430P - A novel survival predicting model for breast cancer brain metastasis based on multimodal data

Background

The prognosis of patients with breast cancer brain metastasis (BCBM) remains significantly poor. Accurate prediction of the prognosis of BCBM patients is very important for formulating treatment strategies.

Methods

We retrospectively collected clinicopathological data and MR images per patient diagnosed with BCBM at our center from 2001 to 2021, and validated the updated breast graded prognostic assessment (uB-GPA) in our cohort. Then cox regression analysis was used to identify prognostic clinical factors. Preprocessed MR images and prognostic clinical factors were used together to develop an end-to-end deep learning network based on the transformer encoder in the training set. After feature encoding by the encoder, output sequences were passed to a multi-task logistic regression model to obtain the patients' risk scores. The prognostic performance of the model was evaluated in the test set.

Results

A total of 700 patients with BCBM were enrolled, with a median OS of 11 months (95%CI: 9.8-12.2m). The concordance index (C-index) of uB-GPA was 0.563 (0.551-0.575). Nine clinical factors were shown to be associated with poorer prognosis of patients with BCBM: Age greater than 48 years at BC diagnosis (P=0.003), KPS less than 80 at BCBM diagnosis (P=0.001), triple-negative subtype (P<0.001), initial metastasis site other than the brain (P<0.001), number of brain metastases ≥3 (P<0.001), concurrent leptomeningeal metastasis (P=0.008), not receiving radiotherapy after diagnosis (P=0.003), liver metastasis (P=0.006), and lung metastasis (P=0.049). A deep learning model based on both MR images and clinical factors was developed in the training set (n=166), yielded the C-index of 0.664 and areas under the receiver operative characteristic curve of 0.705, 0.723 and 0.790 respectively for predicting 3-m, 6-m and 12-m OS in the test set (n=36). The median OS of patients in high and low-risk group were 5m (95%CI: 4-9m) and 14m (95%CI: 13m-?), respectively.

Conclusions

This study developed a novel tool to predict survival of patients with BCBM and achieve good performance. It helps in identifying patients' risk of death and in formulating personalized management strategies.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.