Common Pitfalls and How to Avoid Them
From the course AI for Pharmacists
The recurring mistakes that undermine AI adoption in pharmacy practice — and the professional safeguards that prevent them.
Pitfall 1: Alert Fatigue from Over-Sensitive AI
Mistake: Deploying an AI interaction checker that flags every theoretical drug-drug interaction regardless of clinical significance, resulting in pharmacists overriding 90%+ of alerts without review.
Why it matters: A pharmacist who habitually clicks through alerts will eventually click through a clinically significant one. Over-sensitive AI creates a safety illusion while degrading actual safety.
Fix: Require any clinical AI tool to report its alert specificity rate (percentage of alerts that require clinical action) before deployment. Reject tools with specificity below 20%. Ask vendors for data from comparable practice settings, not just controlled trials.
Pitfall 2: Using AI Output Without Verification for High-Stakes Decisions
Mistake: Accepting AI-generated drug dosing recommendations, pharmacogenomic interpretations, or prior authorization summaries without independent verification against primary references.
Why it matters: AI clinical decision support tools are trained on datasets that may be outdated, incomplete, or not representative of your patient population (age, comorbidities, ethnicity).
Fix: Establish a tiered verification policy:
- Low-stakes tasks (counseling point summaries, refill reminders): AI output acceptable with brief review
- Medium-stakes tasks (prior authorization documentation): verify key clinical claims against source
- High-stakes tasks (dose adjustments, interaction significance for narrow therapeutic index drugs): always verify against primary reference (Micromedex, Clinical Pharmacology) independent of AI
Pitfall 3: HIPAA Compliance Gaps with Consumer AI Tools
Mistake: Entering patient-identifiable information (name, date of birth, medication list) into general-purpose AI tools (ChatGPT, Claude, Gemini) without a Business Associate Agreement (BAA).
Fix: Use only AI tools with an executed BAA for any patient-identifiable data. For general pharmacy education, workflow template generation, or policy drafting, consumer AI tools are appropriate — just never include patient identifiers.
Pitfall 4: Delegating Patient Judgment to AI
Mistake: Using AI adherence predictions to decide which patients "deserve" additional pharmacist time, effectively letting an algorithm determine care priority.
Why it matters: AI adherence models encode historical patterns that may reflect systemic biases in care access, not patient capability or motivation.
Fix: Use AI adherence scores as a prioritization starting point, not a final triage decision. Review the highest-risk flagged patients yourself and override the algorithm when clinical context warrants it.
Pitfall 5: Failing to Document AI-Assisted Decisions
Mistake: Using AI to generate clinical intervention documentation without noting in the patient record that AI tools assisted in the preparation.
Fix: Establish a documentation standard: when AI tools generate content that becomes part of a patient interaction or clinical record, note "AI-assisted documentation reviewed and verified by [Pharmacist Name, PharmD]." This creates an audit trail and assigns clear professional accountability.
Core Principle
AI is a clinical decision support tool, not a clinical decision maker. The pharmacist's professional judgment, license, and legal accountability are not transferable to an algorithm.
Key Takeaways
- •Require vendors to report alert specificity rates — reject clinical AI tools with specificity below 20%
- •Apply tiered verification: brief review for low-stakes, primary reference check for narrow therapeutic index drugs
- •Never enter patient-identifiable data into any AI tool without an executed HIPAA Business Associate Agreement
- •Document 'AI-assisted, reviewed and verified by [Pharmacist]' to maintain clear professional accountability
Progress is kept in this browser only.