Prevalence of type 2 diabetes (T2D) is rapidly increasing worldwide. Although clinical trials of insulin treatment show good results, real world outcomes are poor. This is mainly caused by lack of adherence due to complexity of the treatment. Dose need is highly individual, dose adjustments are empirical, and glycemic targets should be set by clinicians based on physiological risk. In many cases, basal insulin dose adjustments are only performed during clinic visits, and reaching the glycemic target can therefore take years in practice.
We propose an adaptive dose guidance algorithm with automated glycemic target setting. The algorithm uses a dose estimation approach, developed using clinical data from 1.925 insulin naïve people with T2D. Based on self-monitored blood glucose and insulin injection data, the adaptive individual dose estimate and its uncertainty is used to propose a next safe and efficient dose. The glycemic target is automatically chosen to minimize risk of hypoglycemia. We test the performance in silico and compare to a simple standard of care algorithm.
In a simulated scenario with low adherence to dose adjustments, 55% and 79% of participants reached the glycemic target after one year using the standard of care and proposed algorithm, respectively. The number of hypoglycemia events remained the same. Results of a simulated high adherence scenario indicate that the performance of the proposed and standard of care algorithms is similar.
The proposed algorithm has the potential to improve glycemic outcomes in a real-world setting where adherence is sub-optimal.