Enhancing Clinical Documentation: How Guardoc Health Utilizes Amazon Nova Models
Guardoc Health is revolutionizing the world of clinical documentation by processing over a million clinical documents daily through Amazon Nova models and Bedrock. In an industry where precision is paramount, getting it right means the difference between quality patient care and costly errors. A single oversight can lead to denied Medicare claims, audit fines, or even complications in patient treatment. With the right technology, however, the benefits are significant—reducing corrections, minimizing hospital transfers, and ultimately driving down compliance costs.
The Scope of the Challenge
Guardoc Health tackles an immense variety of documents, encountering formats that can make any process cumbersome. Clinical settings generate multi-page PDFs with handwritten notes layered over printed text, authorization forms where a single checkbox can change everything, and patient intake forms that combine typed entries with rubber-stamped approvals.
Research published in BMJ Quality and Safety indicates that around 12 million US outpatients face diagnostic errors annually, attributed largely to information management shortcomings. With Guardoc handling such high volumes, even a 1% error rate could result in thousands of inaccurately recorded conditions each day, with serious implications for patient safety and compliance.
Guardoc boasts impressive statistics: a 46% decline in documentation errors, a 70% reduction in audit fines, and over $400,000 in annual ROI for just one facility. During a quarterly deployment that involved two facilities and 200 patients, their technology identified 847 documentation corrections and flagged 86 issues related to PDPM reimbursement accuracy, leading to a 74% decrease in hospital transfers per 100 admissions.
Crafting an Efficient Retrieval Pipeline
To ensure both cost-effectiveness and accuracy, Guardoc’s architecture employs a sophisticated approach to condition classification. It utilizes retrieval augmented generation, leveraging a patient’s own documentation for evidence before synthesizing this information into actionable insights.
Initially, Amazon Textract extracts both text and structural metadata from incoming documents, achieving efficiency and cost-effectiveness. The extracted information is segmented along clinical boundaries, preserving the integrity of data like medication lists or diagnoses.
Next, each segment is transformed using Amazon Titan Text Embeddings V2 and securely stored in Amazon DynamoDB, thereby ensuring that patient data remains private. A custom pre-filter refines the document type and recency before conducting a k-nearest neighbor search to find the most relevant chunks. This streamlined process minimizes data transfer, maintaining efficiency.
The Amazon Nova 2 Lite component eliminates obvious mismatches, while only the most relevant pages proceed to Amazon Nova Pro. Here, algorithms analyze layout, handwriting, signatures, and stamps to generate precise classifications for downstream systems.
This tiered design not only enhances accuracy but also maximizes cost efficiency by reserving more intensive computational resources for critical stages of the process.
Addressing Complex Documentation Challenges
Guardoc’s research identifies two main document types that previously troubled their earlier systems. The first involves physician attestations on prior authorization forms, where a handwritten note can override a printed checkbox. The second pertains to patient-reported symptoms, often captured in handwriting that might not appear elsewhere in clinical records.
The challenge of extracting medication details adds another layer of complexity. Drug names, dosages, and frequencies can appear in various formats—structured tables, narrative notes, handwritten changes, and faxes that might have been passed through multiple sources. Guardoc’s innovative hybrid pipeline tackles this by first using Amazon Textract for neat tables and then sending both the original PDF and Textract output to Amazon Nova Pro, which resolves ambiguities like wrapped table columns and handwritten notes.
“With the Nova family, we’re enabling healthcare organizations to identify high-risk cases sooner, acting before they escalate into costly problems,” says Assaf Amiaz, Director of Product at Guardoc Health. “By automating workflows that once demanded manual attention, we empower teams to bridge compliance gaps, reduce errors, and dedicate more time to enhancing patient outcomes.”
The marriage of technology and clinical expertise is indeed a transformative journey in the healthcare landscape. Embrace the future of clinical documentation with solutions like Guardoc, where improved efficiency meets better patient care.
Ready to elevate your healthcare practices? Dive into the world of advanced clinical documentation, and discover how thoughtful integration of technology can profoundly impact patient outcomes today!
