291P - Multi-center investigation of a detection model utilizing cfDNA for early-stage breast cancer screening

Background

Early detection contributes to higher survival rates among breast cancer patients, and the integration of cell-free DNA (cfDNA) technology can facilitate tumor malignancy classification and guide clinical precision oncology. Despite promising aspects of cfDNA analysis in advanced breast cancer, its sensitivity for early breast cancer still requires further improvement. Therefore, for the first time, we developed a predictive model using cfDNA LP-WGS (low-pass whole genome sequencing) based on Chinese population, to enhance breast cancer detection and tumor subtype classification.

Methods

We included 306 breast cancer patients from seven medical centers (Stage 0 - Ⅱa 87.9%), along with 176 healthy controls with benign breast disease, whose plasma cfDNA samples were analyzed using whole-genome sequencing (WGS). Multiple features and deep learning algorithms were utilized in the training cohort, and model performance was evaluated in internal and external validation datasets.

Results

The study employed DNN (Deep Neural Networks) deep learning algorithms to analyze multiple features. In the test dataset (n = 312), through 10-fold cross-validation, the model achieved specificities of 95.61% and sensitivities of 87.89% (AUC: 0.9745). In the internal validation cohort (n = 120), the model achieved specificities of 92.10% and sensitivities of 90.20% (AUC: 0.978). In the external validation cohort (n = 50), the model achieved specificities of 92.00% and sensitivities of 96.00% (AUC: 0.98). Additionally, an analysis of relevant features revealed differences in various features, including end motifs and break point features. Notably, our novel deep learning algorithm presented good performance in different molecular subtypes of breast cancer.

Conclusions

We established a deep learning model based on a novel cfDNA pattern for early prediction of early stage breast cancer in Chinese population. This study provides a promising approach for early detection of breast cancer tumors and offers new insights into early cancer diagnosis in clinical practice. Future study will involve large-scale validation studies and explore tumor subtype prediction and staging prediction.

Clinical trial identification

NCT06016790.

Editorial acknowledgement

Legal entity responsible for the study

Chao Ni and Ziao Lin.

Funding

Key research and development program of Zhejiang Province 2024C03050.

Disclosure

All authors have declared no conflicts of interest.