Inside Guardoc Health's Bet on Amazon Nova to Catch Nursing Home Documentation Errors Before They Cost Lives

One million clinical documents a day. That is the volume Guardoc Health now pushes through Amazon's Nova foundation models via Bedrock, an operational scale that reflects just how much paperwork long-term care facilities generate, and how much of it can silently harbor mistakes.

Guardoc builds documentation software specifically for long-term care providers, a sector where charting errors carry outsized consequences. Research published in BMJ Quality and Safety estimates diagnostic errors affect roughly 12 million U.S. adults annually in outpatient settings alone, underscoring why automated review matters beyond simple efficiency gains.

How the pipeline works

The system starts with Amazon Textract, which pulls text and structural metadata from each incoming page, whether it's a nursing note, a lab report, or a physician's order. That extracted content is then embedded using Amazon Titan Text Embeddings V2 and stored in Amazon DynamoDB, partitioned by individual patient.

This patient-level partitioning matters for what comes next: retrieval augmented generation. Rather than relying on generic medical knowledge, Guardoc's condition-classification step pulls evidence directly from a resident's own documentation history, grounding each flagged discrepancy in that specific patient's chart rather than population-level assumptions.

What the numbers show

Guardoc Health cites a 46 percent reduction in documentation errors and a 70 percent drop in audit fines since deploying the Nova-based pipeline. The company also reports more than $400,000 in annual ROI for a single facility, a figure that hints at why long-term care operators, often running on thin margins, would prioritize this kind of tooling.

A separate case study spanning seven facilities and 1,618 residents identified 10,612 issues in total. A narrower, quarterly deployment covering two facilities and 200 patients produced more granular detail: 86 issues tied to PDPM reimbursement accuracy, 847 documentation corrections, and a 74 percent reduction in hospital transfers per 100 admissions.

That last figure is the one likely to draw the most attention from clinicians and administrators alike, since hospital transfers from skilled nursing facilities are both costly and often flagged as a quality-of-care indicator by regulators like CMS.

None of the reporting on this deployment surfaced conflicting figures or disputed claims, so the numbers above represent a consistent picture across both AWS's own writeup and independent trade coverage. What remains unclear is how these results generalize beyond the specific facilities studied, and whether similar gains hold at larger scale as Guardoc's document volume grows.