Skip to main content

Building with Claude: Prompting, Tools & Compliance

Duration: 65 min · Level: Advanced · Module: 5. The Personal Medical Biller · Focus: Claude, Anthropic, HIPAA-BAA, tool-use, prompt-engineering

Learning objectives

By the end of this lesson you will be able to explain and apply:

  • Anthropic HIPAA BAA
  • System prompt design
  • Tool use design
  • PHI in prompts
  • Prompt caching

Why this matters​

Claude (Anthropic) is the recommended LLM for US healthcare agent applications because of its HIPAA Business Associate Agreement availability, strong instruction-following for regulated tasks, and reliable refusal of PHI-mishandling requests.

Overview​

Claude (Anthropic) is the recommended LLM for US healthcare agent applications because of its HIPAA Business Associate Agreement availability, strong instruction-following for regulated tasks, and reliable refusal of PHI-mishandling requests. This lesson covers the specific engineering patterns for Claude-based healthcare agents.

Key concepts​

Key idea

Anthropic HIPAA BAA: Anthropic offers BAAs for Claude API customers via Claude for Enterprise; required before any PHI can be sent to the Claude API; sign before building, not after; covers Claude 3 family models

  • System prompt design: healthcare agent system prompts must explicitly state: role and permitted actions, PHI handling instructions (never output PHI in plaintext logs), required disclaimers (not a medical advice system), escalation triggers, and compliance constraints
  • Tool use design: use Claude's tool_use API to give agents structured access to EHR APIs, payer APIs, and database; tools enforce data access controls (Claude can only read data the tool explicitly returns) — more secure than letting the model interpret raw database dumps
  • PHI in prompts: never include more PHI in a prompt than necessary for the specific task; use patient ID + encounter ID for context rather than full name + SSN; retrieve PHI in the tool response only when needed
  • Prompt caching: Anthropic's prompt caching feature caches long system prompts and reference documents (payer policy library, ICD-10 codebook) — reduces cost and latency for agents that make many calls with the same base context; cache hit rate should be >80%
  • Structured output: use Claude's JSON mode or constrained output for all agent actions that feed downstream systems; unstructured free-text output from a coding agent cannot be safely passed to a claims system

Check your understanding​

Try to recall each answer before expanding it.

Q1. What do you know about Anthropic HIPAA BAA?

Anthropic offers BAAs for Claude API customers via Claude for Enterprise; required before any PHI can be sent to the Claude API; sign before building, not after; covers Claude 3 family models

Q2. What do you know about System prompt design?

healthcare agent system prompts must explicitly state: role and permitted actions, PHI handling instructions (never output PHI in plaintext logs), required disclaimers (not a medical advice system), escalation triggers, and compliance constraints

Q3. What do you know about Tool use design?

use Claude's tool_use API to give agents structured access to EHR APIs, payer APIs, and database; tools enforce data access controls (Claude can only read data the tool explicitly returns) — more secure than letting the model interpret raw database dumps

Q4. What do you know about PHI in prompts?

never include more PHI in a prompt than necessary for the specific task; use patient ID + encounter ID for context rather than full name + SSN; retrieve PHI in the tool response only when needed

Q5. What do you know about Prompt caching?

Anthropic's prompt caching feature caches long system prompts and reference documents (payer policy library, ICD-10 codebook) — reduces cost and latency for agents that make many calls with the same base context; cache hit rate should be >80%

References​

  • Claude API Documentation — Tool Use and HIPAA — Anthropic (2024). docs.anthropic.com

← Previous: H5.1 Architecture: The Personal Medical Biller as a System of Agents · Next: H5.3 Patient-Facing Agent: The Billing Advocate Experience →

Part of Module 5: The Personal Medical Biller.