Skip to main content

​
POST /api/summarize-record/claude

Generate a comprehensive nurse-to-nurse handoff summary from patient data using Claude AI. Supports integration with image analysis and comparison with previous handoff notes.
Requires ANTHROPIC_API_KEY in environment variables.

​
Request

​
patientData
object
required
Complete patient record data
​
previousNotes
string
Previous handoff notes for comparison. When provided, the AI will generate a “Changes Since Last Handoff” section highlighting differences.
​
imageAnalysis
string
Analysis from /api/analyze-image/claude. When provided, the AI will integrate whiteboard/document information throughout the summary.

​
Request Example

curl -X POST http://localhost:3001/api/summarize-record/claude \
  -H "Content-Type: application/json" \
  -d '{
    "patientData": {
      "name": "John Doe",
      "age": 67,
      "sex": "M",
      "mrn": "123456",
      "diagnosis": "Acute MI",
      "condition": "Stable",
      "medications": ["Aspirin 81mg", "Metoprolol 25mg"],
      "allergies": ["Penicillin"],
      "code_status": "Full Code",
      "last_vitals": {
        "bp": "142/88",
        "hr": 78,
        "o2sat": 96
      },
      "tasks": [
        {
          "time": "20:00",
          "description": "Evening medications",
          "priority": "high",
          "type": "medication",
          "completed": false
        },
        {
          "time": "14:00",
          "description": "Troponin draw",
          "priority": "high",
          "type": "lab",
          "completed": true
        }
      ]
    },
    "previousNotes": "Patient admitted with chest pain. Troponin pending.",
    "imageAnalysis": "Whiteboard shows BP trending down from 158/92 this morning."
  }'

​
Response

​
summary
string
AI-generated handoff summary with:
  • Changes since last handoff (if previousNotes provided)
  • Patient overview and Shift-Level Forecast
  • Current clinical status and vital signs
  • Key lab/imaging findings with trends
  • Active medications and treatments
  • Task status (completed and outstanding by priority)
  • Safety concerns
  • Integration of image analysis data (if provided)
​
provider
string
AI provider used. Always returns "claude" for this endpoint.

​
Response Example

{
  "summary": "## Changes Since Last Handoff\n\n**COMPLETED:** Troponin draw (14:00)\n**CHANGED:** BP improved from 158/92 to 142/88 (IMAGE UPDATE: confirmed on whiteboard)\n**NEW:** Evening medications due at 20:00\n\n## Patient Overview\n\nJohn Doe, 67M, MRN: 123456\nDiagnosis: Acute MI\nRoom: 401A\n\n## Current Clinical Status\n\nVitals (Current):\n- BP: 142/88 (↓ from 158/92 - IMPROVING)\n- HR: 78\n- SpO2: 96%\n\nTrend Analysis: Blood pressure showing positive response to treatment\n\n## Active Medications\n\n- Aspirin 81mg daily\n- Metoprolol 25mg BID\n\n## Task Status\n\n### Critical Priority\n⚠ 00:00 - Repeat troponin (CRITICAL)\n\n### High Priority\n⚠ 20:00 - Evening medications\n\n### Completed This Shift\n✓ 14:00 - Troponin draw (HIGH priority, lab)\n\n## Shift-Level Forecast\n\n⛈️ High workload at midnight\n- Troponin results expected\n- Possible MD notification needed\n\n☀️ Stable period 02:00-06:00\n\n## Safety Concerns\n\n- Allergies: Penicillin\n- Code Status: Full Code\n- Fall Risk: Assess on entry",
  "provider": "claude"
}

​
Error Responses

​
Features

​
Task Analysis

The endpoint automatically analyzes tasks in the patient data:
  • Separates completed vs. pending tasks
  • Organizes by priority: Critical → High → Medium → Low
  • Highlights time-sensitive tasks
  • Shows task completion metrics
Task fields used:
  • time: Scheduled time
  • description: Task description
  • priority: critical, high, medium, low
  • type: medication, lab, vitals, imaging, etc.
  • completed: boolean

​
Integration with Image Analysis

When imageAnalysis is provided:
  • Merges whiteboard data with electronic records
  • Prioritizes recent information from images
  • Prefixes conflicts with **IMAGE UPDATE:**
  • Weaves naturally into appropriate sections (not separate)
  • Cross-references between sources

​
Comparison with Previous Notes

When previousNotes is provided:
  • Generates “Changes Since Last Handoff” section
  • Highlights differences with prefixes:
    • **CHANGED:** - Modified information
    • **NEW:** - New items
    • **RESOLVED:** - Issues now resolved
    • **COMPLETED:** - Tasks finished

​
Trend Highlighting

Emphasis on clinical trends:
  • Vital sign changes (↑ increasing, ↓ decreasing)
  • Lab value trajectories
  • Condition progression
  • Response to treatments

​
Shift-Level Forecast

Includes workload prediction:
  • ⛈️ High workload periods with reasons
  • ☀️ Stable periods for planning
  • Clustered events: Meds + labs + vitals
  • Timeline visualization concept

​
Best Practices

Include all available fields in patientData:
  • Complete medication lists with dosages
  • All known allergies
  • Recent vital signs history (not just latest)
  • Detailed task list with accurate priorities
  • Admission date and diagnosis
Ensure tasks include:
  • Accurate completed status
  • Realistic priorities (critical, high, medium, low)
  • Specific times (24-hour format)
  • Clear descriptions
  • Task type classification
For best results, combine all three parameters:
  1. First, call /api/analyze-image/claude with whiteboard photo
  2. Retrieve patientData from your database
  3. Get previousNotes from last handoff
  4. Send all three to /api/summarize-record/claude
This creates comprehensive, context-aware handoffs.
Typical integration flow:
// 1. Analyze whiteboard image
const imageResult = await analyzeImage(whiteboardPhoto);

// 2. Get patient data from database
const patientData = await getPatient(mrn);

// 3. Get previous handoff notes
const previousNotes = patientData.handoff_notes;

// 4. Generate new summary
const summary = await summarizeRecord({
  patientData,
  previousNotes,
  imageAnalysis: imageResult.summary
});

// 5. Save new handoff
await saveHandoff(mrn, summary.summary, imageResult.summary);

​
Additional Notes

This application uses Claude Sonnet 4 exclusively. The implementation does not include OpenAI integration.

Build docs developers (and LLMs) love

Get started for freeTalk to us