Response to Letter by O'Connell et al
We acknowledge the interest of O'Connell et al regarding the GLIAS study1 which allows us to highlight some questions that, due to conciseness, were not completely addressed in the original publication.
Hyperglycemia in acute ischemic stroke (IS) could be the reflection of many physiological and pathological mechanisms. But in our opinion, the key question is the deleterious effect of high glucose levels itself, not the possible causes of hyperglycemia, and we sought to look for the clinically significant glucose threshold independent of other factors that could influence those levels. It is true that reinstitution of feeding could influence glucose levels but, to minimize this effect, capillary glucose determinations were done before meals. Although glucose levels may have been influenced by stroke severity, we adjust for it in a multivariate regression analysis.
O'Connell et al are also concerned by the exclusion criteria in GLIAS study: coma at admission or premorbid disability, but these are common criteria used in several studies, including the GIST study conducted by the group of O'Connell.2
We agree that previous glycemia status measured by glycohemoglobin (HbA1c) is an important point, allowing the identification of unrecognized diabetes mellitus. In fact, we also measured HbA1c and up to 18% of acute IS patients had possible unrecognized diabetes with HbA1c ≥6.2%. Moreover, we reported that known and possible unrecognized diabetic patients had poorer outcome than those nondiabetic. However, these differences disappeared after controlling for maximum glucose levels ≥155 mg/dL within the first 48 hours, suggesting that the prognostic influence of this threshold was independent of the presence of known and unknown diabetes.3
There is some controversy regarding hyperglycemia criteria in IS. Most studies used predefined and arbitrary thresholds ranging from 108 to 150 mg/dL.4 None of those definitions were based on multicenter prospective studies designed specifically to identify the clinically significant threshold of glycemia associated with poor outcome in ischemic stroke patients, and this is the main contribution of the GLIAS study.
We recognize the writing error when referencing the GIST study, and we apologize for it. We are also sorry about our comment that most patients in GIST were included with no hyperglycemia, as it could be misunderstood, but what we meant was that as most of them were included with baseline glucose levels between 108 and 150 mg/dL, they were under the threshold that we found associated with poor outcome. Regarding THIS data, we have not suggested that it “supports” the GLIAS threshold. Our comment was that THIS5 was conducted with patients with a hyperglycemia definition (>150 mg/dL) close to the threshold level found in our study.
We don’t agree with O'Connell et al in the final statement that “a cut-off level with such poor sensitivity and specificity has to be of limited value.” In our opinion, they are confounding 2 concepts in choosing an optimal cut-point. The first is the discriminator capacity of the prediction equation. As we stated in the discussion, our “predictive value was modest.” An AUC of 0.656 means that if we choose at random 1 patient with a good outcome and 1 with a poor outcome, around 2 of every 3 pairs of poor and good patients will be well-ordered. But the question is not whether this predictive capacity is “good” or “bad,” but whether there is another better equation, in the sense of having better long-term predictions with reliable and easy to collect variables. The second concept is the “cost” we are prepared to pay for a false-negative and a false-positive result—if, for example, the cost of a false-positive is higher, we would move to higher specificity values to the detriment of sensitivity. As those costs can be highly local, we proposed as the optimal cut-point that which provides a greater increase in the odds of poor outcome.
Finally, we would like to highlight that the GLIAS study is, to our knowledge, the first prospective multicenter study providing a glucose level threshold for outcome in IS patients. Because intensive glucose reduction has failed in stroke2 and in critically ill patients,6 the next question is whether hypoglycemic treatment with moderate reduction of glucose levels could prevent its deleterious effect in IS patients, and this is the setting in which the threshold provided by the GLIAS study could serve as a reference.
Fuentes B, Castillo J, SanJosé B, Leira R, Serena J, Vivancos J, Dávalos A, Gil-Nuñez A, Egido J, Díez-Tejedor E; for the Stroke Project of the Cerebrovascular Diseases Study Group, Spanish Society of Neurology. The prognostic value of capillary glucose levels in acute stroke. Stroke. 2009; 40: 562–568.
Gray CS, Hildreth AJ, Sandercock PA, O'Connell JE, Johnston DE, Cartlidge NEF, Bamford JM, James OF, Alberti KGMM; for the GIST Trialist Collaboration. Glucose-potassium-insulin infusions in the management of post-stroke hyperglycemia: the UK Glucose Insulin in Stroke Trial (GIST-UK). Lancet Neurol. 2007; 6: 397–406.
Fuentes B, Díez-Tejedor E, Castillo J, Dávalos A, Gil-Nunez A, Vivancos J, Egido J; on behalf of the Stroke Project of the Cerebrovascular Disease Study of the Spanish Society of Neurology: Prognostic value of glucose levels in acute stroke outcome: GLIA Study (abstract). Stroke. 2006; 37: 625.
Bruno A, Kent TA, Coull BM, Shankar RR, Saha C, Becker KJ, Kissela BM, Williams LS. Treatment of hyperglycemia in ischemic stroke (THIS): a randomized pilot trial. Stroke. 2008; 39: 384–389.