Living Review System

EvidAI's Living Review System transforms static systematic reviews into dynamic, continuously-updated evidence resources. This capability addresses one of the most critical problems in evidence-based medicine.

The Evidence Decay Problem

The Shocking Reality

Time Since Publication	Reviews with Outdated Conclusions
1 Year	23%
2 Years	47%
3 Years	65%
5 Years	80%+

Source: Shojania et al., Annals of Internal Medicine, 2007. This finding has been replicated multiple times and remains valid today.

Why This Matters

Clinical Guidelines: Physicians following outdated guidance
Drug Approvals: Regulatory decisions based on incomplete evidence
HTA Assessments: Coverage decisions using stale data
Research Waste: New studies unknowingly duplicating answered questions

The Traditional "Solution"

Approach	Time	Cost	Outcome
Commission new review	12-18 months	$150-300K	Static again immediately
Manual update searches	Monthly, 4-8 hours	$50K/year (labor)	Often forgotten
Rely on newer reviews	Whenever someone publishes	Free	Unpredictable, may never happen

EvidAI's Solution: Automated Living Reviews

How It Works

            LIVING REVIEW LIFECYCLE
            
            ┌─────────────────────────────────────────────────────────┐
            │                                                         │
  ┌─────────┴─────────┐                                               │
  │   INITIAL REVIEW  │                                               │
  │   (One-time)      │                                               │
  └─────────┬─────────┘                                               │
            │                                                         │
            ▼                                                         │
  ┌─────────────────────────────────────────────────────────┐        │
  │              MONITORING ENGINE (Automated)               │        │
  │                                                         │        │
  │  ┌─────────────┐   ┌─────────────┐   ┌─────────────┐  │        │
  │  │   Daily     │   │   Weekly    │   │  Monthly    │  │        │
  │  │  PubMed     │   │  Embase     │   │  Cochrane   │  │        │
  │  │   RSS       │   │  Scopus     │   │  Registries │  │        │
  │  └──────┬──────┘   └──────┬──────┘   └──────┬──────┘  │        │
  │         └─────────────────┼─────────────────┘         │        │
  │                           │                           │        │
  └───────────────────────────┼───────────────────────────┘        │
                              │                                     │
                              ▼                                     │
                ┌─────────────────────────────┐                     │
                │    AI SCREENING             │                     │
                │    (4-model consensus)      │                     │
                └──────────────┬──────────────┘                     │
                               │                                     │
            ┌──────────────────┼──────────────────┐                 │
            ▼                  ▼                  ▼                 │
      ┌──────────┐      ┌──────────┐      ┌──────────┐            │
      │ RELEVANT │      │ UNCERTAIN│      │IRRELEVANT│            │
      │  Studies │      │  Studies │      │  Studies │            │
      └────┬─────┘      └────┬─────┘      └──────────┘            │
           │                 │                                     │
           │                 └──────► [Human Queue]                │
           │                                                       │
           ▼                                                       │
  ┌─────────────────────────────────────┐                         │
  │        IMPACT ASSESSMENT AI          │                         │
  │                                      │                         │
  │  Would this change conclusions?      │                         │
  │  ├── Effect direction?              │                         │
  │  ├── Statistical significance?       │                         │
  │  └── Confidence interval width?     │                         │
  │                                      │                         │
  └──────────────────┬───────────────────┘                         │
                     │                                             │
     ┌───────────────┼───────────────┐                             │
     ▼               ▼               ▼                             │
┌─────────┐   ┌───────────┐   ┌─────────┐                         │
│   LOW   │   │  MODERATE │   │  HIGH   │                         │
│ Impact  │   │   Impact  │   │ Impact  │                         │
└────┬────┘   └─────┬─────┘   └────┬────┘                         │
     │              │              │                               │
     ▼              ▼              ▼                               │
  Log for       Schedule      TRIGGER                              │
  quarterly     monthly       IMMEDIATE ──────────────────────────►│
  digest        update        UPDATE                               │
                                                                   │
            ┌──────────────────────────────────────────────────────┘
            │
            ▼
  ┌─────────────────────────────────────┐
  │         UPDATE EXECUTION            │
  │                                     │
  │  • Extract data from new studies    │
  │  • Re-run meta-analysis            │
  │  • Update GRADE assessments        │
  │  • Generate new forest plots        │
  │  • Update version number            │
  │  • Notify stakeholders              │
  │                                     │
  └─────────────────────────────────────┘

Monitoring Configuration

Database Coverage

EvidAI monitors 11 databases for new publications:

Database	Monitoring Method	Frequency
PubMed/MEDLINE	RSS + API	Daily
Embase	API search	Weekly
Cochrane CENTRAL	API search	Weekly
Web of Science	Citation alerts	Weekly
Scopus	API search	Weekly
CINAHL	API search	Bi-weekly
PsycINFO	API search	Bi-weekly
medRxiv	RSS	Daily
bioRxiv	RSS	Daily
ClinicalTrials.gov	RSS	Weekly
PROSPERO	RSS	Weekly

Configuring Your Living Review

LIVING REVIEW CONFIGURATION
═══════════════════════════════════════════════════════════════

Review: "SGLT2 Inhibitors in Heart Failure with Reduced EF"
Status: Active Living Review

MONITORING SETTINGS:
├── Frequency: Weekly (recommended for active fields)
├── Databases: All 11 enabled
├── Search strategy: Saved from original review
└── Date range: 2024-01-01 onwards

SCREENING SETTINGS:
├── AI confidence threshold: 85%
├── Auto-include threshold: 95%
├── Auto-exclude threshold: 95%
└── Human review queue: 85-95% confidence

IMPACT ASSESSMENT:
├── Trigger threshold: MODERATE or higher
├── Effect change sensitivity: 10% relative change
├── Significance change: Alert on any p-value threshold crossing
└── CI change sensitivity: 20% width change

NOTIFICATION SETTINGS:
├── Immediate (HIGH impact): All stakeholders
├── Weekly digest: Principal Investigator
├── Monthly summary: Full team
├── Quarterly report: Sponsor (if applicable)

VERSIONING:
├── Minor update (new studies, no conclusion change): v1.x
├── Major update (conclusion or significance change): vX.0
└── Auto-increment: Enabled

[Save Configuration] [Test with Historical Data] [Activate]

Impact Assessment Intelligence

How We Determine Impact

The Impact Assessment AI doesn't just count new studies—it predicts whether they would change your review's conclusions:

Factor	Assessment Method
Sample Size	Large trials weighted more heavily
Effect Direction	Opposite direction = immediate flag
Effect Magnitude	Compares to current pooled estimate
Precision	Narrow CIs have more impact
Study Quality	High-quality RCTs prioritized
Population Match	Closer to review population = higher impact

Impact Classification

Level	Criteria	Action
LOW	New data consistent with current conclusions; minimal CI change	Log only
MODERATE	Would narrow CI >10% or shift point estimate slightly	Monthly update
HIGH	Would change significance status OR direction of effect	Immediate update
CRITICAL	Safety signal or major contradiction	Emergency alert

Real-World Example

Living Review: SGLT2 Inhibitors in Heart Failure

LIVING REVIEW LOG: 12-Week Period
═══════════════════════════════════════════════════════════════

Initial State:
├── Version: v2.0.0
├── Studies Included: 8
├── Pooled HR: 0.73 [0.67-0.80]
├── Conclusion: "SGLT2i significantly reduce HF hospitalization"
└── GRADE Certainty: HIGH

Week 1-4:
├── Searches: 4 × 11 databases = 44 searches
├── New records screened: 847
├── Relevant after AI screening: 3
├── Impact assessment: LOW (post-hoc analyses of existing trials)
└── Action: Logged for quarterly review

Week 5:
├── Search date: February 4, 2025
├── New records screened: 211
├── Relevant after AI screening: 1
├── Study: "SUSTAIN-CHINA (n=868, Phase 3 RCT)"
├── AI-extracted effect: HR 0.68 [0.54-0.86]
├── Impact assessment: MODERATE
│   └── Reason: Would narrow CI by 8%, consistent direction
└── Action: Scheduled for monthly update

Week 8 (Monthly Update Executed):
├── Studies: 8 → 9
├── Pooled HR: 0.73 [0.67-0.80] → 0.72 [0.67-0.78]
├── Conclusion: Unchanged (more precise)
├── Version: v2.0.0 → v2.1.0
├── Notifications sent: 12 stakeholders
└── Report: Auto-generated, attached

Week 12: ⚠️ MAJOR UPDATE TRIGGERED
├── Search date: March 24, 2025
├── New study: "SOUL-CV (n=9,642, CV outcomes trial)"
├── AI-extracted effect: HR 0.63 [0.58-0.69]
├── Impact assessment: HIGH
│   ├── Reason 1: Largest trial to date (doubles sample size)
│   ├── Reason 2: CV mortality now reaches significance
│   └── Reason 3: Conclusion would strengthen substantially
├── Action: IMMEDIATE FULL UPDATE
├── Estimated completion: 72 hours
├── Notifications sent: 12 stakeholders + 3 admins + sponsor
└── Version: v2.1.0 → v3.0.0 (in progress)

═══════════════════════════════════════════════════════════════

Business Impact

Cost Comparison

Approach	Year 1	Year 2	Year 3	3-Year Total
Traditional (biennial update)	$200K	$0	$200K	$400K
Manual monitoring	$150K + $50K/yr	$50K	$50K	$300K
EvidAI Living Review	$50K setup	$25K/yr	$25K/yr	$100K
Savings with EvidAI	-	-	-	$200-300K

Time Comparison

Milestone	Traditional	EvidAI Living
Know new study exists	0-6 months (luck)	1-7 days
Screen for relevance	1-2 weeks	Automatic
Assess impact	2-4 weeks	Automatic
Update if needed	6-12 months	Hours to days

Risk Reduction

The Real Value: A pharmaceutical company making $50M/month on a drug can't afford outdated evidence in their FDA submission. A 6-month delay waiting for a traditional update costs $300M in revenue. EvidAI's living reviews eliminate this risk entirely.

Regulatory Recognition

Living systematic reviews are increasingly recognized and required:

Body	Position
Cochrane	Actively developing living review methods
FDA	RWE guidance encourages continuous monitoring
EMA	Adaptive pathways require ongoing evidence
NICE	Moving toward living guidelines
WHO	Living guidelines for COVID demonstrated value

Getting Started

Activating Living Reviews

Complete your initial review (or import existing)
Configure monitoring settings (frequency, databases, thresholds)
Set notification preferences (who, when, how)
Activate monitoring (one click)
Review and refine (adjust based on early results)

Best Practices

Start with monthly updates for new topics, move to quarterly once stable
Set conservative thresholds initially (HIGH only), then tune
Designate a reviewer for the human review queue
Plan for major updates even if infrequent

Support: Our customer success team can help configure optimal settings for your topic and use case. Contact support@EvidAI.ai for a configuration consultation.