The ARPA-H-SOL-25-113 document outlines the requirements for submitting a Solution Summary to the ARPA-H PRECISE-AI solicitation. It emphasizes the necessity of a preliminary submission prior to a full proposal, with specific formatting and length guidelines: a maximum of three pages for a single technical area (TA) submission and six pages for multiple TAs. The Solution Summary must adhere to a structured outline, comprising a summary of the concept, the innovation and impact of the proposed approach, a description of the proposed work including deliverables and milestones, and an overview of team organization and capabilities.
Each section should detail how the proposed solutions address technical challenges, include quantitative metrics for comparison with existing technologies, and evaluate the technical risks involved. Additionally, a Basis of Estimate (BOE) is required, outlining the federal funds requested and a comprehensive cost breakdown. This submission serves to evaluate the potential effectiveness and resource needs of proposed technological innovations in the ARPA-H PRECISE-AI program, ensuring clarity and thoroughness in the proposals submitted for consideration.
The Advanced Research Projects Agency for Health (ARPA-H) has issued a DRAFT Program Solicitation titled Performance and Reliability Evaluation for Continuous Modifications and Usability of AI (PRECISE-AI). The program seeks to develop self-correction techniques for AI decision support tools (AI-DSTs) in healthcare to ensure ongoing efficacy and safety, as current methods largely depend on pre-market evaluations. The solicitation highlights the need for continuous AI monitoring and adaptability to operational changes, emphasizing the significance of automated ground truth extraction, performance degradation detection, root cause analysis, and effective communication with clinicians.
PRECISE-AI will comprise five technical areas focusing on automated label extraction, degradation detection and correction, uncertainty quantification, core data infrastructure, and independent validation. It aims to create an open-source repository of monitoring tools, enhancing AI tool interpretability and clinical outcomes through real-world testing. Selected proposals will address specific use cases within priority clinical tracks, with a robust framework for data sharing and collaboration across diverse healthcare systems. The anticipated outcomes include improved AI tool performance, enhanced clinical decision-making, and risk reduction for patient care. The solicitation is aimed at fostering innovative approaches that ensure AI tools are effective, trustworthy, and transparently integrated into clinical settings.