OpenEvidence Performance Tuning

Table of Contents

• Overview
• Prerequisites
• Instructions
• Output
• Error Handling
• Examples
• Resources

Overview

Optimize OpenEvidence clinical query performance for point-of-care response times. Covers query optimization, intelligent caching with TTL management, connection pooling, request batching, and performance monitoring.

Prerequisites

• OpenEvidence integration running
• Monitoring configured (see openevidence-observability)
• Redis for caching
• Access to performance metrics

Performance Targets

Instructions

Step 1: Optimize Query Construction

Remove filler words from questions, limit context to top 5 conditions and 10 medications, set maxCitations: 5 for faster responses.

Step 2: Implement Intelligent Caching

Build ClinicalQueryCache with SHA-256 key generation. Use dynamic TTL: 5min for stat/urgent, 24hr for pharmacokinetics/mechanisms, 1hr for guidelines, default for others.

Step 3: Enable Connection Pooling

Configure HTTPS agent with keepAlive: true, 10 max sockets, 5 max free sockets. Pre-warm connections on startup.

Step 4: Set Up Request Batching

Use DataLoader to batch concurrent queries (max 5 per batch, 50ms scheduling window) within rate limits.

Step 5: Add Performance Monitoring

Instrument queries with Prometheus histograms for latency by specialty/urgency/cached, plus cache hit/miss counters.

Output

• Optimized query construction
• Intelligent caching with dynamic TTL management
• Connection pooling with keep-alive
• Request batching via DataLoader
• Performance metrics and monitoring

Error Handling

Examples

Cache TTL Strategy

Stat/urgent queries:      5 minutes
Pharmacokinetics/mechanisms: 24 hours
Treatment guidelines:     1 hour
Default clinical queries: configurable (default 1hr)

See detailed implementation for advanced patterns.

Resources

• Redis Caching Best Practices
• DataLoader Documentation