Adoption and response time are part of model performance.
AI value · Time to antibiotic
Verified
1.85 hours earlier when confirmed within three hours
Observational association despite adjustment.
Before
Recognizes possible sepsis through ordinary EHR review and clinical presentation. → Orders antibiotics and other treatment when sepsis is recognized.
After
Continuously monitors EHR data and passively displays a clickable alert icon when criteria are met. → Evaluates the analysis and confirms or dismisses sepsis; timely confirmation initiates treatment sooner.
Human boundary
Provider owns diagnosis and treatment.
Why it matters
An alert only changes care when a clinician confirms it quickly.
How the work changed
Before
How the work ran before the change.
Step 1 of 2
Bedside physician or advanced-practice provider
Recognizes possible sepsis through ordinary EHR review and clinical presentation.
ControlClinical judgment; the study does not decompose a single conventional recognition queue.
Step 2 of 2
Care team
Orders antibiotics and other treatment when sepsis is recognized.
ControlHuman clinical authority and sepsis protocol.
What changed
An alert only changes care when a clinician confirms it quickly.
Decision rightHuman moves from creator to judge
After
How the same work runs now.
Step 1 of 2
TREWS
Continuously monitors EHR data and passively displays a clickable alert icon when criteria are met.
ControlThe provider may open, confirm, dismiss, or leave the alert unaddressed.
Step 2 of 2
Bedside provider
Evaluates the analysis and confirms or dismisses sepsis; timely confirmation initiates treatment sooner.
ControlProvider retains diagnosis and treatment rights; three-hour response timing is the evaluated hinge.
Process model built from the published workflow evidence for Johns Hopkins Health System. Every step, actor, and control appears in full below.Every step, actor, and control
Exception path
Dismissed alerts remain under ordinary surveillance.
Decision authority
Provider owns diagnosis and treatment.
Before
#
Actor
Action
Control
01
Bedside physician or advanced-practice provider
Recognizes possible sepsis through ordinary EHR review and clinical presentation.
Clinical judgment; the study does not decompose a single conventional recognition queue.
02
Care team
Orders antibiotics and other treatment when sepsis is recognized.
Human clinical authority and sepsis protocol.
After
#
Actor
Action
Control
01
TREWS
Continuously monitors EHR data and passively displays a clickable alert icon when criteria are met.
The provider may open, confirm, dismiss, or leave the alert unaddressed.
02
Bedside provider
Evaluates the analysis and confirms or dismisses sepsis; timely confirmation initiates treatment sooner.
Provider retains diagnosis and treatment rights; three-hour response timing is the evaluated hinge.
Work that left the path
Unassisted continuous chart surveillance
Human role before
Bedside providers recognized possible sepsis through clinical presentation and ordinary EHR review, then handed treatment orders to the care team under clinician authority.
Human role after
Providers confirm or dismiss alerts and own antibiotics and treatment.
AI role
Continuously monitors EHR data and posts a sepsis-risk alert.
Outcomes
Time to antibiotic
Verified
Late/dismissed/unaddressed alert→1.85 hours earlier when confirmed within three hours
Two-year deployment · 9,805 sepsis cases; five hospitals
Observational association despite adjustment.
What leaders can reuse
Anti-pattern
Do not present association as randomized causality.
Questions
01Where is the operating threshold set and who can override it?
02What measured result would trigger rollback or retraining?
03Which residual decisions must remain human-owned?
Portability conditions
EHR integration
Named owner
Prospective monitoring
Reputation risk
low
Evidence and authority
What the public record supports.
Current · updated
2 peer reviewed; publication outcomes are verified.
Bundle 1.0.0 · reviewed 2026-09-06 · stable ID 5df4a75ec00584bd
