A brain metastasis prediction model in women with breast cancer

Breast cancer (BC) is a leading cause of mortality and the most frequent malignancy in women, and most deaths are due to metastatic disease, particularly brain metastases (BM). Currently, no biomarker or prediction model is used to predict BM accurately. The objective was to generate a BM prediction model from variables obtained at BC diagnosis.

A retrospective cohort of women with BC diagnosed from 2009 to 2020 at a single center was divided into a training dataset (TD) and a validation dataset (VD). The prediction model was generated in the TD, and its performance was measured in the VD using the area under the curve (AUC) and C-statistic.

The cohort (n = 5009) was divided into a TD (n = 3339) and a VD (n = 1670). In the TD, the model with the best performance (lowest AIC) was built with the following variables: age, estrogen receptor status, tumor size, axillary adenopathy, anatomic clinical stage, Ki-67 expression, and Scarff-Bloom-Richardson score. This model had an AUC of 0.79 (95%CI, 0.76-0.82; p < 0.0001) in the TD. The 10-fold cross-validation showed the good stability of the model. The model displayed an AUC of 0.81 (95%CI, 0.77-0.85; P < 0.0001) in the VD. Four groups, according to the risk of BM, were generated. In the low-risk group, 1.2% were diagnosed with BM (reference); in the medium-risk group, 5.0% [HR 4.01 (95%CI, 1.8 - 8.8); P < 0.0001); in the high-risk group, 8.5% [HR 8.33 (95%CI, 4.1-17.1); P < 0.0001]; and in the very high-risk group, 23.7% [HR 29.72 (95%CI, 14.9 - 59.1); P < 0.0001].

This prediction model built with clinical and pathological variables at BC diagnosis demonstrated robust performance in determining the individual risk of BM among patients with BC, but external validation in different cohorts is needed.