Construction of a predictive model for gemcitabine combined with cisplatin resistance in intrahepatic cholangiocarcinoma based on multidimensional inflammatory indices

Introduction

The treatment options for cholangiocarcinoma (CCA) exhibit significant variability (1). Early-stage localized lesions may achieve potential cure through radical surgical resection, yet the majority of patients lose the opportunity for surgery due to delayed diagnosis, often presenting at an advanced stage with extensive metastasis or local infiltration at the time of diagnosis. For patients who are not candidates for surgery, systemic chemotherapy becomes the primary treatment modality, with gemcitabine combined with cisplatin (CisGem) serving as the standard first-line regimen (2). Although it can transiently delay disease progression, primary resistance is prevalent, resulting in low response rates and limited survival benefits. The core issue of this clinical dilemma lies in the lack of effective tools to identify patient subgroups who may benefit from standardized chemotherapy at an early stage of treatment.

The inflammatory regulatory network within the tumor microenvironment may hold the key to deciphering drug resistance mechanisms (3). Pathological studies have demonstrated a dynamic interplay between infiltrating immune cells in CCA tissues and chemotherapy response. Neutrophils directly interfere with drug activity by releasing toxic mediators, while lymphocyte depletion weakens immune surveillance functions, collectively establishing a microenvironmental barrier to chemotherapy resistance (4). This inflammation-driven drug-resistant phenotype is not confined to local lesions; elevated systemic inflammatory burden is also closely associated with poor prognosis. Cross-cancer evidence supports this hypothesis, as abnormal fluctuations in peripheral blood inflammatory markers have been prospectively demonstrated to predict chemotherapy failure in pancreatic and liver cancers (5), indicating its potential universal applicability.

In recent years, predictive models based on peripheral blood inflammatory markers have demonstrated significant value in tumor treatment response evaluation due to their accessibility and low cost (6). This study focuses on indicators such as carbohydrate antigen 19-9 (CA19-9), albumin (ALB), and systemic immune-inflammation index (SII). CA19-9, as a key biomarker for CCA, not only reflects tumor burden but is also closely related to immune suppression mediated by M2 macrophage polarization in the tumor microenvironment. ALB levels directly reflect the imbalance in systemic nutrition-inflammation status and can influence chemotherapy sensitivity by regulating the hepatic nuclear factor 4α (HNF-4α). SII, which integrates the interactions among neutrophils, platelets, and lymphocytes, has been shown to characterize the dynamic evolution of the inflammation-immune network in the tumor microenvironment.

Currently, research on drug resistance in CCA primarily focuses on the detection of genetic mutations. Although this approach has made some progress in exploring mechanisms, its clinical translation is hindered by technical complexity and cost constraints. In contrast, the assessment of inflammatory markers based on routine peripheral blood tests offers advantages in immediacy and accessibility, particularly suitable for resource-limited healthcare settings. This study aims to construct a CisGem resistance risk stratification tool for patients with advanced CCA by integrating systemic inflammatory parameters with clinical characteristics, providing a new pathway for optimizing individualized treatment strategies. The development of such a tool is expected to improve the prediction and management of chemotherapy resistance in current clinical practice. The secondary objective is to validate the prognostic value of the resistant state on survival, using survival analysis to clarify the overall survival (OS) differences between the resistant and non-resistant groups, providing clinical endpoint-level evidence for the model’s predictive efficacy. To enable early clinical intervention, this study aims to develop a risk scoring model based on low-cost indicators, with the ultimate goal of embedding it into future multicenter trial designs to achieve early resistance screening and personalized treatment optimization. We present this article in accordance with the TRIPOD reporting checklist (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-336/rc).

Methods

Study design and patient cohort

This study is a single-center retrospective cohort study, approved by the Ethics Committee of Xinchang Guanghe Hospital (No. 2025041801), with a waiver of informed consent, conducted in December 2022. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. According to the World Health Organization (WHO) classification of digestive system tumors, intrahepatic cholangiocarcinoma (iCCA) is defined as an epithelial-derived malignant tumor originating from the branches above the secondary bile ducts within the liver, and is confirmed by pathological review. The training set included eligible patients from January 2018 to December 2021, while the validation set consisted of consecutively enrolled patients from January 2022 to December 2022.

The inclusion criteria for this study were: histopathologically confirmed unresectable stage IIIB–IV iCCA [American Joint Committee on Cancer (AJCC) 8th edition staging]; first-line systemic chemotherapy with a CisGem-containing regimen; complete baseline imaging data with at least one measurable target lesion (longest diameter of solid tumor ≥10 mm); and completion of blood routine, liver and kidney function, and tumor marker (CA19-9) tests within one week prior to treatment. Exclusion criteria included: co-existing other subtypes of CCA (hilar or distal CCA) or mixed hepatocellular carcinoma; previous or concurrent other malignancies; active inflammatory bowel disease, autoimmune diseases, or receipt of immunomodulatory agents/glucocorticoids (prednisone equivalent dose >10 mg/day) within 4 weeks prior to treatment; active infections; Child-Pugh class C liver dysfunction or estimated glomerular filtration rate <30 mL/min/1.73 m²; and missing clinical data exceeding 20%.

Patient baseline data were extracted from the electronic medical record system, including: demographic characteristics: age, gender, smoking history, alcohol consumption history, hypertension history, diabetes history; clinical parameters: Eastern Cooperative Oncology Group Performance Status (ECOG PS) score, baseline metastatic sites (none/intrahepatic/extrahepatic), CA19-9 level (cut-off value ≥37 U/mL); liver function indicators: ALB, total bilirubin, alanine aminotransferase (ALT).

Inflammatory markers were calculated based on peripheral venous blood samples collected within one week prior to treatment: neutrophil-to-lymphocyte ratio (NLR) = absolute neutrophil count / absolute lymphocyte count; platelet-to-lymphocyte ratio (PLR) = platelet count / absolute lymphocyte count; SII = platelet count × absolute neutrophil count / absolute lymphocyte count. Blood tests were performed using the Sysmex XN-9000 automated analyzer (Sysmex Corporation, Kobe, Japan), with quality control in compliance with ISO 15189 standards. For patients with non-informative CA19-9 levels (e.g., Lewis antigen-negative individuals constituting approximately 5–10% of iCCA patients), the model maintains predictive utility through ALB and SII indicators. This multi-dimensional design ensures broader applicability across diverse patient populations.

Efficacy evaluation and drug resistance determination

According to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1, the efficacy after two treatment cycles was evaluated by two independent radiologists. Primary resistance was defined as an increase of ≥20% in the sum of the longest diameters of target lesions compared to baseline or the appearance of new lesions; the sensitive group included complete response (CR), partial response and stable disease. In case of inconsistent evaluation results, a third senior physician would arbitrate. In this study, patients were followed up for at least half a year, with OS defined as the interval from diagnosis to death from any cause.

Statistical analysis

The sample size calculation for this study is based on the core objective of developing a predictive model. According to the multivariable prediction model sample size criteria proposed by Riley et al. (7), two key conditions must be met: (I) the number of events per variable (EPV) should be at least 10 times the number of parameters to be estimated; (II) the total sample size should ensure that the model does not overfit. The final model in this study includes three predictors (CA19-9, ALB, SII). Using the number of resistance events in the training set (n=23) as the EPV reference, the EPV is calculated as 23/3≈7.7, which is slightly lower than the ideal value of 10. To compensate for this limitation, we adopted the following strategies: (I) reducing the initial variable set from 9 to 3 using least absolute shrinkage and selection operator (LASSO) regression to simplify the model; (II) performing external validation with an independent validation set (n=30, including 9 resistance events).

The training set data were used for model construction. Baseline variables were compared using t-tests, Mann-Whitney U tests, and χ² tests. The LASSO regression was employed to address multicollinearity and select variables. Finally, a multivariate logistic regression was utilized to build a joint prediction model. The model performance was evaluated using the receiver operating characteristic (ROC) curve, calculating the area under the curve (AUC), sensitivity, specificity, and Youden index. The validation set was used for external validation, assessing the clinical applicability of the model through calibration curves and decision curve analysis (DCA). Survival analysis was conducted using the Kaplan-Meier method to plot survival curves, with differences between groups assessed using the log-rank test. The hazard ratio (HR) and 95% confidence intervals (CIs) were calculated using the Cox proportional hazards model. Statistical analysis was performed using R 4.2.1 software (R Core Team, Vienna, Austria), with the significance threshold set at a two-sided P<0.05.

Results

Characteristics of the study cohort

This study included a total of 80 patients with intrahepatic CCA in the training set, among whom 23 (28.75%) were in the drug-resistant group and 57 (71.25%) were in the non-drug-resistant group, as shown in Table 1. There were no statistically significant differences between the two groups in baseline characteristics such as age, gender, ECOG PS score, history of hypertension, diabetes, smoking, alcohol consumption, tumor metastasis, and tumor staging (P>0.05), indicating that the two groups were comparable in these baseline characteristics.

In terms of inflammatory indicators, there were significant differences between the two groups. The median (interquartile range) of CA19-9 levels in the drug-resistant group was 79.00 (68.00, 92.50) U/mL, significantly higher than that in the non-drug-resistant group, which was 53.00 (38.00, 69.00) U/mL (Z=3.567, P<0.001). The ALB levels in the drug-resistant and non-drug-resistant groups were 3.43±0.35 g/dL and 3.80±0.32 g/dL, respectively (t=−4.506 P<0.001). The NLR in the drug-resistant and non-drug-resistant groups was 3.31±0.92 and 2.81±0.84, respectively (t=2.340, P=0.02). The SII in the resistant group and non-resistant group was 533.35±157.44 and 456.28±105.67, respectively (t=2.161, P=0.04). The lymphocyte-to-monocyte ratio (LMR) in the resistant group and non-resistant group was 2.71±0.53 and 2.17±0.64, respectively (t=3.563, P=0.001). There was no statistically significant difference in PLR between the two groups (t=−1.471, P=0.15).

Survival analysis

The Kaplan-Meier survival curve is shown in Figure 1, with the median OS in the primary resistance group being significantly shorter than that in the sensitive group (10.5 vs. 11.9 months, HR: 2.03, 95% CI: 1.01–4.07, log-rank P=0.045). This finding aligns with established literature indicating poor prognosis in advanced iCCA, particularly for primary resistant cases. As of December 2024, most patients have reached ≥12 months of follow-up, with extended follow-up ongoing to verify the stability of survival outcomes.

Figure 1 Kaplan-Meier curve of OS in the training set according to resistance status. Hazard ratio (95% CI) =2.03 (1.01–4.07), P=0.045. CI, confidence interval; OS, overall survival.

Analysis of factors related to drug resistance

Variables were screened using LASSO regression, and through 10-fold cross-validation, the optimal lambda was ultimately determined to be 0.035. Based on this, 9 independent variables were selected, namely: ALB (g/dL), SII, diabetes, hypertension, smoking, sex, CA19-9 (U/mL), PLR, and age. The coefficients of each variable are detailed in Table 2, and the LASSO selection curves are shown in Figures 2,3.

Figure 2 LASSO regression coefficient curve. The x-axis λ is log-transformed with base 10. LASSO, least absolute shrinkage and selection operator.

Figure 3 The relationship curve between lambda and residuals. The x-axis λ is log-transformed with base 10.

Construction and validation of the prediction model

Based on the variables selected through LASSO regression, univariate Logistic regression was performed, followed by multivariate Logistic regression using the forward stepwise method to construct a multivariate Logistic predictive model, as shown in Table 3 below.

A multifactorial predictive model incorporating CA19-9, ALB, and SII was constructed, and the following formula was used to calculate the risk score: Risk Score = 7.52759462 + (CA19-9 × 0.03439253) + (ALB × −2.94417067) + (SII × 0.00014157).

The model demonstrated excellent predictive performance in the training set, as shown in Figure 4, with an area under the ROC curve (AUC) of 0.823 (95% CI: 0.723–0.923), sensitivity of 91.3%, specificity of 63.2%, and Youden index of 0.545. To investigate the calibration of the combined model, a calibration curve was plotted. The likelihood ratio test of the model showed a chi-squared value of 24.136, with P<0.05, indicating that at least one variable in the model was statistically significant and the model was overall effective. The C-index was 0.823, with P<0.05, indicating that the model had moderate discriminative ability. The Hosmer-Lemeshow goodness-of-fit test yielded a chi-squared value of 7.79, with a P value of 0.46, indicating no significant difference between the model’s predicted values and the actual observed values, demonstrating good calibration. The DCA indicates that the curve formed by this model is significantly distant from the curves of the two extreme scenarios, suggesting that the clinical net benefit obtained by this model within a certain threshold range is higher than the clinical benefits generated under the two extreme conditions. This demonstrates that the model has certain clinical application value.

Figure 4 ROC curve (A), calibration curve (B), and clinical decision curve (C) of the prediction model based on the training set. The chi-squared value was 24.136 (P<0.05). The C-index was 0.823 (P<0.05). AUC, area under the curve; CI, confidence interval; FPR, false positive rate; ROC, receiver operating characteristic; TPR, true positive rate.

Clinically, CA19-9, ALB, and SII are routinely tested before treatment initiation. Physicians can directly interpret these values from standard blood tests at admission, apply the model formula for initial risk stratification, and assist in identifying patients likely to develop primary resistance.

Clinical utility validation of the predictive model

The basic information of the validation cohort is shown in Table 4.

In the validation set, as shown in Figure 5, the AUC of the ROC was 0.910 (95% CI: 0.807–1.000), with a sensitivity of 100.0%, a specificity of 81.0%, and a Youden index of 0.810. The likelihood ratio test of the calibration curve model showed a chi-squared value of 18.706, with P<0.05, indicating that at least one variable in the model was statistically significant and the model was overall effective; the C-index was 0.910, with P<0.05, indicating that the model had moderate discriminative ability; the chi-squared value of the Hosmer-Lemeshow goodness-of-fit test was 4.322, with a P value of 0.83, indicating no significant difference between the model’s predicted values and the actual observed values, and the calibration was good. The DCA shows that the curve formed by this model is distant from the curves of the two extreme scenarios, indicating that the clinical net benefit obtained by this model within a certain threshold range is higher than the clinical benefits generated by the two extreme scenarios, suggesting that this model has certain clinical application value. This model utilizes routine pre-treatment blood markers (CA19-9, ALB, SII) to calculate individual risk scores, which can be implemented through standardized electronic forms or embedded within electronic medical record systems. At a risk score threshold of≥0.6, patients are identified as having high primary resistance risk. We recommend early efficacy assessment or more aggressive treatment strategies before the first radiographic evaluation for these high-risk patients. Future implementations should integrate the model into clinical decision support systems via automated interfaces to enable “real-time scoring and stratification”.

Figure 5 ROC curve (A), calibration curve (B), and clinical decision curve (C) of the prediction model based on the validation set. The C-index was 0.910 (P<0.05). The chi-squared value of the Hosmer-Lemeshow goodness-of-fit test was 4.322 (P=0.83). AUC, area under the curve; CI, confidence interval; FPR, false positive rate; ROC, receiver operating characteristic; TPR, true positive rate.

Discussion

This study successfully constructed a quantitative model for predicting the risk of primary resistance to the gemcitabine-cisplatin regimen in iCCA patients by integrating peripheral blood inflammatory markers with clinical characteristics. The results demonstrated that the combined model of CA19-9, ALB, and SII exhibited excellent discriminative and calibration capabilities in both the training set (AUC: 0.823) and the validation set (AUC: 0.910). This tool, based on routine blood tests, provides an efficient and low-cost technical pathway for the early clinical identification of high-risk populations for drug resistance, with significant translational potential. Survival analysis also confirmed that primary resistance is an independent risk factor for survival prognosis (HR: 2.03, 95% CI: 1.01–4.07). The median OS in the resistant group was significantly shorter than that in the sensitive group (10.5 vs. 11.9 months, P=0.045). This result validates the biological significance of the model’s predictions at the clinical endpoint level—specifically, that high-risk patients identified by the model require more aggressive therapeutic interventions to improve survival. Although the absolute difference in median OS was modest (10.5 vs. 11.9 months), this statistically significant divergence (HR: 2.03, P=0.045) is clinically meaningful given the poor prognosis of advanced iCCA. More importantly, it demonstrates the model’s capacity to reflect long-term prognostic risk, which is particularly valuable for treatment planning.

The significant elevation of CA19-9 (median 79.00 U/mL in the resistant group vs. 53.00 U/mL in the non-resistant group, P<0.001) not only reflects the increased tumor burden but also indicates the systemic activation of the inflammatory network within the tumor microenvironment. Previous studies have pointed out that the abnormal expression of CA19-9 is closely associated with M2-type macrophage polarization (7), which drives epithelial-mesenchymal transition through the secretion of IL-6 and IL-8, enhancing the efflux capacity of tumor cells against chemotherapeutic drugs. Therefore, dynamic monitoring of CA19-9 can serve as an important supplementary indicator for assessing treatment response (8). The elevation of SII and NLR (SII: 533.35±157.44 in the drug-resistant group vs. 456.28±105.67 in the non-drug-resistant group, P=0.04) reveals the synergistic role of neutrophils and platelets in the drug-resistant microenvironment (9). Neutrophils directly degrade the active components of chemotherapeutic drugs by releasing myeloperoxidase (10), while inducing the infiltration of regulatory T cells, thereby weakening immune surveillance function. Platelets promote angiogenesis by releasing platelet-derived growth factor and transforming growth factor-beta, and activate the integrin β3 signaling pathway, further enhancing the drug efflux capacity of tumor cells. This dynamic interaction within the inflammation-immune network constitutes an important biological basis for CisGem resistance. The significant decrease in ALB levels (resistant group 3.43±0.35 vs. non-resistant group 3.80±0.32 g/dL, P<0.001) suggests the impact of systemic inflammation-nutritional imbalance on chemotherapy sensitivity (11). The low ALB state may promote metabolic reprogramming in tumor cells by inhibiting the expression of HNF-4α (12), thereby conferring a survival advantage under chemotherapy pressure (13). In addition, the reduction in ALB may form a positive feedback loop with the activation of hypoxia-inducible factor 1α in the microenvironment, jointly maintaining the stability of the drug-resistant phenotype. Conceptually, this represents a ‘composite biological risk model’ integrating three complementary dimensions: CA19-9 (tumor burden), ALB (systemic nutrition-inflammation balance), and SII (inflammatory-immune network). Their synergistic integration captures multidimensional biological mechanisms, providing enhanced predictive stability compared to single-parameter approaches.

This model, validated by DCA, demonstrates significantly higher clinical net benefit compared to extreme strategies within a certain range of drug resistance probability thresholds. This characteristic enables it to optimize the allocation of medical resources, guiding clinicians to prioritize the gemcitabine plus platinum regimen for patients with potential benefits while reserving a window period for second-line treatment for patients at high risk of drug resistance. Moreover, the biomarkers required by the model (CA19-9, ALB, SII) are cost-effective, reducing detection costs by more than 90% compared to genomic testing, and are rapid, making them particularly suitable for primary medical institutions with limited resources. For immediate clinical implementation, physicians can calculate the risk score using pretreatment blood test results (CA19-9, ALB, SII) through the provided formula during initial evaluation. This enables same-day risk stratification before chemotherapy initiation, allowing timely consideration of alternative regimens for high-risk patients. Critically, the model’s multi-parameter design provides redundancy for CA19-9 non-expressors (e.g., Lewis-negative patients). In such cases, ALB and SII—which reflect fundamental inflammatory-nutritional and immune-microenvironment status—maintain discriminative power. This significantly expands the applicable population beyond CA19-9-dependent prediction systems. This approach is expected to promote the homogeneous development of iCCA diagnosis and treatment globally.

However, this study is a single-center retrospective design, and although time bias was partially corrected through the validation set, multi-center external validation remains a necessary next step. Current follow-up extends through December 2024, with most patients now beyond 12-month observation. We are actively conducting extended survival tracking to validate the stability of OS outcomes. Critically, this model is not intended to replace genomic approaches but provides an alternative strategy for patients lacking access to molecular profiling or targeted sequencing. In clinical practice, these approaches should be complementary: our model serves as a rapid screening tool to identify high-risk patients warranting further molecular investigation. Additionally, the current model is optimized for the CisGem regimen, and its generalizability across different chemotherapy regimens requires further verification. Notably, this model is not intended to replace genomic analysis, but serves as a complementary tool—particularly in resource-limited settings—to provide a transitional or pilot approach for precision medicine. We further recommend integrating it with genomic profiling in future studies to construct a multidimensional prediction system. Due to limitations in follow-up duration and cohort size, this study could not assess the relationship between the model and the duration of objective response. Future prospective studies will validate its predictive capacity for long-term safety and durable treatment responses. Future research could integrate dynamic monitoring of inflammatory markers with machine learning algorithms to build adaptive predictive models that capture critical points in the evolution of drug resistance. By combining genomics or metabolomics data (14), a multi-dimensional drug resistance prediction framework can be refined, and prospective intervention trials can be conducted to evaluate the improvement in patient survival through model-guided personalized treatment. This study has several important limitations. As a single-center study, it lacks representation of racial and geographic diversity. Future multicenter validation should include diverse populations, regions, and healthcare resource settings to assess cross-population stability. Additionally, incorporating variables like race and metabolic profiles could enhance the model’s generalizability within multidimensional frameworks. Given the anticipated expansion of genomic applications in oncology, future studies should prioritize integrating this model with genomic data to develop multi-omics prediction frameworks. Combining accessible inflammatory markers with molecular profiling may achieve synergistic predictive accuracy for individualized therapy. Future studies should expand the cohort size to specifically investigate the “long-term survivor” subgroup (e.g., OS >18 months). Characterizing the inflammatory profiles and clinical features of these exceptional responders will enable refinement of the current model into a more clinically transformative survival prediction tool that identifies patients most likely to derive sustained benefit from therapy.

Conclusions

In summary, the predictive model established in this study provides an innovative tool for the early assessment of gemcitabine plus platinum resistance risk in iCCA patients, demonstrating significant advantages in statistical efficacy, clinical applicability, and cost-effectiveness. With multicenter validation and the integration of multi-omics data, this model is expected to become a standard risk assessment tool for systemic therapy in CCA, highlighting its potential application prospects that require prospective validation.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-336/rc

Data Sharing Statement: Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-336/dss

Peer Review File: Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-336/prf

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2025-336/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Xinchang Guanghe Hospital (No. 2025041801) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin 2024;74:12-49. [Crossref] [PubMed]
2. Valle J, Wasan H, Palmer DH, et al. Cisplatin plus gemcitabine versus gemcitabine for biliary tract cancer. N Engl J Med 2010;362:1273-81. [Crossref] [PubMed]
3. Naro C, Ruta V, Sette C. Splicing dysregulation: hallmark and therapeutic opportunity in pancreatic cancer. Trends Mol Med 2025;31:610-24. [Crossref] [PubMed]
4. Wang T, Xu C, Zhang Z, et al. Cellular heterogeneity and transcriptomic profiles during intrahepatic cholangiocarcinoma initiation and progression. Hepatology 2022;76:1302-17. [Crossref] [PubMed]
5. Rao H, Guo Z, Wen X, et al. Case Report: Immune checkpoint inhibitor-related vitiligo-like depigmentation in non-melanoma advanced cancer: A report of three cases and a pooled analysis of individual patient data. Front Oncol 2022;12:1099108. [Crossref] [PubMed]
6. Goswami M, Toney NJ, Pitts SC, et al. Peripheral immune biomarkers for immune checkpoint inhibition of solid tumours. Clin Transl Med 2024;14:e1814. [Crossref] [PubMed]
7. Riley RD, Ensor J, Snell KIE, et al. Calculating the sample size required for developing a clinical prediction model. BMJ 2020;368:m441. [Crossref] [PubMed]
8. Nishida N. Biomarkers and Management of Cholangiocarcinoma: Unveiling New Horizons for Precision Therapy. Cancers (Basel) 2025;17:1243. [Crossref] [PubMed]
9. Bai S, Shi X, Dai Y, et al. The preoperative scoring system combining neutrophil/lymphocyte ratio and CA19-9 predicts the long-term prognosis of intrahepatic cholangiocarcinoma patients undergoing curative liver resection. BMC Cancer 2024;24:1106. [Crossref] [PubMed]
10. Jin J, Mou H, Zhou Y, et al. Nomogram for Predicting Survival Post-Immune Therapy in Cholangiocarcinoma Based on Inflammatory Biomarkers. Cancer Control 2024;31:10732748241305237. [Crossref] [PubMed]
11. Curran CS, Cui X, Li Y, et al. Meta-analysis of niacin and NAD metabolite treatment in infectious disease animal studies suggests benefit but requires confirmation in clinically relevant models. Sci Rep 2025;15:12621. [Crossref] [PubMed]
12. Feng Y, Zhao M, Wang L, et al. The heterogeneity of signaling pathways and drug responses in intrahepatic cholangiocarcinoma with distinct genetic mutations. Cell Death Dis 2024;15:34. [Crossref] [PubMed]
13. Xu Y, Wang Y, Qi R, et al. Role of connexin 32 in the directional differentiation of induced pluripotent stem cells into hepatocytes. Int J Med Sci 2024;21:508-18. [Crossref] [PubMed]
14. Peng Z, Cao G, Hou Q, et al. The Comprehensive Analysis of Efficacy and Safety of CalliSpheres(®) Drug-Eluting Beads Transarterial Chemoembolization in 367 Liver Cancer Patients: A Multiple-Center, Cohort Study. Oncol Res 2020;28:249-71. [Crossref] [PubMed]