A clinical prediction model for the presence of brain metastases from colorectal cancer.

Background: We conducted this study to explore clinicopathological profiles of brain metastases (BM) and establish a clinical prediction model that predicts the presence of BM in colorectal cancer (CRC) patients.

Methods: Patients with initially diagnosed CRC were reviewed between the year 2010 and 2015. Multiple imputations are used for handling missing values. Prognostic factors were identified by the univariate and multivariate Cox regression model. Univariate and multivariate logistic regression was used to identify the predictive factors for the presence of BM. A nomogram was constructed based on statistically significant risk factors of the presence of BM. The decision curve analysis (DCA) was used to assess the clinical usefulness and net benefits of the nomogram for the presence of BM.

Results: Four hundred ninety-five patients with brain metastasis at the initial diagnosis were identified, representing 0.24% of the whole cohort and 0.91% of the metastatic cohort. Multivariable logistic regression demonstrated that young age, positive CEA, adenocarcinoma, lower differentiated grade, presence of liver metastases, presence of lung metastases, and presence of bone metastases were significantly associated with higher risk of developing BM. The decision curve analysis inform clinical decisions were better than a scenario in which all patients or no patients are screened across a wide range of threshold at ≥ 0.027%.

Conclusions: The risk estimates provided by the nomogram can be extremely useful for earlier diagnosis, especially when discussing screening strategy among high-risk patients.