Pharmacist Alone — The Baseline Error Detection Rate
Solo pharmacist review catches most, but not all, prescription errors.
- ~72%: Baseline detection rate (errors caught unaided)
- 90 sec: Avg review time (per prescription)
- ~50: Errors per 1,000 Rx (industry estimate)
- −15%: Fatigue effect (accuracy after hour 6)
Why solo review misses errors
High volume, interruptions, and fatigue erode attention.
Common missed error types
Dose miscalculations and drug interactions slip through most.
The cost of a miss
Undetected errors can cause harm, readmission, or death.
AI Co-Review Begins — A Second Set of Digital Eyes
The AI scans every prescription in parallel with the pharmacist.
- <1 sec: Scan time per Rx (automated pass)
- 12+: Fields checked (dose, interaction, allergy, duplication)
- 100%: Coverage (every prescription scanned)
- 0%: Human workload change (AI runs alongside, not instead)
How the AI reads a script
Structured fields and free text both get parsed.
What triggers a scan
Every new prescription queues instantly for AI review.
Not a replacement
AI assists judgment, it never makes the final call.
AI Flags Candidate Errors — Suspicious Scripts Highlighted
Flagged prescriptions get a dashed gold outline for attention.
- tunable: Flag threshold (sensitivity slider controls it)
- +2–3×: True-positive lift (vs unflagged baseline catch)
- rises: False-positive risk (with higher sensitivity)
- real: Alert fatigue risk (too many flags reduce trust)
What a flag means
Elevated risk score, not a confirmed error.
Sensitivity trade-off
Higher sensitivity catches more, but flags more falsely.
Design goal
Maximize true catches while keeping false alarms low.
Pharmacist Final Judgment — Reviewing Every AI Flag
The pharmacist inspects each flag and decides, override or confirm.
- ~20 sec: Flag review time (per flagged script)
- ~78%: Flag agreement rate (pharmacist confirms AI flag)
- ~22%: Override rate (pharmacist dismisses false flag)
- human: Final authority (AI never auto-rejects a script)
Human stays in the loop
Pharmacist always makes the dispensing decision.
Confirming true errors
Flagged true errors get caught far more often.
Dismissing false flags
Clinical context lets pharmacists clear safe scripts fast.
Combined Detection Rate — Baseline vs. AI-Assisted
Pairing pharmacist judgment with AI flags lifts overall catch rate.
- ~72%: Baseline detection (pharmacist alone)
- ~91%: Combined detection (pharmacist + AI)
- +19 pts: Absolute lift (errors caught)
- ~9%: Residual miss rate (still requires vigilance)
Why the combination wins
AI recall plus human context outperforms either alone.
Limits of the system
Novel or ambiguous cases still need expert judgment.
Next steps
Continuous monitoring keeps both catch and trust high.