Health History Tracking

Health History Tracking

Beyond conversation memory, Vitamem can serve as the persistent health history layer for your AI application — accumulating a rich, searchable record of a user’s health journey across many sessions over months or years.

What to Track

Vitamem’s fact extraction works best when your extraction prompt focuses on specific categories:

Category	Examples
Conditions	”Has hypertension”, “Diagnosed with anxiety disorder”
Medications	”Takes lisinopril 10mg daily”, “Recently stopped statins”
Vitals & labs	”A1C was 7.2 in March”, “Blood pressure averaging 130/85”
Lifestyle	”Exercises 4x per week”, “Follows a low-carb diet”
Goals	”Wants to lose 15 pounds by summer”, “Aiming for A1C under 7.0”
Care team	”Primary care: Dr. Chen”, “Sees endocrinologist quarterly”
Family history	”Father had heart disease”, “Mother has Type 2 diabetes”

Example: Chronic Condition Tracking

// Extraction prompt tuned for chronic condition management
async extractMemories(messages) {
  const prompt = `Review this health check-in conversation and extract health history facts.

Categories to extract:
- Medical conditions (diagnosed or suspected)
- Medications and supplements (name, dose, frequency)
- Lab values and vital signs (with dates if mentioned)
- Lifestyle factors (diet, exercise, sleep, stress)
- Health goals and progress
- Symptoms mentioned
- Care team and appointments

Conversation:
${messages.map(m => `${m.role}: ${m.content}`).join('\n')}

Return JSON only: [{ "content": "specific factual statement", "source": "confirmed" | "inferred" }]
Be specific — include numbers, dates, and dosages when provided.`;

  const raw = await this.chat([{ role: 'user', content: prompt }]);
  return JSON.parse(raw);
},

Building a Health Timeline

You can reconstruct a user’s health timeline by querying their memories:

// Get a broad snapshot of the user's health history
const healthSnapshot = await mem.retrieve({
  userId: 'user-789',
  query: 'health conditions medications lifestyle diet exercise',
  limit: 20,
});

// Format as a health summary for injection into a new session
const healthSummary = healthSnapshot
  .filter(m => m.score > 0.7) // Only high-relevance memories
  .map(m => `- ${m.content} (${m.source === 'confirmed' ? 'stated' : 'observed'})`)
  .join('\n');

const systemPrompt = `You are a health companion for this user.

Known health history:
${healthSummary}

Use this context to give personalized, relevant support.`;

Handling Health Record Updates

When a user’s health situation changes (new diagnosis, medication change, etc.), Vitamem handles it naturally:

• The new fact is extracted at dormant transition
• Deduplication compares it against existing memories
• If it’s genuinely new (e.g., “Now on insulin” when previously only metformin was recorded), it’s saved
• If it’s an update to an existing fact (e.g., A1C went from 7.4% to 6.8%), the supersede mechanism updates the existing memory in-place with the new value. The storage adapter tracks previousValue for vitals, giving you a history of changes.

Note

Supersede is automatic when supersedeThreshold is configured (default: 0.75). Facts with cosine similarity between the supersede threshold and the deduplication threshold (default: 0.92) update the existing memory rather than creating a duplicate. If your app needs to track that a user stopped a medication, add explicit extraction guidance for that in your prompt.

Privacy Considerations

Health history data is sensitive. When building on Vitamem:

1. Use encrypted storage — SupabaseAdapter with Row Level Security enabled
2. Scope memories to users — Vitamem already scopes all memories by userId
3. Right to deletion — Use deleteUserData(userId) to remove all memories for a user (GDPR/CCPA compliance)
4. Minimize retention — Use closedTimeoutMs with sweepThreads() to automatically close old threads
5. Don’t store PHI without safeguards — if you’re handling Protected Health Information, consult a compliance specialist

See SECURITY.md for our security policy.