United StatesHealthcare and life sciences2026 – ongoingCarta Healthcare
Claude-based evidence extraction for clinical registry data abstraction
Carta Healthcare cut clinical data abstraction time by up to 66% with Claude
Problem
Abstractors had to search medical records by hand for answers that feed clinical registries. The work was slow and costly, and complex surgical cases took 5-6 hours.
Approach
Carta built its Lighthouse platform on Claude through Amazon Bedrock. In a two-phase pipeline, Claude Haiku 3.5 and Sonnet 4 extract information from structured data and clinical notes, and Claude Sonnet then suggests answers and ranks the evidence. Abstractors now validate findings instead of searching for them, and clinical experts can adjust prompts and test changes in production the same day.
Results
Takeaway
Moving staff from searching records to checking suggested answers cut time most on the complex cases.
Categories
- Industry
- Healthcare and life sciences
- Business function
- Production and quality
- Innovation type
- Process innovationProduct and service innovation
- AI technology
- Generative AI
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Anthropic describes Carta Healthcare's Lighthouse platform, which uses Claude to extract and rank evidence for registry abstraction. It reports up to 66% less time, 50% cost savings and 98-99% inter-rater reliability. In one large health system routine cases fell from 30 to 15-22 minutes and complex cases from 5-6 hours to about 90 minutes. The page carries no date; the Internet Archive first saved it on January 13, 2026.
Full details are in the original sources.