An Analysis and Comparison of Current Artificial Pancreas Algorithms in Robust Patient Environments

As Type 1 Diabetes management advances toward Automated Insulin Delivery (AID) systems, physiological delays in insulin absorption and carbohydrate digestion continue to fundamentally limit algorithmic efficacy. To quantify these constraints, this research utilized the in silico UVA/Padova model to systematically evaluate standard closed-loop control architectures under diverse metabolic variances. Across two simulation phases, standard reactive Proportional-Integral-Derivative (PID) control was evaluated against open-loop Basal-Bolus therapy and an enhanced PID controller featuring an Insulin-on-Board (PID-IOB) tracking constraint during unannounced meal challenges. Results demonstrated that standard PID frequently induced severe late-stage hypoglycemia due to delayed insulin stacking, placing 60% of a 10-patient adult cohort into the critical Lower-D Zone of the Control-Variability Grid Analysis. While incorporating the PID-IOB safety constraint successfully mitigated these severe crashes, it concurrently increased peak hyperglycemia and reduced absolute Time in Range (TIR) for highly sensitive profiles. Ultimately, this thesis demonstrates that purely reactive control methodologies cannot simultaneously neutralize rapid meal spikes and prevent delayed insulin crashes, establishing a rigorous physiological validation framework essential for the future development of proactive Artificial Pancreas algorithms.