AI Sepsis Mortality Review — Case Series Playbook
A hospital's Mortality Review Committee is reviewing 12 sepsis deaths in a 6-month period. The medical director suspects that delayed recognition in the emergency department is contributing to mortality. Average time from ED arrival to sepsis recognition is 3.2 hours, against a best-practice target of 1 hour.
When to use this playbook
- Use this playbook when the decision looks like the situation above: A hospital's Mortality Review Committee is reviewing 12 sepsis deaths in a 6-month period.
- It is a fit when you have source files in hand and need a structured, reviewable analysis — not a generic chat answer about "Sepsis Mortality Review — Case Series".
- Do not use it as a substitute for licensed, legal, clinical, or authorized official judgment in the domain.
What you'll need
- De-identified medical records for the 12 cases (ED notes, vital signs, lab results, nursing documentation, physician orders)
- ED arrival-to-recognition timeline for each case
- SEP-1 bundle element documentation for each case
- ED volume and staffing data for the relevant periods
- Published sepsis mortality risk factors and early recognition criteria
Attachments: Documents (Documents)
The Prompt
You are a clinical quality specialist conducting a sepsis mortality case series review for a hospital's Mortality Review Committee. I am attaching: Work only from the attached source files. If a conclusion is not supported, say so. Produce: 1. For each of the 12 cases, identify the earliest clinical signal of sepsis (vital signs, lab values) and calculate how long before recognition that signal was present. 2. Classify each death by preventability: clearly preventable, possibly preventable, or not preventable—with the specific documentation supporting each classification. 3. Identify the common failure patterns across preventable cases: missed documentation, lack of sepsis protocol activation, communication failures between nursing and physicians. 4. Calculate the ED workflow time from arrival to first antibiotic for each case and compare to the 1-hour target. 5. Tell me the committee's presentation structure and the quality improvement recommendations that directly address the identified failures. Call out where independent models are likely to disagree, and list follow-up documents a reviewer should request.
What to expect
- Per-case earliest signal analysis with recognition delay
- Preventability classification for all 12 cases
- Common failure pattern analysis
- Arrival-to-antibiotic timeline comparison
- Committee presentation structure and QI recommendations
Review before you act
- Validate this output against source files before relying on it: For each of the 12 cases, identify the earliest clinical signal of sepsis (vital signs, lab values) and calculate how long before recognition that signal was present.
- Validate this output against source files before relying on it: Classify each death by preventability: clearly preventable, possibly preventable, or not preventable—with the specific documentation supporting each classification.
- Validate this output against source files before relying on it: Identify the common failure patterns across preventable cases: missed documentation, lack of sepsis protocol activation, communication failures between nursing and physicians.
- Validate this output against source files before relying on it: Calculate the ED workflow time from arrival to first antibiotic for each case and compare to the 1-hour target.
- Confirm every cited figure, date, counterparty, or requirement against the attached originals — models compress and can drop a qualifier.
- Treat disagreement between models as a review item, especially on classification, materiality, and recommended next action.
- Do not authorize an operational, clinical, legal, credit, or enforcement action solely because the models agree.
Why compare models on this
For Sepsis Mortality Review — Case Series, running the same attachments across independent models is useful because the hard part is classification and completeness, not fluency. The workflow is already designed to surface per-case earliest signal analysis with recognition delay; preventability classification for all 12 cases; common failure pattern analysis; arrival-to-antibiotic timeline comparison. Those are comparison artifacts — they only exist if more than one model runs. Models disagree on whether an alert is noise, whether a death was sepsis-attributable, and whether a risk model is calibrated. Those disagreements belong in a morbidity-and-mortality style review, not an auto-implemented rule.
See governed multi-model AI on your own prompt
Compare GPT-5, Claude, and Gemini side by side, with human review and a decision record built in.

