Patients characteristics and survival outcomes
In total, this study involved 854 eligible patients diagnosed with PCCA as the sole primary cancer in SEER database from 2004–2015. The descriptive and clinical characteristics of these patients are provided in Table.1. In the training cohort, the median follow-up periods were 14.5 (range, 1–83) months with 76.6% patients being over age of 60 years. Notably, 78.9% patients had fewer number of regional lymph nodes examined than 6 recommended by 8th AJCC TNM stage guideline. Similarly, as mentioned in 8th AJCC TNM stage guideline, the positive number of lymph node beyond 3 nodes is an independent factor that affect the prognosis of patients with PCCA. In general, in terms of prognosis, the 1-,3- and 5- year cancer-specific survival was stratified by age, sex, pathology grade, T, N and M stage (7th edition), surgery, marital status, number of regional nodes examined, positive number of regional nodes, liver metastasis. And according to previous study of our center21, our colleagues revealed the importance of marital status in prognosis of both intrahepatic and extrahepatic bile duct cancer. Therefore, we took marital status into account in this analysis and found that the prognosis of patients who remained in marriage was significantly better than who were unmarried.
Identification Of Optimal Tumor-size Cutoff Value With Prognosis
The cutoff value, 24 mm and 36 mm, of tumor size was identified by X-tile plot based on minimal P-value approach and the maximum of chi-square log-rank values was 66.8 according to cancer-specific survival (Fig. 2). To investigate the impact of tumor-size cutoff value on CSS, we first reclassified patients into three risk groups, 1–24 mm, 25–36 mm and ≥37 mm, using 24 mm and 36 mm as the cutoff value, the size of tumor is an important prognosis factor for PCCA and patients with 37 mm or more had significant poor prognosis and 1–24 mm group showed better survival outcome than 25–36 mm in CSS (p < 0.001, Fig. 2). To better reveal the clinical value of this tumor-size cutoff value, Kaplan-Meier method was used to validate its implications for patients with early and advance stage, the results suggested that tumor size less than 24 mm in patients with early stage of PCCA indicated a better prognosis, and for patients with advanced PCCA, tumor size larger than 37 mm is a significant negative factor affecting prognosis (Fig. 2).
Effect of tumor size counts on survival outcome of PCCA
Initially, univariate analysis revealed that, besides of tumor size, other clinicopathologic characteristics strongly correlated with CSS with age at diagnosis, pathological grade, TNM stage, marital status, number of regional nodes examined, regional nodes positive and liver metastasis (Table.2). Multivariate analysis with cox regression demonstrated that more larger tumor size showed survival disadvantage in CSS (tumor size: 25–36 mm hazard ratio (HR) 1.232; ≥ 37 mm HR 2.025). Notably, we noticed that tumor size larger than 37 mm exhibited a significant effect on CSS prognosis in both univariate and multivariate analysis. Of other variables, age at diagnosis, N stage, M stage, surgery and regional nodes positive remained significantly associated with CSS in the multivariate cox regression model (all P < .05), suggesting they were also the independent predictors for CSS in PCCA patients (Table.2).
Construction And Validation Of Nomogram For Css
With the independent predictors of CSS derived from multivariate analysis, we established a nomogram to predict the 1-,3- and 5-year CSS in patients with PCCA. Age at diagnosis, N stage, M stage, surgery, number of positive regional nodes and tumor size were included into the nomogram (Fig. 3). Notably, the tumor size contributed the most to prognosis except surgery and number of regional positive lymph nodes in the novel nomogram. Each variable in nomogram was assigned a risk score on the point scale. And we were able to calculated the total risk point to estimate the 1-,3- and 5-year survival rate according to the survival probability scales in the nomogram. The nomogram based on tumor size showed good accuracy with c-index of 0.733 for CSS model in the training cohort. Calibration plots for the probabilities of 1-, 3-, and 5-year CSS showed an optimal agreement between prediction by the nomogram and the actual observation regarding both training and external validation sets (Fig. 3). To further explore the discriminatory accuracy of the nomogram with that of 7th edition TNM staging systems in the training set. The nomogram discriminatory accuracy of the nomogram for CSS prediction was superior to that of either the 7th edition TNM stage systems (C-index: 0.733 vs 0.621). The external prognostic validation using the above cut-points revealed that prognosis of patients with tumor size less than 24 mm had more better survival outcome, and the calibration plots also performed the good correlations between the nomogram prediction and actual observation for 1-, 3- and 5-year CSS and C-index of the nomogram was also higher than the 7th edition TNM stage systems (C-index: 0.600 vs 0.556, Fig. 3).
Comparison of the Values of Area Under the ROC Curve
Comparisons of the novel nomogram and the 7th edition TNM staging system was performed using the AUC values. For the whole study cohort, the AUC values of the nomogram for predicting 1-,3- and 5- year CSS were 0.836, 0.845 and 0.891. These results were all significantly better than the corresponding 1-,3- and 5-year AUC values of CSS in TNM staging system (0.624, 0.695 and 0.669, Fig. 4). In the external cohort, the1-, 3- and 5-year AUC value of the nomogram also showed better accuracy for predicting the CSS than that with the 7th TNM stage system (0.613 vs 0.572, 0.634 vs 0.530 and 0.666 vs 0.576, Fig. 4). DCA was performed to compare the clinical usability and benefits of the nomogram with that of the traditional AJCC stage. As shown in Fig. 4, compared to the AJCC stage model, the new nomogram’s 5-year DCA curves showed larger net benefits across a range of death risk in the training and external validation cohort.
To enhance the clinical application of the nomogram, we divided all the patients of this study into low (score 0–22 for CSS), middle (score 23–33 for CSS) and high (score 34–43 for CSS) risk subgroup according to cut-point analysis by X-tile program. As shown in Fig. 5, the nomogram showed superior discriminative capacity for predicting CSS compared with the 7th TNM stage system and patients in the high-risk group had lower CSS compare with patients in low and middle risk groups(p < 0.001).