How We Reduced Sepsis Mortality by 18% With Better Clinical Decision Support

Two years ago, our sepsis mortality rate was above the national benchmark. Leadership wanted an improvement initiative. The IT team wanted to build an alert. I wanted to try something different.

The standard approach to sepsis detection in EHR systems is an alert that fires when certain criteria are met — temperature, heart rate, white blood cell count, suspected infection. The problem is that by the time all the criteria are met and the alert fires, the bedside nurse often already knows something is wrong. The alert confirms what they've already recognized. What nurses needed wasn't a late confirmation — it was earlier pattern recognition support.

Designing for the Workflow

Instead of building another pop-up alert, we embedded sepsis risk information directly into the nursing assessment flowsheet. When a nurse documents vital signs and assessment findings, the system calculates a running SIRS score and a modified qSOFA score in real time, visible right next to the vital signs display.

No pop-up. No interruption. Just a quiet indicator that turns yellow or red as the patient's clinical picture evolves. The nurse sees it in context, during the work they're already doing.

We added one more element: when the indicator turns red, a single button appears: "Initiate Sepsis Bundle." One click opens a pre-built order set with the SEP-1 bundle components — blood cultures, lactate, IV fluids, and antibiotics — ready for the physician to review and sign.

What Changed

We measured outcomes for 12 months after deployment, compared to the 12 months prior:

• Time from first abnormal vital signs to first antibiotic decreased from 4.2 hours to 2.1 hours
• SEP-1 bundle compliance improved from 61% to 89%
• Sepsis mortality decreased by 18% (from 22.4% to 18.3%)
• False positive rate was 12% — significantly lower than the previous pop-up alert system, which had a 34% false positive rate

We didn't achieve these results by being more aggressive with alerts. We achieved them by making the right action easy at the right time.

Why This Approach Worked

Three principles drove the design:

Ambient, not interruptive. The information was always visible but never demanded attention. This avoided alert fatigue while keeping sepsis risk in the nurse's peripheral awareness.

Workflow-integrated, not bolt-on. We didn't create a new screen or a new process. We enhanced the screen the nurse was already using. Zero additional clicks for awareness; one click for action.

Actionable, not just informational. The "Initiate Sepsis Bundle" button removed the cognitive overhead of remembering which orders to place. The nurse identifies the problem; the system makes the response frictionless.

Lessons for Other CDS Projects

The sepsis project taught us a framework we now apply to all CDS design: embed intelligence where clinicians already work, make the right action the easiest action, and measure outcomes — not just alert acceptance rates.

Most CDS projects measure whether clinicians interact with the alert. We measured whether patients lived. That's the standard every CDS initiative should be held to.