The metrics of glycaemic control in critical care.

INTRODUCTION Trials of tight glucose control have compared measures of central tendency, such as average blood glucose, and yielded conflicting results. Other metrics, such as standard deviation, reflect different properties of glucose control and are also associated with changes in outcome. It is possible, therefore, that the conflicting results from interventional studies arise from effects on glycaemic control that have not been reported. METHODS Using glucose measurements from patients admitted to four adult intensive care units in one UK hospital, we sought to identify metrics of glycaemic control, examine the relationship between them and identify the metrics that are both independently and most strongly associated with outcome. RESULTS We examined nine previously described metrics and identified a further four. Cluster analysis classified these metrics into two families, namely, those reflecting measures of central tendency and those reflecting measures of dispersion. A measure of minimum glucose was also identified but related to neither family. Plots of the quintiles of these metrics against hospital mortality revealed population-specific relationships. Areas under receiver-operating characteristic curves could not identify an optimum metric of central tendency or dispersion. Using odds ratios, we were able to show that the effect of these metrics is independent of one another. CONCLUSION Our results suggest that glycaemic control is associated with outcome on the basis of three independent metrics, reflecting measures of central tendency, measures of dispersion and a measure of minimum glucose.