Biliary tract cancer (BTC), encompassing intrahepatic cholangiocarcinoma (ICC), extrahepatic cholangiocarcinoma (ECC), gallbladder cancer (GBC), and ampullary carcinoma (AC), represents the second most common hepatobiliary malignancy, accounting for approximately 3% of all gastrointestinal cancers (1,2). Despite its relative rarity, BTC is characterized by aggressive behavior and poor prognosis, often diagnosed at advanced stages that limit therapeutic efficacy (3). Although surgical resection remains the only potentially curative treatment and recent advances in chemotherapy have improved outcomes, high rates of recurrence and metastasis continue to challenge long-term survival, underscoring the urgent need for improved prognostic tools (2,4,5).

The Tumor-Node-Metastasis (TNM) staging system serves as the cornerstone for evaluating disease extent and guiding treatment decisions in BTC (6). However, its reliance solely on anatomical features fails to capture critical biological and host-related factors that significantly influence clinical outcomes. Patients with identical TNM stages frequently exhibit divergent survival trajectories, suggesting that tumor biology and systemic responses play pivotal roles in BTC progression (7). This limitation highlights the inadequacy of TNM staging as a standalone prognostic tool, particularly in this heterogeneous and aggressive malignancy.

Mounting evidence indicates that the tumor microenvironment's inflammatory response plays a crucial role in the development, progression, and metastasis of solid tumors, including BTC (8). Multiple pretreatment inflammation-based indices, such as the neutrophil-to-lymphocyte ratio (NLR), prognostic nutritional index (PNI), and C-reactive protein-to-albumin ratio (CAR), have been proposed as prognostic markers in BTC (9-11). However, their individual prognostic value remains inconsistent across studies, and consensus on the most informative marker for risk stratification is lacking (12). This knowledge gap underscores the need for evaluation of multiple inflammatory indices in the same cohort to objectively identify the optimal prognostic marker in BTC.

To address these challenges, this study aimed to comprehensively evaluate eight preoperative inflammatory indices and develop a prognostic nomogram incorporating the most informative marker alongside clinicopathological factors in patients with BTC undergoing resection. By deliberately excluding TNM components, we sought to capture biological and host-related prognostic factors orthogonal to anatomical staging, thereby providing a complementary tool to enhance postoperative risk stratification for overall survival (OS) and improve clinical decision-making in this aggressive malignancy.

Patients and data collection. This retrospective cohort study included 286 consecutive patients with histologically confirmed BTC who underwent surgical resection at our institution between January 2010 and September 2024. Exclusion criteria were palliative-intent surgery (n=25), incomplete clinicopathological or laboratory data (n=11), and non-adenocarcinoma histology (n=5; e.g., neuroendocrine carcinoma), with some overlap among excluded cases, yielding 247 patients: 53 with ICC, 120 with ECC, 33 with GBC, and 41 with AC. Clinical and demographic data - age, sex, American Society of Anesthesiologists Physical Status (ASA-PS), body mass index (BMI), and surgical procedure - were extracted from electronic medical records. Of these, 149 (60.3%) had biliary obstruction and underwent preoperative stenting or drainage to relieve jaundice. Serum carbohydrate antigen (CA) 19-9 levels were measured within one month before surgery, post-stenting/drainage, using chemiluminescent enzyme immunoassay (Fujirebio, Tokyo, Japan). Histopathological assessment included TNM staging (7th edition, Union for International Cancer Control) (6), tumor differentiation (well, moderate, or poor), perineural invasion, and resection margin status (R0: no residual tumor; R1: microscopic; R2: macroscopic). The 7th edition was used throughout, as the 8th requires invasion depth for T classification, often unavailable due to incomplete records. Eleven patients without lymphadenectomy - eight after transduodenal ampullectomy for AC and three after simple cholecystectomy for GBC - were classified as N0 based on preoperative imaging and intraoperative findings. During surgery, 19 patients (7.7%) had resectable limited metastases (liver, n=8; para-aortic lymph nodes, n=6; peritoneum, n=5) and underwent simultaneous resection of primary and metastatic lesions. Adjuvant chemotherapy (e.g., S-1) was given at physician discretion, typically for N1 disease or R1/R2 resection without contraindications. Follow-up duration was estimated using the reverse Kaplan-Meier (KM) method (13). OS, the primary endpoint, was defined from surgery to last follow-up or death from any cause. Blood samples for inflammation-based indices were collected a median of two days (range=0-7 days) before surgery, with no active infection or systemic inflammation per clinical assessment. The following inflammation-based indices were calculated per established methods: NLR: Neutrophil count (/mm3)/lymphocyte count (/mm3); Platelet-to-lymphocyte ratio (PLR): Platelet count (/mm3)/lymphocyte count (/mm3); Lymphocyte-to-monocyte ratio (LMR): Lymphocyte count (/mm3)/monocyte count (/mm3); PNI: 10 × serum albumin (g/dl) + 0.005 × lymphocyte count (/mm³); Modified Glasgow prognostic score (mGPS) 0: Albumin ≥3.5 g/dl and C-reactive protein (CRP) ≤1.0 mg/dl; 1: Either albumin <3.5 g/dl or CRP >1.0 mg/dl; 2: Albumin <3.5 g/dl and CRP >1.0 mg/dl).; CAR: CRP (mg/dl)/albu-min (g/dl); Systemic-immune-inflammation index (SII): Platelet count (/mm3) × neutrophil count (/mm3)/lymphocyte count (/mm3); Systemic-inflammation-response index (SIRI): Neutrophil count (/mm3) × monocyte count (/mm³)/lymphocyte count (/mm3).

Ethical approval. This study was approved by the Institutional Ethics Committee (approval no.: 2025-40) and adhered to the Declaration of Helsinki. Patient consent was waived due to the retrospective design and use of de-identified data.

Statistical analysis. Inflammation-based indices were dichotomized using optimal cutoffs from maximally selected rank statistics to maximize log-rank differences in KM curves. The same method stratified patients by nomogram score into four risk groups. Univariate Cox regression identified OS-associated factors. To address multicollinearity and select variables for the nomogram, Least Absolute Shrinkage and Selection Operator (LASSO) regression was applied, explicitly excluding TNM components (T stage, N stage, M stage, TNM stage), with λ optimized by 10-fold cross-validation. Selected variables entered multivariable Cox regression for hazard ratios and nomogram construction. Model performance was evaluated via KM analysis, Harrell’s concordance (C-) index, time-dependent area under the receiver operating characteristic curve (tdAUC), decision curve analysis (DCA), and calibration plots for 1-, 3-, and 5-year OS. Internal validation used 1,000 bootstrap resamples to estimate 95% confidence intervals (CIs) and assess overfitting. All analyses used R (v4.4.1), with significance at p<0.05 (two-tailed).

Patient characteristics and univariate analysis. Table I summarizes the baseline characteristics of 247 patients. The median follow-up was 63.8 months (95%CI=53.7-74.9 months). Factors significantly associated with OS included biliary drainage, CA19-9 >37 U/ml, advanced T stage, N1 status, M1 status, higher TNM stage, moderate/poor tumor differentiation, perineural invasion, R1/R2 resection, and adjuvant chemotherapy. Among the inflammation-based indices, all except PLR showed significant associations with OS: NLR >4.5, LMR ≤2.5, PNI ≤40.4, mGPS 1 and 2, CAR >0.18, SII >547, and SIRI >2.15.

Variable selection and multivariate analysis. LASSO regression, applied to 20 variables from Table I excluding T stage, N stage, M stage, and TNM stage, selected seven OS predictors: CA19-9 >37 U/ml, CAR >0.18, PNI ≤40.4, BMI ≥25 kg/m2, moderate/poor tumor differentiation, perineural invasion, and R1/R2 resection (Figure 1A-C). Multivariable Cox regression retained six independent prognostic factors: BMI ≥25 kg/m2, CA19-9 >37 U/ml, moderate/poor tumor differentiation, perineural invasion, R1/R2 resection, and CAR >0.18. PNI was not an independent predictor in the final model (p=0.067) (Table II).

Nomogram development and survival analysis. A nomogram incorporating the six prognostic factors was developed (Figure 2). KM analysis stratified patients into four distinct risk groups (Low, Medium-low, Medium-high, High) based on optimal nomogram score cutoffs (p<0.001, Figure 3A). TNM stage-based stratification showed less distinct separation, particularly between Stages III and IV (Figure 3B). Notably, nomogram-based reclassification assigned 50.6% (78/154) of TNM Stage I-II patients to Medium-high or High-risk groups, while 16.1% (15/93) of Stage III-IV patients were categorized as Low or Medium-low risk.

Discrimination and calibration of the nomogram. The nomogram achieved a C-index of 0.722 (95%CI=0.676-0.768) versus 0.659 (95%CI=0.608-0.706; p=0.022) for TNM staging. tdAUCs were consistently higher for the nomogram across all evaluated time points (Figure 4A). DCA demonstrated slightly higher net benefit for the nomogram versus TNM staging across a range of threshold probabilities (Figure 4B). Calibration plots confirmed reliable OS prediction at one, three, and five years (Figure 5).

Subgroup analysis of nomogram score prognostic impact. Stratified analysis by BTC subtype and TNM stage evaluated the nomogram's prognostic consistency across subgroups (Figure 6). The nomogram score demonstrated consistent association with OS across all subgroups, with no significant interaction (p for interaction=0.913 and 0.266 for tumor location and TNM stage, respectively).

In this study, we developed a prognostic nomogram for patients with resected BTC, incorporating CAR selected from eight inflammation-based indices. The nomogram demonstrated acceptable discrimination, clinical utility, and good calibration. Subgroup analysis showed consistent prognostic performance across BTC subtypes and TNM stages without significant interactions, suggesting broad applicability.

The role of inflammation and nutritional status in cancer progression has been increasingly recognized, particularly through effects on the tumor microenvironment and systemic cytokine dynamics (8,14). Chronic inflammation promotes tumor proliferation and metastasis via mediators such as interleukin-6, which activates STAT3 signaling, and tumor necrosis factor-alpha, which drives matrix remodeling (8,15). These mechanisms are reflected in systemic inflammation-based indices (16). Numerous studies have examined the prognostic value of such indices in BTC, often focusing on specific subtypes. For instance, in ICC, a meta-analysis by Cui et al. reported NLR to be associated with worse prognosis, and other studies identified SII as significant (17-19). In ECC, Miyahara et al. assessed NLR and LMR as potential prognostic markers (20). In GBC, several studies investigated NLR, SII, SIRI, and CAR for their prognostic relevance (21-24). Additionally, meta-analyses support SII, mGPS, PNI, and CAR as prognostic across BTC types (10,11,25,26). However, studies systematically comparing a broad set of indices to identify the most informative remain scarce.

Among inflammation-based indices, CAR integrates CRP, a marker of systemic inflammation, and albumin, an indicator of nutritional status. This dual profile may enhance its prognostic value compared to inflammation-only indices like NLR or SII. CAR has shown prognostic relevance across various solid tumors (27,28). Systemic inflammation, reflected by elevated CRP, contributes to tumor progression through cytokine-mediated pathways. Low albumin, indicating malnutrition, further impairs immune function and worsens outcomes, making CAR a strong prognostic marker in BTC. In our LASSO regression, CAR and PNI - both including albumin - were selected, emphasizing albumin’s role in capturing nutritional decline. PNI was excluded in multivariable Cox analysis, likely due to redundancy with CAR, which had greater prognostic strength. Unlike mGPS, which uses categorical thresholds, or PNI, which reflects nutrition alone, CAR’s continuous scale and dual nature improve prognostic precision. Though we dichotomized CAR at >0.18 for survival stratification, its continuous structure distinguishes it from mGPS’s fixed categories. This combination of nuance and practicality may explain CAR’s advantage in our analysis.

Other nomogram components hold prognostic value. Elevated CA19-9 signals tumor burden and aggressiveness (29). Clinicopathological factors - tumor differentiation, perineural invasion, and residual tumor status - reflect disease aggressiveness (30-32). Notably, BMI ≥25 kg/m² independently predicted poor prognosis despite univariable insignificance, retained by LASSO for multivariable contribution. Obesity may promote tumor progression locally via adipose inflammation and microenvironment changes, and systemically through metabolic and inflammatory mediators (33). However, its prognostic role in BTC remains unclear: some studies link obesity to higher mortality, while others suggest survival benefit in advanced cases treated with chemotherapy (34,35). These inconsistencies warrant further exploration of metabolic factors and body composition in BTC progression and treatment response.

Several nomograms incorporating inflammatory indices for resected BTC exist. Sun et al. proposed a GBC model combining SII, CA19-9, and TNM stage; Zhu et al. developed an ICC model integrating PIIN score (based on ALBI grade, NLR, PNI, and SII) with N stage and tumor number (22,36). Though outperforming TNM alone, their TNM reliance limits capturing heterogeneity beyond anatomy. Our nomogram excludes TNM, using CAR and clinicopathological factors for biologically driven, complementary postoperative stratification - useful when TNM is uncertain (e.g., incomplete lymph node assessment or unclear invasion depth). Like TNM models, it is postoperative-only due to variables like perineural invasion and residual status. Yet preoperative CAR, BMI, and CA19-9 enable preliminary estimates, with the full nomogram refining prognosis post-resection to guide follow-up and adjuvant therapy.

Study limitations. First, its retrospective, single-center design and small sample size risk selection bias and residual confounding. Second, and most critically, despite bootstrap internal validation, lack of external validation limits model robustness and generalizability - future studies must address this. Third, inflammatory indices could be affected by preoperative biliary drainage (common here), despite sampling without active inflammation. Fourth, BTC anatomical diversity and surgical invasiveness - including complications like liver failure after major resection - may confound OS. Fifth, classifying 11 patients without full lymphadenectomy as N0 constrains precise N staging. Yet, since the nomogram excludes TNM by design, this imprecision does not impair performance and highlights its utility when conventional staging is unreliable. Sixth, including limited metastatic disease adds prognostic heterogeneity but reflects real-world BTC surgery and aligns with the model’s TNM-independent goal. Lastly, evolving surgical techniques, care, and therapies over the study period risk time-related bias, reducing applicability to current practice.

In conclusion, CAR was the most informative inflammatory marker for prognosis in resected BTC. This prognostic nomogram, integrating CAR with key clinicopathological factors while excluding TNM components, complements TNM staging by enhancing prognostic stratification, especially for patients with heterogeneous outcomes within the same anatomical stage. However, external validation is required to confirm its clinical utility.

All Authors declare no conflicts of interest in relation to this study.

Shinichi Ikuta: Conceptualization, Methodology, Writing - original draft; Tsukasa Aihara and Takayoshi Nakajima: Data curation, Investigation; Masataka Fujikawa: Formal analysis, Visualization; Naoki Yamanaka: Supervision, Project administration, Writing - review and editing.

The Authors thank the staff at our institution for their support during data collection.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

AI tools were used solely to assist with statistical coding and English text proofreading. The AI did not contribute to the study design, data collection, data interpretation, or the conclusions of the research.