298P - Enhancing prognostic accuracy for breast adenoid cystic carcinoma using machine learning models

Background

Adenoid cystic carcinoma (ACC) of the breast is a rare subtype of breast cancer that is characterized by more favorable outcomes than extramammary ACC. Its rarity makes the establishment of standardized treatment protocols challenging. This study aimed to develop a machine learning (ML)-based prognostic model that improves the accuracy of survival rate predictions for patients with ACC.

Methods

The SEER database provided the data used for this study’s analysis (2000–2020). Patients who met any of the following criteria were excluded: diagnosis not confirmed by histology, previous history of cancer or other concurrent malignancies, or unknown data. To identify prognostic variables, we conducted Cox regression and constructed prognostic models using five ML algorithms to predict the 5-year survival. Patient records were randomly divided into training (70 %) and validation (30 %) sets. A validation method incorporating the area under the curve (AUC) of the receiver operating characteristic curve was used to validate the accuracy and reliability of the ML models.

Results

The study population comprised 1212 patients. The median patient age was 60 years, and the median tumor size was 1.9 cm. The largest racial group was white (79.5 %). Furthermore, most cases were localized (95.1%), T1 (56.2%), and N0(96.4%). The most common type of surgery performed was partial mastectomy (74.8%). Analysis of the therapeutic options revealed significantly different survival rates. While surgery + RT had a 94.9% overall 5-year survival rate, surgery alone had a rate of 91%. For surgery + CTX, the survival rate was 83.5% and 80.1% for surgery + CTX + RT. The corresponding cancer specific survival rates were 68.5%, 63.5%, 75%, and 70.8%, respectively. The ML models demonstrated strong predictive capabilities, with random forest classifier performing the best. ML identified that the most contributing factors were age, followed by the N component of the AJCC stage and ER status.

Conclusions

ML approaches provide superior predictive performance compared with traditional statistical methods. The developed models can significantly aid personalized treatment planning and improve the clinical outcomes of patients with ACC.

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.