Glucose management is an important element of intensive care management. In general, glucose dysregulation is related to negative outcomes in critically ill patients and is a multi-system complication (1). A few authors have explored the effectiveness of glucose levels on the prognosis of glucose dysregulation with different glucose management protocols; however, the results are contradictory, especially regarding the question of whether strict control of glucose levels is linked to improve clinical prognosis (2, 3). Increased glucose variability (GV) increases the risk of poor outcomes, which may explain the contradictory results in the literature (4).

In 2006, Egi et al. (5) demonstrated for the first time that GV is closely related to poor outcomes in critically ill adults. In recent years, a growing number of studies have confirmed that GV is independently associated with unfavorable outcomes in critically ill adults, regardless of its definition (6–10). However, fewer studies, which explored the impact of GV on negative prognosis have focused on pediatric populations, compared to adults, and most of the previous studies (11, 12) were retrospectively conducted in a single center. Thus, prospective multicenter studies are needed on the association between adverse prognosis and GV in pediatric patients. Subsequently, we conducted this study with a prospective analysis of a pediatric cohort to address this clinical issue.

The GV, also known as glucose fluctuation, represents the fluctuation of glucose levels over time and refers to a non-stationary state in which glucose levels fluctuate between high and low values, such as speed, magnitude, and frequency. A few GV metrics appear to be correlated with worse outcomes in critically ill patients, such as standard deviation (7, 8, 13), coefficient of variation (7, 13, 14), mean amplitude of glycemic excursion (7, 13, 15), mean absolute glucose (13, 16), and glycemic lability index (GLI) (15). These measurements differ in the requirements for calculations, such as the frequency or time interval between glucose measurements. Nevertheless, there is currently no consensus on the definition of GV and the “best” index.

Our retrospective study (17) also demonstrated that GV was independently associated with unfavorable outcomes. In addition to GV, hyperglycemia and hypoglycemia have been previously reported to be associated with poor prognosis in pediatric patients (11, 18–20). Therefore, in this study, we prospectively investigated whether the correlation between GV and poor prognosis remained after controlling for hyperglycemia and hypoglycemia. This study aimed to broaden the understanding of the ways in which GV, hypoglycemia, and hyperglycemia are associated with adverse clinical outcomes in children admitted to the pediatric intensive care units (PICUs) and to improve our understanding of glucose management.

We consecutively monitored a total of 2,725 patients in the five participating units during the study period in which 780 patients met the nadir criteria for entry into the study. A flowchart of the case screening is shown in Figure 1.

In total, 97 (12.4%) of the 780 participants died within the 28-day period after PICU admission. At the primary endpoint (28-day mortality), participants were divided into survival and non-survival groups. The primary reasons for ICU admission included 353 cases of respiratory diseases (such as severe pneumonia and asthma), 35 cases of cardiovascular diseases (such asshock), 130 cases of neurological diseases, 85 cases of trauma, 14 cases of postoperative care, 58 cases of sepsis, and 105 other cases (such as oncologic, renal, metabolic, and hematologic cases). A total of 9,142 glucose values were collected of which 65.9% were from arterial blood gas analysis, 3.4% from venous blood, and 30.7% from capillary blood, with a median frequency of 11 (IQR: 8.25–14) glucose monitoring per patient. Demographic information, clinical features, and disease severity between the survival and non-survival groups are shown in Table 1.

Table 2 illustrates the comparative differences in glucose metrics between the survival and non-survival groups, showing that all glucose metrics were significantly lower in the survivors group than in the non-survivors group (p < 0.05).

The AUCs are shown in Table 3, which were plotted to evaluate the ability of glucose metrics and PIM2 score in predicting 28-day PICU mortality. These results separately correlated with PICU 28-day mortality (AUC > 0.700, p < 0.05). However, MAG (AUC: 0.762, 95% CI: 0.705–0.819, p < 0.001) has the largest areas compared to other GV indices in the prediction of PICU 28-day mortality. MAG had significantly better discrimination of PICU 28-day mortality than the other metrics (p < 0.05). Therefore, we used MAG to represent GV and explore its impact on adverse outcomes. Additionally, we utilized an optimal cut-off value for 28-day mortality prediction.

We sequentially included the baseline variables in the Cox regression model with p ≤ 0.1 in Table 1, followed by hypoglycemia and MAG, respectively. The results of the multivariate regression model showed that hypoglycemia was not associated with 28-day mortality (Table 4, Model 2) after adjusting for significant confounders. However, GV represented by MAG (adjusted HR: 2.795, 95% CI: 1.687–4.632, p < 0.001) was still related to PICU mortality after adjustment for hypoglycemia (Table 4, Model 3). The addition of MAG to the model for hypoglycemia led to further quantitative evolution of the model statistics (AUC: from 0.616 to 0.668, p < 0.001; IDI: 0.030, p = 0.006; NRI: 0.325, p = 0.022). There was no interaction between MAG and hypoglycemia.

We constructed the same model described above by replacing hypoglycemia with hyperglycemia. After adjustment for confounding factors, patients with hyperglycemia experienced increased rates of 28-day mortality (adjusted HR for hyperglycemia, 2.026; 95% CI: 1.240–3.310; p = 0.005) (Table 5, Model 2). Subsequent addition of MAG to the model for hyperglycemia resulted in an early incremental improvement in model statistics (AUC: from 0.651 to 0.681, p = 0.001; IDI: 0.017, p = 0.044; NRI: 0.224, p = 0.186). After adjusting for MAG (adjusted HR: 2.455, 95% CI: 1.411–4.270, p = 0.001), the effect of hyperglycemia (adjusted HR: 1.419, 95% CI: 0.815–2.471, p = 0.216) (Table 5, Model 3) on the adverse outcome was attenuated and no longer statistically relevant. In the final model, MAG had a greater prognostic contribution than hyperglycemia (χ²-df (30): 9.115 and 0.528 for MAG and hyperglycemia, respectively).

However, the interaction between hyperglycemia and MAG was statistically significant (p = 0.016). We used Cochran's and Mantel–Haenszel statistics to further explore the relationship between hyperglycemia, MAG, and mortality (Figure 2). Higher mortality was observed in the high GV group than in the low GV group (p < 0.001), regardless of the presence of hyperglycemia. The test of homogeneity of the odds ratio (OR) was p = 0.022; therefore, after stratifying by MAG, in the low GV group, hyperglycemia was relevant to PICU mortality (OR = 2.910, 95% CI: 1.513–5.596, p = 0.001). In the high GV group, hyperglycemia did not affect mortality (OR = 0.846; 95% CI: 0.363–1.974, p = 0.699). This means that patients with hyperglycemia will have different mortality rates depending on the MAG. In contrast, in the high GV group, hyperglycemia did not play a role in mortality events. These results support the possibility that GV may be a stronger predictor of 28-day mortality than hyperglycemia.

Table 6 shows the effect of blood fluctuations on ICU-free days to day 28, MVfree days to day 28, and the incidence of MODSs. The results show that the median ICU-free days in the high variability group were lower than that of the low group (15 [0.0–22.9] vs. 22 [15.1–24.5], p = 0.001). Similar results were observed in the ventilator-free days (21.5 [0.0–26.1] vs. 25.7 [21.0–28.0], p < 0.001). The incidence of MODs was significantly higher in the high-GV group than in the low-GV group (62.9 vs. 29.5%, p < 0.001).

To our knowledge, this is the first multicenter prospective study of pediatric dysglycemia (such as hypoglycemia, hyperglycemia, and GV) conducted in Southwest China, which attempted to establish an association between dysglycemia and 28-day mortality in PICUs. Several important relationships were also observed. First, GV was independently associated with unfavorable outcomes, and it may play a more predictive role in PICU mortality than hyperglycemia and hypoglycemia. These findings emphasize the important role of GV in children admitted to the PICU. Also, MAG may be a dominant index reflecting glucose fluctuations, which considers the speed and magnitude of change, and the time series among glucose values.

A strong relationship between increasing GV and poor prognosis has been consistently reported in adult patients (6–10, 31). Few studies in this field have been conducted in pediatrics; however, similar insights have been obtained. In 2006, Wintergerst et al. (11) retrospectively observed that increased GV was related to increased hospitalization stay and mortality rates in 1,094 PICU admissions in a single center; however, the severity of the disease was ignored. Hirshberg et al. (22) performed a retrospective study of 863 non-diabetic children in the PICU during a 1-year period and found that the incidence of glucose fluctuations (defined as having both hyperglycemia and hypoglycemia) was 6.8%, and glucose fluctuations were still associated with high mortality after adjusting for PRISM score. In another retrospective cohort of 101 critically ill children, researchers found that increased GV was associated with increased mortality (12). Similarly, Bhutia et al. (21) prospectively observed 170 critically ill children and found that GV was related to multi-organ dysfunction. Pinchefsky et al. (32) found that GV was independently associated with the deterioration of brain function in neonates with encephalopathy. Although these studies involved children in a single clinical setting, they add to the evidence that GV is associated with adverse outcomes. In our series, we further prospectively investigated the relationship between GV and deleterious effects in multiple centers, and similar results were reported in our findings. Based on our results, GV was associated with mortality, MODs, and ventilator/ICU-free days, which confirmed the strong predictive role of GV in children in the PICU. This association may be attributed to oscillatory glucose induction that exacerbates oxidative stress, promotes inflammatory responses, and regulates ROS-mediated NFκB/RAGE activation, resulting in endothelial cell injury (33–35). Taken together, these results emphasize the dangerous role of GV in critically ill patients. Indeed, management of GV may be warranted in glucose management regimens.

Our data are also complementary to findings from observational studies that evaluated the association between hypo/hyperglycemia and GV during critical illness and mortality. The results of our study suggest that GV may be a more powerful independent risk factor of mortality in patients in the PICU than hyperglycemia and hypoglycemia. Consistent with our findings, Bagshaw et al. (36) found that glucose fluctuation, defined as having both hypoglycemia and hyperglycemia, was relevant to a larger OR of ICU (1.5 vs. 1.2 vs. 1.0, p < 0.05) and hospital (1.4 vs. 1.2 vs. 1.0, p < 0.05) mortality, compared to only hypoglycemia or hyperglycemia, or neither. Several in vitro or animal experiments (37–39) have demonstrated that oscillating glucose levels aggravate oxidative stress compared to persistent hyperglycemia, resulting in accelerated cell injury and apoptosis. Quagliaro et al. (37) demonstrated that in umbilical vein cells, the levels of protein kinase C-β were higher in the rapid glucose fluctuation group than in the sustained hyperglycemia group.

From the perspective of clinical treatment, a single blood glucose value can be influenced by various factors, such as medication, diet, and stress status. Therefore, hyperglycemia or hypoglycemia determined by a single value may not accurately reflect the metabolic status of critically ill patients. In contrast, GV is calculated using a few glucose values and may reveal dynamic glucose changes. This may be one explanation that glucose fluctuation may be a stronger predictor of mortality than hyperglycemia and hypoglycemia. Therefore, GV should be considered as an important predictor in future studies of prognostic models in critically ill patients, and clinicians should pay more attention to dynamic fluctuations of glucose versus single glucose values.

In contrast to the previous study, which only used a single index that defined GV, our study selected four indicators (SD, GLI, MAG, and ACACP). ROC analysis showed that three of the four GV indicators (SD, GLI, and MAG) had a moderate predicted value for 28-day PICU mortality (AUC > 0.7, p < 0.05). Among all four indices, MAG obtained the largest AUC and was found to be a better indicator of GV than SD. Hermanides et al. (16) illustrated that the glucose profile demonstrated different MAGs even in the presence of the same SD value, and MAG was superior to SD in terms of predicting mortality. Variability should be considered in terms of the speed and magnitude of change, as well as the time interval between glucose measurements; meanwhile MAG does account for this variability, but SD does not. Our retrospective study (17) found that GLI, a variability indicator that also accounts for the arrangement of estimations and time, had increased predictive power for PICU mortality compared to SD, CV, and MAGE (AUC: 0.626 vs. 0.601, 0.599, 0.573). Based on the results of previous studies (7, 15) and our own results, GV index which accounts for the range, speed, time series, should be used; MAG, or GLI may be a good choice. Given its clinical practicality, we introduced ACACP as a GV indicator, which is simple to apply to a treatment protocol (24). This index focused on the comparison of changes in subsequent blood glucose values. Although the AUC of ACACP was smaller than that of MAG in our dataset, its unique calculation and clinical significance suggest that GV metrics should reflect the change from one value to the next and guide clinical management to minimize this change, thereby facilitating glycemic management and improving prognosis in clinical practice. Finding indicators that contain information, such as time series, rate, and range of change, and that are convenient for clinical application is a potential practical target for future treatment.

The American College of Critical Care Medicine suggested that a target BG that is in the range of 100–180 mg/dL may be a reasonable goal for critically ill pediatric patients (40). In Surviving Sepsis Campaign international guidelines, there is a consensus to target glucose levels below 180 mg/dL (10 mmol/L) for children with septic shock (41). These guidelines focus only on the level of absolute glucose values. The Clinical Practice Guideline (40) states that GV needs to be used as an indicator of clinical glucose control management. Our findings also suggest that glucose management should not only consider the glucose value, but also GV. The available literature was inadequate to support recommendations regarding GV measurements and standard indices in pediatric patients. The glucose regimen and monitoring system should be designed to minimize glycemic variability in clinical practice. Tracking and minimizing differences from one blood glucose measurement to the next is a potential practical target for future treatment. ACACP as a practical indicator, combined with continuous glucose monitoring (CGM), will likely increase the safety and acceptance of GV management protocols. CGM may be a good choice for practical, real-time treatment goals, regardless of its high price and requirements for use.

The multicenter, prospective design is a major strength of this study, particularly the choice to assess four GV indices in a large cohort of unselected children in the PICU and to explore their relationship with adverse outcomes. The results have a wider applicability than those of single-center studies. In addition, GV in the early stages may reflect a physiological stress response; indeed, as treatment duration increases, it may be influenced by many factors related to treatment, such as nutritional intake and drug usage (42). In the present study, GV metric calculations included the first 72 h of glucose values and excluded children using glucocorticoids or TPN, reducing the impact of medical management. Thus, GV in our study may predict mortality. However, our data on glucose included only the first 72 h after admission to the PICU, which represents a short period; thus, there may be a bias in the results. However, our large cohort may have eliminated this effect to some extent.

Our study has some limitations. First, there may be ascertainment bias, as patients with abnormal blood glucose levels may have more frequent measurements, while those with normal blood glucose levels have a comparatively reduced number of measurements. As a result, we may miss the record of hyperglycemia or hypoglycemia in some patients. Ideally, CGM should be performed at predetermined time points; however, this will increase the amount of blood sampled and, thereby, the level of discomfort experienced by young children. Advances in CGM technology will facilitate the management or monitoring of early GV in critically ill children. In addition, our glucose measurements were obtained from different blood specimens, and homogeneity could not be ensured. Finally, our study cannot draw conclusions about causality, since the results of observational studies can only assume that an association exists.

Our results endorse and expand on previous literature findings by showing that fluctuations in blood glucose control negatively affect clinical prognosis in children in the PICU, and GV may be more strongly associated with mortality outcomes compared to hypoglycemia and hyperglycemia. Clinicians in PICUs should pay more attention to GV. MAG may be the dominant indicator of GV, which contains information on the speed and magnitude of blood glucose changes. These results emphasize the importance of monitoring GVs in children in the PICU, such as the use of the CGM system. Exploring the mechanisms of GV and optimizing personal glycemic management in PICUs will be the focus of future research to minimize the potential risk of healthcare-acquired events.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

Written informed consent was obtained from the individual(s), and minor(s)' legal guardian/next of kin, for the publication of any potentially identifiable images or data included in this article.

JL participated in the study design, project management, and supervision. MD and WL were responsible for the conceptualization and design of forms, data management and quality control, writing, and editing. JL, BH, YZ, GZ, JJ, and HF were responsible for the project implementation organization and management and data quality control. MD and YL were responsible for the statistical analysis and documentation. MD, WL, FL, JC, LD, DX, and KY were responsible for the recruitment of participants and documentation of data. All authors were contributed to this article and approved the submitted version.

This study was supported by the General Project Fund for National Clinical Research Center for Child Health and Disorders (grant number: NCRC-2019-GP-12).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.