Agent Memory for Healthcare & Medical AI

1. Diagnosis history (temporal + semantic)
   → AI knows what this patient has, what's active, what's resolved

2. Current medications & allergies (keyword)
   → AI checks safety before suggesting treatments

3. Treatment protocols & outcomes (temporal)
   → AI avoids treatments already tried that failed or caused complications

4. Patient preferences & lifestyle (semantic)
   → AI tailors recommendations to patient values and constraints

import { Brain } from "brain-ai";

const memory = new Brain({
  namespace: "healthcare",
});

// When patient presents with symptoms:
async function handlePatientEncounter(patientId: string, chief_complaint: string) {
  // 1. Get diagnosis history
  const diagnoses = await memory.retrieve({
    agentId: `patient-${patientId}`,
    query: "What diagnoses and comorbidities does this patient have?",
    topK: 5,
    filters: { type: "diagnosis_history" },
  });

  // 2. Get medications and allergies
  const medsAndAllergies = await memory.retrieve({
    agentId: `patient-${patientId}`,
    query: "current medications allergies adverse reactions",
    topK: 10,
    filters: { type: "medication_fact" },
  });

  // 3. Get treatment history
  const treatments = await memory.retrieve({
    agentId: `patient-${patientId}`,
    query: chief_complaint, // Semantic: find similar past presentations
    topK: 5,
    filters: { type: "treatment_protocol" },
  });

  // 4. Get patient preferences
  const preferences = await memory.retrieve({
    agentId: `patient-${patientId}`,
    query: "patient preferences lifestyle constraints decision-making",
    topK: 3,
    filters: { type: "patient_preference" },
  });

  // Build clinical context
  const clinicalContext = {
    diagnoses,
    medsAndAllergies,
    treatments,
    preferences,
    chiefComplaint: chief_complaint,
  };

  // AI generates differential diagnosis and safe treatment plan
  const assessment = await medicalAI.diagnose(chief_complaint, clinicalContext);

  // Store encounter summary in memory
  await memory.store({
    agentId: `patient-${patientId}`,
    content: `Encounter (${new Date().toISOString()}): CC: ${chief_complaint}\nAssessment: ${assessment.diagnosis}\nPlan: ${assessment.plan}`,
    type: "diagnosis_history",
    metadata: {
      patientId,
      timestamp: new Date().toISOString(),
      chief_complaint,
      diagnosis: assessment.diagnosis,
      icd_codes: assessment.icd_codes,
      confidence: assessment.confidence,
    },
  });

  return assessment;
}

Chief Complaint: "I'm having joint pain and fatigue."

AI Memory Lookup:
1. Diagnosis history:
   - Type 2 diabetes (since 2018, well-controlled HbA1c 7.2%)
   - Hypertension (on lisinopril, BP 130/80)
   - Prior MI (2023, angioplasty + stent placed)
   - Resolved: Seasonal allergies

2. Current medications:
   - Metformin 500mg BID
   - Lisinopril 10mg daily
   - Atorvastatin 20mg daily
   - Aspirin 81mg daily
   - No drug allergies

3. Treatment history:
   - Knee pain (2020): Physical therapy → improved
   - Shoulder pain (2022): Corticosteroid injection → worked for 18 months
   - Prior imaging: X-rays 2024 (normal), no MRI contraindications

4. Patient preferences:
   - Prefers non-pharmaceutical options first
   - Lives rural, limited access to specialists
   - Values education and shared decision-making

AI Reasoning:
"Joint pain + fatigue in diabetic patient could be rheumatoid arthritis, diabetic neuropathy, or medication side effect. Given:
- No prior rheumatologic history
- Well-controlled diabetes
- Recent atorvastatin dose stable
Consider: Check RF/CCP (screen for RA), TSH (hypothyroidism), check if statin myopathy

Safe recommendations: Physical therapy first (patient preference + prior success), avoid NSAIDs (cardiac history), consider rheumatology referral if serology positive."

interface HealthcareMemory {
  agentId: string; // "patient-{patientId}"
  namespace: "healthcare";

  // What was stored
  content: string; // The clinical note, diagnosis, or fact

  // How to retrieve it
  type: "diagnosis_history" | "medication_fact" | "treatment_protocol" | "patient_preference";
  metadata: {
    patientId: string;
    encounterId?: string;
    timestamp: string;
    
    // For diagnosis_history
    icd_codes?: string[]; // ICD-10 codes
    status?: "active" | "resolved" | "past_medical_history";
    onset_date?: string;
    
    // For medication_fact
    med_name?: string;
    dosage?: string;
    frequency?: string;
    contraindications?: string[];
    
    // For treatment_protocol
    procedure_type?: string;
    outcome?: "success" | "partial" | "failure";
    
    // For patient_preference
    category?: "communication" | "lifestyle" | "decision_style";
  };
}