Feasibility Study Confirms Potential of AI-Driven Insulin Therapy

In preparation for the MELISSA clinical trial a feasibility study was successfully completed by Debiotech with the support of University of Bern and the University Hospital of Geneva. The feasibility study evaluated the Adaptive Basal-Bolus Algorithm (ABBA) – an AI-powered decision-support system designed to optimise insulin therapy for people living with type 1 or type 2 diabetes on multiple daily injections therapy.

ABBA applies a reinforcement learning to analyse glucose, insulin, and carbohydrate information, providing personalised recommendations for daily basal adjustments and on-demand bolus dosing. Building on earlier in-silico evaluations, which demonstrated improvements in time-in-range and reductions in both hyper- and hypoglycaemia, the feasibility study has now assessed the algorithm’s performance under real-world conditions.

The feasibility study confirmed that the reinforcement learning–based algorithm is safe, usable, and capable of supporting insulin management in adults with type 1 diabetes. In the four-week study, participants with high adherence to ABBA’s personalised dose recommendations showed outcomes that consistently trended in a positive direction: time-in-range increased slightly, hyperglycaemia decreased, and hypoglycaemia did not increase. Although the short intervention period limited statistically significance, the study demonstrates that ABBA can be effectively integrated into daily diabetes management and supports progression to the next phase of MELISSA clinical evaluation.

The successful completion of this feasibility study marks an important milestone for MELISSA. It not only confirms that AI-supported therapy recommendations can be safely integrated into daily diabetes management but also provides the evidence needed to advance the project to its next major phase: the clinical study of the MELISSA application.