Establishment and validation of diagnostic and prognostic prediction models for liver metastasis in patients with rectal cancer: a SEER based study.

Abstract Rectal cancer is one of the most common gastrointestinal tumors, among which the liver is the most common site of distant metastasis and liver metastasis leads to poor prognosis. We aimed to develop and validate a diagnostic nomogram to predict the occurrence of rectal cancer with liver metastasis (RCLM) and a prognostic nomogram to predict the cancer-specific survival (CSS) in RCML patients. Data on patients with rectal cancer diagnosed between 2010 and 2013 were collected from the Surveillance, Epidemiology, and End Results (SEER) database. Univariate logistic regression, the area under receiver operating characteristic curve (AUC), and multivariate logistic regression were used to determine the independent risk factors of RCLM. Univariate Cox proportional hazards regression and multivariate Cox proportional hazards regression were used to identify independent prognostic factors for RCLM. We then developed two novel nomograms, and the results were evaluated by receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). A total of 29367 patients with rectal cancer were included, with an average age of 66.71 ± 12.47 years old. Among them, 3403 patients (11.59%) had liver metastases at the time of diagnosis. The independent risk factors of RCLM included AJCC N, chemotherapy, CEA, DX-lung (Distant metastasis to the lung) and surgical sites. Age, chemotherapy, total number (from the primary tumors), surgery sites, histological type were independent prognostic factors of patients with RCLM. The results of ROC curves, calibration curves, DCA, C-indexes and Kaplan–Meier (K-M) survival curves in the development, validation and testing sets confirmed that two nomograms can precisely predict occurrence and prognosis of RCLM. Two nomograms are expected to be effective tools for predicting the risk of liver metastasis for patients with rectal cancer and personalized prognosis prediction for patients with RCLM, which may benefit clinical decision-making.