Deploying a Sepsis Alert in the EHR
A 210-bed hospital plans to go live with an electronic sepsis screening alert in 60 days. The alert fires when two or more systemic inflammatory response criteria plus a suspected infection trigger are documented. Vendor validation data suggest a positive predictive value of about 12% in a general inpatient population. The chief medical information officer supports go-live on schedule. Nursing leadership reports that the current alert burden is already high - clinicians override 84% of existing alerts - and asks for a delay. The informatics team has not yet built an override-reason field, and there is no plan for measuring whether the alert changes time-to-antibiotics. The sepsis committee chair wants the alert live because the hospital's sepsis bundle compliance is 61% against a 75% target.
1. Should the alert go live on schedule? Justify your answer using alert design and measurement principles.
Read the worked answer
Going live on schedule with the current design is hard to defend. Key considerations:
- Positive predictive value of ~12% means roughly seven of eight alerts will be false positives in this population. That is not automatically disqualifying - screening tools trade specificity for sensitivity - but it must be paired with a workflow that makes false positives cheap to dismiss and true positives easy to act on.
- Override rate of 84% on existing alerts signals alert fatigue. Adding a high-volume, low-specificity alert into that environment risks being ignored, which is worse than no alert because it creates a false sense of safety.
- No override-reason field means the organization cannot learn why clinicians dismiss alerts, cannot distinguish bad logic from bad workflow, and cannot improve the tool.
- No outcome measure means the project cannot demonstrate benefit or harm.
A defensible position: delay go-live by a defined period to add an override-reason field, define a small set of outcome measures (time to antibiotics, bundle completion, alert-to-order conversion, override rate by reason), and pilot on two units before house-wide rollout.
2. What would a minimal, credible measurement plan look like?
Read the worked answer
A minimal plan should be small enough to actually run and specific enough to change decisions:
Process measures
- Alert firing rate per 1,000 patient-days
- Override rate, with reasons categorized
- Percentage of alerts leading to an order within 60 minutes
Outcome measures
- Median time from trigger to antibiotic administration
- Sepsis bundle compliance (the existing 61% figure, tracked the same way)
- ICU transfer and mortality among alerted patients, reviewed cautiously because of confounding
Balancing measures
- Total alert volume per clinician shift
- Time spent in alert-related documentation
Build the measures into the pilot so the two-unit comparison is available before house-wide rollout. Report to the sepsis committee monthly, with a pre-agreed threshold for stopping or revising the alert.
3. The sepsis committee chair argues that delay costs lives. How do you respond as a leader?
Read the worked answer
Take the concern seriously rather than dismissing it. The chair is right that sepsis mortality is time-sensitive and that a 61% bundle compliance rate is a real gap.
Respond by separating the goal from the tool. The goal - faster recognition and treatment - can be advanced during the delay through interventions that do not depend on the alert: nurse-driven screening protocols, standing order sets, pharmacist involvement in antibiotic timing, and education on the existing bundle. These can start immediately.
Then set a firm, short delay with named deliverables: override-reason field built, pilot units selected, measures defined, review date set. A delay measured in weeks with a visible plan is different from an indefinite postponement, and saying so explicitly protects the relationship with the committee and keeps pressure on the informatics team to deliver. If the chair still objects, escalate the tradeoff to the medical executive committee with the measurement plan attached rather than resolving it informally.