Carta Healthcare cuts clinical data processing time by 66% with Claude

Hospitals generate huge volumes of clinical documentation, and manual clinical data abstraction is slow: trained abstractors spend about an hour per routine case (5-6 hours for complex cases), and one health system needed over 11,000 labor hours yearly to abstract 22,000+ surgical cases. Carta Healthcare built its Lighthouse platform using Claude Haiku 3.5 and Sonnet 4 via Amazon Bedrock in a two-phase pipeline: extracting relevant information from records, then synthesizing evidence to answer registry questions for abstractors to validate. Case time dropped 66% overall (routine cases from 30 to 15-22 minutes; complex cases from 5-6 hours to 90 minutes), saving 3,667 to 6,050 annual hours for the large health system, with 99% inter-rater reliability, over 50% cost savings on abstraction work, and 100% customer retention with 90% of customers expanding their engagement.

Anthropic builds the Claude model family, including a dedicated Claude for Life Sciences push used across drug discovery, regulatory writing and clinical documentation.

This sits in the Build stage of the AI use-case lifecycle: Named examples from the foundation-model labs and platforms pharma teams use to actually construct, train and generate the model or application.