Six Naloxone Reversals in Eight Weeks
At a 400-bed community hospital, a review of anesthesia and PACU records identified six episodes in eight weeks in which a patient was found unresponsive or had a POSS score of 0 after general anesthesia and received naloxone. In the preceding two years the department averaged about one such event every eight weeks. All six patients received intraoperative fentanyl plus hydromorphone and PACU hydromorphone; four were taking gabapentin preoperatively and two had known obstructive sleep apnea. Total anesthesia volume was about 9,000 cases in the period. The department chair wants to remove hydromorphone from the PACU formulary today. The operating room manager argues that six events out of 9,000 cases is simply noise and that the review has become 'anesthesia's hobby horse.'
1. Is this signal or noise? Show the reasoning you would use, and respond to both the chair and the manager.
Read the worked answer
Do the arithmetic before choosing a side. If the true underlying rate were unchanged at one event per eight weeks, the probability of observing six or more in an eight-week window is about 0.0006 - roughly one in 1,800. Under a Poisson model with that baseline, six events is not plausible chance. A run chart with six consecutive points above the centreline would also meet a standard special-cause rule.
But the numerator needs interrogation before it drives action:
- Did the denominator actually hold at 9,000 cases, and did case mix shift (more major spine, more OSA, more outpatient to inpatient)?
- Did surveillance change - a new POSS scoring mandate or a new naloxone dispensing policy will increase detection without any change in harm. That is measurement artifact, not improvement failure.
- Are all six events genuinely comparable, or is one a distinct cause (residual neuromuscular blockade, a dropped oxygen supply, a PCA programming error)?
To the manager: six in 9,000 is the wrong denominator. The relevant denominator is not all anesthetics but patients receiving a specific postoperative opioid regimen, and the outcome - an unresponsive patient requiring reversal - is a never-event class, not a statistical curiosity.
To the chair: the signal is real, but removing one drug from a formulary is a solution that does not match the problem. Hydromorphone is a marker for a delivery system, not the cause.
2. Design the improvement effort. What are your outcome, process, and balancing measures, and which changes would you test first?
Read the worked answer
Frame it as a system problem with a key driver diagram. Aim: zero unplanned opioid-related rescue events; secondary aim: reduce high-risk opioid exposure.
Measures
- Outcome: naloxone administrations per 1,000 anesthetics; unplanned escalations of care.
- Process: proportion of patients with a preoperative OSA screen and a risk-stratified monitoring plan; proportion receiving a multimodal regimen with non-opioid adjuvants; proportion with documented sedation scores at required intervals; time from last opioid to first sedation assessment.
- Balancing: mean and worst pain scores, total opioid consumption, PACU length of stay, patient satisfaction, and time to first ambulation. Without these you will simply create undertreated pain.
Changes to test, in order
- Structured case review of all six events with a standard template - this will identify whether the dominant failure is prescribing, monitoring, or handoff.
- A risk-stratified PACU monitoring order set: continuous pulse oximetry with capnography for patients with OSA or receiving neuraxial opioids or gabapentinoids, and mandatory sedation scoring before and after every parenteral opioid.
- A default multimodal order set (acetaminophen, NSAID where appropriate, regional block, dexmedetomidine-sparing strategy) to lower the opioid doses that created the exposure.
- A preoperative gabapentinoid review, since four of six patients were on one.
- A PDSA cycle on a single unit rather than a hospital-wide formulary ban, so you learn before you impose.
3. How will you know whether the intervention worked, and what would make you change course or abandon it?
Read the worked answer
Judge it with a control chart, not a before-and-after percentage. Plot naloxone events per 1,000 anesthetics on a p-chart with the historical baseline as the centreline, and annotate the date of each intervention. You need a stable baseline of roughly 20–25 points, then look for the conventional signals: a point beyond control limits, a run of eight points on one side of the centreline, or a sustained shift.
Interpretation traps
- With events this rare, monthly rates swing wildly; do not declare success after two quiet months. Pre-specify that a change in the special-cause pattern will be required before you claim effect, and review at 6 and 12 months.
- A fall in naloxone use with a simultaneous rise in unrelieved pain scores or in calls to the pain service means you have suppressed the indicator, not the harm.
- A rise in capnography use without a fall in events suggests monitoring alone is insufficient and the prescribing pathway is the true driver.
Change course if: events continue at baseline after the monitoring and order-set changes are reliably in place - then escalate to a formulary restriction, which becomes an evidence-based step rather than a first guess. Abandon an element if it produces clear harm on a balancing measure (for example, capnography alarms driving alarm fatigue and desensitization elsewhere). Stop only after review, publish the reasoning, and keep the measurement running.