Case Study
Clinical-Grade AI Validated for NSQIP Abstraction
Layer Health consistently matches and outperforms human accuracy in multi-system NSQIP abstraction.
98%+
Average accuracy on complex NSQIP fields
40%
Reduction in abstraction variability
About ACS NSQIP
Layer Health evaluated AI-guided abstraction against expert manual abstraction using 300 NSQIP cases from multiple health systems. AI performance was measured against a gold-standard reference dataset created through independent chart review by expert NSQIP abstractors. AI-guided abstraction achieved human-level or better performance across 40 NSQIP fields and significantly outperformed manual abstraction on two of the most clinically complex variables.
The challenge
NSQIP abstraction is one of the most clinically demanding chart review workflows in healthcare.
Experienced abstractors must review longitudinal patient records, interpret complex registry definitions, and reconcile information documented across operative notes, pathology reports, progress notes, and other structured and unstructured clinical documentation.
As registry programs grow, timely, consistent abstraction becomes harder to sustain. Manual abstraction is resource intensive, and variation in applying complex registry definitions can affect data consistency.
The solution
Layer Health evaluated AI-guided abstraction using 300 historical NSQIP cases sampled across three health systems.
Rather than relying on traditional data extraction, Layer Health uses a two-step clinical reasoning process:
Targeted extraction identifies clinically relevant evidence across structured and unstructured documentation in the longitudinal patient record.
Clinical reconciliation resolves conflicting or temporally separated documentation and synthesizes that evidence into a single registry-ready abstraction aligned with NSQIP definitions.
Performance was evaluated against expert manual abstraction using statistical significance testing, before prospective deployment into live abstraction workflows.
“I really like having Layer Health's support. It prompts me to go back to the definitions and surfaces details in little notes that I wouldn’t routinely see or think to look for. I’m happy to have it to validate my answers.”
NSQIP Abstractor
The results
Layer Health achieved human-level or better performance across all 40 NSQIP variables:
Significantly higher accuracy than manual abstraction on Case Acuity (96% vs. 90%) and Operative Approach (97% vs. 92%)
Remaining variables demonstrated statistically equivalent performance
Review of cases where AI and manual abstraction differed revealed that most differences reflected the complexity of applying registry definitions to real-world clinical documentation
Looking ahead
NSQIP validates the clinical AI capabilities that power a much broader set of healthcare applications.
Layer Health demonstrated that its AI can interpret complex clinical information, reconcile conflicting documentation, and consistently apply detailed clinical definitions.
Those same capabilities support quality improvement, patient identification, clinical decision support, and the next generation of clinical intelligence across the health system.
About Layer Health
Layer Health helps healthcare organizations reduce abstraction burden, scale quality programs, and improve consistency with AI.
Founded by AI and clinical leaders from MIT, Harvard Medical School, Microsoft and Google. Built on a foundation of 200+ peer-reviewed publications and 40,000+ academic citations in AI and healthcare.
Leading healthcare organizations choose Layer Health's clinical AI platform to automate chart review, improve quality reporting, and unlock new applications from their data:
50-60%+ time savings
98%+ validated accuracy
Standardized abstraction practices
A scalable foundation for new registries, quality initiatives, and clinical AI
