Risk factors, prognostic predictors, and nomograms for pancreatic cancer patients with initially diagnosed synchronous liver metastasis

Abstract

Background: Liver metastasis (LM) remains a major cause of cancer-related death in patients with pancreatic cancer (PC) and is associated with a poor prognosis. Therefore, identifying the risk and prognostic factors in PC patients with LM (PCLM) is essential as it may aid in providing timely medical interventions to improve the prognosis of these patients. However, there are limited data on risk and prognostic factors in PCLM patients.

Aim: To investigate the risk and prognostic factors of PCLM and develop corresponding diagnostic and prognostic nomograms.

Methods: Patients with primary PC diagnosed between 2010 and 2015 were reviewed from the Surveillance, Epidemiology, and Results Database. Risk factors were identified using multivariate logistic regression analysis to develop the diagnostic mode. The least absolute shrinkage and selection operator Cox regression model was used to determine the prognostic factors needed to develop the prognostic model. The performance of the two nomogram models was evaluated using receiver operating characteristic (ROC) curves, calibration plots, decision curve analysis (DCA), and risk subgroup classification. The Kaplan-Meier method with a log-rank test was used for survival analysis.

Results: We enrolled 33459 patients with PC in this study. Of them, 11458 (34.2%) patients had LM at initial diagnosis. Age at diagnosis, primary site, lymph node metastasis, pathological type, tumor size, and pathological grade were identified as independent risk factors for LM in patients with PC. Age > 70 years, adenocarcinoma, poor or anaplastic differentiation, lung metastases, no surgery, and no chemotherapy were the independently associated risk factors for poor prognosis in patients with PCLM. The C- index of diagnostic and prognostic nomograms were 0.731 and 0.753, respectively. The two nomograms could accurately predict the occurrence and prognosis of patients with PCLM based on the observed analysis results of ROC curves, calibration plots, and DCA curves. The prognostic nomogram could stratify patients into prognostic groups and perform well in internal validation.

Conclusion: Our study identified the risk and prognostic factors in patients with PCLM and developed corresponding diagnostic and prognostic nomograms to help clinicians in subsequent clinical evaluation and intervention. External validation is required to confirm these results.

Keywords: Liver; Neoplasm metastasis; Nomograms; Pancreatic neoplasms; Prognosis; Surveillance, Epidemiology, and End Result program.

Figure 1

Nomogram for predicting liver metastasis from pancreatic cancer patients. ^a P < 0.001. LN: Lymph node; LM: Liver metastasis.

Figure 2

Validation of the diagnostic nomogram in the training and validation sets. A: The receiver operating characteristic curve of the training set; B: The calibration curve of the training set; C: The decision curve analysis of the training set; D: The receiver operating characteristic curve of the validation set; E: The calibration curve of the validation set; F: The decision curve analysis of the validation set. AUC: Area under curve.

Figure 3

The least absolute shrinkage and selection operator regression used to select prognostic factors for overall survival. A: Least absolute shrinkage and selection operator (LASSO) coefficient profiles of 16 variables for overall survival (OS); B: LASSO Cox analysis identified 7 variables for OS. The LASSO regression analysis run in R runs 10 times K cross-validation for centralization and normalization of included variables and then selects the most appropriate lambda value depending on the type measure of -2 Log-likelihood and binomial family. “Lambda.lse” gives a model with good performance but the least number of independent variables.

Figure 4

A prognostic nomogram for pancreatic cancer patients with liver metastasis. ^a P < 0.01; ^bP < 0.001. Surg prim: Surgical treatments of the primary site; Surg dis: Surgical treatments of the distant site.

Figure 5

The calibration curves and decision curve analysis of the prognostic nomogram in the training set. A: The calibration curve of the nomogram for 6 mo in the training set; B: The calibration curve of the nomogram for 12 mo in the training set; C: The calibration curve of the nomogram for 18 mo in the training set; D: The decision curve analysis of the nomogram for 6 mo in the training set; E: The decision curve analysis of the nomogram for 12 mo in the training set; F: The decision curve analysis of the nomogram for 18 mo in the training set.

Figure 6

The calibration curves and decision curve analysis of the prognostic nomogram in the validation set. A: The calibration curve of the nomogram for 6 mo in the validation set; B: The calibration curve of the nomogram for 12 mo in the validation set; C: The calibration curve of the nomogram for 18 mo in the validation set; D: The decision curve analysis of the nomogram for 6 mo in the validation set; E: The decision curve analysis of the nomogram for 12 mo in the validation set; F: The decision curve analysis of the nomogram for 18 mo in the validation set.

Figure 7

Time-dependent receiver operating characteristic curve analysis and Kaplan-Meier survival curves of prognostic nomogram. A: Time-dependent receiver operating characteristic curve of the prognostic nomogram for 6, 12, and 18 mo in the training set; B: Time-dependent receiver operating characteristic curve of the prognostic nomogram for 6, 12, and 18 mo in the validation set; C: The Kaplan-Meier survival curves of the patients in the training set; D: The Kaplan-Meier survival curves of the patients in the validation set. AUC: Area under curve.