Ochsner Health achieves top-decile Epic alert-to-action ratio with AI-driven sepsis clinical decision support

Sepsis is the leading cause of inpatient mortality and carries significant financial burden ($27.7B–$41.5B annually for Medicare patients). Despite international initiatives, sepsis mortality rates had not declined significantly. Prior EHR-integrated clinical decision support tools showed mixed results due to alert fatigue, poor interaction rates, and clinician dismissiveness toward alerts.

Ochsner deployed a novel suite of EHR-integrated clinical decision support tools built on Epic's sepsis early detection predictive algorithm, validated against the Ochsner patient population (AUC noninferior to Epic's broader population). The tools included risk-stratified best practice alerts (BPAs) for providers and nurses with color-coded severity tiers, a real-time shared sepsis checklist and timer visible across all care settings, and an ED triage screening questionnaire. The interface was designed using Nielsen's 10 usability heuristics with iterative PDSA cycles. Deployed stepwise across 14 facilities from June 2021 through November 2022.

The sepsis CDS tools achieved a 16.46% alert-to-action ratio, compared to the Ochsner system-wide average of 8.4%–12.1% for all best practice alerts — placing performance in the top decile of all Epic users. Over the 6-month study period (March–September 2022), alert-to-action ratios improved from 8.7% to 12.2% while total BPAs per inpatient day decreased, indicating reduced alert fatigue alongside higher engagement. Of 1,959 sepsis screenings initiated, 237 patients were confirmed as triple-positive and 39 (16.46%) had the inpatient treatment protocol initiated.

• Intuitive UI design using established heuristics (Nielsen's 10 principles) and iterative PDSA cycles significantly improved clinician engagement with AI-driven alerts compared to legacy tools.
• Suppression logic targeting already-treated or end-of-life patients is critical to reducing false positives and alert fatigue without sacrificing sensitivity.
• Team-wide visualization of clinical decisions via a shared checklist and timer fostered multidisciplinary accountability and outperformed siloed provider-only alert systems.