How Carta Healthcare Uses Claude to Automate Clinical Data Abstraction
Carta Healthcare deployed Claude to automate clinical data abstraction from hospital records, replacing a manual process requiring roughly one hour per case by trained abstractors. The AI-powered hybrid intelligence system pairs Claude with human clinical expertise to achieve 98–99% inter-rater reliability. The result was a 66% reduction in abstraction time and over 50% cost savings.
Impact
66%
Reduction in clinical data abstraction time
50%+
Cost savings on abstraction work
98–99%
Inter-rater reliability score achieved
Challenge
Manual clinical data abstraction required approximately one hour per case by skilled human abstractors, making it costly and difficult to scale while maintaining the 98–99% inter-rater reliability required by clinical registries.
Solution
Carta Healthcare integrated Claude into a hybrid intelligence workflow that automates document comprehension and data extraction across complex clinical records, with human abstractors serving as validators rather than primary data collectors.
Tools & Technologies
Sign up to read complete case studies, access detailed metrics, and unlock all use cases.
Full Story
Carta Healthcare was founded out of work at Stanford Children's Hospital to address one of healthcare's most persistent operational bottlenecks: manual clinical data collection. Trained abstractors were required to sift through physician notes, lab results, medication records, and surgical reports to answer registry questions — a process consuming approximately one hour per case and demanding deep domain expertise.
The core challenge was not simply automating a task but doing so at the accuracy levels required by clinical registries, where inter-rater reliability (IRR) scores of 98–99% represent the gold standard. Previous automation approaches failed to handle clinically complex, multi-step questions across heterogeneous documentation formats.
Carta Healthcare integrated Claude via Anthropic's API into a hybrid intelligence workflow, where Claude handles document comprehension and initial data extraction while human abstractors serve as validators rather than primary data hunters. The system was designed for the full breadth of clinical registry question types, including those previously considered too complex to automate.
The results demonstrated both efficiency and quality gains simultaneously. Abstraction time dropped by 66%, costs fell by more than 50%, and the solution consistently achieved 98–99% IRR — meeting or exceeding the industry standard. The approach also expanded the scope of automatable questions beyond what any prior solution had achieved.