The CRUMB study: closed-loop response to unannounced mixed and carbohydrates-rich breakfasts. A randomized controlled...

Unannounced meals pose a major challenge to type 1 diabetes patients. This study compared the performance of two automated insulin delivery (AID) algorithms, the Medtronic MiniMed 780G's proportional-integral-derivative (PID) and the Tandem t: slim X2's model predictive control (MPC), following unannounced breakfasts (ClinicalTrials.gov identifier: NCT07455643, retrospectively registered the 3rd of February 2026).

In a randomized crossover study, we enrolled 20 children between 11 and 18 years using the MiniMed 780G or the Tandem t: slim X2 AID system. Endpoints included 2-hour and 4-hour blood glucose difference (ΔBG), glucose peak, time to peak, and time spent above (TAR), below (TBS), in range (TIR), and in tight range (TITR). Two meals were tested: a carbohydrate (CHO) meal and a mixed one, both containing 30 g of carbohydrates, and additional 15 g proteins for the mixed meal. Announced (AM) and unannounced (UM) meals were analyzed.

AM, compared to UM, showed significant higher 2-hourΔBG; conversely 4-hourΔBG did not differ significantly. In CHO AM, algorithms were comparable. In mixed AM, PID gained lower peaks and TAR, with higher TIR and TITR. In UM, PID obtained lower 2-hourΔBG than MPC, with reduced peak, TAR, and improved TIR and TITR. Differences in 4-hourΔBG and time-at-peak were not significant.

Both AID algorithms mitigated postprandial glycemia and returned glucose to baseline levels within 4 h without safety concerns. The PID demonstrated higher reactivity to unannounced meals, while performances were comparable when meal boluses were properly announced.

NCT07455643.