INTEGRATING PERSONAL, HEREDITARY, AND POLYGENIC PROFILES VIA DEEP LEARNING-BASED RISK ESTIMATION FOR PRECLINICAL BREAST CANCER DIAGNOSIS satudominguez@gmail.com

3103-01

INTEGRATING PERSONAL, HEREDITARY, AND POLYGENIC PROFILES VIA DEEP LEARNING-BASED RISK ESTIMATION FOR PRECLINICAL BREAST CANCER DIAGNOSIS

1

Panama

Oral only

Breast cancer is the most prevalent malignancy in women worldwide, accounting for approximately 2.3 million new diagnoses and causing 670,000 deaths in 2022 alone. Advanced-stage disease (stages III–IV) not only dramatically worsens patient prognosis, with five-year survival plummeting to 32% for metastatic cases, but also imposes a substantial economic burden. Our study introduces a deep learning–based agent that integrates personal, familial, and polygenic risk profiles to improve preclinical detection of breast cancer, aiming to shift diagnosis to earlier, more treatable stages and reduce both mortality and cost.

This retrospective study analyzed 2021–2023 electronic health records from Panama’s Social Security system. An AI-based breast cancer risk detection agent was developed to leverage patient demographic, clinical, and imaging data and flag individuals at high risk of undiagnosed malignancy. Performance was evaluated against standard care for early cancer detection by calculating the model’s sensitivity, specificity, and increase in detection rate.

The AI agent increased early-stage (stage I–II) breast cancer detection by 47% relative to historical data. It achieved 96% sensitivity and 98% specificity. This level of accuracy is comparable to state-of-the-art AI diagnostic models. The 47% improvement substantially exceeds previously reported gains from AI-assisted screening (~17–30%)

The AI risk detection agent markedly improved early breast cancer detection in this Panamanian cohort. Its high sensitivity and specificity indicate strong potential as an adjunct to screening programs. Wider adoption of such technology could enable earlier diagnoses and reduce breast cancer mortality.