Added value of DCER-features to clinicopathologic model for predicting metachronous metastases in rectal cancer patients

Abstract

Rationale and objectives: The study aimed to investigate whether dynamic contrast-enhanced MRI parameters and preoperative radiological features (DCER-Features) add value to the clinicopathologic model for predicting metachronous metastases in rectal cancer patients.

Materials and methods: From January 2014 to December 2020, 859 patients in the PACS system were retrospectively screened. Of the initial 722 patients with surgically confirmed rectal cancer and no synchronous metastases, 579 patients were excluded for various reasons such as lack of clinicopathological or radiological information. 143 patients were finally included in this study. And 73 Patients of them developed metachronous metastasis within five years. After stepwise multiple regression analyses, we constructed three distinct models. Model 1 was developed solely based on clinicopathological factors, and model 2 incorporated clinicopathological characteristics along with DCE-MRI parameters. Finally, model 3 was built on all available factors, including clinicopathological characteristics, DCE-MRI parameters, and radiological features based on rectal magnetic resonance imaging. The radiological features assessed in this study encompass tumor imaging staging, location, and circumferential resection margin (CRM) for primary tumors, as well as the number of visible lymph nodes and suspected metastatic lymph nodes. Receiver operating characteristic (ROC) and decision curve analysis (DCA) were conducted to evaluate whether the diagnostic efficiency was improved.

Results: The performance of model 3 (including clinicopathologic characteristics and DCER-Features) was the best (AUC: 0.856, 95% CI 0.778-0.886), whereas it was 0.796 (95% CI 0.720-0.828) for model 2 and 0.709 (95% CI 0.612-0.778) for model 1 (DeLong test: model 1 vs model 2, p = 0.004; model 2 vs model 3, p = 0.037; model 1 vs model 3, p < 0.001). The decision curves indicated that the net benefit of model 3 was higher than the other two models at each referral threshold. The calibration plot of the three models revealed an excellent predictive accuracy.

Conclusion: This study suggests that DCER-Features have added value for the clinicopathological model to predict metachronous metastasis in patients with rectal cancers.

Keywords: DCE-MRI; Metachronous metastases; Predictive model; Rectal cancer.