People with type 1 diabetes mellitus rely on a basal-bolus insulin
regimen to roughly emulate how a non-diabetic person’s body delivers insulin.
Adjusting such regime is a challenging process usually conducted by an expert
clinical. Despite several guidelines exist for such purpose, they are usually
impractical and fall short in achieving optimal glycemic outcomes. Therefore,
there is a need for more automated and efficient strategies to adjust such regime.
This paper presents, and in silico validates, a novel technique to automatically
adapt the basal insulin profile of a person with person with type 1 diabetes. The
presented technique, which is based on Run-to-Run control and Case-Based
Reasoning, overcomes some of the limitations of previously proposed
approaches and has been proved to be robust in front of realistic intra-day
variability. Over a period of 5 weeks on 10 virtual adult subjects, a significant
reduction on the percentage of time in hyperglycemia (<70mg/dl) (from 14.3±5.6
to 1.6±1.7, p< 0.01), without a significant increase on the percentage of time in
hypoglycemia (>180mg/dl) (from 10.2±5.9 to 1.6±1.7, p=0.1), was achieved.