Ninety-Four Percent Override Rate and Five Suspect Rules

The proposal conflates volume with value. Alert burden and clinical risk are separate axes. Turning off the five busiest rules would cut volume sharply and could also remove the only protection against the highest-consequence interaction in the list. Evaluate each rule on the right information (evidence-based, patient-specific) and right format (interruptive versus passive), asking whether an override carries any information.

Rule by rule

• Duplicate PPI: low clinical value, high volume, near-universal override. Suppress the interruptive alert entirely; if cost matters, address it through formulary and order-set defaults. This is the clearest removal.
• SSRI plus NSAID, class-wide: real risk but poorly discriminating. Replace the interruptive alert with a passive note in the order composer, or narrow firing to patients with additional risk (age, anticoagulant, prior GI bleed) if the editor allows any logic.
• Meperidine in patients over 65: a Beers criterion. Do not soften it; better, remove meperidine from the formulary or from order sets so the alert rarely needs to fire. Alerting on a drug that should not be on the shelf is a design failure, not a fatigue problem.
• Simvastatin above 40 mg with amlodipine: genuine labeling restriction (20 mg maximum with amlodipine). Keep it, but make it a dosing-guidance alert with the correct maximum and a one-click dose change, rather than a generic warning.
• Warfarin plus TMP-SMX: highest risk, and the 71% override rate with three INR values above 6 shows the alert is firing into a situation where monitoring is not happening. Keep it interruptive and add a monitoring order (INR at 3-5 days) inside the alert, or convert to a hard stop with an embedded alternative.

Governance point. The Five Rights also include the right person, channel, and time. A 1,180-alert month reaching the same prescriber repeatedly at the point of order signing is a channel problem. Route low-severity content to a daily digest or order-set default instead of a modal.

Build a small, repeatable measurement kit rather than a dashboard.

Data

• A monthly extract from the vendor's reporting tool with four fields per alert: rule name, action taken, prescriber, patient. That alone yields fires, overrides, and override rate by rule.
• A 20-chart monthly manual review, stratified by rule, done by pharmacy residents or a staff pharmacist for 30 minutes a week. Review whether the override was clinically defensible and whether any monitoring was ordered.
• A safety query for INR results above 5, bleeding-related diagnoses, and admissions in patients on warfarin plus an interacting drug, run against data the pharmacy already has access to.

Analysis Plot override rate per rule on a p-chart with the pre-change months as baseline. A change is real when a sustained shift appears, not when one month dips.

Evidence that the warfarin rule is working

• The override rate falls while the absolute number of warfarin/TMP-SMX co-prescriptions also falls, or while the proportion with an INR ordered within 5 days rises.
• INR values above 6 in that population decline.
• Chart review shows documented reasoning and monitoring, not just fewer firings.

Evidence the fix backfired

• Prescribers switch to a less effective antibiotic for urinary tract infection in anticoagulated patients, delaying needed treatment.
• Override rate for the warfarin rule stays high while firings fall only because the rule fires less often.
• Overrides concentrate in a few prescribers, indicating a local workaround rather than a system improvement.

Sustainment. Report per-rule volume, override rate, and one safety measure to the governance committee monthly, and retire any rule that cannot show a safety signal within two quarters. Set a standing agenda item so rule creep is caught before volume rebuilds.

Acknowledge the legitimate concern, then separate two different designs. A hard stop that blocks the order with no path forward does create a denial risk and invites a workaround such as ordering the drug through a free-text field. A hard stop with a designed alternative does not. The distinction matters more than the severity label.

Design for the pilot

• The alert presents three one-click paths: choose an alternative agent with dosing guidance; continue TMP-SMX with automatic INR orders at 3 and 7 days plus a follow-up task; or cancel and return to the order.
• Continuing is always available but requires a brief structured reason (e.g., "no alternative based on susceptibilities"), which produces data instead of a reflexive dismissal.
• Place the content at the order set level as well, so the combination surfaces while the prescriber is choosing an antibiotic for UTI, not only after the choice is signed.
• Run the pilot on two units for eight weeks with the prior alert as a comparison, and include a balancing measure: time from presentation to first antibiotic dose in patients on warfarin. If that interval lengthens materially, the design is causing delay and should be revised.

Reasoning to state explicitly

• The current alert already permits the combination; what it does not do is prompt monitoring. Three INR values above 6 in one quarter is evidence that a dismissible warning is not achieving the intent.
• Escalation to a hard stop is justified by consequence, not by frequency. The Five Rights frame this as right format for right information: high-severity, low-frequency, actionable content belongs in an interruptive, guided format.
• Make the decision reversible. Predefine the criteria for stopping the pilot (override rate above 60% with no increase in INR monitoring, or a safety event traced to the alert) so the committee is not defending a decision it cannot revisit.