Case study: clinical operations
Calder Regional Health
Six hospitals. One warning that arrived too late.
How Calder rebuilt sepsis detection around the people who have to act on it, and what measurably changed in the eleven months that followed.
At a glance
Client
Calder Regional Health
Setting
Six acute-care hospitals, 2,140 beds, one shared electronic record
Scope
Redesign of the sepsis early-warning workflow, from alert logic to bedside escalation
Duration
Eleven months, February to December 2024
Team
Four clinical informaticists, two service designers, one data scientist
01 / The situation
The signal existed. It never reached the bedside.
Calder’s electronic record had been generating sepsis alerts for years. They were accurate often enough to matter. They landed in a central monitoring inbox reviewed by a rotating team of two.
By the time an alert became a phone call, and the phone call became a decision, the median patient had waited most of a shift. Nobody was ignoring the data. The data simply had nowhere useful to go.
Flip the switch to compare
Before
After
Before: alerts routed to a central inbox
After: alerts routed to the bedside nurse
02 / The turn
Three moves. None of them were technology projects.
Move 01
Move the alert to the bedside
Move 02
Give the alert a decision, not a number
Move 03
Make the escalation impossible to miss
Select a move above to read the detail.
03 / The delta
What changed, measured the same way twice.
Flip the switch to compare
Before
After
Time to antibiotics
5.8 hrs
2.1 hrs
Median hours from the first abnormal vital sign to antibiotic administration.
64% faster
Alerts acted on
31%
86%
Share of early-warning alerts with a documented clinical response inside thirty minutes.
Up 55 points
Sepsis mortality
18.4%
10.9%
In-hospital mortality among patients meeting sepsis criteria on admission.
41% lower
“We didn’t buy a better algorithm. We finally gave the one we already had somewhere to land.”
Dr Amara Okonjo, Chief Quality Officer, Calder Regional Health
The outcome, no longer withheld
41%
Fewer sepsis deaths across the network.
Eleven months after network rollout, adjusted for case mix and admission source. Measurement window October to December 2024, n = 4,318 patients.
Method & detail
The parts that usually get left out.
Approach
Eleven weeks of shadowing before a single line of alert logic changed. We rode night shifts in three of the six hospitals and logged every alert from the moment it fired to the moment somebody acted.
Data sources
Alert audit logs, medication administration records and the network’s own sepsis registry. Response times came from system timestamps. Nobody was asked to remember how long they took.
Limitations
This is a before-and-after comparison inside one health system, not a controlled trial. Two hospitals also expanded their rapid-response teams during the same period, which we could not fully isolate.
Timeline
Discovery February to April. Pilot on two wards May to July. Network rollout August to October. Measurement window October to December, after a ninety-day washout before the first reading.