Intensive care patients often experience hyperglycemia, insulin resistance (low insulin sensitivity), and high endogenous glucose production due to their critical situation. STAR is a model-based glycemic control protocol that uses insulin sensitivity (SI) identified on hourly bases to define patient variability. The numerical calculation during the identification phase of SI may result in negative SI value, which is an indication of high insulin resistance or another pathological patient state. Negative values of SI are physiologically not possible and are prevented in the parameter identification phase by a nonnegative constraint. These cases, when SI is forced to take a non-negative value, potentially result in poor blood glucose (BG) fitting and signaling some model limitations like an estimated low EGP. Using clinical data of 717 patients from three independent ICUs (Malaysia, New Zealand, and Hungary), the time occurrence and durations of constrained SI situations are analyzed, and different practical scenarios were suggested to estimate and handle patient's EGP levels in clinical application. An EGP estimation method is used to estimate the most suitable EGP value based on model fitting. By setting different EGP higher limit values, the fitting error and remaining constrained SI values are also analyzed and assessed. Results show that 96% of these constrained SI situations happen within the first 96H, and 95% of it lasts for 3h. Results also confirm that using an EGP limit higher than 3.5 s shows no further improvement in terms of modeling accuracy. Based on results, the most practical scenario to handle these situations is to keep the increased EGP until four days of treatment passed; after that, if it happens again, we may set back EGP to the initial value after 3h each time we increase it.