Circadian variability of insulin sensitivity: Physiological input for in silico artificial pancreas

Closed-loop control clinical research trials have been considerably accelerated by in silico trials using the Food and Drug Administration-accepted type 1 diabetes mellitus (T1DM) simulator. We have recently demonstrated that postprandial insulin sensitivity (S_I) in T1DM subjects was lower at breakfast (B) than lunch (L) and dinner (D), but not significantly, because of the small population size. The goal of this study was therefore to incorporate this novel information into the University of Virginia/Padova T1DM simulator and to reproduce in silico the observed circadian variability. Twenty T1DM subjects received an identical mixed meal at B, L, and D. S_I was calculated for each meal using the oral glucose minimal model. Seven S_I daily patterns were identified, and their probabilities were estimated. Each in silico subject was linked to a time-varying S_I profile, while random deviations of up to 40% were allowed. Simulations were compared with experimental data. The integrated area above the basal glucose curve values were 2.60±0.91 (B), 1.38±0.91 (L), and 1.44±1.07 (D) 10⁴ min·mg/dL in silico versus 2.87±1.65 (B), 1.98±1.56 (L), and 2.16±2.00 (D) 10⁴ min·mg/dL in vivo. Incremental peak glucose values were 109±33 (B), 80±29 (L), and 81±30 (D) mg/dL in silico versus 136±39 (B), 126±37 (L), and 125±48 (D) mg/dL in vivo. The incorporation of a time-varying S_I into the simulator makes this technology suitable for running multiple-meal scenarios, thus enabling a more robust design of artificial pancreas algorithms.