Nomogram Predicts Risk and Prognostic Factors for Bone Metastasis of Pancreatic Cancer: A Population-Based Analysis

Abstract

Background: The overall survival (OS) of pancreatic cancer (PC) patients with bone metastasis (BM) is extremely low, and it is pretty hard to treat bone metastasis. However, there are currently no effective nomograms to predict the diagnosis and prognosis of pancreatic cancer with bone metastasis (PCBM). Therefore, it is of great significance to establish effective predictive models to guide clinical practice.

Methods: We screened patients from Surveillance Epidemiology and End Results (SEER) database between 2010 and 2016. The independent risk factors of PCBM were identified from univariable and multivariable logistic regression analyses, and univariate and multivariate Cox proportional hazards regression analyses were used to determine independent prognostic factors affecting the prognosis of PCBM. In addition, two nomograms were constructed to predict the risk and prognosis of PCBM. We used the area under the curve (AUC), C-index and calibration curve to determine the predictive accuracy and discriminability of nomograms. The decision curve analysis (DCA) and Kaplan-Meier(K-M) survival curves were employed to further confirm the clinical effectiveness of the nomogram.

Results: Multivariable logistic regression analyses revealed that risk factors of PCBM included age, primary site, histological subtype, N stage, radiotherapy, surgery, brain metastasis, lung metastasis, and liver metastasis. Using Cox regression analyses, we found that independent prognostic factors of PCBM were age, race, grade, histological subtype, surgery, chemotherapy, and lung metastasis. We utilized nomograms to visually express data analysis results. The C-index of training cohort was 0.795 (95%CI: 0.758-0.832), whereas that of internal validation cohort was 0.800 (95%CI: 0.739-0.862), and the external validation cohort was 0.787 (95%CI: 0.746-0.828). Based on AUC of receiver operating characteristic (ROC) analysis, calibration plots, and decision curve analysis (DCA), we concluded that the risk and prognosis model of PCBM exhibits excellent performance.

Conclusion: Nomogram is sufficiently accurate to predict the risk and prognostic factors of PCBM, allowing for individualized clinical decisions for future clinical work.

Keywords: Cox regression; Surveillance Epidemiology and End Results (SEER) database; bone metastasis; logistic regression; nomogram; pancreatic cancer; predictors.

Figure 1

Nomogram to estimate the risk of bone metastasis in patients with pancreatic cancer. HP, head of pancreas; BP, body of pancreas; OLP, overlapping lesion of pancreas; OSPP, other specified parts of pancreas; TP, tail of pancreas; PD, pancreatic duct; AC, Adenocarcinoma; IDC, Infiltrating duct carcinoma; NC, neuroendocrine carcinoma.

Figure 2

ROC curves, calibration plots and DCA of the nomogram for the risk of pancreatic cancer with bone metastasis. (A) The area under ROC curve was utilized to judge the advantages and disadvantages of nomogram. (B) Calibration plot for the diagnostic nomogram. The diagonal 45-degree line indicates perfect prediction. (C) Decision curve analysis for the diagnostic nomogram. The net benefit calculated by adding true positive and minus the false positive corresponds to the measurement of Y-axis; X-axis represents the threshold probability. (D) The area under ROC curve of external validation cohort. (E) Calibration plot for diagnostic nomogram in external validation cohort. (F) Decision curve analysis for diagnostic nomogram in external validation cohort.

Figure 3

Comparison of area under the receiver operating characteristic curves and DCA curves between nomogram and TNM stage in the training cohort (A, B) and external validation cohort (C, D).

Figure 4

Nomogram for predicting the overall survival of patients with pancreatic cancer presenting with bone metastasis. To use this nomogram, the specific point for each variable of the patient lies on each variable axis. Draw a vertical line upward to determine the point at which each variable accepts; the sum of these points is located on the Total Points axis, and draw a vertical line down to the survival axis to determine the probability of 1-, 2- and 3- year overall survival.

Figure 5

ROC curves of the ability of nomogram and TNM stage to predict 1-, 2- and 3-year overall survival in (A–C) training cohort, (D–F) internal validation cohort, and (G–I) external validation cohort.

Figure 6

Calibration curves of the nomograms. Calibration curves of 1-, 2- and 3-year overall survival for PCBM patients in (A–C) training cohort, (D–F) internal validation cohort, and (G–I) external validation cohort. The dotted line represents the ideal reference line, where the predicted probability would match the observed survival rate. The blue dots are calculated by bootstrapping (resample:100) and represent the nomogram performance. The closer the solid blue line is to the dotted line, the more accurate the model is in predicting overall survival.

Figure 7

Decision curve analysis of the nomogram and TNM stage for survival prediction of PCBM patients. (A) 1-, 2- and 3-year survival benefit in the training cohort. (B) 1-,2- and 3-year survival benefit in the internal validation cohort. (C) 1-, 2- and 3-year survival benefit in the external validation cohort.

Figure 8

Kaplan-Meier curves of OS for patients in low-risk and high-risk groups. (A) the training cohort; (B) the internal validation cohort; (C) the external validation cohort.