(Stroke. 2000;31:448.)
© 2000 American Heart Association, Inc.
Original Contributions |
A Predictive Risk Model for Outcomes of Ischemic Stroke
Presented in part at the First Neurology Outcomes Research Conference of the American Neurological Association, Montreal, Canada, October 17, 1998.
From the Departments of Neurology (K.C.J., E.C.H.) and Health Evaluation Sciences (K.C.J., A.F.C., D.P.W., W.A.K., X.-Q.W.), University of Virginia, Charlottesville.
Correspondence to Karen C. Johnston, MD, University of Virginia Health System, Department of Neurology, No. 394, Charlottesville, VA 22908. E-mail kj4v{at}virginia.edu
 |
Abstract |
|---|
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Background and Purpose—The great variability of outcome seen in stroke patients has led to an interest in identifying predictors of outcome. The combination of clinical and imaging variables as predictors of stroke outcome in a multivariable risk adjustment model may be more powerful than either alone. The purpose of this study was to determine the multivariable relationship between infarct volume, 6 clinical variables, and 3-month outcomes in ischemic stroke patients.
Methods—Included in the study were 256 eligible patients from the Randomized Trial of Tirilazad Mesylate in Acute Stroke (RANTTAS). Six clinical variables and 1-week infarct volume were the prespecified predictor variables. The National Institutes of Health Stroke Scale, Barthel Index, and Glasgow Outcome Scale were the outcomes. Multivariable logistic regression techniques were used to develop the model equations, and bootstrap techniques were used for internal validation. Predictive performance of the models was assessed for discrimination with receiver operator characteristic (ROC) curves and for calibration with calibration curves.
Results—The predictive models had areas under the ROC curve of 0.79 to 0.88 and demonstrated nearly ideal calibration curves. The areas under the ROC curves were statistically greater (P<0.001) with both clinical and imaging information combined than with either alone for predicting excellent recovery and death or severe disability.
Conclusions—Combined clinical and imaging variables are predictive of 3-month outcome in ischemic stroke patients. Demonstration of this relationship with acute clinical variables and 1-week infarct information supports future attempts to predict 3-month outcome with all acute variables.
Key Words: models, statistical • prognosis • stroke, ischemic • stroke outcome
|
Introduction |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The annual incidence of stroke is approximately 700 000 per year in the United States.1 Ischemic strokes account for 80% of these strokes.1 The high incidence and prevalence of stroke as well as the great variation in clinical outcome have led to an interest in identifying accurate predictors of outcome.
Numerous clinical variables have been identified as potential predictors of clinical outcome.2 3 4 5 6 7 8 9 10 11 12 13 Age2 4 5 6 7 9 10 11 12 and severity of presenting clinical deficit2 3 4 6 7 10 11 12 13 are consistently found to be predictive of outcome. Many other predictors have been reported to have a univariable relationship with outcome, but their multivariable relationship to outcome is less clear because of conflicting results in the litera-ture.3 4 5 6 10 11 12 13 Infarct volume, as measured by CT, has also been shown to correlate with clinical outcome.14 15 This relationship appears to be modest and to account for only a limited amount of the variance in outcome. Very few studies have examined a combination of clinical and imaging information in an attempt to predict clinical outcome in acute ischemic stroke patients.16 17 18 19 The combination of clinical variables and imaging variables as predictors of stroke outcome in a multivariable risk adjustment model may be useful in evaluating stroke outcomes in both prospective and retrospective clinical research as well as in patient care.
The purpose of this study was to determine the multivariable relationship between infarct volume, several clinical patient characteristics, and outcomes 3 months after ischemic stroke. This study is a test of the hypothesis that imaging and clinical information together could predict outcome with the use of subacute infarct volume data.
|
Subjects and Methods |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Study Population
The stroke patients used for this analysis were participants in the Randomized Trial of Tirilazad Mesylate in patients with Acute Stroke (RANTTAS).20 The RANTTAS trial was a multicenter, randomized, double-blinded, vehicle-controlled trial to evaluate the efficacy and safety of intravenous tirilazad mesylate in patients with acute ischemic stroke. Eligible patients were treated within 6 hours of symptom onset at 1 of 27 North American centers between May 1993 and December 1994. The primary outcome measures for this trial were the Barthel Index (BI) and Glasgow Outcome Scale (GOS) scores collected at 3 months. National Institutes of Health Stroke Scale (NIHSS) scores were also collected at 3 months as an additional key outcome measure. Three-month outcome measurements were obtained by study investigators who were trained to perform the measurements accurately and consistently. A random sample of half of all fully eligible patients were selected at study entry to submit head CT scans for volumetric measurement of infarct size at 7 to 10 days with planimetric techniques.20 A total of 556 fully eligible patients were enrolled in the trial, and 256 submitted acute and follow-up head CT scans for central evaluation. All 256 fully eligible patients with prospectively collected CT and clinical measures from the RANTTAS trial were included in this study. Since treatment with tirilazad mesylate did not have an effect on outcome,20 the 2 treatment groups were combined for this analysis.
Independent Variables
Independent variables for the model were identified
prospectively to avoid increased risk of type I errors and overfitting
of the predictive model. First, a detailed literature review was
performed to identify the published relationships between clinical
predictors of stroke outcome. More than 60 variables have been
examined in the literature for their relationship to stroke
outcome.2 3 4 5 6 7 8 9 10 11 12 14 15 16 17 18 19 The articles were read and their
bibliographies reviewed. We then evaluated the frequency of specific
stroke outcomes in the data set. In an attempt to avoid overfitting, we
limited the number of candidate variables to 7 on the basis of the
size of the study population and number of outcomes.21
These variables were selected on the basis of the strength of the
relationship to stroke outcome in the published literature,
availability in the RANTTAS data set, and investigator judgment. Six
clinical variables and 1 imaging variable were chosen. These
prespecified variables and any dichotomization of these
variables were identified before we examined any relationships in
the data set, and at no time during the analysis were any
changes made to the variables.
The variables are shown in Table 1
.
The NIHSS is a valid and reliable rank order scale that measures 11
categories of neurological deficit by neurological examination in
stroke patients.22 The NIHSS score was determined by
personnel trained and certified in the proper use of the scale.
Premorbid stroke and diabetes mellitus were dichotomized as present
or absent by history. The degree of prestroke disability was estimated
according to clinical history with the GOS, a standard and validated
scale that measures 5 categories of disability in patients with
neurological injury.23 The GOS was then dichotomized as
the presence or absence of any disability before stroke to reduce the
total number of degrees of freedom in the final model. Ischemic
stroke subtype was determined at 7 to 10 days after stroke according to
the Trial of ORG 10172 in Acute Stroke Treatment (TOAST)
criteria.24 This was then dichotomized as small-vessel
occlusive disease or not small-vessel occlusive disease on the basis of
evidence that patients with small-vessel infarcts have better
outcome.8 13 The imaging variable was infarct volume,
as determined by noncontrast head CT scan obtained 7 to 10 days after
stroke and measured in cubic centimeters.
|
Outcome Variables
The outcomes for the model were prespecified. These included 3
commonly used clinical outcome measures in stroke trials. The
NIHSS22 and the GOS23 are both
well-established, validated measures of clinical outcome. The BI is a
standard and well-validated scale that measures ability to complete 10
basic activities of daily living.25 Each of the 3 outcome
scales was prospectively dichotomized at 2 different points on the
scale. One point identifies excellent outcome (full or nearly full
recovery) versus worse than excellent outcome (all else). The second
cut point identifies very poor outcome (nursing home level disability
or death) versus better than very poor outcome (all else). The
definitions are shown in Table 1
. Dichotomization provides
outcomes that are easily understood by clinicians and patients, and the
literature provides significant face validity for all but 1 of these
defined outcomes, which have been used in numerous clinical
trials.12 20 26 We are unaware of any clinical trials that
have used severe disability or death measured by NIHSS as an outcome.
The >20 cutoff for the NIHSS in the very poor outcome model was based
on the clinical judgment of the authors as well as other
analyses that used 20 as a cut point.12
Missing Variables
If the outcome variable required by the model was
missing, the subject was eliminated from the analysis (NIHSS,
n=35; BI, n=27; GOS, n=27). If either the initial NIHSS score was
missing (n=5) or the CT infarct volume was missing (n=3), the subject
was eliminated from the analysis. This resulted in a total of
221 patients with an NIHSS outcome score, but only 215 subjects were
analyzed in the NIHSS models since 1 or both of the
aforementioned missing variables were missing (2 patients were
missing both initial NIHSS and infarct volume). A total of 229 patients
had BI outcome scores and GOS outcome scores, but only 222 were used in
the BI models and GOS models because of missing predictor variables
(1 patient was missing both initial NIHSS and infarct volume in both
sets of models).
Follow-up CT infarct volume measures that occurred outside of the window of 7 to 10 (±1) days were imputed as infarct volumes for 21 subjects who were imaged slightly outside of this window. These included 10 patients imaged before the window (earliest was day 3) and 8 patients imaged after the window, with all but 1 being within 30 days (latest was day 78). The remainder of the patients had incomplete data, and exact timing of imaging is unknown.
Analysis
The rigor of variable selection (described above) and the
analysis were designed to avoid overfitting the models in the
data set. After prespecification of the predictor and outcome
variables, we developed and tested 6 distinct models using the
RANTTAS data set. We used restricted cubic splines to allow nonlinear
covariate relationships with age to be assessed. This is a method of
allowing a continuous variable to be nonlinear in the
analysis, therefore avoiding linearity assumptions. Because of
limitations in size of our data set, we did not allow for nonlinearity
of the NIHSS or infarct volume predictors. A model was developed for
each of the 3 outcome measures (NIHSS, BI, GOS) for each of the 2
levels of outcome (excellent, very poor). For each of the 6 models,
comparison models including only the clinical variables and only
the imaging variable were also evaluated.
We compared the performance of the combined model (clinical and
imaging variables) with the clinical model (clinical variables
only) and the imaging model (infarct volume only). The weights of each
of the predictive variables for each model were determined by
multivariable logistic regression techniques. Model discrimination
was assessed by area under the receiver operator characteristic (ROC)
curves, which was computed by the nonparametric
method.27 The area under the ROC curve is a measure of
overall predictive discrimination, which is defined in this study as
the ability to separate those patients who had excellent outcome from
those who did not have excellent outcome, or those who had very poor
outcome from those who did not have very poor outcome. An ROC curve
area of 0.5 indicates no discrimination, and an ROC curve area of 1.0
indicates perfect discrimination. The area under the ROC curves for the
nested models (combined versus clinical or imaging models) were
compared with likelihood ratio 
2
statistics.28 Calibration was assessed with calibration
curves.21 Calibration is the ability to predict
probabilities across all ranges of risk and reflects the reliability or
degree of bias of the model. It is illustrated with calibration graphs
that plot predicted probability (using the model) against actual
probability. In a calibration graph, the line on the 45-degree angle
represents perfect calibration (the predicted probability
equals the actual probability). The closer the model calibration curve
is to the 45-degree line, the better is the calibration.
Bootstrap validation of the combined model for each of the outcomes was also performed.29 This is a technique, described in detail by Efron and Gong,29 in which the model is developed with all subjects and then reanalyzed on repeated random samples of the data set. This method of internal validation assesses how accurately the models will predict outcome in a new similar sample of stroke patients. It assesses whether or not the model is biased because of overfitting of the model. Resampling occurred 100 times for each bootstrap validation, and age was allowed to be nonlinear in each model by using a restricted cubic spline with 3 knots. All analyses were done with S-Plus 4.5 software (MathSoft Inc).
|
Results |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The patient predictor and outcome characteristics are displayed in Table 2
. The mean age was nearly 69
years, and the median NIHSS score of 10 suggested moderate severity of
initial neurological deficit. Approximately one fourth of patients had
a history of diabetes mellitus, one fourth had a history of stroke, and
one fourth had the small-vessel stroke subtype. Few patients were
disabled before the enrolling stroke event. Stroke severity was broadly
distributed as measured by both the NIHSS and the infarct volume
(Figure 1). Eighty-four patients had no
visible infarct by head CT obtained at 7 to 10 days. As shown in Table 2, 14% of patients were dead at 3 months. Excellent outcomes by
each of the 3 measures (NIHSS, BI, GOS) were quite common. Very poor
outcome, as defined by the NIHSS, was very infrequent.
|
|
The multivariable relationship between the 7 predictor
variables and each of the 6 outcome variables is shown in Table 3. For excellent outcome as determined by
the NIHSS score of 
1 at 3 months, the baseline NIHSS score,
small-vessel infarct, history of previous stroke, history of diabetes,
history of prestroke disability, and infarct volume at 7 to 10 days
were all significant predictors. For excellent outcome as determined by
the BI (score 
95) and the GOS (score=1) at 3 months, age, NIHSS,
diabetes, disability, and infarct volume were all significantly
predictive of outcome. For very poor outcome, only infarct volume was a
significant predictor of NIHSS score of 20 or death at 3 months. For
very poor outcome as determined by the BI (score <60 or dead) and by
the GOS (score >2 or dead), age, initial NIHSS score, prior disability
(for BI only), and infarct volume were significantly predictive of
outcome. The information needed to calculate each regression model is
given in the Appendix.
|
Discrimination, as indicated by bootstrap-corrected area under
the ROC curve, is shown in Table 4 for
models using imaging variables only, clinical variables only,
or both clinical and imaging variables combined. The combined model
had an area under the ROC curve greater than either of the other 2
models for 5 of the 6 outcomes. For prediction of excellent outcome,
the model using NIHSS as a measure of excellent outcome had the best
discrimination, with an ROC area of 0.87. For very poor outcome, the
model using BI as a measure of very poor outcome had the best
discrimination, with an ROC area of 0.88. The uncorrected area under
the ROC curve was similar to the bootstrap-corrected data, with the
maximum area under the ROC curve being 0.89 for excellent outcome in
the NIHSS model and 0.89 for very poor outcome in both the BI model and
the GOS model (data not shown). For both the original and the
bootstrap-corrected areas, the combined model (clinical and imaging
variables) consistently had the greatest ROC area compared
with the other 2 models (clinical alone, imaging alone). The likelihood
ratio tests were significant for all but 1 comparison of the
models.
|
Calibration graphs for the combined model when predicting excellent
outcome and very poor outcome, as measured by NIHSS, are shown in
Figure 2. For excellent outcome as
measured by NIHSS (Figure 2A), the proximity of the
bias-corrected calibration curve to the line of identity (ideal)
demonstrates that the combined model is well calibrated. The proximity
of the bias-corrected line to the apparent line (model
performance before bootstrap) demonstrates that overfitting was
minimal. Although the calibration curve of the original model is closer
to the ideal, the bias-corrected curve (bootstrap validated) is the
more valid representation of the likely relationship in other
similar data sets since it penalizes for the fact that we derived and
validated the models in the same patients. The excellent outcome
calibration curves for both BI and GOS were very similar (data not
shown). For very poor outcome as measured by the NIHSS (Figure 2B), the calibration was not quite as good. The very poor
outcome model calibration curves for the BI and GOS were similar to the
excellent outcome curves (data not shown).
|
To demonstrate that the excellent performance of the
combined model was not due to a limited distribution of the probability
of an excellent outcome, Figure 3
displays the broad distribution of predicted probabilities for
excellent outcome as measured by the NIHSS. The other measures of
excellent outcome were distributed similarly (data not shown).
|
|
Discussion |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
There is a strong multivariable relationship between the 7 variables (6 clinical and 1 imaging) and excellent outcome (full or nearly full recovery) and the 7 variables and very poor outcome (nursing home level disability or death) at 3 months after ischemic stroke. The models that include both clinical and imaging predictor information are statistically superior in performance to the models using clinical variables alone or imaging variables alone. The combined models were tested in a heterogeneous stroke population who participated in a multicenter, randomized acute stroke trial. They had a broad distribution of predicted outcome. Bootstrap validation demonstrated the internal validity of the models, which showed excellent discrimination and calibration. Five of the 6 models (excellent outcome models and BI, GOS of the very poor outcome models) had excellent ability to discriminate between outcomes as demonstrated by the area under the ROC curves (see Table 4
). Model predictions with ROC
areas >0.80 have been endorsed for use in individual predictions. The
ability of these models to discriminate outcomes for a disease that has
such heterogeneity of characteristics and outcomes is
encouraging. The calibrations of the excellent outcome models and 2 of
the very poor outcome models (BI, GOS) are nearly ideal, as
demonstrated by the calibration graphs. The NIHSS-measured very poor
outcome model had an ROC area of 0.79 and poorer calibration than the
other models. We believe that the limited number of outcomes by this
measure (only 36, as seen in Table 2) is responsible for the
lesser performance of this model. Very few previous studies have combined clinical and imaging variables for the purpose of predicting stroke outcomes. Hénon et al17 combined clinical predictors with information from head CT but did not use infarct volume, which is now well recognized as a predictor of outcome. In this study the investigators evaluated a relatively large number of predictor variables (18) despite a low death rate of 16% (the specified outcome variable), thereby potentially increasing the risk of an overfitted model with poor external validity.21 Toni et al18 also evaluated a model that combined clinical and imaging information. Variables in their analysis were chosen by stepwise techniques, which makes the resulting model very sensitive to the characteristics of the development data and is prone to overfitting.21 A similar approach was taken by Finocchi et al,19 who combined clinical and imaging variables but categorized infarct volume into 3 categories (no lesion, medium lesion, large lesion). They suggested that since the final size of an infarct cannot be detected on CT for several days after the event, the infarct volume does not improve the predictive ability of such a model. Current availability of new imaging techniques, such as diffusion-weighted MRI, that identify lesion volume in the acute setting30 31 32 33 34 35 may improve our ability to predict stroke outcomes in the acute setting.
There are several limitations of this study. First, this model has only been tested in the data set from which it was derived. It has not been prospectively validated with independent data. We were careful, however, to avoid overspecification of the model by prospectively identifying the independent variables, avoiding stepwise variable selection techniques, retaining all candidate variables in the final model, and limiting candidate variables to no more than 1 for each 10 observations of the least frequent outcome.21 Our model used each of these techniques, making it unlikely that we have significantly overmodeled the data set. The use of the bootstrap validation to internally validate our model, which penalizes for optimism caused by any overfitting present, demonstrates model stability and suggests that we have not overfit to this single data set.
The fact that infarct volume and stroke subtype data were not collected acutely in the RANTTAS trial limits our ability to use the models in the acute ischemic stroke setting. However, this analysis does provide evidence that imaging and clinical information together can be predictive of 3-month outcome and that together they are more powerful than either alone. These data support future attempts to predict 3-month outcome using measures of infarct volume and stroke subtype, obtained with new imaging technology, in the acute stroke setting.
The small size of the data set and specifically of the outcomes of
interest was also limiting. We believe that the small sample limited
our power to demonstrate statistically significant relationships
between some predictors and outcome (Table 3) in several of the
models despite our hypothesis that clinically important relationships
exist. We also recognize, however, that some of the differences in
strength of relationships of predictors in different models may be
related to the fact that we are measuring different outcomes. In
addition, sample size likely limited our ability to demonstrate a
stronger predictive relationship in the very poor outcome as measured
by NIHSS model (with only 36 outcomes).
A model that can predict clinical outcome in acute stroke patients would be potentially useful in a number of ways. Risk adjustment of acute stroke populations could improve clinical research as well as potentially improve the stroke management process. The great heterogeneity of the ischemic stroke population has been well described,36 37 38 39 40 41 42 43 as has the difficulty such heterogeneity has imposed on the evaluation of new therapies for stroke patients.13 44 45 46 Patients with ischemic stroke vary considerably in clinical characteristics such as infarct size and location, vascular anatomy, premorbid conditions, age, severity of deficit, previous brain function, and genetic makeup, to name only a few. This heterogeneity is a concern since drug effect may vary across the severity spectrum (treatment by severity interaction), as has been demonstrated in other diseases.47 Even with strict eligibility criteria, such variables cannot be controlled for in clinical trials without risking the generalizability of the results.43 The considerable variability in the stroke population, therefore, makes it much more difficult to detect treatment effects since imbalances in important patient characteristics may occur by chance, especially in small trials. DeGraba et al13 recently suggested the need for stratification by stroke severity in clinical stroke trials since disease severity may affect the clinical outcome. Since the effect of an intervention on outcome may also vary with stroke severity and since the great heterogeneity of the stroke population may conceal treatment effects, risk adjustment models may increase our ability to detect the effects of therapy in clinical research by statistically accounting for this heterogeneity or by providing a single stratification variable (predicted probability of outcome) that allows the consideration of individual severity of disease at the time of clinical trial enrollment.
Knowing expected outcome may also allow improved selection of patients
anticipated to respond to therapy. As illustrated in Figure 3, there may be a proportion of patients who at entry to a trial have a
high likelihood of full recovery or a high likelihood of a devastating
outcome. Depending on the agent being evaluated, such extreme risk
patients may be included or excluded to increase the statistical power
for detecting treatment effects. In addition, the appropriateness of
eligibility criteria in a trial may depend on outcome and complication
risks. A patient who is highly likely to have a full recovery, for
example, may not be a good candidate to receive an experimental
intervention that has a significant risk of complications.
We have demonstrated a strong multivariable relationship between 7 predictor variables and 3-month outcome in ischemic stroke patients. This analysis provides the initial evidence ("proof of concept") that such a multivariable relationship exists. This analysis suggests that future attempts to predict stroke outcome using acute clinical and acute imaging predictor variables would be valuable. Validation of these models in other similar patient populations is needed.
Appendix
|
|
Acknowledgments |
|---|
The RANTTAS study was supported in part by the National Institutes of Health/National Institute of Neurological Disorders and Stroke (R01-NS31554) and Pharmacia and Upjohn Company (Kalamazoo, Mich).
|
Footnotes |
|---|
The RANTTAS investigators are listed in the Appendix of Reference 20.
Received September 3, 1999; revision received November 12, 1999; accepted November 17, 1999.
|
References |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
-
Broderick J, Brott T, Kothari R, Miller R, Khoury
J, Pancioli A, Gebel J, Mills D, Minneci L, Shulka R. The Greater
Cincinnati/Northern Kentucky Stroke Study: preliminary first-ever and
total incidence rates of stroke among blacks. Stroke. 1998;29:415–421.
[Abstract/Free Full Text] -
Kotila M, Waltimo O, Neimi ML, Laaksonen MA, Lempinen
M. The profile of recovery from stroke and factors influencing outcome.
Stroke. 1984;15:1039–1044.
[Abstract/Free Full Text] -
Dove HG, Schneider KC, Wallace JD. Evaluating and
predicting outcome of acute cerebral vascular accident.
Stroke. 1984;15:858–864.
[Abstract/Free Full Text] - Allen CMC. Predicting the outcome of acute stroke: a prognostic score. J Neurol Neurosurg Psychiatry. 1984;47:475–480.[Abstract]
- Wade DT, Wood VA, Hewer RL. Recovery after stroke: the first 3 months. J Neurol Neurosurg Psychiatry. 1985;48:7–13.[Abstract]
-
Jongbloed L. Prediction of function after stroke: a
critical review. Stroke. 1986;17:765–776.
[Abstract/Free Full Text] -
Chambers BR, Norris JW, Shurvell BL, Hachinski VC.
Prognosis of acute stroke. Neurology. 1987;37:221–225.
[Abstract/Free Full Text] -
Sacco SE, Whisnant JP, Broderick JP, Phillips SJ,
O’Fallon M. Epidemiological characteristics of lacunar infarcts in a
population. Stroke. 1991;22:1236–1241.
[Abstract/Free Full Text] - Lefkovits J, Davis SM, Rossiter SC, Kilpatrick CJ, Hopper JL, Green R, Tress BM. Acute stroke outcome: effects of stroke type and risk factors. Aust N Z J Med. 1992;22:30–35.[Medline] [Order article via Infotrieve]
- Censori B, Camerlingo M, Casto L, Ferraro B, Gazzaniga GC, Cesana B, Mamoli A. Prognostic factors in first-ever stroke in the carotid artery territory seen within 6 hours after onset. Stroke. 1993;24:532–535.[Abstract]
- Fiorelli M, Alpérovitch A, Argentino C, Sacchetti ML, Toni D, Sette G, Cavalletti C, Gori MC, Fieschi C, for the Italian Acute Stroke Study Group. Prediction of long-term outcome in the early hours following acute ischemic stroke. Arch Neurol. 1995;52:250–255.[Abstract]
-
The NINDS t-PA Stroke Study Group. Generalized efficacy
of t-PA for acute stroke: subgroup analysis of the NINDS t-PA
Stroke Trial. Stroke. 1997;28:2119–2125.
[Abstract/Free Full Text] -
DeGraba TJ, Hallenbeck JM, Pettigrew KD, Dutka AJ,
Kelly BJ. Progression in acute stroke value of the initial NIH Stroke
Scale score on patient stratification in future trials.
Stroke. 1999;30:1208–1212.
[Abstract/Free Full Text] -
Brott T, Marler JR, Olinger CP, Adams HP, Tomsick T,
Barsan WG, Biller J, Eberle R, Hertzberg V, Walker M. Measurements of
acute cerebral infarction: lesion size by computed tomography.
Stroke. 1989;20:871–875.
[Abstract/Free Full Text] -
Saver JL, Johnston KC, Homer D, Wityk R, Koroshetz W,
Truskowski LL, Haley EC, for the RANTTAS Investigators. Infarct volume
as a surrogate or auxiliary outcome measure in ischemic stroke
clinical trials. Stroke. 1999;30:293–298.
[Abstract/Free Full Text] -
Bonita R, Fore MA, Stewart AW. Predicting survival
after stroke: a three-year follow-up. Stroke. 1988;19:669–673.
[Abstract/Free Full Text] -
Hénon H, Godefroy O, Leys D, Mounier-Vehier F,
Lucas C, Rondepierre P, Duhamel A, Pruvo JP. Early predictor of death
and disability after acute cerebral ischemic event.
Stroke. 1995;26:392–398.
[Abstract/Free Full Text] - Toni D, Fiorelli M, Bastianello S, Falcou A, Sette G, Ceschin V, Sacchetti ML, Argentino C. Acute ischemic strokes improving during the first 48 hours of onset: predictability, outcome, and possible mechanisms: a comparison with early deteriorating strokes. Stroke. 1997;22:10–14.
- Finocchi C, Gandolfo C, Gasparetto B, Del Sette M, Croce R, Loeb C. Value of early variables as predictors of short-term outcome in patients with acute focal cerebral ischemia. Ital J Neural Sci. 1996;17:341–346.
-
The RANTTAS Investigators. A Randomized Trial of
Tirilazad Mesylate in Patients With Acute Stroke (RANTTAS).
Stroke. 1996;27:1453–1458.
[Abstract/Free Full Text] - Harrell FE, Lee KL, Mark DB. Tutorial in biostatistics: multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:367–387.
- Lyden P, Brott T, Tilley B, Welch KM, Mascha EJ, Levine S, Haley EC, Grotta J, Marler J, for the NINDS TPA Stroke Study Group. Improved reliability of the NIH Stroke Scale using video training. Stroke. 1994;25:2220–2226.[Abstract]
- Jennett B, Bond M. Assessment of outcome after severe brain damage. Lancet. 1975;1:480–484.[Medline] [Order article via Infotrieve]
- Madden KP, Karanjia PN, Adams HP Jr, Clarke WR. Accuracy of initial stroke subtype diagnosis in the TOAST study: Trial of ORG 10172 in Acute Stroke Treatment. Neurology. 1995;45:1975–1979.[Abstract]
- Mahoney FT, Barthel DW. Functional evaluation: Barthel Index. Md Med J. 1965;14:61–65.
-
The National Institute of Neurological Disorders and
Stroke rt-PA Stroke Study Group. Tissue plasminogen
activator for acute ischemic stroke. N
Engl J Med. 1995;333:1581–1587.
[Abstract/Free Full Text] -
Hanley JA, McNeil BJ. The meaning and use of the area
under a receiver operating characteristic (ROC) curve.
Radiology. 1982;143:29–36.
[Abstract/Free Full Text] - Buse A. The likelihood ratio, Wald, and Lagrance multiplier tests: an expository note. Am Stat. 1982;36:153–157.
- Efron B, Gong G. A leisurely look at the bootstrap, the jackknife, and cross-validation. Am Stat. 1983;37:36–48.
- Warach S, Gaa J, Siewert B, Wielopolski P, Edelman RR. Acute human stroke studied by whole brain echo planar diffusion-weighted magnetic resonance imaging. Ann Neurol. 1995;37:231–241.[Medline] [Order article via Infotrieve]
- Gonzalez G, Schaefer P, Buonanno F, Schwamm L, Budzik R, Rordorf G, Wang B, Sorensen G, Koroshetz W. Clinical sensitivity and specificity of diffusion weighted MRI in hyperacute stroke. Stroke. 1997;28:243. Abstract.
-
Singer MB, Chong J, Lu D, Schonewille WJ, Tuhrim S,
Atlas S. Diffusion-weighted MRI in acute subcortical infarction.
Stroke. 1998;29:133–136.
[Abstract/Free Full Text] - Baird, AE, Benfield A, Schlaug G, Siwert B, Lovblad KO, Edelman RR, Warach S. Enlargement of human cerebral ischemic lesion volumes measured by diffusion-weighted magnetic resonance imaging. Ann Neurol. 1997;41:581–589.[Medline] [Order article via Infotrieve]
- Lee LJ, Kidwell C, Alger J, Starkman S, Saver JL. Impact upon stroke subtype diagnosis of early diffusion-weighted MR and MRA imaging. Neurology. 1998;50:A298. Abstract.
-
Fisher M, Albers GW. Applications of
diffusion-perfusion magnetic resonance imaging in acute
ischemic stroke. Neurology. 1999;52:1750–1756.
[Abstract/Free Full Text] - Fisher M, Takano K. The penumbra, therapeutic time window and acute ischaemic stroke. Baillieres Clin Neurol. 1995;4:279–295.[Medline] [Order article via Infotrieve]
-
Lang EW, Daffertshofer M, Daffertshofer A, Wirth SB,
Chesnut RM, Hennerici M. Variability of vascular territory in stroke:
pitfalls and failure of stroke pattern interpretation.
Stroke. 1995;26:942–945.
[Abstract/Free Full Text] - Rosenthal GE, Shah A, Way LE, Harper DL. Variations in standardized hospital mortality rates for six common medical diagnoses: implications for profiling hospital quality. Med Care. 1998;36:955–964.[Medline] [Order article via Infotrieve]
- Wardlaw JM, Warlow CP, Counsell C. Systematic review of evidence on thrombolytic therapy for acute ischaemic stroke. Lancet. 1997;350:607–614.[Medline] [Order article via Infotrieve]
-
Kwakkel G, Wagenaar RC, Koelman TW, Lankhorst GJ,
Joetsier JC. Effects of intensity of rehabilitation after stroke: a
research synthesis. Stroke.. 1997;28:1550–1556.
[Abstract/Free Full Text] -
Kwakkel G, Wagenaar RC, Kollen BJ, Lankhorst GJ.
Predicting disability in stroke: a critical review of the literature.
Age Ageing. 1996;25:479–489.
[Abstract/Free Full Text] - Ortells ML, Mostacero E. Clinical pathogenic heterogeneity and subcortical infarcts prognosis. Rev Neurol. 1996;24:179–182.[Medline] [Order article via Infotrieve]
- Rothwell PM. Can overall results of clinical trials be applied to all patients? Lancet.. 1995;345:1616–1619.[Medline] [Order article via Infotrieve]
- Del Zoppo GJ. Why do all drugs work in animals but none in stroke patients? I: drugs promoting cerebral blood flow. J Intern Med. 1995;273:79–88.
- Grotta J. Why do all drugs work in animals but none in stroke patients? II: neuroprotective therapy. J Intern Med. 1995;237:89–94.[Medline] [Order article via Infotrieve]
- Bath P. Alteplase not yet proven for acute ischaemic stroke. Lancet. 1998;352:1238–1239.[Medline] [Order article via Infotrieve]
- Knaus WA, Harrell FE Jr, LaBrecque JF, Wagner DP, Pribble JP, Draper EA, Fisher CJ Jr, Soll L. Use of predicted risk of mortality to evaluate the efficacy of anticytokine therapy in sepsis. Crit Care Med. 1996;24:46–56.[Medline] [Order article via Infotrieve]
This article has been cited by other articles:
 |
|
 |
|
  M. I. Dallas, S. Rone-Adams, J. L. Echternach, L. M. Brass, and D. M. Bravata Dependence in Prestroke Mobility Predicts Adverse Outcomes Among Patients With Acute Ischemic Stroke Stroke, August 1, 2008; 39(8): 2298 - 2303. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I. R. Konig, A. Ziegler, E. Bluhmki, W. Hacke, P. M.W. Bath, R. L. Sacco, H. C. Diener, C. Weimar, and on behalf of the Virtual International Stroke Tria Predicting Long-Term Outcome After Acute Ischemic Stroke: A Simple Index Works in Patients From Controlled Clinical Trials Stroke, June 1, 2008; 39(6): 1821 - 1826. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Ois, E. Cuadrado-Godia, J. Jimenez-Conde, M. Gomis, A. Rodriguez-Campello, J. E. Martinez-Rodriguez, E. Munteis, and J. Roquer Early Arterial Study in the Prediction of Mortality After Acute Ischemic Stroke Stroke, July 1, 2007; 38(7): 2085 - 2089. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. C. Johnston, D. P. Wagner, X.-Q. Wang, G. C. Newman, V. Thijs, S. Sen, S. Warach, and for the GAIN, Citicoline, and ASAP Investigators Validation of an Acute Ischemic Stroke Model: Does Diffusion-Weighted Imaging Lesion Volume Offer a Clinically Significant Improvement in Prediction of Outcome? * Definitions and Explanations Stroke, June 1, 2007; 38(6): 1820 - 1825. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. M. Kent, H. P. Selker, R. Ruthazer, E. Bluhmki, and W. Hacke The Stroke-Thrombolytic Predictive Instrument: A Predictive Instrument for Intravenous Thrombolysis in Acute Ischemic Stroke Stroke, December 1, 2006; 37(12): 2957 - 2962. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. K. Schiemanck, G. Kwakkel, M. W.M. Post, L. J. Kappelle, and A. J.H. Prevo Predicting Long-Term Independency in Activities of Daily Living After Middle Cerebral Artery Stroke: Does Information From MRI Have Added Predictive Value Compared With Clinical Information? Stroke, April 1, 2006; 37(4): 1050 - 1054. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Saposnik, S. Di Legge, F. Webster, and V. Hachinski Predictors of major neurologic improvement after thrombolysis in acute stroke Neurology, October 25, 2005; 65(8): 1169 - 1174. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. M. Garbusinski, M. A.B. van der Sande, E. J. Bartholome, M. Dramaix, A. Gaye, R. Coleman, O. A. Nyan, R. W. Walker, K. P.W.J. McAdam, and G. E. Walraven Stroke Presentation and Outcome in Developing Countries: A Prospective Study in The Gambia Stroke, July 1, 2005; 36(7): 1388 - 1393. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. B. Goldstein, G. P. Samsa, D. B. Matchar, and R. D. Horner Charlson Index Comorbidity Adjustment for Ischemic Stroke Outcome Studies Stroke, August 1, 2004; 35(8): 1941 - 1945. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. Georgiadis, J. Oehler, S. Schwarz, V. Rousson, M. Hartmann, and S. Schwab Does acute occlusion of the carotid T invariably have a poor outcome? Neurology, July 13, 2004; 63(1): 22 - 26. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. Rundek, K. Nielsen, S. Phillips, K. C. Johnston, M. Hux, D. Watson, and for the GAIN Americas Investigators Health Care Resource Use After Acute Stroke in the Glycine Antagonist in Neuroprotection (GAIN) Americas Trial Stroke, June 1, 2004; 35(6): 1368 - 1374. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. M. Kent, J. Hinchey, L. L. Price, S. R. Levine, and H. P. Selker In Acute Ischemic Stroke, Are Asymptomatic Intracranial Hemorrhages Clinically Innocuous? Stroke, May 1, 2004; 35(5): 1141 - 1146. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. C. Johnston, A. F. Connors Jr, D. P. Wagner, and E. C. Haley Jr Risk Adjustment Effect on Stroke Clinical Trials Stroke, February 1, 2004; 35 (2): e43 - e45. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. L. Brown, K. C. Johnston, D. P. Wagner, and E. C. Haley Jr Predicting Major Neurological Improvement With Intravenous Recombinant Tissue Plasminogen Activator Treatment of Stroke Stroke, January 1, 2004; 35(1): 147 - 150. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. C. Johnston, A. F. Connors Jr, D. P. Wagner, and E. C. Haley Jr Predicting Outcome in Ischemic Stroke: External Validation of Predictive Risk Models Stroke, January 1, 2003; 34(1): 200 - 202. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K.-C. Chang, M.-C. Tseng, H.-H. Weng, Y.-H. Lin, C.-W. Liou, and T.-Y. Tan Prediction of Length of Stay of First-Ever Ischemic Stroke Stroke, November 1, 2002; 33(11): 2670 - 2674. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. C. Johnston, D. P. Wagner, E. C. Haley Jr, and A. F. Connors Jr Combined Clinical and Imaging Information as an Early Stroke Outcome Measure Stroke, February 1, 2002; 33(2): 466 - 472. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
O. Wu, W. J. Koroshetz, L. Ostergaard, F. S. Buonanno, W. A. Copen, R. G. Gonzalez, G. Rordorf, B. R. Rosen, L. H. Schwamm, R. M. Weisskoff, et al. Predicting Tissue Outcome in Acute Human Cerebral Ischemia Using Combined Diffusion- and Perfusion-Weighted MR Imaging Stroke, April 1, 2001; 32(4): 933 - 942. [Abstract] [Full Text] [PDF]
|
|