**Guardoc Health Transforms Clinical Documentation with Amazon Nova on Bedrock**
In the high-stakes world of long-term care, clinical documentation is both a foundational practice and a significant liability risk. Errors in patient records can cascade into denied Medicare claims, audit fines, and, most critically, compromised patient outcomes. Guardoc Health, a specialist in documentation software for long-term care providers, is tackling this challenge head-on by deploying Amazon Nova models via Amazon Bedrock to process over one million clinical documents daily. Their approach demonstrates how AI, when implemented with a clear understanding of risk and cost, can fundamentally improve compliance and patient safety.
### The High Cost of Clinical Documentation Errors
Bringing AI into clinical workflows is not a simple technical upgrade; it is a risk management strategy. A single error in documentation can lead to a cascade of negative consequences, including denied claims under the Patient-Driven Payment Model (PDPM), regulatory audit fines, potential litigation, and, in the most severe scenarios, a misdiagnosis or inappropriate treatment plan. The sheer volume of data in long-term care makes manual processing unsustainable and error-prone.
Guardoc Health’s deployment figures point to a quantifiable return on investment. The company reports a 46 percent reduction in documentation errors, a 70 percent drop in audit fines, and more than $400,000 in annual ROI for a single facility. In a pilot involving two facilities and 200 patients, the system drove 847 documentation corrections and flagged 86 issues specifically tied to PDPM reimbursement accuracy. Furthermore, it was associated with a 74 percent reduction in hospital transfers per 100 admissions. A separate case study covering seven facilities and 1,618 residents identified over 10,000 discrepancies, highlighting the scale of the underlying documentation problem.
### Solving the Problem of Complex Clinical Records
The challenge Guardoc Health addresses is the immense variability of clinical documentation. Patient records are not simple text files; they are a mix of multi-page PDFs, handwritten physician annotations, checkbox-based forms, and free-text notes, all often combined on a single page. Research from *BMJ Quality and Safety* estimates that around 12 million outpatients in the US are affected by diagnostic errors annually, with information-handling failures being a key contributing factor. At Guardoc’s processing scale, even a 1% error rate in condition detection could generate thousands of incorrect records daily, each carrying potential patient safety or compliance risks.
### A Cost-Aware, Multi-Stage Retrieval Architecture
Guardoc’s solution is built on a sophisticated yet cost-effective architecture centered on retrieval augmented generation (RAG). The pipeline is designed with a clear cost-tiering logic, ensuring that the most computationally expensive resources are used only when necessary.
1. **Ingestion and Structuring:** Amazon Textract is used first to extract text and structural metadata from every page. This step is treated as the lowest-cost point in the pipeline.
2. **Chunking and Embedding:** The extracted text is chunked along clinical boundaries (e.g., a complete medication list) and embedded using Amazon Titan Text Embeddings V2. These embeddings are stored in Amazon DynamoDB, partitioned by patient to ensure data isolation.
3. **Intelligent Retrieval and Filtering:** A custom pre-filter narrows the search by document type and recency. A k-nearest neighbour search then retrieves the most relevant text chunks for a specific query.
4. **Multi-Model Reasoning:** Amazon Nova 2 Lite performs an initial text-based pass to eliminate obvious non-matches. For the final, most critical stage, the raw PDF bytes—along with the Textract output—are passed to Amazon Nova Pro. This multimodal model reasons over layout, handwriting, signatures, and stamps to produce the final classification.
This design ensures that cheap, high-throughput components handle the bulk of the work, while the advanced multimodal reasoning of Nova Pro is reserved for the complex cases where it is most needed.
### Tackling the “Hard Cases” in Healthcare
Two document types have proven to be the most challenging for earlier pipeline versions: physician attestation fields on prior authorization forms and patient-reported symptom sections. In both cases, crucial information is often contained in handwriting that overrides printed checkboxes or exists outside of structured data fields.
Medication extraction presents another layer of complexity. Drug names, dosages, routes, and frequencies can be found in structured tables, buried in physician notes, added by hand to printed lists, or included in scans of faxed documents. Guardoc’s hybrid pipeline uses Amazon Textract to parse clean tables and then passes both the original PDF and the Textract output to Amazon Nova Pro. This allows the model to resolve wrapped table columns, handwritten additions, and other non-standard formats that Optical Character Recognition (OCR) alone cannot handle correctly.
“The Nova family is making it easier for healthcare organisations to detect high-risk cases earlier and act before issues become costly,” said Assaf Amiaz, Director of Product at Guardoc Health. “By automating workflows that once required manual oversight, the Nova family helps teams reduce compliance gaps, prevent errors, and focus more of their time on improving patient outcomes.”
—
### FAQ Section
**Q: What specific problems does Guardoc Health’s AI system solve for long-term care providers?**
A: The system is designed to reduce documentation errors, lower audit fines, decrease hospital transfers, and improve compliance with reimbursement models like PDPM. It automates the review of complex clinical documents to catch issues that could lead to financial penalties or patient harm.
**Q: How does Guardoc Health’s architecture manage costs while maintaining accuracy?**
A: The system uses a cost-tiered RAG architecture. It employs low-cost Amazon Textract for initial data extraction and Amazon Titan for embedding and coarse filtering. More expensive, multimodal reasoning with Amazon Nova Pro is reserved for the final stage, where it is needed to resolve the most complex cases involving handwriting, signatures, and non-standard formats.
**Q: Which document types are the most difficult for clinical AI to process?**
A: According to Guardoc, the most challenging documents are physician attestation fields on prior authorization forms, where handwritten notes can override printed checkboxes, and patient-reported symptom sections, where handwriting may contain information not found elsewhere in the record. Medication lists that combine structured tables with handwritten additions are also difficult to parse accurately.
**Q: What measurable outcomes has Guardoc Health reported from its deployments?**
A: Reported outcomes include a 46% reduction in documentation errors, a 70% drop in audit fines, over $400,000 in annual ROI per facility, a 74% reduction in hospital transfers, and the identification of over 10,612 issues across seven facilities.
—
### Conclusion
Guardoc Health’s deployment of Amazon Nova models on Bedrock represents a powerful example of how to implement AI in a high-risk, regulated industry. By focusing on a cost-aware architecture and a deep understanding of clinical documentation challenges, they have created a system that not only automates a tedious process but actively enhances patient safety and financial compliance. The results—from significant reductions in errors and audit fines to fewer hospital transfers—demonstrate that AI, when applied strategically, can be a transformative tool for healthcare providers, allowing them to focus less on correcting documentation and more on improving patient outcomes.
