Smart Insulin Pumps and Automated Dosing Systems

Abstract

Automated insulin delivery has undergone a fundamental technological transition over the past decade, moving from simple sensor-augmented pumps that notify patients of glucose thresholds, through predictive suspend systems that prevent hypoglycaemia autonomously, to fully closed-loop systems that manage both basal and bolus insulin delivery without requiring the patient to count carbohydrates or announce meals. Each step along this continuum has produced measurable clinical improvements over its predecessor, but the magnitude of gain varies substantially across patient populations and clinical contexts, and the user burden implications of each system level have been inconsistently characterised in the literature. This study presents the Smart Insulin Delivery System Assessment Framework (SIDSAF), evaluating five automated dosing system types -- sensor-augmented pump with threshold suspend (SAP-TS), predictive low glucose suspend (PLGS), hybrid closed-loop with meal announcement (HCL-MA), advanced hybrid closed-loop with automated basal-bolus (AHCL-ABB), and fully closed-loop with no meal announcement required (FCL) -- across four clinical contexts: type 1 diabetes adults, type 1 diabetes paediatrics, type 1 diabetes with hypoglycaemia unawareness, and type 2 diabetes on intensive insulin therapy. Performance was scored using the Automated Dosing Effectiveness Score (ADES), a weighted composite of time-in-range improvement (0.30), hypoglycaemia reduction (0.20), HbA1c reduction (0.20), user burden reduction (0.15), and safety profile (0.15). Fully closed-loop systems achieved the highest ADES (0.931) across all four contexts, with the best time-in-range improvement (0.960), hypoglycaemia reduction (0.940), HbA1c improvement (0.940), and user burden reduction (0.960). Advanced hybrid closed-loop systems ranked second (0.914). The performance gradient was consistent: more automation produced better clinical outcomes and lower user burden across all four patient populations evaluated, with the largest absolute gain occurring in the hypoglycaemia unawareness subgroup.