Prognostic models for mucinous and non-specific adeno cholangiocarcinoma: a population-based retrospective study

Background Clinically, the diagnosis and treatment of cholangiocarcinoma are generally different according to the location of occurrence, and the studies rarely consider the differences between different pathological types. Cholangiocarcinomas in large- and middle-sized intrahepatic bile ducts are mostly mucinous, while in small sized bile duct are not; mucinous extrahepatic cholangiocarcinomas are also more common than mucinous intrahepatic cholangiocarcinoma. However, it is unclear whether these pathological type differences are related to the prognosis. Methods Data of total 22509 patients was analyzed from Surveillance, Epidemiology, and End Results program database out of which 22299 patients were diagnosed with common adeno cholangiocarcinoma while 210 were diagnosed with mucinous cholangiocarcinoma. Based on the propensity score matching (PSM) analysis, between these two groups’ clinical, demographic, and therapeutic features were contrasted. The data were analyzed using Cox and LASSO regression analysis and Kaplan-Meier survival curves. Ultimately, overall survival (OS) and cancer specific survival (CSS) related prognostic models were established and validated in test and external datasets and nomograms were created to forecast these patients’ prognosis. Results There was no difference in prognosis between mucinous cholangiocarcinoma and adeno cholangiocarcinoma. Therefore, we constructed prognostic model and nomogram that can be used for mucinous and adeno cholangiocarcinoma at the same time. By comparing the 9 independent key characteristics i.e. Age, tumor size, the number of primary tumors, AJCC stage, Grade, lymph node status, metastasis, surgery and chemotherapy, risk scores were calculated for each individual. By integrating these two pathological types in OS and CSS prognostic models, effective prognosis prediction results could be achieved in multiple datasets (OS: AUC 0.70–0.87; CSS: AUC 0.74–0.89). Conclusion Age, tumor size, the number of primary tumors, AJCC stage, Grade, lymph node status, metastasis, surgery and chemotherapy are the independent prognostic factors in OS or CSS of the patients with mucinous and ordinary cholangiocarcinoma. Nomogram that can be used for mucinous and adeno cholangiocarcinoma at the same time is of significance in clinical practice and management of cholangiocarcinoma.


Introduction
Cholangiocarcinomas are a vast group of malignancies which are believed to have their origin in the biliary tract epithelium, either inside the liver or the biliary tract.These malignancies are usually hard to diagnose, their pathogenesis is not very well understood yet, their poor prognosis has been the main reason that their management has taken a nihilistic turn (1,2).Cholangiocarcinoma is the collective term used for malignancies arising within the intrahepatic and extrahepatic biliary tract.A large proportion (90%) of cholangiocarcinomas are adenocarcinomas, with various already reported histological variants, including adenocarcinoma, papillary adenocarcinoma, intestinal-type adenocarcinoma, and mucinous adenocarcinoma (3).This huge difference of number lead cholangiocarcinomas to be classified broadly into two groups adenocarcinomas and 'rare variants' (4).Mucinous carcinoma presents with the presence of large extracellular mucus lakes which have floating cancer cells, accounting for more than half of neoplasm (5).Regardless of kind, it is unquestionably acknowledged that cholangiocarcinoma is a fatal cancer that typically manifests late, is famously difficult to identify, and is associated with a very high fatality rate.The occurrence of intrahepatic cholangiocarcinoma is increasing around the globe (6)(7)(8)(9)(10).The reason for this increase is not clear, although it could be related to an interplay between predisposing genetic factors and environmental triggers.MRI and CT with endoscopic ultrasound and PET has made it easier to get the important diagnostic information in certain group of patients.Surgical resection of the tumor is the only chance for cure, with results depending on personal surgical skills of the surgeon and patient selection.Studies have reported that liver transplantation could be a choice for long-term survival in selected patients when combined with neoadjuvant chemoradiotherapy.Chemotherapy and radiotherapy have been ineffective for patients which had unresectable tumors (11).As the global incidence has remarkably increased, mortality is high, treatment options are very few, prognosis is poor and overall survival is not promising (12, 13), clinical practice is truly in need of a method to calculate the survival probability of these patients based on different prognostic factors.Analyzing mucinous and non-specific adenocarcinoma subtypes within cholangiocarcinoma is crucial for comprehensively understanding the disease spectrum, guiding clinical decision-making, and improving patient outcomes.This approach underscores the importance of a multidisciplinary approach integrating histopathological, molecular, and clinical data to refine disease classification and treatment strategies.But unfortunately, currently available studies focused on limited types and didn't cover the rare variants.So our study focused on analyzing and establishing a prognostic model which will give insights about both adenocarcinomas and mucinous cholangiocarcinomas.
Developed in the United States, the Surveillance, Epidemiology and End Results (SEER) database contains population-based clinical survival information from registries that cover 34.6% of the country's population.We therefore concentrated on creating a predictive nomogram for adenocarcinomas as well as an uncommon variation, mucinous cholangiocarcinoma, in this work based on statistics from the SEER database.

Data source and case selection
The National Cancer Institute's SEER cancer database (http:// www.seer.cancer.gov),which was released in November 2021, was used to extract information about patients with cholangiocarcinoma (version 8.4.0.1).In the United States, 17 population-based cancer registries provide information on cancer that is gathered by the SEER program.Because the data in the SEER database was de-identified and tagged for public availability, the data from SEER database was relieved from the necessity to get ethical approval from the Second Xiangya Hospital of Central South University Evaluation Board.
Using the use of the site-specific International Classification of Oncological Diseases 3 (ICD-O-3) code, C221, C240, C248, C249, cholangiocarcinoma in the SEER program was recognized from 1998 to 2016.The diagnosis of adenocarcinoma or cholangiocarcinoma (hereinafter referred to as adeno cholangiocarcinoma) was made using the ICD-O-3 codes 8140/3 and 8160/3, while the diagnosis of mucinous cholangiocarcinoma was made using the ICD-O-3 codes 8453/3 (intraductal papillary-mucinous carcinoma), 8470/3 (mucinous cystadenocarcinoma), 8471/3(papillary mucinous cystadenocarcinoma), 8480/3(mucinous adenocarcinoma), 8481/3 (mucin-producing adenocarcinoma).Using the sequence number for a single primary tumor or the first of two or more primaries, the number of primary tumors was determined.The following were the exclusion criteria: (1) the tumor was not primary; and (2) the case lacked full follow-up information or household income.The final group of patients included 22299 with adeno cholangiocarcinoma and 210 with mucinous cholangiocarcinoma.The standards of the SEER database coding and staging manual were followed for creating the codes for case collection (Figure 1).

Clinical information acquisition and processing
Demographic data (such as patient ID and age), tumor characteristics (such as tumor size, the number of primary tumors, AJCC stage group, tumor grade, metastasis status, and lymph node status), treatment data (such as surgery, radiotherapy, and chemotherapy status), and follow-up for survival were all extracted from the SEER database (survival months, cause-specific death, and vital status).
Age was used as a demographic indicator (≤62 years, 63-75 years, and ≥76 years).Tumor size (≤29mm, 30mm-75mm, ≥76mm, and unknown), the number of primary tumors (one primary only versus the first of two or more primaries), the AJCC stage (I/II, III/ IV, or unknown), the grade (I/II, III/IV, or unknown), the status of the lymph nodes (negative, positive, or unknown), and metastasis were all noted as characteristics of the tumor (M0, M1, Mx or unknown).Surgery (yes or no) and chemotherapy (yes or no/ unknown) were the two types of treatment.

Survival analysis before and after the propensity score matching
Propensity score matching (PSM) was used to modify the baseline characteristics of patients with mucinous and adeno cholangiocarcinoma because the clinical characteristics between cases with mucinous and adeno cholangiocarcinoma in the SEER database were heterogeneous via the analysis of Chi-square or The time from diagnosis to death from any cause is known as overall survival (OS).The time between a cholangiocarcinoma diagnosis and death was called "cancer-specific survival" (CSS).These two indices served as the study's result endpoints.The Kaplan-Meier analysis was used to create the survival plot, and the log-rank test was used to compare patients with mucinous and adeno cholangiocarcinoma before and after PSM.

COX and LASSO regression for prognostic models and validation
The possible prognostic factors on the OS and CSS of patients with cholangiocarcinoma were identified using univariate and multivariate Cox regression models.
Least absolute shrinkage and selection operator (LASSO)-COX analyses were conducted to identify the ideal weighting coefficients for these features and construct a model to test whether various pathological types and clinical traits can collectively predict the prognosis of patients with mucinous and adeno cholangiocarcinoma after PSM.Using the R software's 'glmnet' package, ten-fold crossvalidation was used to create LASSO-COX regression models for OS and CSS.Moreover, in OS and CSS, the lambda parameter's ideal values were 0.29 and 0.022, respectively.
Using the 'pROC' package in R, ROC analysis of the follow-up outcomes and risk scores over ten years was conducted based on the aforementioned model, and the area under the curve (AUC) and confidence interval were analyzed (CI).Patients were divided into high-and low-risk groups based on the ideal cutoff or median of risk scores, and the prognostic differences between the two groups were further evaluated using the "survival" software.The log-rank test was used to determine the significance of the prognostic difference between the two groups.
As for validation, we first validated the model by using PSM excluded samples in the SEER database (n=22097).Furthermore, the study validated the OS and CSS prognostic models through an external validation cohort, which included 94 patients with cholangiocarcinoma from the Second Xiangya Hospital of Central South University.This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Xiangya Second Hospital, Central South University (Changsha, China) with the informed consent of all patients.All patients were divided into high-risk and low-risk groups based on the optimal cutoff risk score calculated by the prognostic models.The Kaplan-Meier analysis and the log-rank test were used to examine differences in survival between the two groups.The ROC was used to determine the accuracy and discriminability of the prediction model.

Nomogram
Finally, the R 'rms' package was used to incorporate the survival data from the LASSO-COX analysis to create nomograms that predicted the OS and CSS of patients at 1, 3, and 5 years.A nomogram quickly determines the risk of sickness or the likelihood that an individual will survive by integrating numerous predictors and showing multiple lines to scale.The C-index is used to gauge the power of the nomogram.

Comparison of the baseline clinical characteristics between mucinous and adeno cholangiocarcinoma
A total of 22509 patients including 22299 with adeno cholangiocarcinoma and 210 with mucinous cholangiocarcinoma in the SEER database, were enrolled in the study (Figure 1).The differences in most of the baseline clinical characteristics between mucinous and adeno cholangiocarcinoma e.g.age, year of diagnosis, LNR, LODDS, tumor size, sequence number, summary stage, AJCC stage group were statistically significant (p <0.05) (Table 1).Surgery status, chemotherapy, total number of tumors, race, median house hold income and marital of the patients with mucinous cholangiocarcinoma were similar to those with adeno cholangiocarcinoma (P>0.05).

Propensity score matching and survival analysis
The PSM method was used to balance the baseline clinical characteristics between patients with mucinous and adeno cholangiocarcinoma (all standard deviations ≤ 0.05; Table 1).Patients were first classified into AD and MUC patient groups, and then logistic regression was used to calculate the probability of MUC for each patient with MUC as the dependent variable (Y) and all other known clinical characteristics as the independent variable (X).We then matched the experimental and control groups according to probability.In the actual software operation process, you need to provide a caliper value (caliper), caliper value is the experimental group and the control group in the matching, the probability of the allowable error, the caliper value of this study is 0.05.MUC and AD were paired according to this caliper value range in a 1: 1 ratio.In the last resort, 412 patients (mucinous group versus adeno group = 1: 1) were included in the following analysis after PSM.
Patients with mucinous and adeno cholangiocarcinoma preand post-PSM were appraised by Kaplan-Meier analysis.A total of 22509 patients including 22299 with adeno cholangiocarcinoma and 210 with mucinous cholangiocarcinoma were enrolled in the analysis before PSM.The median OS was 2.0 months in the mucinous and adeno group, while the median CSS was 2 months in the adeno group and 3.0 months in the mucinous group.So, no significant difference in outcome was observed in both groups.A similar result could be obtained for patients with mucinous and adeno cholangiocarcinoma after PSM.The median OS was 2.0 months in the mucinous and adeno group, while the median CSS was 2.0 and 3.0 months in the adeno group and mucinous group, respectively (Figures 3A-D) (1-year OS rates: 34.86% versus 35.51%, 3-year OS rates: 10.70% versus 15.06%, and 5-year OS rates: 5.79% versus 9.66%; 1-year CSS rates: 43.04% versus 42.46%, 3-year CSS rates: 16.14% versus 20.56%, and 5-year CSS rates: 10.46% versus 14.71%, after PSM, p > 0.05).

Univariate and multivariate analysis
Univariate and multivariate Cox regression analyses were performed to sift out the potential clinical characteristics of patients with mucinous cholangiocarcinoma which may influence the prognosis.In univariate regression analysis, age, year of diagnosis, LNR, LODDS, surgery, chemotherapy, tumor size, sequence number, tumor grade, summary stage, AJCC stage group and marital status were prognostic risk factors for both OS and CSS in patients with mucinous cholangiocarcinoma.(p <0.05,Table 2).
In multivariate regression analysis, older age (≥ 76 vs. ≤ 62, OR 1.37, p = 0.04) (where older age was found to be associated with poorer prognosis than younger age), tumor size (≤29mm vs ≥76mm/ unknown OR 1.44 p=0.04) (where larger smaller tumor size was found to be associated with better prognosis than the larger tumor size)were determined as independent characteristics associated with OS of mucin group and year of diagnosis (2008-2013 OR 1.36, p=0.05) was determined as independent characteristic associated with CSS of mucin group.Meanwhile LNR (≤0.1 vs >0.3/unknown OR 4.3, p=0) (where lower LNR was found to be associated with better prognosis than higher LNR), chemotherapy (no chemotherapy/unknown vs. chemotherapy, OR 0.44, p = 0)(where receiving chemotherapy was associated with better prognosis than not receiving chemotherapy), number of primary tumors (one primary only vs. first of 2 or more primaries, OR 2.3, p = 0.007)(where less number of primary tumors was found to be associated with better prognosis than higher number of primary tumors), marital status (single vs married OR 0.5, p=0.002) (where being married was found to be associated with better prognosis than being single), summary stage (localized vs regional/unknown OR 0.17, p=0)(having localized tumor was associated with better prognosis than having regional metastasis) and AJCC stage group were determined as independent characteristics associated with both OS and CSS of mucinous cholangiocarcinoma.It was also noticed that LODDS, surgery, radiation therapy and tumor grade did not directly influence overall survival nor cancer specific survival (p ≥0.05).

Construction of predictive models for OS and CSS
Besides the aforementioned prognostic characteristics (Table 3), it is necessary to assess whether the adeno and mucinous cholangiocarcinoma groups can also predict the prognosis of patients.Therefore, LASSO-COX analysis was performed after univariate COX regression analysis to construct OS (Figures 4A,  B) and CSS (Figures 5A, B) predictive models based on abovementioned factors and groups.After 10-fold cross-validation, the optimal l values 0.029 and 0.022 were obtained in OS and CSS models, respectively.It is worth noting that for non-hierarchical variables with more than 3 categories, these variables need to be converted into dummy variable matrix for analysis.Finally, the group and 9 prognostic factors were sifted for the predictive model of OS, including LNR, surgery status, chemotherapy status, tumor size, sequence number, age, tumor grade, AJCC stage group and metastasis M stage (Figures 4A, B).
Additionally, a survival study based on the risk score was performed on 412 individuals who had had PSM screening.For the OS model and the CSS model, the ideal cutoff value was calculated as 2.5165 and 2.7889, respectively.Based on their cutoff values, the included patients could be divided into high-risk and lowrisk categories.The predictive model of OS was able to identify between patients with favorable or unfavorable prognoses, according to the Kaplan-Meier curve analysis.The high-risk group manifested a shorter OS than the low-risk group (p = 9.9e-22) (Figure 4C; Supplementary Figure 1A).Time-dependent ROC analysis showed that AUC of risk score for the prediction of 1,3,5,7,9-year OS was 0.77, 0.77, 0.80, 0.82 and 0.80, respectively (Figure 4D).
Subsequently, through the same modeling process, 11 prognostic factors were determined in the predictive model of CSS, including year of diagnosis, LNR, chemotherapy status, tumor size, sequence number, marital status, tumor grade, summary stage, AJCC stage group, lymph nodes status and metastasis M stage (Figures 5A, B).The high-risk group likewise manifested a shorter CSS than the lowrisk group in Figure 5C (p = 3.6e-31) and Supplementary Figure 1B (p = 2.7e-23).Time-dependent ROC analysis showed that AUC of risk score for the prediction of 1,3,5,7,9-year CSS was 0.81, 0.82, 0.84, 0.86 and 0.85, respectively (Figure 5D).

Validation of predictive models for OS and CSS
Firstly, validated the models by using samples excluded by PSM in the SEER database.The remaining 22097 patients excluded by PSM were again subjected to KM analysis.The included patients in predictive model of OS could be divided into high-risk (n=10773) and low-risk (n=11324) groups based on the median value of OS risk score.The Kaplan-Meier curve analysis revealed that the prognosis between high-risk group and low-risk group defined by predictive model of OS were significantly different.The high-risk group manifested a shorter OS i.e., 2 months than the low-risk group i.e., 13months (p < 0.0001) (Figure 6A; Supplementary Figure 2A).Furthermore, AUC based on time-dependent ROC of risk score for the prediction of 1,3-and 5-year OS was 0.80, 0.80 and 0.80, respectively (Figure 6B).Similarly, the included patients in predictive model of CSS could be classified into high-risk (n =11017) and low-risk (n =11080) groups through the median value of CSS risk score.Through the product-limited method (KM analysis), the high-risk group manifested a shorter CSS i.e., 3 months than the low-risk group i.e.,15momths (p < 0.0001), defined by the predictive model of CSS (Figure 6C; Supplementary Figure 2B).Time-dependent ROC analysis showed that AUC of risk score for the prediction of 1,3-and 5-year CSS was 0.78, 0.79 and 0.79, respectively (Figure 6D).
Moreover, the external validation cohort comprised 94 patients with cholangiocarcinoma (adeno or mucinous), who were divided into high-risk and low-risk groups based on the optimal cutoff value of the risk score in the validation cohort.The Kaplan-Meier curve and the log-rank test were used to examine differences in survival between the two groups.In both the CSS prediction model and OS prediction model, patients in the low-risk group had a better prognosis than those in the high-risk group in the external validation cohort (OS and CSS: P<0.0001) (Figures 7A, B).Furthermore, the ROC curve showed that the AUC value of the  OS prediction model at 1, 2, and 3 years were 0.87, 0.79, and 0.70, respectively, in the validation cohort (Figure 7C).Similarly, the AUC of the ROC in the CSS prediction model at 1, 2, and 3 years were 0.89, 0.82, and 0.74, respectively (Figure 7D).

Nomograms for OS and CSS using prognostic factors
In our investigation, nomograms were employed to enhance the practicality of the OS (Figure 8) or CSS (Supplementary Figure 3) predictive models by providing more vivid illustrations.Using the aforementioned scale, each characteristic's score was calculated.The final score was determined as the total of these qualities' scores.For patients with adeno and mucinous cholangiocarcinomas, we were able to forecast the prognosis of 1-, 3-, and 5-year OS or CSS using the perpendicular line connecting the total point axis and the two outcomes axis.

Discussion
Cholangiocarcinomas are a vast group of malignancies which are believed to have their origin in the epithelial cells either inside the liver or the biliary tract.These malignancies are usually hard to diagnose, their pathogenesis is not very well explained yet, their poor prognosis has been the main reason that their management has taken a nihilistic turn (1, 2)..A large proportion (90%) of biliary tract carcinoma are adenocarcinomas, with various histological subtypes, including adenocarcinoma (not otherwise specified), intestinal-type adenocarcinoma, papillary adenocarcinoma, and mucinous adenocarcinoma (3,14,15).
In the United States, the Surveillance, Epidemiology and End Results (SEER) database was formed and it contains populationbased clinical survival data from registries that cover 34.6% of the country's population.So, based on SEER database statistics, in this study we focused to develop a prognostic nomogram which can be used widely for adenocarcinomas as well as rare variant i.e., mucinous cholangiocarcinoma.
Previously reported studies have mostly been focusing on location of tumor as N. A. van der Gaag et al., in 2012 introduced a prognostic nomogram for patients which were undergoing resection of extrahepatic cholangiocarcinoma.According to them, tumor location of extrahepatic CCA does not independently predict CSS after resection and constructed a prognostic model based on lymph node status, residual tumor levels of resection margins, and tumor differentiation status, that predicted CSS better than TNM staging (16).But this study is limited only for extrahepatic tumors and did not give any clues if these outcomes apply to different pathological subvariants.Also, Alfredo Guglielmi et al., in 2009  (17).They claimed lymph node status to be the most important prognostic factor after resection.These findings are consistent in our study too such as we also found that lymph node status as our analysis also states that having less number of  Most patients with CCA are diagnosed at advanced phase of the disease and have poor OS (19)(20)(21)(22), however, its therapeutic and diagnostic targets are still unknown.In an effort to find a more useful prospective diagnostic target for CCA, we constructed and validated a prognostic model and nomogram.Initially we screened 22509 CCA patients including 22299 with adeno CCA and 210 with mucinous CCA using SEER database and collected their clinical information including age, year of diagnosis, LNR, LODDS, tumor size, sequence number, summary stage, and AJCC stage group.The baseline clinical characteristics such as total number of tumors, race, median household income, and marital status between mucinous and adeno CCA patients were found to be similar (P>0.05).Then, the PSM analysis was incorporated to omit the bias selection between with mucinous and adeno CCA patients.Meanwhile the Kaplan-Meier survival analysis allowed us to determine the percentage of patients who survive a specific event (23).Before and after the PSM analysis, the patients were assessed for Kaplan-Meier survival analysis and showed no significant difference.The Kaplan-Meier survival analysis prior to the PSM analysis was performed on a total of 22509 patients including 22299 with adeno CCA and 210 with mucinous CCA.We found that both groups (mucinous and adeno CCA) have similar median OS i.e., 2.0 months.However, the median CSS was different in both groups as adeno CCA group (2 months) and mucinous CCA group (3.0 months).Overall, there was a non-significant difference for both OS and CSS rate between adeno CCA and mucinous CCA groups (Figures 3A-D).
To avoid overfitting of the prediction model and to solve clinical decision problems, the established and constructed nomogram's predictive accuracy and predictive validity were thoroughly assessed while finding prognostic markers (24).The relationship between CCA patients' prognostic factors and their relative effects on OS and CSS were determined using univariate and multivariate Cox regression analyses.Meanwhile, univariate and multivariate analysis helps to explore the distribution frequency of an independent key factor between both groups of CCA patients (18).In this study, the univariate regression analysis showed similar prognostic factors for both OS and CSS in mucinous CCA group.However, multivariate regression analysis revealed independent set of prognostic variables as risk factors associated with OS and CSS of mucinous CCA group.In addition, we observed that LODDS, surgery, radiation therapy and tumor grade did not directly influence OS nor cancer specific survival (p ≥ 0.05).In addition, the predictive OS and CSS models were constructed using LASSO-COX analysis.Anyhow, LASSO-COX analysis is a durable tool for data analysis and integrating the significant prognostic factors as well (25).Following the 10-fold cross-validation, optimum l values as 0.029 and 0.022 were obtained in OS and CSS models, respectively.Finally, the group and 9 prognostic factors including LNR, surgery status, chemotherapy status, tumor size, sequence number, age, tumor grade, AJCC stage group and metastasis M stage were determined in the predictive model of OS.Besides, depending upon the risk score, PSM analysis patients (n = 412) were exposed to survival analysis and at standard cut off values given as OS model = 2.5165 and CSS model = 2.7889.This analysis allowed us to segregate the patients into two groups as high-risk and low-risk groups.Meanwhile, The Kaplan-Meier curve analysis revealed that predictive OS model distinguished patients with good or bad prognoses.The predictive model of CSS showed 11 prognostic factors affiliated with the pathogenesis of CCA.Next, to validate the OS and CSS predictive models, we incorporated PSM excluded patients (n=22097) data to survival analysis based upon their risk score.We developed two groups of patients as high-risk (n = 10773) and low-risk (n = 11324) groups based on their median value in OS predictive model.The high-risk group manifested a shorter OS i.e., 2 months than the low-risk group i.e., 13months with significant statistical difference (p < 0.0001) (Figure 6A; Supplementary Figure 2A).Time-dependent ROC analysis showed that AUC of risk score for the prediction of 1,3 and 5year OS was 0.80, 0.80 and 0.80, respectively (Figure 5B).Also, the OS predictive model discriminated the patients with good or bad prognoses.Correspondingly, CSS predictive model also helped us to classify the patients into high-risk (n = 11017) and low-risk (n = 11080) groups based on their median value.The high-risk group manifested a shorter CSS i.e., 3 months than the low-risk group i.e.,15 months (p < 0.0001) (Figure 6C; Supplementary Figure 2B).Time-dependent ROC analysis showed that AUC of risk score for the prediction of 1,3-and 5-year CSS was 0.78, 0.79 and 0.79, respectively (Figure 5D).Here, we examined that the patients in high-risk groups had poor prognosis and shorter OS and CSS than low-risk groups.
While encountering the risk factors permitted us to explore and anticipate the future risk of the disease.To best of our knowledge, we attempt to deliver a significant and more practical prognostic model and nomogram for mucinous and adeno CCA.In conclusion, it is stated that there is no difference in prognosis between mucinous CCA and ordinary CCA.Cumulatively, all prognostic variables such as age tumor size, the number of primary tumors, AJCC stage, Grade, lymph node status, metastasis, surgery and chemotherapy are the independent factors in prognosis and overall survival of the patients with mucinous and ordinary CCA.So, our study will give future insights about the clinical management of adeno and mucinous cholangiocarcinoma and help the clinicians to rely on these prognostic factors while managing these patients and assessing prognosis based on these factors.As recent studies are focusing on finding out prognostic biomarkers for cholangiocarcinoma, it would be a favorable idea to update our model by also including prognostic biomarkers cholangiocarcinoma to further enhance its accuracy and applicability.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material.Further inquiries can be directed to the corresponding authors.The nomogram of OS-associated predictive models.The sum of the scores indicated by the red arrows represents the survival probability corresponding to 1, 3, and 5 years.

FIGURE 1 Study
FIGURE 1Study flow chart.

2
FIGURE 2 Identification of optimal cutoff values for the year of diagnosis, age, LNR, LODDS and tumor size via X-tile software analysis.(A, B) Optimal minimum and maximum cutoff values for age: 62 and 75 years and survival curve.(C, D) Optimal minimum and maximum cutoff values: 0.1~0.3 for LNR and survival curves.(E, F) Optimal minimum and maximum cutoff values: -0.8~-0.2 for LODDS and survival curves.(G, H) Optimal minimum and maximum cutoff values: 29 and 75 mm for tumor size and survival curves.(I, J) Optimal minimum and maximum cutoff values: 2007 and 2013 for the years of diagnosis and survival curves.

3
FIGURE 3 Survival results in pre-propensity score matching (PSM) and post-PSM.(A) Kaplan-Meier analysis of overall survival (OS) or (B) cancer-specific survival (CSS) based on patients with mucinous and adeno cholangiocarcinoma before PSM.Gehan-Breslow tests used to calculate P-values.(C) Survival analysis of OS or (D) CSS based on patients with mucinous and adeno cholangiocarcinoma after PSM.Log-rank tests used to calculate P-values.

4 5
FIGURE 4    Construction and estimate of overall survival (OS)-associated predictive models.(A, B) The LASSO coefficient and deviance profiles represent the optimal l value and risk factors, respectively.According to the risk score calculated by predictive models after PSM, KM curves of OS shown in (C) indicating that the prognosis of the low-risk group is significantly better than the high-risk score group.(D) According to the risk score, ROC curves of OS at 1, 3, 5, 7, 9 years was shown in the predictive model.

7
FIGURE 7 Validation of OS and CSS-associated predictive models in external datasets.According to the risk score of the patients excluded by PSM, KM curves of OS (A) and CSS (B) indicating that the prognosis of the low-risk group is significantly better than the high-risk score group.And ROC curves of OS and CSS at 1,2,3 years were shown in (C, D), respectively.

TABLE 1
Demographic and clinical characteristics comparing adeno and mucinous cholangiocarcinoma (pre-PSM and post-PSM).

TABLE 2
Univariable Cox Regression for analyzing the associated factors for mucinous cholangiocarcinoma.

TABLE 3 Multivariable
Cox Regression for analyzing the associated factors for mucinous cholangiocarcinoma.