Case Study
How the American Cancer Society Beat Human Accuracy and Cut Registry Abstraction From Weeks to Hours
Layer Health gave ACS a validated data foundation to build on as its cancer research expands to new types and variables.
98%+
Average accuracy across 40 oncology fields
750
Patients abstracted across 4 cancer types
Weeks > Hours
Abstraction turnaround time
About The American Cancer Society
The American Cancer Society (ACS) is a national nonprofit working to end cancer through advocacy, research, and patient support. ACS partnered with Layer Health on its Cancer Prevention Study-3 (CPS-3), a long-term study following more than 300,000 participants across the United States to understand the causes and prevention of cancer. CPS-3 depends on abstracting registry-grade data from thousands of patient records, spanning multiple cancer types, health systems, and EHRs.
The challenge
Research-grade abstraction forces a tradeoff. Teams can abstract thoroughly, or they can abstract at scale. Rarely both.
Standard NAACCR registry fields, plus custom ACS research variables, on every patient
Records spread across different health systems and EHRs, in inconsistent formats
Unstructured data: clinician notes, imaging reports, and scanned PDFs
Abstraction timelines measured in weeks, not hours
CPS-3 needed abstraction that could hold research-grade accuracy while covering more patients, more fields, and more cancer types than manual review could reach.
The solution
ACS partnered with Layer Health to test whether AI could abstract standard registry fields and custom research variables at the accuracy its research demands.
Layer Health reviews the full, longitudinal patient record and returns a suggested answer for each field, every one linked to direct evidence in the chart. Researchers validate the evidence rather than searching for it.
The results
Accuracy: 98%+ AI Accuracy, Exceeding Human Abstraction
AI accuracy matched or exceeded human abstraction across 40 fields. Across 750 patients and four cancer types, Layer Health averaged 98%+ accuracy on 40 fields that human abstractors completed at 96%.
Speed and Scale: Abstraction Time Cut from Weeks to Hours
Research-scale volume, without the wait. Manual abstraction for a study this size runs into weeks. Layer Health processes hundreds of patient charts per hour, turning the same workload around in hours and accelerating tissue retrieval cycle times along the way.
Consistency: Consistent Results, Free from Human Variability
The same field, abstracted the same way, every time. Human abstraction varies from person to person and day to day. Layer Health applies one standard across every record, removing the inconsistency that undermines research-grade data.
Generalizability: Built for Custom Fields and Any EHR
Standard registry elements plus ACS's own research variables. Layer Health abstracted 40 fields across breast, ovarian, colon, and hematologic cancers directly from multi-institutional PDFs that ACS provided, resolving OCR noise, missing metadata, and inconsistent formatting. The platform can also pull structured data via FHIR on Epic and Cerner when it's available.
Research Impact: Evidence-Linked Data Opens New Research Applications
Every abstracted field traces back to the chart. Because each suggested answer links to direct evidence, ACS gains a foundation it can trust and reuse. That accuracy and traceability support real-time data acquisition for time-sensitive research, faster and larger studies, and new applications like clinical trial matching.
Looking ahead
With its Cancer Prevention Study-3, ACS showed AI-guided abstraction holds research-grade accuracy at scale, matching and exceeding human performance across four cancer types and 40 complex fields. Validating a new cancer type in roughly two weeks turns that result into a repeatable path for expanding coverage, not a one-time proof point.
As CPS-3 continues following its cohort of more than 300,000 participants, the same platform extends to additional cancer types, new research variables, and a growing pool of patients, giving ACS an evidence-linked foundation to move faster on the research questions ahead.
