artificialintelligence-news.com
·
Jul 28
Guardoc Health processes clinical documentation using Amazon Nova models
Guardoc Health processes over one million clinical documents daily using Amazon Nova models via Bedrock, aiming to reduce errors in long-term care documentation. The system addresses complex inputs like handwritten annotations and mixed-format forms, which often lead to denied Medicare claims and audit fines. Guardoc reports a 46 percent reduction in documentation errors and a 70 percent drop in audit fines. In a quarterly deployment across two facilities, the system drove 847 corrections and flagged 86 reimbursement issues, correlating with a 74 percent reduction in hospital transfers. A separate study of seven facilities identified 10,612 issues. The architecture uses retrieval augmented generation, employing Amazon Textract for initial extraction and Amazon Titan embeddings for storage. Cost-tiering logic reserves computationally intensive Amazon Nova Pro reasoning for final classification stages, while cheaper components handle high-volume filtering. This approach targets specific pain points like physician attestation fields and medication extraction, where handwriting often overrides printed data. Guardoc claims over $400,000 in annual ROI per facility, though baseline methodologies remain unpublished. The technology seeks to mitigate diagnostic errors affecting millions of US outpatients annually by improving information handling accuracy in clinical records.
💡
By isolating expensive multimodal reasoning for only the most complex handwritten elements, Guardoc demonstrates a viable path to scaling AI in healthcare without prohibitive costs. This efficiency directly tackles the financial and safety risks inherent in the Patient-Driven Payment Model, where documentation accuracy dictates reimbursement and patient outcomes.