Case Study

Transform Quality Improvement with 98%+ Abstraction Accuracy
Layer Health delivers 98%+ validated accuracy in registry abstraction, matching or beating human abstractors on every field tested.

98%+
Validated accuracy
Every
registry, abstractor and health system
About this case study
Registry accuracy shapes quality benchmarks, accreditation reviews and the clinical feedback physicians act on. This case study draws on Layer Health implementations at leading health systems and research organizations, where Layer Health matched or exceeded human accuracy across stroke, cardiovascular, surgical and oncology registries.
The challenge
Registry accuracy has real consequences. An incorrect abstraction doesn't just affect a data submission - it affects a quality benchmark, an accreditation review, a clinical feedback loop physicians act on. Getting it right matters.
Manual abstraction is also inconsistent. Knowledge passes from abstractor to abstractor without standardization. Teams at the same health system answer the same questions differently - and 70% report observing manual errors or discrepancies.
The solution
Layer Health ensures accuracy before go-live, not after. Rigorously validated on each health system's own data before roll-out.
Validated for stroke, cardiovascular, surgical, oncology registries and more
On Epic, Cerner, and data in any format
Better-than-human accuracy across every variable tested
Dramatically less variability - abstractors aligned to the same clinical standard
How Layer Health delivers validated accuracy.
Comprehensive patient chart review
Analyze the entire, longitudinal patient record
Pull from all data types - PDFs, structured tables, unstructured clinician notes, imaging reports
Works across Epic and Cerner, including health systems mid-EHR transition
Customize to each health system
Tailor AI to site specific rules and custom registry questions
Run AI outputs against historical registry submissions
Identify every point where AI and human abstraction disagree
Train the AI models on every disagreement
Establish and maintain performance
Capture every post-go-live override, classify and incorporate feedback into the model
Monitor models continuously to catch drift pre-submission
Eliminate the “gray areas” of individual interpretation and standardize data at scale
“Before Layer, if the answer wasn't in the first few notes, I had to move on, because I didn't have the time to dig. Now, I’m submitting better cases because Layer Health surfaces things from the whole chart that I would have had to skip.“
GWTG Abstractor, Northwestern
The results
From field-level accuracy to actionable data.
98.4% accuracy, validated before go-live
AI is at or above human accuracy across fields requiring:
Unstructured data (e.g,. PMH, meds, imaging, tests)
Structured data (e.g,. labs, ADT, demographics)

What accurate data makes possible.
Empower your quality teams to act on gold standard data
Build deep abstractor trust in the clinical data
Codify best practices by aligning to one clinical standard
Protect and advance registry accreditation
Here’s what that looks like at research scale.

Better than human accuracy. Across every element. Every cancer type.
The American Cancer Society (ACS) runs longitudinal Cancer Prevention Studies across hundreds of institutions. Its research depends on data quality, but manual abstraction at that scale means working through thousands of fragmented, inconsistently formatted patient records from many health systems. ACS needed a solution that could match human accuracy and scale beyond it.
Layer Health deployed AI models to abstract 40 fields across four cancer types (breast, ovarian, colon and hematologic), covering standard NAACCR cancer registry elements plus custom research variables. Each model was validated against gold-standard answers and deployed in weeks. ACS achieved:
Accuracy that matched or exceeded human abstractors on every element
Dramatically less variability than human abstraction
750 patients abstracted in under 6 hours
Research-grade outputs with direct links to chart evidence

