Duplicate-Order Alert With a 94% Override Rate
A 400-bed community hospital's CPOE system fires an interruptive alert for duplicate laboratory orders: it triggers when a provider orders a chemistry panel within 12 hours of a prior panel or a CBC within 24 hours. The alert fires about 1,150 times per month, and 94% are overridden. The laboratory director wants it disabled because it delays care and generates calls. Nursing reports that some providers order panels at 4 a.m. because the alert makes them think the previous order expired. The EHR analyst notes the alert has no filter for the emergency department or ICU, where repeat testing is expected. A vendor offers a laboratory stewardship module that suppresses duplicate alerts and predicts which repeat tests are unlikely to change management. The quality committee wants a recommendation within 30 days.
1. How would you diagnose why the alert is failing, and what data would you collect before changing or disabling it?
Read the worked answer
Diagnosis. A 94% override rate means the alert has low positive predictive value and is poorly targeted. The 4 a.m. ordering behavior suggests the message is ambiguous and is creating a workaround rather than preventing duplicates.
Data to collect
- Override reasons and free-text comments; provider, unit, and shift.
- Whether a repeat result changed management (e.g., transfusion, medication change) and time to result.
- Rate of repeat testing by location; ED and ICU should be analyzed separately.
- Harm signals: delayed results, extra venipuncture, patient complaints.
- Sample chart review of 50–100 overridden alerts.
Likely root causes
- No clinical context or indication.
- No location filter for ED or ICU.
- Alert text does not offer an action such as combining with the existing order or ordering at the next scheduled draw.
- No feedback to high utilizers.
- The alert is interruptive for a low-risk scenario.
Do not simply disable it. Duplicate testing can cause iatrogenic anemia and wasted resources, but alert fatigue is also a safety hazard. The right question is whether a redesigned, targeted intervention can preserve benefit at lower burden.
Options
- Replace the interruptive alert with inline display of the last result and collection time.
- Require an indication for repeat testing, with expected-repeat options for ICU and ED.
- Reserve interruptive alerts for high-risk combinations, such as a CBC within four hours without indication.
- Add order-set logic and laboratory stewardship feedback.
2. Recommend a redesigned alert strategy and the governance and metrics needed for a pilot.
Read the worked answer
Recommended design: a two-tier, indication-aware approach.
- Tier 1 (non-interruptive): show the last result, collection time, and delta in the order screen for all repeats.
- Tier 2 (interruptive): fire only for high-risk combinations, such as a CBC within four hours on a non-ICU patient without an indication, or repeat chemistry after a critical value, and only when the clinician has not documented a reason.
- Allow structured indications (expected repeat, monitor therapy, pre-procedure) that suppress the alert but remain auditable.
- Add one-click actions: combine with existing order, defer to next draw, or cancel duplicate.
Governance. Convene a laboratory stewardship group with laboratory medicine, hospital medicine, ED, nursing, and informatics. Define success metrics before the pilot: alert volume, override rate, repeat testing rate, rate of repeat results that change management, and harm events. Pilot on one unit for 8–12 weeks using PDSA cycles.
Why not hard-stop or disable. A hard stop would convert a low-value alert into a workflow blocker and likely increase workarounds. Disabling entirely would remove the small safety benefit. The tiered design matches alert intrusiveness to clinical risk.
Communication. Explain the change to providers, publish the rationale, and give feedback to high utilizers. Document the decision and re-review at three months.
3. How would you evaluate the vendor's predictive laboratory stewardship module before purchase?
Read the worked answer
Do not purchase on the vendor's AUC alone. An AUC of 0.82 from external validation does not establish local performance, calibration, or workflow benefit.
Evaluation plan
- Local silent validation for 3–6 months on historical and prospective data.
- Assess calibration, positive predictive value at the alert threshold, and alert burden per clinician per shift.
- Subgroup performance: ED, ICU, oncology, immunocompromised, and pediatric if applicable.
- Compare against current rules and against a simple baseline; measure incremental value.
- Workflow: where the alert appears, what action it recommends, and whether the recommended action is feasible.
- Explainability and clinician trust; plan for drift monitoring and retraining.
- Contract terms: data rights, update frequency, performance guarantees, and exit rights.
Decision rule. Adopt only if the module improves a predefined outcome (e.g., time to antibiotics or repeat-test reduction) without increasing alert burden or causing alert fatigue, and if local validation confirms adequate positive predictive value. Otherwise, optimize the existing rules first.