GPT-4 Visit Notes System

summer_health · Healthcare · 2024 · source Agentic: no · promotional bias: high · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: pediatric visit documentation

Summer Health needed to turn pediatricians' shorthand visit observations into clear, parent-readable medical visit notes while reducing provider administrative burden and preserving healthcare compliance requirements.

Constraints:

• Summer Health provides pediatric care via text message at any time of day or night
• visit notes must document what was discussed, the care plan, and required follow-ups
• pediatricians normally translate medical shorthand into complete summaries manually
• medical providers spend over 50% of their time on administrative tasks such as writing visit notes
• parents can experience delayed notes that contain medical jargon
• the platform needed HIPAA compliance support through a Business Associate Agreement

Architecture

Summer Health used GPT-4 to generate pediatric visit notes from clinician-written observations. After a pediatrician records shorthand observations during a text-based visit, the LLM summarizes them into clear, jargon-free notes that can be shared with parents. OpenAI's platform was selected because it combined strong medical-language understanding with Business Associate Agreement support for HIPAA compliance.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  parent["Parent"] --> summer_health_text_care_workflow["Summer Health text-care workflow"]
  summer_health_text_care_workflow --> pediatrician["Pediatrician"]
  pediatrician --> clinician_observations["Clinician observations"]
  clinician_observations --> gpt_4_note_generator["GPT-4 note generator"]
  gpt_4_note_generator --> openai_hipaa_capable_platform["OpenAI HIPAA-capable platform"]
  openai_hipaa_capable_platform --> gpt_4_note_generator
  gpt_4_note_generator --> parent_facing_visit_note["Parent-facing visit note"]
  parent_facing_visit_note --> parent

Components

Data flow

A pediatrician provides care over Summer Health's text-message workflow, writes observations during the visit, and passes those observations into an LLM-powered summarization step. GPT-4 converts the shorthand clinical observations into a complete, plain-language visit note covering the discussion, care plan, and follow-ups for parents.

Why this architecture

The system keeps clinical fact creation with the pediatrician and uses GPT-4 for the high-friction transformation step: converting shorthand medical observations into complete, plain-language notes. That architecture targets the provider time sink while using a platform capable of HIPAA-aligned contracting.

Models used: GPT-4

Scale

• Medical care providers spend over 50% of their time on administrative tasks such as writing visit notes.
  “Medical care providers spend over 50% of their time on administrative tasks like writing visit notes”article, stated
• Summer Health offers pediatric care at any time of day or night via text message.
  “Summer Health provides parents with access to fast, reliable pediatric care at any time of the day (or night), all via text message.”article, stated

Alternatives considered (roads not taken)

Manual pediatrician-written visit-note summaries

Rejected because: Manual note writing consumes provider time, contributes to burnout, delays parents, and can leave parents with jargon-heavy notes.

Choose a model platform without healthcare contracting support

Rejected because: Summer Health needed a platform that could provide a Business Associate Agreement for HIPAA compliance.

Use a less capable language model for medical shorthand translation

Rejected because: The use case required understanding complex medical language and adapting output to parent-readable requirements.

Patterns

structured-output-contract (clinical note generation)

The output is not open-ended prose; it must produce a visit note covering what was discussed, the care plan, and necessary follow-ups in parent-readable language.

human-in-the-loop-gate (clinical grounding)

The pediatrician remains the source of clinical observations; GPT-4 transforms those provider-written notes rather than inventing the visit content.

safety-policy-enforcement (healthcare compliance)

Model/platform selection was constrained by the need for Business Associate Agreement support to fulfill HIPAA compliance.

model-routing (capability and compliance fit)

Summer Health chose GPT-4 because the visit-note task required medical-language understanding, user-requirement adaptability, and compliant platform support.

Tradeoffs

Automate the summarization step.

Chose LLM-generated parent-readable notes from pediatrician observations over Pediatricians manually writing full visit summaries from shorthand.

Rationale: The LLM targets the administrative transformation burden while preserving provider-authored observations as the clinical source.

Use GPT-4 through OpenAI.

Chose GPT-4 on a platform with BAA support over A model/platform lacking medical-language strength or HIPAA contracting support.

Rationale: Summer Health needed both robust medical-language understanding and the ability to satisfy HIPAA compliance requirements.

Gotchas (what bit them)

Clinical notes are a provider time sink even after the visit ends.

Root cause: Pediatricians must translate shorthand observations into complete visit summaries, taking time away from patient care.

Resolution: Use GPT-4 to summarize provider observations into complete, shareable notes.

Parent-facing medical communication fails if notes are delayed or jargon-heavy.

Root cause: Traditional notes often arrive late and are hard for parents to understand.

Resolution: Generate clear, jargon-free notes that can be shared with parents.

Healthcare model adoption requires compliance support, not just language quality.

Root cause: The system handles healthcare documentation, so HIPAA-aligned contracting was a gating platform requirement.

Resolution: Select a platform that offered Business Associate Agreements in addition to GPT-4 capabilities.

Evaluation

The source states the goal of increasing pediatric visit-note accuracy and explains why GPT-4 was selected, but it does not provide a detailed evaluation harness, benchmark size, or measured accuracy result.

Metrics: medical providers spend over 50% of their time on administrative tasks such as writing visit notes · GPT-4 selected for medical-language understanding and adaptability · BAA support selected for HIPAA compliance

Outcomes

• Summer Health identified visit notes as a post-visit pain point for both pediatricians and parents.
• The proposed LLM workflow converts provider observations into clear, jargon-free visit notes.
• The architecture targets reduced administrative time for providers and faster, more understandable notes for parents.
• OpenAI/GPT-4 was selected because it combined medical-language capability with BAA support for HIPAA compliance.

The source gives no deployment count or measured production volume; the only quantified scale is that medical providers spend over 50% of their time on administrative tasks such as visit-note writing.