TrialsgrambetaRCTs are beautiful

Nimri AI insulin dosing

2020

Insulin Dose Optimization Using an Automated Artificial Intelligence-Based Decision Support System in Youths with Type 1 Diabetes (ADVICE4U)

In 108 youths with type 1 diabetes on insulin pump therapy, AI-guided remote insulin dose adjustment every three weeks was non-inferior to physician-guided adjustment for time in target glucose range (70–180 mg/dL) over six months.

Result

Percentage of time in target glucose range 70–180 mg/dL
Absolute difference
1.4 points

The absolute difference is derived from the event rates.

Also reported: Readings <54 mg/dL 1.3 ± 1.4% vs 1.0 ± 0.9% (non-inferior, P<0.0001). Three severe diabetes-related adverse events (two severe hypoglycaemia, one DKA) in the physician arm, none with the AI-DSS.

Original publication
Subtopic
Diabetes care
Design
Multicentre, multinational, parallel-group non-inferiority RCT, 6 months
Relevance
Important: one of several pillars.
Source
Nimri R, et al. Nat Med. 2020;26(9):1380-1384.doi 10.1038/s41591-020-1045-7

Discussion & critique

First randomised evidence that an automated algorithm can titrate insulin pump settings as well as specialists at academic diabetes centres: the case for scaling scarce expertise. Non-inferiority against expert physicians in 108 youths over six months; neither arm reached glycaemic targets (time in range around 50%), so the tool matched rather than surpassed specialist care.