OBJECTIVE: This study evaluated the predictive capability of simple linear extrapolation of continuous glucose data in postsurgical patients undergoing intensive care. METHODS: Twenty patients, both with or without an established diagnosis of diabetes mellitus, scheduled to undergo cardiothoracic surgery were included. Glucose was continuously monitored in the intensive care unit with a microdialysis-based subcutaneous glucose monitoring system. The prediction horizon (PH) with respect to a given glucose reading was calculated by extrapolating the linear trend of the glucose signal and subjected to both analytical and clinical assessment (by calculation of the average duration of consecutive positive and negative glucose signal trends, the root mean squared error [RMSE], and by insulin titration error grid [ITEG] analysis, respectively). RESULTS: In total, 609 h of continuous glucose data from 17 patients were analyzed. The average duration of consecutive positive and negative glucose signal trends was 7.97 (3.99-19.98) min (median, interquartile range). An increase in the RMSE of 0.5 mmol/L (9 mg/dL) was associated with a PH of 37 min. A strong increase in the number of data points in the unacceptable violation zone of the ITEG was associated with a PH of approximately 20 min. CONCLUSIONS: Our data provide evidence that simple linear extrapolation of glucose trend information obtained by continuous glucose monitoring can be used to predict the course of glycemia in critically ill patients for up to 20-30 min. This "glimpse into the future" can be used to proactively prevent the occurrence of adverse events.
OBJECTIVE: This study evaluated the predictive capability of simple linear extrapolation of continuous glucose data in postsurgical patients undergoing intensive care. METHODS: Twenty patients, both with or without an established diagnosis of diabetes mellitus, scheduled to undergo cardiothoracic surgery were included. Glucose was continuously monitored in the intensive care unit with a microdialysis-based subcutaneous glucose monitoring system. The prediction horizon (PH) with respect to a given glucose reading was calculated by extrapolating the linear trend of the glucose signal and subjected to both analytical and clinical assessment (by calculation of the average duration of consecutive positive and negative glucose signal trends, the root mean squared error [RMSE], and by insulin titration error grid [ITEG] analysis, respectively). RESULTS: In total, 609 h of continuous glucose data from 17 patients were analyzed. The average duration of consecutive positive and negative glucose signal trends was 7.97 (3.99-19.98) min (median, interquartile range). An increase in the RMSE of 0.5 mmol/L (9 mg/dL) was associated with a PH of 37 min. A strong increase in the number of data points in the unacceptable violation zone of the ITEG was associated with a PH of approximately 20 min. CONCLUSIONS: Our data provide evidence that simple linear extrapolation of glucose trend information obtained by continuous glucose monitoring can be used to predict the course of glycemia in critically illpatients for up to 20-30 min. This "glimpse into the future" can be used to proactively prevent the occurrence of adverse events.