Tight glycemic control (TGC) in intensive care unit (ICU) patients represents an active research field as it has been proved its mortality and cost reduction effects. Previous works demonstrated that insulin sensitivity plays an important role in this question. The paper investigates by two defined metrics patient's variability in insulin sensitivity based on a previously introduced (ICING) glucose-insulin model. These metrics grasp the deviations of the actual insulin sensitivity data from their predictions; hence, characterizing a patient's variability at a given time. We also introduce and examine a way to characterize variability across longer time periods and across different patients. Investigations are applied to an actual longitudinal database consisting of n = 261 patients with 47836 hours of measurement in total, with patients grouped according to diagnosis groups formed from their Apache III codes. Data was also segregated according to time spent in ICU (in days). Kruskal-Wallis-test was then employed (separately for different days) to assess whether patients in different diagnosis groups exhibit significantly different variability. Differences were further analyzed with Tukey HSD post-hoc testing. Results show that insulin sensitivity decreases with time and that differences between diagnosis groups diminish. However, there are significant differences on the first two days of stay in the ICU according to one of the metrics, with cardiac patients being more variable and gastric patients (especially non-operative ones) being less variable.