Evaluation of Bowhead health patient navigator for breast cancer treatment in India: A pilot study.

e13659 Background: Access to guideline-concordant cancer treatment is a significant challenge in India due to limited resources and variability in clinical care practices. The Bowhead Health Patient Navigator (BHPN) is a digital tool designed to deliver evidence-based, personalized treatment recommendations for breast and lung cancer. Built with a secure and privacy-preserving architecture, BHPN leverages advanced AI models to support clinical decision-making while maintaining strict compliance with data governance. This pilot study evaluated the accuracy and applicability of BHPN recommendations compared to official treatment guidelines, local tumor board decisions, and real-world practices for breast cancer. Methods: Matching of 40 synthetic breast cancer cases to India’s cancer care guidelines was conducted using BHPN. Synthetic data was developed by two surgical medical oncologists from India, encompassing demographics, medical history, cancer staging, pathology reports, and treatment outcomes. BHPN’s processing pipeline operates entirely on-premise within secure cloud infrastructure, ensuring data localization and compliance. Patient data was normalized using an open-source Bedrock LLM in a secured containerized environment with whitelisted IPs, preventing external data exposure. Clinical trial information, updated daily from sources like ClinicalTrials.gov, was integrated and matched in real time. The primary objective was to compare BHPN recommendations to standard treatment guidelines, with a target accuracy of 95%. Secondary objectives focused on assessing real-world treatment concordance, technical adaptability, and areas requiring optimization in decision-making algorithms. Results: BHPN achieved 92% concordance with guideline-based recommendations and 89% with tumor board decisions. The platform generated personalized treatment plans, such as hormone therapy for premenopausal ER-positive breast cancer patients, while accounting for comorbidities. Discrepancies were observed in cases involving complex multi-drug regimens where real-world practices deviated from standard protocols. The localized data processing approach ensured end-to-end privacy while enabling real-time clinical trial matching. Feedback emphasized the need to enhance contextual reasoning for edge-case scenarios. Conclusions: BHPN demonstrates significant potential in transforming cancer care delivery by providing secure, guideline-concordant, and personalized recommendations. Its on-premise design and compliance-focused architecture ensure adaptability and scalability in resource-constrained settings. Further validation with larger datasets and real-world implementations is recommended to confirm its impact on oncology outcomes in underserved regions.