Letter by Weimar and König Regarding Article “Initial Lesion Volume Is an Independent Predictor of Clinical Stroke Outcome at Day 90: An Analysis of the Virtual International Stroke Trials Archive (VISTA) Database”
To the Editor:
We read with great interest the article by Vogt et al.1 In their retrospective analysis of 1938 patients with ischemic stroke from the Virtual International Stroke Trials Archive (VISTA) database, lesion volume obtained within 72 hours after stroke onset was identified as the single most important predictor of stroke outcome at Day 90 assessed on the National Institutes of Health Stroke Scale (NIHSS), modified Rankin Scale, and mortality. In addition to age and initial stroke severity at admission assessed on the NIHSS, inclusion of lesion volume in the present analysis increased the explained proportion of variance (R 2) from 0.26 to 0.41, meaning that an additional 15% of outcome variability is explained by baseline parameters alone.
There are some remarks of caution concerning the methodology of the presented study. In building the models, possible interactions between variables (ie, initial NIHSS and lesion size) were not considered. Although recognized to have a nonlinear relationship with the outcome, continuous variables such as NIHSS at baseline were entered on a linear scale, thus possibly underestimating the effect. Also, using a linear regression for the prediction of NIHSS is questionable, because the developed model easily renders predictions out of the range of the scale.
The goodness of the developed models is evaluated by the probability values of the single variables and by the explained proportion of variance of the regression model. For the prediction of outcome, however, the accuracy with which patients are classified needs to be evaluated, for example, in terms of sensitivity and specificity. In the study presented here, in which a novel predictor is added to existing models, indices based on the reclassification of patients are recommended.2
By showing that the same variables were selected for varying outcomes and patient groups, the authors deduce that their models are valid. However, validation means application of the model using the estimated parameters to independent data sets and thus showing that the classification accuracy is similar.3 Considering that model development was based on <10% of the VISTA database and a large majority of patients with ischemic stroke had anterior circulation infarcts, it would certainly be interesting how well the newly developed models perform on external validation in patients with acute stroke included in other clinical trials or thrombolysis registries.
Furthermore, additional clinical variables (including infarct location and modified Rankin Scale within 48–72 hours) have been previously identified as independent predictors in patients with acute stroke and an external validation of the resulting logistic regression model explained R 2=52.4% of the complete variation.4 Although these variables may have a lower interrater reliability than quantitative measures, they are more readily accessible and do not require a standardized MRI protocol nor early follow-up imaging. The main value of the models therefore could be a prognostic indication in patients with moderate to severe ischemic stroke, who are typically included in clinical trials.
The authors propose that future acute stroke trials should incorporate an outcome model consisting of age, lesion volume, and NIHSS at baseline to increase power for effect detection. To this date, however, none of the acute stroke trials using a predefined adapted end point design or outcome correction has been able to show any significant treatment effect allowing to prospectively validate this approach. The use of validated outcome models for clinical stroke research therefore remains to be shown before their use as part of an end point definition can be recommended in clinical stroke trials.
Christian Weimar, MD
Department of Neurology
University of Duisburg-Essen
Inke R. König, PhD
Institut für Medizinische Biometrie und Statistik
Universität zu Lübeck
C.W. is a member of the VISTA Steering Committee.
Stroke welcomes Letters to the Editor and will publish them, if suitable, as space permits. Letters must reference a Stroke published-ahead-of-print article or an article printed within the past 3 weeks. The maximum length is 750 words including no more than 5 references and 3 authors. Please submit letters typed double-spaced. Letters may be shortened or edited. Include a completed copyright transfer agreement form (available online at http://stroke.ahajournals.org and http://submit-stroke.ahajournals.org).
- © 2012 American Heart Association, Inc.
- Vogt G,
- Laage R,
- Shuaib A,
- Schneider A
German Stroke Study Collaboration. Predicting outcome after acute ischemic stroke: an external validation of prognostic models. Neurology. 2004; 62: 581– 585.