The long-term metabolic outcomes of dysglycemia during critical illness

Gornik and colleagues should be commended for providing the first report on the long-term glucose metabolism sequelae in survivors of critical illness with no apparent dysglycemia shortly after hospital discharge [1]. The authors hypothesized that hyperglycemia in the intensive care unit unmasks predisposition for future development of diabetes. The interpretation of their findings, however, is limited by several aspects of their methodology. 
 
No systematic control for covariates and confounders was performed. Their cohort was divided, as acknowledged by the authors, by an arbitrary (for the purpose of long-term prediction) glycemic cutoff value. While intuitively appealing, the latter is not necessarily the optimal predictor of future risk of impaired glucose metabolism. When considering the whole cohort of patients with a normal first postdischarge oral glucose tolerance test (the eligible cohort), what was the predictive role for traditional predictors of diabetes mellitus and prediabetes mellitus (DM/pre-DM) (for example, age, family history, body mass index) and for inhospital variables (for example, severity of illness)? It would also be instructive to explore the role of previously studied glycemic indices - such as admission, maximal and mean glucose values - to determine the best predictive cutoff value for future development of DM/pre-DM. 
 
In addition, almost 40% of the eligible cohort was excluded from analysis due to death or discontinuation of follow-up. How many of these patients who had at least the first 12-month follow-up were diagnosed with DM/pre-DM? 
 
Because DM/pre-DM are commonly under-recognized [2], these analyses can provide initial predictive modeling, which may inform future development of public health policy for earlier detection and intervention in survivors of critical illness.

Gornik and colleagues should be commended for providing the fi rst report on the long-term glucose metabolism sequelae in survivors of critical illness with no apparent dysglycemia shortly after hospital discharge [1]. Th e authors hypothesized that hyperglycemia in the intensive care unit unmasks predisposition for future development of diabetes. Th e interpretation of their fi ndings, however, is limited by several aspects of their methodology.
No systematic control for covariates and confounders was performed. Th eir cohort was divided, as acknowledged by the authors, by an arbitrary (for the purpose of long-term prediction) glycemic cutoff value. While intuitively appealing, the latter is not necessarily the optimal predictor of future risk of impaired glucose metabolism. When considering the whole cohort of patients with a normal fi rst postdischarge oral glucose toler ance test (the eligible cohort), what was the predic tive role for traditional predictors of diabetes mellitus and prediabetes mellitus (DM/pre-DM) (for example, age, family history, body mass index) and for inhospital variables (for example, severity of illness)? It would also be instructive to explore the role of previously studied glycemic indices -such as admission, maximal and mean glucose values -to determine the best predictive cutoff value for future development of DM/pre-DM.
In addition, almost 40% of the eligible cohort was excluded from analysis due to death or discontinuation of follow-up. How many of these patients who had at least the fi rst 12-month follow-up were diagnosed with DM/ pre-DM?
Because DM/pre-DM are commonly under-recognized [2], these analyses can provide initial predictive modeling, which may inform future development of public health policy for earlier detection and intervention in survivors of critical illness.

© 2010 BioMed Central Ltd
The long-term metabolic outcomes of dysglycemia during critical illness

Ivan Gornik
We appreciate Dr Oud's sound comments to the limitations of our study [1]; we will try to address them all.
Analysis of covariates was performed, although the results were not included in the paper. Family history of diabetes and body mass index were, in univariate analyses, associated with higher 5-year incidence of diabetes; higher Acute Physiology and Chronic Health Evaluation II score was associated with lower 5-year incidence of diabetes. A logistic regression model that included age, family history of diabetes, body mass index, Acute Physiology and Chronic Health Evaluation II score and intensive care unit hyperglycemia revealed only the fi nal variable as an independent predictor of future diabetes ( Table 1).
Severity of illness was associated with lower risk of diabetes. We have also previously reported the incidence of diabetes following hyperglycemia in sepsis according to the severity of disease [3]. Relative risks for DM in patients who had hyperglycemia in mild sepsis and septic shock were 6.94 and 2.20, respectively. Th e hypothesis is that more severe illness can produce hyperglycemia more frequently even in patients without a predisposition to diabetes, therefore reducing the discriminating capability.
Besides the chosen threshold of 7.7 mmol/l, other possible thresholds and glycemic variables (admission, highest and mean glucose) were possible predictors of future diabetes, but none showed better results in our popu lation. Future studies may show their value, especially in popu lations other than medical intensive care unit patients.
Th e dropout rate was indeed high and the numbers of patients decreased after the fi rst years of follow-up. Th e diff erence in the incidence of diabetes became signifi cant only after 2 years: 3.1% of patients (9/290) in the hyperglycemia group and 0.7% of patients (4/539) in the normo glycemia group were diagnosed with diabetes (P = 0.023). Th e 2-year period could therefore be the minimal follow-up target in future studies.

Competing interests
The author declares that he has no competing interests.