AUTHOR=Dong Milan , Liu Wenjun , Luo Yetao , Li Jing , Huang Bo , Zou Yingbo , Liu Fuyan , Zhang Guoying , Chen Ju , Jiang Jianyu , Duan Ling , Xiong Daoxue , Fu Hongmin , Yu Kai TITLE=Glycemic Variability Is Independently Associated With Poor Prognosis in Five Pediatric ICU Centers in Southwest China JOURNAL=Frontiers in Nutrition VOLUME=Volume 9 - 2022 YEAR=2022 URL=https://www.frontiersin.org/journals/nutrition/articles/10.3389/fnut.2022.757982 DOI=10.3389/fnut.2022.757982 ISSN=2296-861X ABSTRACT=Background: GV is a common complication of dysglycemia in critically ill patients. However, there are few studies on the role of GV in the prognosis of pediatric patients, and there is no consensus on the appropriate method for GV measurement. The objective of this study was to determine the “optimal” index of GV in non-diabetic critically ill children. Also, we aimed to confirm the potential association between GV and unfavorable outcomes and whether this association persists after controlling for hypo-or hyperglycemia. Methods: Glucose values were recorded for the first 72 h and were used to calculate the GV. Four different metrics (SD, GLI, MAG and ACACP) were considered and compared to identify the "best" GV index associated with poor prognosis in non-diabetic critically ill children. The primary outcome of this study was the 28-day mortality. ROC curve analysis was conducted to compare the predictive power of different metrics of GV for the primary outcome. The GV index with the largest area under ROC curve (AUC) was chosen for subsequent multivariate analyses. Multivariate Cox regression analysis was performed to identify the potential predictors of the outcome. To compare the contribution in 28-day mortality prognosis between glycemic variability and hyper-or hypoglycemia, performance metrics were calculated, which included AUC, NRI, and IDI. Results: Among 780 participants, 12.4% (n=97) died within 28 days after admission to the PICU. Statistically significant differences were found between survivors and non-survivors in terms of four GV metrics, in which MAG (AUC: 0.762, 95% CI: 0.705–0.819, P<0.001) achieved the largest AUC. Subsequent addition of MAG to the multivariate Cox model for hyperglycemia resulted in further quantitative evolution of the model statistics (AUC 0.651–0.681, P=0.001; IDI: 0.017, P=0.044; NRI: 0.224, P=0.186). The impact of hyperglycemia on outcome was attenuated and no longer statistically relevant after adjustment for MAG (aHR 2.455, 95% CI 1.411–4.270, P=0.001). Conclusions: GV is strongly associated with poor prognosis independent of mean glucose level, demonstrating more predictive power compared with hypo-or hyperglycemia after adjusting for confounding factors. MAG may be a good choice for future research.