In this multi-center, prospective cohort of 293 critically ill patients, we observed a significant association between higher levels of CGI, assessed by 14-day CGM, and lower risk of both short- and long-term mortality. Even after further adjustment for HbA1c, mean glucose during hospitalization and CV, the association between CGI and mortality remained significant. Our data suggests that CGI holds promise to serve as a new marker for poor prognosis in critically ill patients.

The introduction of CGM technology has provided an opportunity for novel and in-depth insights into glucose regulation. As a new indicator derived from CGM data through time series analysis, glucose complexity may provide complementary—and perhaps more powerful—information on glucose regulation than conventional glycemic analysis. [15, 24,25,26,27,28,29,30] However, currently evidence linking glucose complexity to adverse clinical outcomes is still very limited. For critically ill patients, in 2010, a pilot study involving 42 patients found that glucose complexity, evaluated by detrended fluctuation analysis, is associated with higher mortality [31]. Later in 2012, this association were further confirmed in a post-hoc analysis of 174 patients from two randomized controlled trials [32]. Unlike these two studies with either small sample size or retrospective design, our study was conducted in a multi-center, prospective cohort of 293 patients. In addition, a new CGM technology as well as a new calculation method for glucose complexity were used in the current study and patients were monitored for longer period, with a median of 12 days. Moreover, we considered long-term mortality as an outcome, and found a significant association between CGI and long-term mortality as well.

The underlying mechanisms under the close association between CGI and mortality in critically ill patients can only be speculated. A seductive explanation could be based on the the mild to moderate correlations of CGI with hyperglycemia, hypoglycemia and glycemic variability measures. In line with this, the predictive power of CGI was partially diminished following adjustments for HbA1c, mean glucose during hospitalization, and CV in our study. Therefore, on the one hand, CGI may offer a valuable view of abnormal glucose regulation reflected by these conventional glycemic metrics, which has been well established to be associated with mortality by excessive counterregulatory hormones, production of inflammatory cytokines and reactive oxygen species [33, 34]. On the other hand, it should be noted that although attenuated, the association between CGI and mortality remained largely significant after further adjustment for HbA1c, mean glucose during hospitalization and CV. This implies that CGI may provide added value in mortality prediction among critically ill patients beyond these conventional glycemic metrics. Additionally, as patients with higher CGI had presumably healthier glucose regulatory system, they may have better glucose control after ICU discharge, which may partially explain the higher long-term survival rate in these subjects.

It is interesting that lower CGI (instead of higher CGI) was observed to be related to worsen glucose control and poorer outcomes. Lundelin et al. [31] hypothesized that a healthy glucose regulatory system should be able to detect small changes in glucose concentrations and makes continuous small adjustments, to maintain glucose homeostasis. Thus, for patients with normal physiological conditions, the tracing of glucose concentration would be characterized by frequent small ups and downs, displaying high glucose complexity. In contrast, in disease status, the glucose regulatory system may require bigger changes in glucose concentration to launch a counterregulatory response, displaying low glucose complexity.

Beyond critically ill patients, existing evidence showed that decreasing CGI is correlated with deteriorating glucose regulation [15]. Moreover, CGI may help identify the residual risk of mortality in people with seemingly well-controlled diabetes [16]. Recently, CGI was found to have a more pronounced association with cognitive dysfunction in type 2 diabetes, compared to HbA1c, SD and time in range [35]. Therefore, CGI holds the promise of complementing the current glycemic assessment system evaluated by CGM as an early and sensitive metric. However, further researches are still warranted to explore the possible role of CGI in diverse populations with abnormal glucose metabolism, and its relationships with different outcomes.

The main strengths of this study include a multi-center, prospective study design, use of 14-day CGM in accordance with recommendation from international consensus and guideline [22], and long-term follow-up. However, there are several limitations that should be noted. First, the diabetes status adjusted in the models may not have been completely accurate. Although we prospectively determined diabetes status at the onset of ICU admission based on all available information, the possibility of undiagnosed diabetes influencing our findings remains. Second, because of a limited scope of the medical records used in our study, the nutrition supply data was not available in this study. Finally, the data on potential risk factors of CGI including mood fluctuations, physical activity, and other environmental factors was not available in the current study. Therefore, the possibility of residual confounding could not be completely excluded.

In conclusion, the present study found that lower CGI is significantly and independently associated with increased short- and long-term mortality in critically ill patients. CGI quantifies the complex non-linear features of glucose regulation, which offers an innovative mean of understanding the systemic properties. Therefore, CGI holds promise to serve as a new marker for assessing glucose homeostasis and predicting poor prognosis in critically ill patients. Its clinical value deserves to be further investigated in different populations.