The prognostic value of domain-specific cognitive abilities in acute first-ever stroke

Subjects.

The population consisted of 228 consecutive patients with first-ever symptomatic stroke admitted to stroke units of three hospitals in the Netherlands (University Medical Center Utrecht, Tweesteden Hospital Tilburg, and St.-Elisabeth Hospital Tilburg) between December 2001 and January 2003. Patients were eligible if they had either ischemic stroke or primary intracerebral hemorrhage. The diagnosis of stroke was based on the presence of both an acute focal deficit and an associated lesion on CT or MRI. Patients with a normal scan underwent a second scan within the first week poststroke. Exclusion criteria were 1) pre-existing impairment or history that might influence cognitive or functional outcome, i.e., history of drug abuse, pre-existent dependence in activities of daily living, or pre-existent cognitive decline (as defined by a score of 3.6 or higher on the short Informant Questionnaire on Cognitive Decline in the Elderly–IQCODE Dutch version),¹⁷ 2) age older than 85 years, and 3) inability to be examined within the first 21 days poststroke due to severe disturbances in consciousness or inability to comprehend task instructions. This exclusion procedure resulted in a study population of 190 patients with acute stroke, of which 168 patients (88%) could be neuropsychologically examined.

Finally, patients with a recurrent stroke or who developed comorbidity that might affect outcome (i.e., cancer, myocardial infarction, CABG, or psychiatric illnesses) between the first and the second examination were excluded from follow-up examination. Figure 1 presents a chart showing the number of patients who were included and excluded from this study.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Characterization of patient cohort at baseline and at follow-up.

A control group was assembled as a reference sample for the neuropsychological examination, consisting of 75 subjects living in the community. The controls were either spouses or family of patients, or volunteers who came to our attention through advertising in newspapers or by word of mouth. Control subjects were carefully matched to the stroke patients with respect to age, education, and sex. Moreover, they performed both examinations with the same time interval as the patients to control for potential practice effects.

Predictor variables.

On admission, demographic factors, vascular risk factors, factors associated with medical status at admission, CT/MRI characteristics, and domain-specific cognitive abilities were obtained as candidate predictor variables. All except the continuous predictor variables were dichotomized in order to reduce the number of variables and for clinical clarity. Demographic factors included age (years), level of education (scored with 7 categories ranging from 1 = did not finish primary school to 7 = university degree, and dichotomized at the median),¹⁸ and sex. Vascular risk factors comprised previously diagnosed diabetes mellitus, hypertension, hypercholesterolemia, TIA, smoking during the last 5 years, and alcohol consumption of more than 2 units per day. Factors associated with the patients’ medical status on admission were body temperature > 38°C in the first week during hospital stay, admission glucose level, total serum cholesterol level at admission, systolic and diastolic blood pressure, stroke severity (assessed by means of the NIH Stroke Scale [NIHSS],¹⁹ and categorized as severe if NIHSS > 7),²⁰ functional dependence (assessed by means of the modified Barthel Index [MBI],²¹ and categorized as present if MBI < 19),²² and weakness of either arm or leg (categorized as present if NIHSS items for either arm or leg > 1). An experienced stroke neurologist (H.B.v.d.W.) who was blind to the clinical data evaluated the patients’ stroke type (categorized as cerebral infarct or intracerebral hemorrhage) from CT or MRI, in addition to the location of the lesion (classified into three categories: involvement of left hemisphere, right hemisphere, and brain stem/cerebellum), the main affected territory of vascular blood supply (classified as anterior, middle, or posterior cerebral artery, or vertebrobasilar arteries), the presence of silent infarcts (classified as present when an infarct was found in the brain without the description by the patient, family, or medical record of a prior stroke episode),²³ the presence of cerebral atrophy (classified as present when a generalized dilatation of cortical sulci was found, with or without associated ventricular enlargement),²³ leukoariosis (scored as present if patients obtained a score > 0 on the van Swieten scale),²⁴ and the volume of the lesion, calculated using Leica Q500 MCP image analysis software by manual tracing of the lesion on each slice showing the infarct or hemorrhage, followed by multiplying lesion area by slice thickness in all slices showing the lesion. This method has been shown to have a high inter-rater reliability and is described in detail elsewhere.²⁵

A detailed cognitive examination was performed within 3 weeks poststroke (mean time from onset to examination, 8.3 ± 4.5 days). An extended evaluation took place after a minimum of 6 and a maximum of 10 months poststroke. Both the baseline and the follow-up examination covered seven major cognitive domains consisting of verbal and nonverbal neuropsychological tasks (tasks added to the follow-up examination are italicized).¹⁶ Reasoning was assessed with the Raven Advanced Progressive Matrices (short form) and Similarities (Wechsler Adult Intelligence Scale [WAIS]-III). Verbal memory was measured by means of the Rey Auditory Verbal Learning Test, the Digit span (WAIS-III), and the Wechsler Memory Scale–story recall A. Executive functioning was assessed with the Brixton Spatial Anticipation Test, the Visual Elevator (Test of Everyday Attention),²⁶ letter fluency, the Stroop Color Word test, Semantic Fluency, and the Zoo test (BADS). Visual perception and construction was assessed with the Judgment of Line Orientation (short form), the Test of Facial Recognition (short form), the Rey-Osterrieth Complex Figure–copy, and the WAIS-III block patterns. Visual memory was measured with the Corsi Block span, the Rey-Osterrieth Complex Figure–delay, the Wechsler Memory Scale–Visual Reproduction, and the Location Learning Task.²⁷ Language was assessed with the Token Test (short form), the Boston Naming test (short form), and the Chapman reading task. Finally, unilateral neglect was measured with the Star cancellation (BIT) and the Line Bisection Schenkenberg.

The neuropsychological examination in the early phase of stroke was considered feasible if patients were able to perform at least 10 of the 15 tasks, which allowed evaluation of the majority of the cognitive domains. Patients who were able to reach this criterion but who were unable to perform one or more tasks were given the minimum score on the tasks they could not perform. In this way, both patients with aphasia and unilateral neglect could be included in our study. The procedure of administering multiple neuropsychological tasks within one cognitive domain allowed us to transform raw test scores of patients on individual tasks into compound z-scores based on the means and standard deviations of the control group on the first and second examination to control for potential practice effects. Subsequently, we averaged z-scores of tasks belonging to the same cognitive domain. Cutoff scores for cognitive impairment within each domain were determined by a performance that differed from the control mean at the 0.05 level of significance (Z < −1.65).²⁸ This approach was used for both (acute and follow-up) neuropsychological examinations.

Outcome measures.

Two distinct types of outcome were used in the present study, i.e., long-term cognitive outcome and functional outcome in terms of ADL.

Cognitive outcome was assessed with the abovementioned neuropsychological examination at follow-up and classified as cognitively intact (defined as no impairment on any of the cognitive domains) vs cognitively impaired (defined as an impairment in at least one cognitive domain).

Functional outcome was assessed with two separate ADL measures: basic and instrumental ADL. Basic ADL instruments assess straightforward activities such as personal hygiene and dressing, whereas instrumental ADL instruments assess more complex ADL, such as grocery shopping, household management, and social activities. In this study, basic ADL was determined with the modified Barthel Index (MBI).²¹ A MBI value ≥19 was used as an indication of intact basic ADL function.²² The Frenchay Activities Index (FAI)²⁹ was used to assess instrumental ADL. In total, the scale comprises 15 individual activities summed to give an overall score ranging from 0 (inactive) to 45 (very active). Intact instrumental ADL function was defined as FAI ≥ 15.³⁰

All outcome measures were dichotomized in order to provide clear and interpretative information for both clinicians and patients. Outcomes were assessed blind to the predictor variables.

Statistical analyses.

First, to determine whether any selection bias occurred between patients who were re-examined at follow-up and those who were not, we performed Student’s t-tests for continuous data, Mann–Whitney U test for ordinal data, and χ² analyses for categorical data on a range of baseline characteristics of patients.

Second, univariate analyses of potential predictor variables were undertaken for the three outcome measures. Those variables with a univariate association at p ≤ 0.1 were considered for entry into forward stepwise multiple logistic regression analysis. Three multivariate prediction models were produced for each outcome measure. A standard medical model was developed entering all variables associated with outcome in the univariate analyses, except for the early cognitive predictors. The cognitive model included only cognitive factors associated with outcome in the univariate analyses. The combined model included both medical and cognitive predictors. Receiver-operator characteristic (ROC) curves were used to compare the information content of the different models for all outcome measures.³¹ The larger the area under the ROC curve (AUC), the higher the information content of the model. The Hosmer-Lemeshow goodness-of-fit statistic³² was used to determine if the models provided a good fit for the data. Models are well calibrated and fit the data if the goodness of fit statistic is large (p > 0.05).

Clinical characteristics of patients not examined at follow-up.

As shown in figure 1, 111 of 168 (66%) patients were re-examined. The interval between the early examination and follow-up did not differ between controls (7.4 ± 1.3 months) and patients (7.4 ± 1.0 months). Patients who were not examined at follow-up were older, more impaired on the MBI at baseline, and experienced more deficits in executive functioning, visual memory, and visual perception and construction at baseline. The failure to be re-examined was also related to the presence of silent infarcts and white matter lesions, whereas no association with vascular risk factors or characteristics directly related to the stroke lesion could be demonstrated (table 1).

Prevalence of adverse outcome on the long term.

Of the included patients, 48.6% showed cognitive impairments in the first 3 weeks poststroke, whereas 30.6% of patients showed cognitive impairment at follow-up in at least one cognitive domain, indicating a decline in the number of patients with cognitive impairment (p < 0.001). For descriptive purposes, the prevalence of domain-specific cognitive disorders at baseline and at follow-up is shown in table 2.

Remaining basic ADL disturbances were found in 18.7% of patients, whereas 24.3% demonstrated instrumental ADL disability. Associations among the three outcome measures are presented in figure 2. A total of 54.9% of patients demonstrated a complete recovery with respect to all outcome measures, whereas 10.8% of patients demonstrated an incomplete recovery with regard to all three outcome measures.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Association of outcome measures. ADL = activities of daily living.

Determinants of cognitive impairment on the long term.

Univariate analyses showed that cognitive impairment on the long term was associated with older age, lower education, female sex, no regular alcohol consumption, lower MBI at baseline, body temperature > 38°C, glucose level at admission, large lesion volume, and impairments in all cognitive domains at baseline (all p ≤ 0.1) (see table E-1 on the Neurology Web site at www.neurology.org). These predictor variables were introduced into multivariate stepwise logistic regression analysis to find independent predictors of long-term cognitive impairment. In the medical model, cognitive impairment was associated with a large lesion volume, female sex, and older age, whereas the cognitive model yielded two independent predictors, i.e., impairments in executive functioning and abstract reasoning (see table E-2). In the combined model cognitive impairment on the long term was associated with impairments in executive functioning and abstract reasoning, and female sex. Lesion volume and age dropped from this model because these variables failed to reach statistical significance. Patients with executive impairments at baseline nearly had sevenfold greater odds of being cognitively impaired 6 months later, whereas patients with early reasoning disturbances had fourfold greater odds of obtaining an adverse cognitive outcome (see table E-2).

All three models demonstrated an acceptable fit by the Hosmer-Lemeshow test. The combined model yielded the largest AUC, followed by the cognitive model (see table E-2). However, the difference among the three models did not reach significance.

Determinants of functional impairment on the long term.

An MBI score < 19 at follow-up was univariately associated with older age, lower education, a higher score on the baseline NIHSS, a lower score on the baseline MBI, body temperature > 38°C, admission glucose level, weakness in either arm or leg, large lesion volume, presence of cerebral atrophy, unilateral neglect, and impairments in language, verbal memory, executive functioning, visual memory, and visual perception and construction at baseline (all p ≤ 0.1) (see table E-1). In the medical model, remaining basic ADL disturbances were independently associated with stroke severity at admission, older age, body temperature > 38°C, and admission glucose level. The cognitive model retained two independent predictors: visual memory and unilateral neglect (see table E-2). In the combined model, incomplete basic ADL recovery was associated with impairments in visual memory and unilateral neglect, and with admission glucose level. All other predictor variables dropped from this model because they failed to reach statistical significance. Patients with acute unilateral neglect had 18-fold greater odds and patients with acute visual memory impairments showed 22-fold greater odds for remaining basic ADL disturbances (see table E-2).

All models demonstrated an acceptable fit by the Hosmer-Lemeshow test. The combined and the cognitive model demonstrated the largest AUC (see table E-2). However, the difference among the three models did not reach significance.

Instrumental ADL impairment on the long term was associated with a low level of education, poorer NIHSS and MBI at baseline, body temperature > 38°C, weakness in arm or leg, stroke location in the right hemisphere, lesion volume, unilateral neglect, and impairments in language, executive functioning, visual memory, and visual perception and construction at baseline (all p ≤ 0.1) (see table E-1). In the medical model, instrumental ADL impairment was independently associated with a poor MBI at admission and a lesion location in the right hemisphere, whereas the cognitive model yielded only one predictor, i.e., visual perception and construction. In the combined model, instrumental ADL impairment was associated with impairments in visual perception and construction at baseline, poor MBI at baseline, and a lesion location in the right hemisphere. Patients with acute visual perceptual disorders demonstrated 27-fold greater odds of obtaining an adverse instrumental functional outcome (see table E-2).

The models demonstrated an acceptable fit by the Hosmer-Lemeshow test (no goodness-of-fit statistic could be calculated for the cognitive model as this model consisted of only one predictor). The combined model demonstrated a larger AUC than the medical model (p < 0.05) and the cognitive model (p < 0.05) (see table E-2).