Bounded closed-loop control can adjust insulin while patients and clinicians retain targets and safety response.
Collections: Regulated autonomy
Executive brief
Automating a frequent decision can improve outcomes when authority boundaries and safety fallback are explicit.
AI value · HbA1c
Verified7.3% AID vs 7.7% control at week 13; adjusted difference −0.6 percentage points
Open device intervention funded by Tandem; one severe hypoglycemia event occurred in the AID group.
Before
Uses the pretrial insulin-delivery method and continuous glucose information to manage insulin. → Adjust therapy using ordinary care and respond to high/low glucose.
After
Uses CGM data to modulate insulin delivery between patient interactions. → Manage meals, device operation, supplies, alarms, and clinical review.
Human boundary
The algorithm controls bounded dose delivery; patients and clinicians retain therapy selection, targets, and safety response.
Why it matters
Bounded closed-loop control can adjust insulin while patients and clinicians retain targets and safety response.
Before
How the work ran before the change.
Patient
Uses the pretrial insulin-delivery method and continuous glucose information to manage insulin.
ControlControl participants continued pretrial delivery and received CGM; the article does not impose one manual dosing sequence.
Patient and diabetes clinician
Adjust therapy using ordinary care and respond to high/low glucose.
ControlClinical oversight and hypoglycemia safety plan.
What changed
Bounded closed-loop control can adjust insulin while patients and clinicians retain targets and safety response.
Decision rightAI acts within a human backstop
After
How the same work runs now.
Automated insulin-delivery algorithm
Uses CGM data to modulate insulin delivery between patient interactions.
ControlConfigured glucose targets, pump limits, CGM availability, and safety alarms.
Patient and clinician
Manage meals, device operation, supplies, alarms, and clinical review.
ControlHumans retain treatment enrollment and emergency response; severe events follow trial safety procedures.
Exception path
Sensor/pump failure, alarms, hypoglycemia, or unsuitable conditions revert to manual safety procedures.
Decision authority
The algorithm controls bounded dose delivery; patients and clinicians retain therapy selection, targets, and safety response.
| # | Actor | Action | Control |
|---|---|---|---|
| 01 | Patient | Uses the pretrial insulin-delivery method and continuous glucose information to manage insulin. | Control participants continued pretrial delivery and received CGM; the article does not impose one manual dosing sequence. |
| 02 | Patient and diabetes clinician | Adjust therapy using ordinary care and respond to high/low glucose. | Clinical oversight and hypoglycemia safety plan. |
| # | Actor | Action | Control |
|---|---|---|---|
| 01 | Automated insulin-delivery algorithm | Uses CGM data to modulate insulin delivery between patient interactions. | Configured glucose targets, pump limits, CGM availability, and safety alarms. |
| 02 | Patient and clinician | Manage meals, device operation, supplies, alarms, and clinical review. | Humans retain treatment enrollment and emergency response; severe events follow trial safety procedures. |
Work that left the path
Human role before
Patients and clinicians calculate or adjust insulin through their existing delivery method.
Human role after
Patients oversee the device and clinicians manage therapy while the algorithm adjusts routine delivery.
AI role
Closed-loop modulation of insulin from CGM data.
8.2% AID and 8.1% control7.3% AID vs 7.7% control at week 13; adjusted difference −0.6 percentage points
13 weeks · 319 randomized adults with insulin-treated type 2 diabetes
Open device intervention funded by Tandem; one severe hypoglycemia event occurred in the AID group.
Anti-pattern
Do not infer that autonomous dosing removes patient workload or severe-event risk.
Questions
Portability conditions
Reputation risk
high