Data Strategy & Personalization

Loop Health’s platform is built on a unified patient data architecture that enables true 1:1 personalization and generates valuable real-world evidence for clinical research.

Overview

Every piece of health data you generate—lab results, wearable metrics, dose logs, conversations with Luna AI—flows into a unified Patient Graph. This comprehensive health profile enables personalized recommendations, tracks your progress over time, and contributes to aggregate research insights.

Core Capabilities

✅ Currently Implemented

The following data collection and personalization features are live in production:

Unified Patient Identity
Longitudinal Data Collection
1:1 Personalization Engine
Real-World Evidence Generation

🚧 In Development

Enhanced data preservation and research capabilities are being implemented. See Future Enhancements below.

1. Unified Patient Identity

Problem: A single patient exists across multiple systems—Clerk (authentication), BigCommerce (orders), Rimo Health (prescriptions), Stripe (payments), lab providers, wearable devices. How do we know they’re the same person?

Solution: Identity Resolution System

How It Works

The identity_mappings table links all identifiers to one canonical patient:


Patient "Will" =
  ├─ clerk_user_id: user_2xyz...
  ├─ email: will@loop.health
  ├─ bigcommerce_customer_id: 12345
  ├─ stripe_customer_id: cus_abc...
  ├─ hubspot_contact_id: 67890
  └─ phone: +1234567890

Key Features:

Automatic linking: When you place an order, your BigCommerce ID is linked to your Clerk account
Conflict resolution: If an email appears in two systems, identities are merged
Cross-system queries: “Show me all orders for this patient” works across platforms

Implementation:


// packages/patient-graph/src/identity/repository.ts
await identityRepo.resolveIdentity({
  type: 'email',
  value: 'will@loop.health'
});
// Returns: clerk_user_id that links to all other systems

Status: ✅ Fully Implemented

Identity resolution working in production
Supports Clerk, BigCommerce, Stripe, HubSpot, phone, email
Automatic merge on conflicts
Used by all API endpoints

2. Longitudinal Data Collection

Problem: Health optimization requires tracking changes over time—weeks, months, years. One-time snapshots aren’t enough.

Solution: Comprehensive timeline tracking across all health data types.

What We Track

Clinical Data

Data Type	Frequency	Retention	Details
Lab Results	As uploaded	Forever	Parsed biomarkers + reference to PDF
Protocol Doses	Daily	Forever	Scheduled doses + adherence logs
Adherence Events	Per dose	Forever	Completed/skipped with reasons
Weekly Check-ins	Weekly	Forever	Subjective outcomes, side effects
Patient Events	Continuous	Forever	Complete clinical timeline

Biometric Data

Data Type	Frequency	Retention	Source
Wearable Readings	Continuous	2 years hot, forever cold	Oura, Whoop, Dexcom, Libre
Daily Aggregates	Daily	Forever	Sleep, HRV, recovery scores
Glucose Readings	Every 5 min	Forever	CGM devices

Behavioral Data

Data Type	Frequency	Retention	Details
Luna AI Conversations	Per interaction	Forever	Full transcripts + tool calls
Protocol Changes	On update	Forever	Who changed, when, what
Subscription Events	On change	Forever	Started, paused, cancelled

Timeline Reconstruction

At any point, we can reconstruct a patient’s complete health state:


// Get patient state on January 15, 2025
const timeline = await patientGraph.reconstructTimeline(
  customerId,
  new Date('2025-01-15')
);
 
// Returns:
{
  activeProtocols: [...],
  recentLabs: [...],
  wearableMetrics: {...},
  conversationHistory: [...]
}

Status: ✅ Fully Implemented

All clinical data timestamped and preserved
Complete event timeline with patient_events table
Wearable data stored with recordedAt timestamps
Conversation history preserved across channels

Known Gaps (see Future Enhancements):

Lab PDFs not yet uploaded to permanent storage
Wearable syncs currently disabled (jobs exist, need to be enabled)
Protocol edit history not versioned
Some data can be hard-deleted

3. 1:1 Personalization Engine

Problem: Generic health advice doesn’t work. Recommendations must account for YOUR specific labs, medications, conditions, goals, and history.

Solution: Patient context service that loads complete health profile before making recommendations.

How Luna AI Uses Your Data

When you ask Luna a question, here’s what happens:


// 1. User sends message
"Should I take BPC-157 for my knee pain?"
 
// 2. Luna loads your complete patient context
const context = await getPatientContext(userId);
 
// Returns:
{
  profile: {
    age: 34,
    biologicalSex: 'male',
    subscriptionTier: 'premium'
  },
  currentMedications: ['metformin', 'lisinopril'],
  knownConditions: ['type 2 diabetes', 'hypertension'],
  recentLabResults: [
    { code: 'A1C', value: 6.2, status: 'normal', date: '2025-01-15' },
    { code: 'creatinine', value: 1.1, status: 'normal', date: '2025-01-15' }
  ],
  activeProtocols: [
    { title: 'Blood Sugar Management', compounds: ['berberine'] }
  ],
  emergencyFlags: [] // Critical lab values or urgent events
}
 
// 3. Luna uses this context to give PERSONALIZED advice
// - Checks drug interactions (BPC-157 + your current meds)
// - Reviews contraindications (your diabetes, hypertension)
// - References your recent labs (kidney function OK for peptides)
// - Considers your active protocols (already using berberine)
// - Avoids generic responses

Personalization Features

Protocol Recommendations

Based on YOUR lab results, not population averages
Considers YOUR medications for interaction checking
Accounts for YOUR conditions and contraindications
References YOUR goals and preferences

Health Scoring

Weighted algorithm using YOUR data:
- 40% wearable metrics (YOUR sleep, HRV, recovery)
- 30% lab results (YOUR biomarkers vs optimal)
- 20% adherence (YOUR dose logging consistency)
- 10% check-ins (YOUR subjective feedback)

Dose Optimization

Tracks YOUR adherence patterns
Suggests timing based on YOUR schedule
Adjusts for YOUR side effects
Monitors YOUR progress over time

Implementation:


// apps/luna/src/services/patient-context.ts
export async function getPatientContext(userId: string): Promise<PatientContext> {
  const repos = createRepositories(db);
 
  // Load complete patient profile
  const profile = await repos.customerProfiles.findByExternalId(userId);
  const labs = await repos.labResults.list({ customerId, limit: 50 });
  const protocols = await repos.protocols.list({ customerId, status: 'active' });
  const events = await repos.patientEvents.list({ customerId, limit: 100 });
 
  // Extract medications, conditions, emergency flags
  const medications = protocols.flatMap(p => p.items.map(i => i.compound));
  const emergencyFlags = detectCriticalLabs(labs);
 
  return {
    customerId,
    profile,
    medications,
    conditions,
    labResults,
    activeProtocols,
    emergencyFlags
  };
}

Status: ✅ Fully Implemented

Patient context service operational
Luna AI uses patient data for all responses
Protocol recommendations personalized
Health scoring operational
Drug interaction checking enabled

4. Real-World Evidence Generation

Problem: Clinical trials are expensive, slow, and often don’t reflect real-world usage. Pharma companies need data on how their compounds perform in actual patients.

Solution: Aggregate de-identified patient data to generate population-level insights.

Cohort Analysis Capabilities

Currently Available Queries:


// Example: Semaglutide effectiveness analysis
const cohort = await analytics.getCohort({
  compound: 'semaglutide',
  minDuration: 365, // 1 year
  minAdherence: 0.80, // 80%+ adherence
  includeOutcomes: ['weight', 'A1C', 'adherence']
});
 
// Returns:
{
  totalPatients: 8247,
  avgWeightLoss: -14.2, // kg
  avgA1CReduction: -1.3,
  avgAdherence: 0.87,
  dropoutRate: 0.12,
  sideEffects: {
    'nausea': 0.34,
    'fatigue': 0.18,
    ...
  }
}

Real Example - AI Correlations:

Loop already analyzes aggregate data to find patterns:


"Users taking BPC-157 + TB-500 for joint pain report 40% better
outcomes than BPC-157 alone."

Sample size: 247 users
Timeframe: 90 days
Confidence: High

What This Enables

For Pharma Partners (e.g., Novo Nordisk):

“Here’s 10,000 patients on compounded semaglutide for 1+ year”
“Average weight loss: 14.2 kg at 12 months”
“Adherence rate: 87% (vs 50% in clinical trials)”
“Side effect profiles by demographic cohort”
“Dose titration patterns and outcomes”
“Real-world dropout predictors”

For Researchers:

Compound combination synergies
Protocol optimization patterns
Biomarker trends over time
Adherence predictors
Outcomes by demographic groups

Privacy & Ethics:

Minimum 100 patients per cohort
All data de-identified
Aggregate insights only, never individual records
Patient consent for research use

Implementation:


// Current: AI Correlations Feature
// docs/ai-correlations-feature.md
- Weekly GPT-4 analysis of aggregate data
- Stored in Payload CMS for review
- Human approval required before publication
 
// Future: Pharma Partnership API
// Will provide structured queries for authorized research partners

Status: ✅ Partially Implemented

AI correlations feature live (weekly insights)
Adherence analytics API operational
Cohort query capabilities exist
De-identification and privacy controls in place

Gaps (see tickets):

Data completeness (lab PDFs, wearable syncs)
Immutability guarantees (soft deletes needed)
Formal data retention policy

Future Enhancements

The following improvements are prioritized for pharma-readiness and research partnerships:

🚧 In Active Development

Critical Data Preservation (4-6 weeks)

Lab PDF permanent storage (Vercel Blob/S3)
Raw wearable API response storage
Soft delete pattern for all clinical data
Enable wearable sync jobs
Protocol version history tracking
Data retention and archival policy

See Linear Tickets for detailed implementation plans.

Why This Matters:

Ensures data completeness for longitudinal analysis
Prevents accidental data loss
Enables reprocessing with improved algorithms
Meets regulatory retention requirements
Builds confidence with pharma partners

📋 Planned

Advanced Analytics (Q2 2026)

Cohort survival analysis (Kaplan-Meier curves)
Propensity score matching for comparisons
Time-series forecasting for outcomes
Causal inference (protocol effect estimation)

Pharma Partnership API (Q2 2026)

Structured query interface for research partners
Real-time cohort analysis dashboard
Automated report generation
Regulatory-compliant data export

Enhanced Personalization (Q3 2026)

Predictive adherence modeling
Outcome forecasting based on similar patients
Automated protocol optimization suggestions
Real-time intervention triggers

Data Flow Diagram


┌─────────────────────────────────────────────────────────────┐
│                    DATA COLLECTION                           │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  Patient Actions                     External Systems        │
│  ├─ Dose logging (mobile app)       ├─ Oura Ring API        │
│  ├─ Lab PDF upload                  ├─ Whoop API            │
│  ├─ Weekly check-ins                ├─ Dexcom CGM           │
│  ├─ Luna AI conversations           ├─ BigCommerce orders   │
│  └─ Protocol adherence              └─ Rimo prescriptions   │
│                                                               │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│                  IDENTITY RESOLUTION                         │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  Link all identifiers to one canonical patient:              │
│  Clerk ID → Email → BigCommerce → Stripe → Rimo → Phone    │
│                                                               │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│                    PATIENT GRAPH                             │
│               (Unified Data Storage)                         │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  ┌─────────────────┐  ┌──────────────────┐                 │
│  │ Clinical Data   │  │ Biometric Data   │                 │
│  ├─────────────────┤  ├──────────────────┤                 │
│  │ • Lab results   │  │ • Wearable       │                 │
│  │ • Protocols     │  │   readings       │                 │
│  │ • Adherence     │  │ • Daily stats    │                 │
│  │ • Check-ins     │  │ • CGM glucose    │                 │
│  │ • Events        │  │                  │                 │
│  └─────────────────┘  └──────────────────┘                 │
│                                                               │
│  ┌─────────────────┐  ┌──────────────────┐                 │
│  │ Behavioral Data │  │ Commercial Data  │                 │
│  ├─────────────────┤  ├──────────────────┤                 │
│  │ • Conversations │  │ • Orders         │                 │
│  │ • Protocol      │  │ • Subscriptions  │                 │
│  │   changes       │  │ • Prescriptions  │                 │
│  └─────────────────┘  └──────────────────┘                 │
│                                                               │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│                   PERSONALIZATION LAYER                      │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  1:1 Patient Context → Personalized Responses:              │
│                                                               │
│  ┌─────────────┐   ┌──────────────┐   ┌─────────────────┐ │
│  │ Luna AI     │   │ Protocol     │   │ Health Scoring  │ │
│  │ (full       │   │ Recommenda-  │   │ (YOUR metrics)  │ │
│  │  context)   │   │ tions        │   │                 │ │
│  └─────────────┘   └──────────────┘   └─────────────────┘ │
│                                                               │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│              AGGREGATE RESEARCH LAYER                        │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  De-identified cohort analysis:                              │
│                                                               │
│  • 10,000 patients on semaglutide for 1+ year              │
│  • Average outcomes, adherence, side effects                │
│  • Protocol optimization patterns                            │
│  • Real-world evidence for pharma partners                  │
│                                                               │
└─────────────────────────────────────────────────────────────┘

Technical Implementation

Key Packages

@loop/patient-graph - Data access layer with repositories
@loop/shared - Drizzle schema and Zod validations
@loop/database - Supabase client
@loop/core - Result type, error handling, logging
@loop/biomarker-parser - Lab PDF parsing
@loop/health-score - Health score calculation
@loop/correlations - Aggregate analysis

Architecture Patterns

Append-Only Event Sourcing

adherence_events - immutable dose logs
patient_events - complete timeline
conversation_history - full transcripts
Never update, only insert new records

Timestamp Everything

Every record has created_at and updated_at
Can reconstruct state at any point in time
Enables longitudinal analysis

RBAC & Audit Logging

Every data access logged in rbac_logs
Who accessed what, when, why
HIPAA compliance audit trail

Data Strategy & Personalization

Overview

Core Capabilities

✅ Currently Implemented

🚧 In Development

1. Unified Patient Identity

How It Works

2. Longitudinal Data Collection

What We Track

Clinical Data

Biometric Data

Behavioral Data

Timeline Reconstruction

3. 1:1 Personalization Engine

How Luna AI Uses Your Data

Personalization Features

4. Real-World Evidence Generation

Cohort Analysis Capabilities

What This Enables

Future Enhancements

🚧 In Active Development

📋 Planned

Data Flow Diagram

Technical Implementation

Key Packages

Architecture Patterns

See Also