic
Discriminative Performance and Calibration for Tumor Recurrence
The proposed nomogram exhibited robust predictive fidelity for tumor recurrence following comprehensive internal validation via 1000 bootstrap resamples. In the primary cohort, the model achieved a C-index of 0.881, indicating high precision, a performance that was well-sustained in the external validation cohort with a C-index of 0.811. Calibration analysis revealed an optimal correspondence between predicted probabilities and observed recurrence rates, evidenced by Brier scores of 0.125 and 0.119 for the primary and external cohorts, respectively (Fig. 2B and C). This reliability was further corroborated by receiver operating characteristic (ROC) analysis, where the area under the curve (AUC) reached 0.881 (95% CI: 0.827–0.934) internally and 0.811 (95% CI: 0.744–0.878) externally (Fig. 2D and E).
When juxtaposed with established clinical staging systems, the nomogram demonstrated superior discriminatory capacity. Within the primary cohort, the model significantly outperformed the 8th edition American Joint Committee on Cancer (AJCC) staging (AUC 0.602), as well as the Liver Cancer Study Group of Japan (LCSGJ), Japanese Society of Hepato-Biliary-Pancreatic Surgery (JSHBPS), Fudan, Zhou, and Metastatic-Extrahepatic-Grade-Node-Albumin (MEGNA) scoring systems, which yielded AUCs ranging from 0.658 to 0.754. Notably, the composite nomogram provided greater prognostic accuracy than its individual key components, surpassing both radiologically confirmed hilar LNM (AUC 0.626) and preoperative anemia (AUC 0.557) in isolation (Fig. 2F). This pattern of superiority was consistent in the external validation cohort, where the nomogram maintained a dominant AUC advantage over all comparator systems (Fig. 2G).
Performance of the nomogram in predicting OS and RFS
The model’s prognostic utility extended effectively to long-term survival endpoints. Internal bootstrap validation yielded C-indices of 0.802 (95% CI: 0.750–0.854) for OS and 0.758 (95% CI: 0.712–0.804) for RFS in the primary cohort, with comparable efficacy observed in external validation (OS: 0.782; RFS: 0.735). Temporal calibration curves confirmed excellent agreement between predicted and observed survival probabilities at 1, 3, and 5 years across both cohorts (Figs. 3B and C and 4B and C). Time-dependent ROC analysis further highlighted that the nomogram consistently exceeded the predictive accuracy of all standard staging systems and singular biomarkers across all time points, with AUCs for 1-, 3-, and 5-year OS ranging from 0.798 to 0.832 in the primary cohort, significantly overshadowing the comparative range of 0.314–0.769 (Figs. 3D and E and 4D and E).
Applying nomogram-derived risk scores allowed for the distinct stratification of patients into low-, intermediate-, and high-risk categories, revealing profound disparities in survival outcomes. Kaplan-Meier analysis demonstrated that high-risk patients in the primary cohort faced drastically inferior outcomes compared to their low-risk counterparts, with 5-year OS rates dropping from 86.6% to 19.6% (P < 0.001) and median OS plummeting from “not reached” to just 16.7 months. A similar result was evident for RFS, where the high-risk group exhibited a median recurrence-free interval of only 6.1 months versus “not reached” in the low-risk group (Figs. 3F and 4F). These risk-dependent survival gradients were strictly replicated in the external validation cohort, where the high-risk designation continued to define a subset with significantly truncated median OS (14.4 months) and RFS (8.1 months) (Figs. 3G and 4G).
In sharp contrast to the granular risk separation achieved by the nomogram, conventional staging systems displayed limited prognostic discrimination. Although most systems achieved statistical significance, they failed to provide clinically meaningful separation between risk groups. Specifically, the AJCC 8th edition and MEGNA score exhibited overlapping survival curves for OS, while the LCSGJ and MEGNA systems showed curve convergence for RFS, indicating a lack of discriminatory power for specific patient subsets (Figures S1–S4).
Discussion
In this multicenter study, we developed and validated a composite postoperative prognostic framework integrating radiological hilar lymph node status and preoperative anemia for patients undergoing curative resection of ICC. The resulting nomogram demonstrated robust discriminative performance, achieving C-indices of 0.881, 0.802, and 0.758 for predicting recurrence, OS, and RFS, respectively, consistently outperforming conventional TNM-based staging systems. Risk-stratified analysis revealed profound outcome disparities, with 5-year OS rates declining from 86.6% in low-risk patients to 19.6% in high-risk patients (P < 0.001). These findings underscore the prognostic value of synthesizing anatomical staging with systemic physiological reserve to capture the biological complexity of ICC.
Addressing this clinical deficit, the present study synthesized a composite postoperative prognostic tool integrating radiological hilar nodal status with systemic hemoglobin reserves. The resulting framework exhibited high discriminative fidelity across training and validation cohorts, consistently outperforming conventional TNM staging systems. Notably, the magnitude of risk associated with the identified predictors was profound. Hilar LNM emerged not merely as a statistical correlate but as a dominant driver of recurrence, conferring a nearly ten-fold increase in risk. Even more striking was the impact of preoperative anemia, which demonstrated an exceptionally high HR for OS and RFS. By stratifying patients into distinct risk groups, our model validates the hypothesis that the convergence of radiographic burden and hematological depletion characterizing a biological phenotype of extreme aggression. Identifying this phenotype post-resection necessitates an aggressive departure from standard observation protocols toward individualized adjuvant interventions.
The conceptual advance of this work lies in harmonizing anatomical staging with physiological reserve markers to bridge the existing gap in multimodal data fusion. Unlike previous investigations that analyzed LNM or hematological indices in isolation, approaches that often fail to capture the complexity of tumor-host interactions [10, 20, 21], our integrated model captures the synergistic lethality of these factors [22]. For instance, earlier work emphasized the impact of LNM on metastatic risk but rarely conducted systematic radiological evaluations of specific regions [22]. Similarly, investigations into hemoglobin primarily examined its association with overall cancer prognosis, overlooking ICC-specific interactions [23]. This synergy is biologically plausible. The pathogenesis of hilar LNM involves the invasion of the hepatic portal lymphatic network [24], establishing micrometastases that likely disrupt local immune surveillance and secrete pro-inflammatory cytokines such as interleukin-6 [13, 14]. Crucially, anatomical obstruction of the biliary tree by metastatic nodes can precipitate cholestasis and liver dysfunction, accelerating disease trajectory.
Parallel to this anatomical progression, the systemic impact of preoperative anemia reflects a profound, adverse remodeling of the tumor microenvironment [25, 26]. Reduced circulating hemoglobin directly compromises oxygen delivery and precipitates profound intratumor hypoxia [27, 28]. This hypoxic tension acts as a major upstream trigger that stabilizes hypoxia-inducible factor-1alpha (HIF-1α) within the core of ICC lesions. The persistent activation of HIF-1α initiates a transcriptional program that upregulates vascular endothelial growth factor, fueling aberrant, leaky angiogenesis and accelerated tumor expansion [29]. More critically for ICC, this hypoxia-driven pathway induces epithelial-mesenchymal transition and stimulates lymphangiogenesis, providing a clear biological explanation for why anemic patients in our cohort displayed a significantly higher propensity for early hilar LNM. Beyond these molecular kinetics, the clinical relevance of systemic anemia is firmly grounded in hepatobiliary oncology [30]. Chronic oxygen deprivation impairs cellular immunity by driving Fas/FasL-mediated T-cell apoptosis [31], thereby facilitating a permissive niche for occult micrometastatic survival after surgery. This creates a self-perpetuating, vicious clinical cycle, the underlying ICC-driven systemic inflammation impairs iron utilization, while the resulting anemia-induced hypoxia accelerates lymphatic and distant dissemination [32–34]. Our clinical findings align with recent large-scale retrospective series confirming that preoperative anemia independently predicts compromised survival following radical hepatectomy for hepatobiliary malignancies [30, 35]. By structurally uniting preoperative anemia with hilar LNM status, our model achieves a highly complementary and synergistic predictive performance.
From a translational perspective, this risk stratification tool holds potential utility for informing post-resection oncological care. For patients exhibiting the high-risk phenotype, defined by concurrent hilar LNM and hemoglobin < 10 g/dL, our data indicate an exceptionally high propensity for early relapse, suggesting that standard surgical resection alone may achieve an unsatisfactory therapeutic ceiling. Consequently, these individuals may warrant consideration for intensified postoperative adjuvant chemotherapy strategies rather than routine surveillance alone, although this hypothesis requires validation in prospective clinical trials. Furthermore, for this specific high-risk stratum, more frequent surveillance intervals (e.g., every 6 to 8 weeks during the critical first two years post-surgery) could be explored to facilitate early radiological detection of recurrent lesions, pending confirmation in appropriately designed prospective studies. Moreover, the reliance on routinely available metrics ensures the model’s scalability, particularly in resource-constrained settings where complex molecular profiling is inaccessible. This facilitates a transition from empirical prognostication to evidence-based risk quantification, advancing the precision oncology agenda for ICC.
Several limitations of our study warrant carefully balanced consideration. First, this study is inherently limited by its retrospective design, which may introduce potential selection bias. Although strict inclusion and exclusion criteria were applied, unmeasured confounding factors, such as variations in preoperative nutritional interventions or subtle differences in surgical techniques over the study period, cannot be completely ruled out. Second, formal statistical power analyses were not performed for all reported post-resection endpoints prior to the study. While our sample size was sufficient to construct a stable prognostic model with promising predictive performance, the absence of a power analysis may limit the robustness and interpretability of some secondary conclusions, particularly within smaller patient subgroups. Additionally, the cohort was exclusively drawn from Chinese centers, where ICC pathogenesis is predominantly driven by chronic hepatolithiasis, Clonorchis sinensis infection, and hepatitis B virus-related cirrhosis [1]. These factors induce a unique background of chronic systemic inflammation, hypersplenism, and preoperative anemia, characteristics that differ substantially from Western populations, where primary sclerosing cholangitis, alcohol-related liver disease, or metabolic dysfunction-associated steatohepatitis typically prevail [3]. Consequently, while our nomogram outperformed staging systems like the AJCC TNM within this etiological context, our findings should be interpreted with caution. Whether this synergistic integration of host-physiological reserves maintains predictive fidelity in Western cohorts, where driver mutations and inflammatory substrates diverge remains to be determined. Finally, our model remains a static baseline assessment. Incorporating longitudinal biomarker trajectories into future iterations could further refine its dynamic prognostic precision. In summary, while our combined model provides a practical tool for postoperative risk stratification, these findings should be interpreted with caution. Well-designed, prospective, and ideally large-scale multicenter studies are fundamentally required to externally validate and confirm the clinical utility of this risk-stratification model before its widespread implementation into routine clinical decision-making.
Conclusions
This study establishes preoperative radiological hilar lymph node status and systemic anemia as powerful, independent prognostic determinants for patients undergoing curative-intent resection for ICC. This model offers a practical, evidence-based instrument to risk-stratify patients’ post-resection and personalize postoperative adjuvant management, potentially improving survival outcomes in this challenging patient population.
Supplementary Information
Supplementary Material 1. Kaplan-Meier survival analyses comparing OS predictions across conventional staging systems and clinical factors in the primary cohort. (A) The 8th edition of the AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. These curves illustrate the discriminatory ability of established staging systems and key clinical variables in predicting OS outcomes.AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; OS, overall survival.
Supplementary Material 2. Kaplan-Meier survival analyses comparing OS predictions across conventional staging systems and clinical factors in the external validation. (A) The 8th edition of the AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. These curves illustrate the discriminatory ability of established staging systems and key clinical variables in predicting OS outcomes. AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; OS, overall survival.
Supplementary Material 3. Kaplan-Meier survival curves demonstrating RFS discrimination across conventional staging systems and clinical risk factors in the primary cohort. (A) The 8th edition AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. This comparative analysis illustrates the prognostic discriminatory power of established staging classifications and key clinical variables in predicting RFS outcomes, providing benchmark comparisons for model validation.AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; RFS, recurrence-free survival.
Supplementary Material 4. Kaplan-Meier survival curves demonstrating RFS discrimination across conventional staging systems and clinical risk factors in the external validation. (A) The 8th edition AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. This comparative analysis illustrates the prognostic discriminatory power of established staging classifications and key clinical variables in predicting RFS outcomes, providing benchmark comparisons for model validation.AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; RFS, recurrence-free survival.
Acknowledgements
The authors acknowledge and express their deepest gratitude to the participants of this research.
Abbreviations
- AJCC 8th
the 8th edition of the american joint committee on cancer
- AUC
Area under the curve
- CEA
Carcinoembryonic antigen
- C-index
Concordance index
- CI
Confidence interval
- CRP
C-reactive protein
- CT
Computed tomography
- ECOG
Eastern cooperative oncology group
- HR
Hazard ratio
- ICC
Intrahepatic cholangiocarcinoma
- IQR
Interquartile range
- JSHBPS
Japanese society of hepato-biliary-pancreatic surgery
- LCSGJ
Liver cancer study group of japan
- LNM
Hilar lymph node metastasis
- MEGNA
Metastatic, extrahepatic, grade, node, albumin score
- MRI
Magnetic resonance imaging
- MVI
Microvascular invasion
- NLR
Neutrophil-to-lymphocyte ratio
- TBS
Tumor burden score
- OR
Odds ratio
- OS
Overall survival
- PLR
Platelet-to-lymphocyte ratio
- RFS
Recurrence-free survival
Authors' contributions
Min Deng: Conceptualization, Data curation, Formal analysis, Methodology, Software, Visualization, Funding acquisition, Writing-original draft. Benyi He: Data curation, Methodology, Formal analysis, Writing-original draft. Zenghai Lin: Conceptualization, Data curation, Methodology, Writing-original draft. Deyao Zhang: Formal analysis, Data curation, Methodology. Yuxiong Qiu: Formal analysis, Software & Visualization. Dong Li: Methodology. Yajun Tang: Methodology. Xianchao Guo: Methodology, Software & Visualization. Guojia Yuan: Methodology. Wei Qin: Methodology. Hao Cai: Methodology, Software, Visualization & Funding acquisition. Yudong Qiu: Methodology. Zubing Chen: Methodology. Di Tang: Conceptualization, Formal analysis, Methodology, Supervision, Writing-review & editing. Wei Wei: Conceptualization, Formal analysis, Methodology, Writing-review & editing. Rongping Guo: Conceptualization, Formal analysis, Funding acquisition, Methodology, Supervision, Writing-review & editing.
Funding
This study was supported by the National Science and Technology Major Project of China (2024ZD0520402); National Natural Science Foundation of China (No. 82303156, No. 82172579); Guangdong Provincial Natural Science Foundation (2024A1515012581); Shenzhen Science and Technology Innovation Committee (JCYJ20250604143712016). Clinical Trials Project (5010 Project) of Sun Yat-sen University (No. 5010-2023001).
Data availability
All data used during the study are available from the corresponding author upon reasonable request.
Declarations
Ethics approval and consent to participate
All participants provided informed consent, and the study was approved by the Clinical Research Ethics Committee of SYSUSH (No. 20250729218) and adhered to the Declaration of Helsinki. This study was a retrospective review of medical records.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Min Deng, Benyi He, Zenghai Lin and Deyao Zhang contributed equally to this work.
Contributor Information
Di Tang, Email: tangdi@sysush.com.
Wei Wei, Email: weiwei@sysucc.org.cn.
Rongping Guo, Email: guorp@sysucc.org.cn.
References
- 1.Siegel RL, Kratzer TB, Wagle NS, Sung H, Jemal A. Cancer statistics, 2026. CA Cancer J Clin. 2026;76(1):e70043. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Moris D, Palta M, Kim C, Allen PJ, Morse MA, Lidsky ME. Advances in the treatment of intrahepatic cholangiocarcinoma: An overview of the current and future therapeutic landscape for clinicians. CA Cancer J Clin. 2023;73(2):198–222. [DOI] [PubMed] [Google Scholar]
- 3.Banales JM, Rodrigues PM, Affo S, Andersen JB, Aspichueta P, Boulter L, Bridgewater J, Calvisi DF, Cardenas A, Cardinale V, et al. Cholangiocarcinoma 2026: status quo, unmet needs and priorities. Nat Rev Gastroenterol Hepatol. 2026;23(1):65–96. [DOI] [PubMed] [Google Scholar]
- 4.European Association for the Study of the L. EASL-ILCA Clinical Practice Guidelines on the management of intrahepatic cholangiocarcinoma. J Hepatol. 2023;79(1):181–208. [DOI] [PubMed] [Google Scholar]
- 5.Macias RIR, Cardinale V, Kendall TJ, Avila MA, Guido M, Coulouarn C, Braconi C, Frampton AE, Bridgewater J, Overi D, et al. Clinical relevance of biomarkers in cholangiocarcinoma: critical revision and future directions. Gut. 2022;71(8):1669–83. [DOI] [PubMed] [Google Scholar]
- 6.Kosaka H, Ueno M, Komeda K, Hokuto D, Iida H, Hirokawa F, Matsui K, Sekimoto M, Kaibori M. The impact of a preoperative staging system on accurate prediction of prognosis in intrahepatic cholangiocarcinoma. Cancers (Basel). 2022;14(5):1107. [DOI] [PMC free article] [PubMed]
- 7.Rhee H, Lim HJ, Han K, Yeom SK, Choi SH, Park JH, Cho ES, Park S, Lee MJ, Choi GH, et al. A preoperative scoring system to predict lymph node metastasis in intrahepatic cholangiocarcinoma. Hepatol Int. 2023;17(4):942–53. [DOI] [PubMed] [Google Scholar]
- 8.Lapitz A, Azkargorta M, Milkiewicz P, Olaizola P, Zhuravleva E, Grimsrud MM, Schramm C, Arbelaiz A, O’Rourke CJ, La Casta A, et al. Liquid biopsy-based protein biomarkers for risk prediction, early diagnosis, and prognostication of cholangiocarcinoma. J Hepatol. 2023;79(1):93–108. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Jareanrat A, Prajumwongs P, Titapun A, Thanasukarn V, Khuntikeo N, Rattanarak K, Namwat N, Klanrit P, Wangwiwatsin A, Chindaprasirt J, et al. Identification of serum peptide biomarkers for cholangiocarcinoma diagnosis and staging via MALDI-TOF MS and LC-MS/MS. Br J Cancer. 2026;134(2):342–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Milana F, Polidoro MA, Soldani C, Franceschini B, Famularo S, Di Tommaso L, Terracciano LM, Lleo A, Donadon M, Torzilli G. Unveiling the prognostic role of blood inflammatory indexes in a retrospective cohort of patients undergoing liver resection for intrahepatic cholangiocarcinoma. Int J Surg. 2024;110(11):7088–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Shannon AM, Bouchier-Hayes DJ, Condron CM, Toomey D. Correction of anaemia through the use of darbepoetin alfa improves chemotherapeutic outcome in a murine model of Lewis lung carcinoma. Br J Cancer. 2005;93(2):224–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Liu CX, Wong CC. Stratifying cholangiocarcinoma: tumor microenvironment, molecular drivers, and novel immunotherapeutic approaches. Clin Mol Hepatol. 2026;32(1):127–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Banales JM, Marin JJG, Lamarca A, Rodrigues PM, Khan SA, Roberts LR, Cardinale V, Carpino G, Andersen JB, Braconi C, et al. Cholangiocarcinoma 2020: the next horizon in mechanisms and management. Nat Rev Gastroenterol Hepatol. 2020;17(9):557–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Li H, Lan T, Liu H, Liu C, Dai J, Xu L, Cai Y, Hou G, Xie K, Liao M, et al. IL-6-induced cGGNBP2 encodes a protein to promote cell growth and metastasis in intrahepatic cholangiocarcinoma. Hepatology. 2022;75(6):1402–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Kesharwani P, Dash D, Koiri RK. Deciphering the role of hepcidin in iron metabolism and anemia management. J Trace Elem Med Biol. 2025;87:127591. [DOI] [PubMed] [Google Scholar]
- 16.Nemeth E, Tuttle MS, Powelson J, Vaughn MB, Donovan A, Ward DM, Ganz T, Kaplan J. Hepcidin regulates cellular iron efflux by binding to ferroportin and inducing its internalization. Science. 2004;306(5704):2090–3. [DOI] [PubMed] [Google Scholar]
- 17.Deng M, Lei Q, Wang J, Lee C, Guan R, Li S, Wei W, Chen H, Zhong C, Guo R. Nomograms for predicting the recurrence probability and recurrence-free survival in patients with hepatocellular carcinoma after conversion hepatectomy based on hepatic arterial infusion chemotherapy: a multicenter, retrospective study. Int J Surg. 2023;109(5):1299–310. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Huang T, Liu H, Lin Z, Kong J, Lin K, Lin Z, Chen Y, Lin Q, Zhou W, Li J, et al. Preoperative prediction of intrahepatic cholangiocarcinoma lymph node metastasis by means of machine learning: a multicenter study in China. BMC Cancer. 2022;22(1):931. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, Ghassemi M, Liu X, Reitsma JB, van Smeden M, et al. TRIPOD + AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Tsilimigras DI, Mehta R, Aldrighetti L, Poultsides GA, Maithel SK, Martel G, Shen F, Koerkamp BG, Endo I, Pawlik TM, et al. Development and Validation of a Laboratory Risk Score (LabScore) to Predict Outcomes after Resection for Intrahepatic Cholangiocarcinoma. J Am Coll Surg. 2020;230(4):381–e391382. [DOI] [PubMed] [Google Scholar]
- 21.Tang Y, Xu L, Zhang G, Li K, Shi A, Shu L, Zhao L, Li E, Sun K, Pan G, et al. Survival analysis and prognostic nomogram for patients with cholangiocarcinoma after radical resection in Asia. Eur J Surg Oncol. 2024;50(12):108659. [DOI] [PubMed] [Google Scholar]
- 22.Zhu J, Wang D, Liu C, Huang R, Gao F, Feng X, Lan T, Li H, Wu H. Development and validation of a new prognostic immune-inflammatory-nutritional score for predicting outcomes after curative resection for intrahepatic cholangiocarcinoma: A multicenter study. Front Immunol. 2023;14:1165510. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Toshida K, Itoh S, Nakayama Y, Tsutsui Y, Kosai-Fujimoto Y, Tomino T, Yoshiya S, Nagao Y, Harada N, Kohashi K, et al. Preoperative HALP score is a prognostic factor for intrahepatic cholangiocarcinoma patients undergoing curative hepatic resection: association with sarcopenia and immune microenvironment. Int J Clin Oncol. 2023;28(8):1082–91. [DOI] [PubMed] [Google Scholar]
- 24.Brindley PJ, Bachini M, Ilyas SI, Khan SA, Loukas A, Sirica AE, Teh BT, Wongkham S, Gores GJ. Cholangiocarcinoma. Nat Rev Dis Primers. 2021;7(1):65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Toi M, Asao Y, Matsumoto Y, Sekiguchi H, Yoshikawa A, Takada M, Kataoka M, Endo T, Kawaguchi-Sakita N, Kawashima M, et al. Visualization of tumor-related blood vessels in human breast by photoacoustic imaging system with a hemispherical detector array. Sci Rep. 2017;7:41970. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Ossandon MR, Phatak DS, Sorg BS, Kalpakis K. Forecasting new development of tumor areas using spatial and temporal distribution profiles of hemoglobin saturation in a mouse model. J Med Imaging (Bellingham). 2014;1(1):014503. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Harris EE, Kasi S, Hwang WT, Vapiwala N, Solin LJ. Role of anemia prior to radiation treatment in local recurrence and survival after breast conservation treatment for early-stage breast cancer. Clin Breast Cancer. 2010;10(1):74–80. [DOI] [PubMed] [Google Scholar]
- 28.Terman DS, Viglianti BL, Zennadi R, Fels D, Boruta RJ, Yuan H, Dreher MR, Grant G, Rabbani ZN, Moon E, et al. Sickle erythrocytes target cytotoxics to hypoxic tumor microvessels and potentiate a tumoricidal response. PLoS ONE. 2013;8(1):e52543. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Park SY, Jeong KJ, Lee J, Yoon DS, Choi WS, Kim YK, Han JW, Kim YM, Kim BK, Lee HY. Hypoxia enhances LPA-induced HIF-1alpha and VEGF expression: their inhibition by resveratrol. Cancer Lett. 2007;258(1):63–9. [DOI] [PubMed] [Google Scholar]
- 30.Schnitzbauer AA, Eberhard J, Bartsch F, Brunner SM, Ceyhan GO, Walter D, Fries H, Hannes S, Hecker A, Li J, et al. The MEGNA Score and Preoperative Anemia are Major Prognostic Factors After Resection in the German Intrahepatic Cholangiocarcinoma Cohort. Ann Surg Oncol. 2020;27(4):1147–55. [DOI] [PubMed] [Google Scholar]
- 31.Zhang J, Han C, Dai H, Hou J, Dong Y, Cui X, Xu L, Zhang M, Xia Q. Hypoxia-Inducible Factor-2alpha Limits Natural Killer T Cell Cytotoxicity in Renal Ischemia/Reperfusion Injury. J Am Soc Nephrol. 2016;27(1):92–106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Rosato BE, D’Onofrio V, Marra R, Nostroso A, Esposito FM, Iscaro A, Lasorsa VA, Capasso M, Iolascon A, Russo R, et al. RAS signaling pathway is essential in regulating PIEZO1-mediated hepatic iron overload in dehydrated hereditary stomatocytosis. Am J Hematol. 2025;100(1):52–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Teng H, Yang Y, Wei H, Liu Z, Liu Z, Ma Y, Gao Z, Hou L, Zou X. Fucoidan Suppresses Hypoxia-Induced Lymphangiogenesis and Lymphatic Metastasis in Mouse Hepatocarcinoma. Mar Drugs. 2015;13(6):3514–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Du X, She E, Gelbart T, Truksa J, Lee P, Xia Y, Khovananth K, Mudd S, Mann N, Moresco EM, et al. The serine protease TMPRSS6 is required to sense iron deficiency. Science. 2008;320(5879):1088–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Catalano G, Alaimo L, Endo Y, Chatzipanagiotou OP, Ruzzenente A, Aldrighetti L, Weiss M, Bauer TW, Alexandrescu S, Poultsides GA, et al. Development and Validation of a Predictive Risk Score for Blood Transfusion in Patients Undergoing Curative-Intent Surgery for Intrahepatic Cholangiocarcinoma. J Surg Oncol. 2025;131(2):242–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Material 1. Kaplan-Meier survival analyses comparing OS predictions across conventional staging systems and clinical factors in the primary cohort. (A) The 8th edition of the AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. These curves illustrate the discriminatory ability of established staging systems and key clinical variables in predicting OS outcomes.AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; OS, overall survival.
Supplementary Material 2. Kaplan-Meier survival analyses comparing OS predictions across conventional staging systems and clinical factors in the external validation. (A) The 8th edition of the AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. These curves illustrate the discriminatory ability of established staging systems and key clinical variables in predicting OS outcomes. AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; OS, overall survival.
Supplementary Material 3. Kaplan-Meier survival curves demonstrating RFS discrimination across conventional staging systems and clinical risk factors in the primary cohort. (A) The 8th edition AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. This comparative analysis illustrates the prognostic discriminatory power of established staging classifications and key clinical variables in predicting RFS outcomes, providing benchmark comparisons for model validation.AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; RFS, recurrence-free survival.
Supplementary Material 4. Kaplan-Meier survival curves demonstrating RFS discrimination across conventional staging systems and clinical risk factors in the external validation. (A) The 8th edition AJCC staging system; (B)The LCSGJ classification; (C) The JSHBPS staging; (D) The Fudan scoring system; (E) The Zhou prognostic score; (F) The MEGNA scoring system; (G) LNM status; and (H) Preoperative anemia. This comparative analysis illustrates the prognostic discriminatory power of established staging classifications and key clinical variables in predicting RFS outcomes, providing benchmark comparisons for model validation.AJCC, American Joint Committee on Cancer; JSHBPS, Japanese Society of Hepato-Biliary-Pancreatic Surgery; LCSGJ, Liver Cancer Study Group of Japan; LNM, lymph node metastasis; MEGNA, metastatic, extrahepatic, grade, node, albumin; RFS, recurrence-free survival.
Data Availability Statement
All data used during the study are available from the corresponding author upon reasonable request.