Chapter 7: Alert Fatigue Prevention

"Alert fatigue is the greatest enemy of CDSS. Research reporting that clinicians override 49–96% of drug-safety alerts shows how serious the problem is."

— Drawing on van der Sijs et al., JAMIA 2006

7.1 What Is Alert Fatigue?

Alert fatigue is the phenomenon in which an excessive number of alerts leads clinicians to ignore even important warnings.

📊 How Serious the Problem Is

⚠️ Consequences of Alert Fatigue

• Missed critical warnings — even life-threatening DDIs are ignored
• Delayed decisions — time spent reviewing alerts
• More burnout — stress on clinical staff
• Loss of trust in the CDSS — the system itself is distrusted
• Reduced patient safety — paradoxically, safety decreases

7.2 Alert Optimization Strategies

1️⃣ Severity-Based Tiering

Do not present every alert the same way; differentiate by severity. WIA-MED-015 defines five alert levels (critical, high, medium, low, info). DDI severity from Chapter 4.2 maps onto them: contraindicated → critical, major → high, moderate → medium, minor → low or info.

2️⃣ Context-Aware Alerts

Filter alerts by taking the patient and situational context into account.

Example: Warfarin-Aspirin interaction

❌ Unconditional alert (bad):
"Warfarin and Aspirin interaction warning"
→ the patient has already been taking both safely for a year

✅ Context-aware (good):
IF the patient has been on the combination for 3+ months
AND no bleeding events
AND INR stable
THEN suppress the alert (or downgrade to informational)

IF new prescription (combination just starting)
THEN show a major warning

3️⃣ Focus on Clinical Significance

Concentrate on real clinical risk rather than theoretical interactions.

Principles:

• Filter along the scale "possible" < "probable" < "highly probable"
• DDIs supported only by isolated literature reports get low priority
• Emphasize only interactions with a high real-world rate of adverse events

4️⃣ De-duplication

Preventing the same alert from repeating:

❌ Bad:
On order entry:     "DDI warning"
On saving order:    "DDI warning" (duplicate!)
On signing order:   "DDI warning" (duplicate again!!)

✅ Good:
On order entry:     "DDI warning" (once only)
Afterwards: the user has already acknowledged → no further alerts
(re-warn only if there is a significant change)

5️⃣ Offer Alternatives

Do not merely report the problem — provide a solution alongside it.

❌ Ineffective alert:
┌─────────────────────────────┐
│ ⚠️ Drug interaction detected │
│ Warfarin + Aspirin          │
│ [OK]                        │
└─────────────────────────────┘

✅ Effective alert:
┌──────────────────────────────────────────────┐
│ ⚠️ Warfarin + Aspirin interaction            │
│                                              │
│ Risk: significantly increased bleeding       │
│                                              │
│ 🔄 Recommended alternatives:                 │
│  [ Review antiplatelet need ]  (1 click)     │
│  [ Reduce aspirin dose ]       (81 mg)       │
│                                              │
│ 📋 If combined, monitor:                     │
│  • INR twice a week                          │
│  • Watch for signs of bleeding               │
│                                              │
│ [Choose alternative] [Approve] [Details]     │
└──────────────────────────────────────────────┘

7.3 User Customization

👤 Per-Physician Preference Settings

Customization must never allow critical (contraindicated) alerts to be switched off; it adjusts presentation and lower-severity tiers only.

📚 Learning-Based Optimization

The CDSS learns clinicians' patterns and optimizes alerts:

• Override-pattern analysis — alerts that are always overridden → lower priority
• Acceptance-pattern learning — always accepted under certain conditions → change to informational
• Time of day — only the important ones during busy periods
• Patient type — inpatient vs outpatient, acuity

7.4 Monitoring Alert Performance

📊 Key Performance Indicators (KPIs)

📈 Dashboard

Acceptance and override thresholds follow the WIA-MED-015 specification; its integration phase sets the tighter operational override KPI and treats alerts overridden more than 50% of the time for three months or more as candidates for removal. The missed-problem rate is a recommended internal measure. The figures below are sample values illustrating how the dashboard is read, not measured results.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Alert Performance Dashboard - January 2025
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Total alerts: 1,250

⛔ Critical:  45 (3.6%)   acceptance 95% ✅
⚠️  High:    320 (25.6%)  acceptance 42% ⚠️
⚡ Medium:  580 (46.4%)  acceptance 18% ❌
ℹ️  Info:    305 (24.4%)  acceptance  8% ℹ️

By alert type:
Drug DDI:          850   override 62% ⚠️
Allergy:            85   override  5% ✅
Dose error:        120   override 15% ✅
Duplicate therapy:  95   override 45% ⚠️
Guideline:         100   override 78% ❌

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Areas needing improvement:
1. Too many DDI alerts → re-calibrate severity
2. Guideline alerts → improve context
3. Medium severity → consider downgrading to low/info
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

7.5 WIA Alert Design Principles

✅ Ten Principles

1. Only what truly matters — avoid the "boy who cried wolf" effect
2. Actionability — give the physician an action they can take
3. Context awareness — consider the patient's situation
4. Timing — show at the right moment
5. Brevity — only the essentials, clearly
6. Alternatives — provide solutions alongside the warning
7. Learning — improve based on usage patterns
8. Transparency — explain why this alert fired
9. Respect — acknowledge the physician's expertise
10. Measure and improve — continuous optimization

Chapter Summary

弘益人間 · Benefit All Humanity

© 2025 WIA - World Certification Industry Association