A Sepsis Alert With 94% Overrides and Two Missed Cases
Mercy Regional (420 beds) deployed a sepsis best-practice advisory 11 months ago. It fires when two SIRS criteria plus a suspected-infection order are present. Current volume is 1,240 alerts per month across ED and med-surg. The override rate is 94%, median time to override is 14 seconds, and chart review of 200 alerts found a positive predictive value of 6% for confirmed sepsis within 24 hours. Two patients with sepsis were missed in the ED last quarter; both died, and both had alerts that were overridden. The sepsis coordinator wants a hard-stop order set for any positive screen. Nursing leadership reports that alerts already interrupt workflow more than 40 times per shift on med-surg. The CMIO proposes disabling the advisory entirely. Analysts note that 38% of overrides carry the comment "not sepsis," and 12% of those patients met sepsis criteria within 12 hours.
1. Interpret these metrics. What do they tell you, and what do they not tell you?
Read the worked answer
The metrics describe a low-value interruptive alert, but they do not support either proposed extreme.
What they tell you:
- A 6% PPV means roughly 15 of every 250 alerts are true positives. The alert is mostly wrong.
- A 14-second median override and 94% override rate indicate the alert is being dismissed before it is read. At that speed, clinicians are not evaluating; they are clearing.
- The 12% figure among "not sepsis" overrides is the most important number: a subset of dismissed alerts were clinically significant within 12 hours. The alert is not useless; it is mistimed or poorly specified, firing before the clinical picture is clear.
- Forty interrupts per shift on med-surg is a workflow tax with almost no yield, and it erodes response to other alerts.
What they do not tell you:
- Whether the two deaths were preventable had the alert been followed. Override is correlated with the outcome, not proven to cause it.
- Whether the trigger is wrong or the timing is wrong. SIRS plus a suspected-infection order is broad and fires early.
- Whether clinicians would respond to a better-targeted alert. The override rate reflects this alert, not all alerts.
A hard stop on 1,240 alerts per month would create roughly 1,165 unnecessary forced stops. That is not a safety intervention; it is a guarantee of workarounds.
2. Recommend a redesign and specify what you would measure to know it worked.
Read the worked answer
I would neither hard-stop nor disable. Redesign in three moves.
Narrow the trigger. SIRS plus a suspected-infection order is too permissive. Add evidence of organ dysfunction or a lactate above threshold, or require a second trigger within a time window. The goal is to raise PPV by trading volume for precision.
Tier the response. Make the first-tier alert non-interruptive - a banner or worklist entry visible to the charge nurse and rapid response team - and reserve an interruptive alert for the higher-specificity trigger. Route it to someone with the authority and time to act, rather than to whoever happens to open the chart.
Make override informative. Require a structured reason (not sepsis, already treating, outside scope, other) with an optional free-text field, and suppress re-firing for a defined period after a documented clinical decision. The 38% "not sepsis" comments are useful signal; structured reasons make them analyzable.
Measures: PPV target above 25%; alert volume target below 300 per month; override rate below 70%; median time from trigger to antibiotic order; sepsis bundle compliance; and two balance measures - time to override and the number of missed sepsis cases identified in review. I would run it as a small test on two med-surg units and the ED for 60 days before scaling, with a pre-specified stopping rule if missed cases rise.
3. The sepsis coordinator is a physician champion and the CMIO outranks her. How do you handle the governance conflict?
Read the worked answer
Treat it as a decision that needs a shared evidence base, not a hierarchy contest, and make the decision rule explicit before the meeting.
First, separate the goals. Both want fewer sepsis deaths; they disagree on mechanism. State that the objective is to reduce missed sepsis without increasing alert burden, and that both hard-stop and disable are being evaluated against that objective.
Second, put the numbers in front of the group in a common frame: at 6% PPV, a hard stop produces about 1,165 forced interruptions a month, and the two missed cases occurred with an alert that was ignored in 14 seconds. That reframes the question from "are alerts good or bad" to "what trigger earns an interruption."
Third, propose a time-boxed pilot with pre-agreed success and failure criteria, so neither side has to concede in the abstract. Give the coordinator a role in designing the higher-specificity trigger and the review of missed cases; that keeps clinical ownership and uses her credibility with prescribers.
Fourth, escalate only if the CMIO blocks a pilot that meets the criteria. If the CMIO wants to disable immediately, ask for the same time-boxed evaluation of the replacement workflow - a worklist or manual screen - so disabling is also tested rather than asserted.
Document the decision, the owner, and the review date. Governance failures here tend to recur as alert-by-alert fights unless the decision rule is written down.