Poststroke Dementia: Title and subTitle BreakInfluence of Hippocampal Atrophy FREE

THE RISK of dementia is increased in patients who had a stroke, with prevalence rates of poststroke dementia (PSD) ranging from 13.6% to 32%^{1 - 4} and incidence reaching 33.3% within 4 years after stroke.⁵ The pathophysiology of PSD includes direct consequences of the vascular lesions of the brain⁶ and possible Alzheimer disease (AD)–associated changes⁷ : among stroke patients with PSD about one third are considered to have AD and stroke because cognitive decline was present before the index stroke, as well as having a progressive time course without any evidence of a new stroke.⁸ Medial temporal lobe atrophy (MTLA) is strongly associated with AD^{9 - 11} and PSD¹² ; however, the influence of MTLA on the risk of new-onset dementia after stroke remains unknown. The aim of our study was to determine the influence of MTLA on the risk of new-onset dementia in stroke patients.

The study population was recruited over a 24-week period (December 1, 1995, through May 12, 1996) among all consecutive patients admitted to the Stroke Unit, Lille University Hospital, Lille, France. Patients of this study belong to the Lille Dementia/Stroke study whose methodology and first results have been reported elsewhere.^{12 - 14} From this study we excluded patients (1) who had transient ischemic attacks or cerebral venous thrombosis, (2) who were younger than 40 years, (3) who were nonwhite or not fluent French speakers, (4) who were without a reliable informant, (5) who had a medical history of severe head trauma or cerebral surgery, (6) who were referred from another hospital or who did not live in the urban community of Lille, because the follow-up is usually performed at other centers, and (7) who had prestroke dementia. Preexisting dementia was diagnosed using a French translation of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE),¹⁵ with a cut-off score set at 104.^{13 ,16}

Patients were examined according to a previously reported procedure^{12 - 14} : on admission to the hospital patients were examined by a neurologist (H.H., F.P., or D.L.). Medical history was determined from all available records and sources (ie, patient, family, and/or general practitioner). Patients underwent standard blood chemistry tests and urinalysis, a 12-lead electrocardiogram, and noncontrast computed tomographic (CT) scan. Doppler ultrasonography and B-mode echo tomography of the cervical arteries and bidimensional transthoracic echocardiography were performed within 24 hours. Delayed CT scan was performed within 8 to 10 days. In selected patients, we performed a cerebral magnetic resonance imaging (MRI) scan, magnetic resonance angiography, 24-hour electrocardiography, transesophageal echocardiography, cerebral angiography, and hypercoagulability testing. We prospectively collected the following data using previously described criteria^{12 - 14} : age; sex; educational level; presence of arterial hypertension, diabetes mellitus, or hyperlipidemia; mean alcohol consumption exceeding 300 g/wk; cigarette smoking; high-risk cardiopathy according to TOAST (Trial of Org 10172 in Acute Stroke Treatment) criteria; and significant stenosis of the cervical arteries, defined as a narrowing of 50% or more of the lumen documented by Doppler ultrasonography and B-mode echo tomography of the cervical arteries, magnetic resonance angiography, or conventional angiography. The severity of the clinical deficits at admission was scored according to the Orgogozo scale.¹⁷ Data were recorded within 24 hours after stroke onset. The presumed causes of stroke were defined at discharge, according to the TOAST criteria.¹⁸

Stroke subtype (ischemic or hemorrhagic), the presence of silent infarcts,¹⁹ and the presence and severity of leukoaraiosis^{20 - 21} and cerebral atrophy²² were determined on noncontrast medium–enhanced CT scans performed at admission using 5-mm contiguous slices. We added to the usual CT procedure performed at admission by investigating the medial temporal lobe according to the procedure of Jobst et al.¹⁰ We used planes 20° caudal to the orbitomeatal line with 2-mm contiguous slices through the posterior fossa. Left and right medial temporal lobe widths (combined hippocampal formation and parahippocampal gyri) were measured in millimeters from the film by a neuroradiologist (J.-P.P.), who was blinded to the clinical diagnosis. A cut-off of 11.5 mm was used to differentiate a patient with MTLA (defined as at least one medial temporal lobe ≤11.5 mm) from a patient without MTLA.¹⁰ In patients with temporal mass effect due to bland or hemorrhagic stroke, the measure was performed only on the unaffected side.

Patients were followed up at 6 months after stroke onset, then annually over a 3-year period, by a visit with a neurologist or by telephone contact with the patient's family or general practitioner. The scheduling of annual visits was based on the date of the stroke. Patients who kept their appointments with the neurologist underwent a battery of neuropsychological tests. We performed a global evaluation of cognitive functions using the Mattis Dementia Rating Scale²³ and assessed the following cognitive functions according to a previously reported procedure¹³ : executive functions, short-term and long-term verbal memory, visual memory, language ability, gestual praxis, gnosia, constructional and visuospatial functions, concept formation, and reasoning. The Mini-Mental State Examination (MMSE)²⁴ was also performed. At the year 3 visit, cognitive functions were evaluated only in patients with an MMSE score of 24 or less, or when dementia was suspected by the neurologist. An interval of 6 months after stroke was chosen for neuropsychological testing to have an optimal decline of the short-term consequences of stroke, since a physical stable course is often obtained at 6 months. All available information gathered from the neurological, neuropsychological, and functional assessments was reviewed at a diagnostic case conference attended by the examining neurologist and 2 neuropsychologists experienced in the diagnosis of dementia. They were all blinded to the CT data concerning medial temporal lobe thickness. The diagnosis of dementia was based on International Classification of Diseases, 10th Revision (ICD-10) criteria.²⁵ To diagnose dementia in aphasic patients able to undergo the neuropsychological evaluation, nonverbal memory impairment was required. For patients who could not or refused to undergo the neuropsychological testing, information about the patient's cognitive status, behavioral and abilities in daily living activities was obtained from the patient's closest relative and an IQCODE score was obtained. A cut-off of 104 was required for the patient to have a diagnosis of dementia. Criteria for AD were those of the NINCDS-ADRDA (National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association) Work Group,²⁶ and criteria for vascular dementia (VaD) were those of the NINDS-AIREN (National Institute of Neurological Disorders and Stroke–Association International pour la Recherche et l'Enseignement en Neurosciences) Work Group.²⁷ We excluded from the statistical analysis intercurrent hypoxic disorders because in many cases, we could not precisely determine the temporal relationship between the date of onset of dementia and intercurrent hypoxic disorders.

The first step of the statistical analysis consisted of a description of the prevalence of MTLA, with a 95% confidence interval (CI). Because of the high mortality rate before the end of the follow-up period, we had to use survival methods that consider cases lost to follow-up. These methods are appropriate when the event of interest does not occur in all patients during the follow-up period and when the actual period of follow-up is not the same for all patients. Start time for survival was the date of stroke onset. The date of onset of dementia was considered the date of the visit at which the diagnosis of dementia was given. The diagnosis of dementia was based on ICD-10 criteria, which require, for a confident diagnosis of dementia, decline in memory and other cognitive abilities to have been clearly present for at least 6 months. As patients were followed up at month 6 and then at year 1, year 2, and year 3, the error concerning the date of onset of dementia is necessarily less than 6 months, which is acceptable considering the aim of the study. The interval used for analysis was 1 month. Kaplan-Meier survival analysis using the log rank test was used to determine the proportion of patients surviving free of dementia in patients with and without MTLA. The explanatory variables were then tested one by one using the Kaplan-Meier survival analysis using the log rank test to evaluate the influence of the following variables: demographic characteristics (sex and primary educational level), vascular risk factors for stroke (arterial hypertension, diabetes mellitus, hyperlipidemia, excessive alcohol consumption, cigarette smoking, a history of stroke, a history of transient ischemic attack, high-risk cardiopathy, and significant stenosis of the cervical arteries), stroke subtype (ischemic or hemorrhagic), cause of stroke according to the TOAST criteria, CT data (presence of silent infarcts on CT scan), and stroke recurrence. The relative risk (RR) was computed using the Cox proportional hazards regression model. Cox proportional hazards regression analysis was used for continuous variables (age, IQCODE score, Orgogozo score at admission, cerebral atrophy score, and leukoaraiosis score). Multicollinearity was assessed using the analysis of correlations among the continuous independent variables.²⁸ Thereafter, multivariate analysis was performed using Cox proportional hazards regression analysis including demographic variables (age and sex) and other potential predictors of PSD for which we obtained a P value less than .20 in previous analysis (diabetes mellitus, dyslipemia, history of stroke, high-risk cardiopathy, IQCODE score, Orgogozo score, MTLA, silent infarcts on CT, and severity of leukoaraiosis) . The α level for all analysis was .05. The statistical analyses were computed using SPSS Version 9.0 for Windows (SPSS, Chicago, Ill).

During the 24-week period of recruitment, 171 consecutive patients who had a stroke were older than 40 years and did not meet any of the exclusionary criteria were admitted in the Stroke Unit, University of Lille, Lille, France: they all underwent positioned medial temporal lobe scans at admission. Twenty-seven patients (15.8%) were excluded because of an IQCODE score of 104 or higher. Therefore, the study population consisted of 144 patients (66 women and 78 men), with a median age of 72 years (age range, 42-100 years). Twelve patients (8.3%) had a deep intracerebral hemorrhage and 132 patients (91.7%) had an ischemic stroke. One hundred patients (69.4%) had a first stroke.

Death occurred in 53 patients (36.8%). Of the 113 survivors at month 6, 95 patients (85%) had at least 1 visit with a neurologist (H.H., F.P., or D.L.) within the follow-up period. After exclusion of patients who died before each planned follow-up visit, the proportion of survivors who were examined was 88 (78%) of 113 patients at month 6, 73 (71%) of 103 at year 1, 56 (58%) of 97 at year 2, and 61 (67%) of 91 at year 3. Of the 88 survivors who were examined at month 6, 83 patients (94%) underwent neuropsychological examinations (4 refused and 2 were too severely aphasic); the proportion was 66 (90%) of 73 at year 1 (5 refused and 2 were too severely aphasic) and 53 (95%) of 56 at year 2. At year 3, of the 61 patients who were examined, 15 had an MMSE score of 24 or less: 10 underwent neuropsychological testing; 5 refused. Among patients who refused or could not undergo neuropsychological testing, the IQCODE score was obtained in all patients. Among survivors who did not keep their appointment follow-up visit, the IQCODE score was obtained in all patients but 1 at month 6, and in all but 3 per year (1 patient was severely physically impaired; 2 were lost to follow-up each year) at years 1, 2, and 3. Information on cognitive status prior to death was obtained in all patients who died more than 6 months after the last visit or the last telephone contact. The presence of dementia could be assessed in all patients but 1 at month 6, in all but 3 at subsequent contacts. The presumed cause of dementia could only be determined in patients who had dementia and who completed at least 1 visit with the neurologist.

Of the 144 patients, 71 (49.3%) (95% CI, 41.1-57.5) had MTLA on CT performed at admission. Of the 71 patients with MTLA, 53 (75%) had an IQCODE score at admission ranging from 79 to 103 (indicating cognitive impairment without dementia), and 18 (25%) had an IQCODE score of 78 (no cognitive impairment). Of the 73 patients without MTLA, 35 (47.9%) had an IQCODE score at admission ranging from 79 to 103, and 38 (52.1%) had an IQCODE score of 78.

Using ICD-10 criteria, 34 patients (23.6%) were diagnosed as having PSD: 19 (56%) fulfilled the NINDS-AIREN²⁷ criteria for probable VaD and 10 (29%) fulfilled the NINCDS-ADRDA²⁶ criteria for possible AD. In the remaining 5 patients, the cause of dementia remained undetermined because they did not have any follow-up visit. Dementia was diagnosed at the month 6 visit in 27 patients (79.5%), at the year 1 visit in 3 patients (8.8%), at the year 2 visit in 3 patients (8.8%), and at the year 3 visit in 1 patient (2.9%).

Seventeen (15.5%) of the 110 patients who did not have dementia had 1 or more recurrent strokes, while 7 (20.6%) of the 34 patients with incident dementia did have 1 or more recurrent strokes. Stroke had been diagnosed before dementia in 2 patients and after the onset of dementia in 5 patients. Of the 24 patients who had stroke recurrence, 7 (29.2%) had dementia; of the 120 patients without stroke recurrence, 27 (22.5%) had dementia (P = .27).

At the end of the follow-up period, the cumulative proportion of patients surviving without dementia was 57.6% in patients with MTLA and 80.9% in patients without MTLA (RR, 2.3; 95% CI, 1.1-4.7; P = .02) (Figure 1). Results concerning the influence of the other variables on the risk of PSD are listed in Table 1.

Assessment of multicollinearity revealed poor information redundancy between the different factors as none of the correlations reached an r higher than 0.6 (Table 2). However, because correlations between cerebral atrophy and age (r = 0.58, P<.001) and between cerebral atrophy and leukoaraiosis (r = 0.57, P<.001) were greater than 0.5,²⁸ the variable cerebral atrophy was not entered in the model. The multivariate analysis found the following to be independent predictors of new-onset dementia: age (RR, 1.04; 95% CI, 1.0-1.08; P = .04), diabetes mellitus (RR, 3.53; 95% CI, 1.45-8.58; P = .005), Orgogozo score at admission (RR, 0.98; 95% CI, 0.96-0.99; P = .002), and leukoaraiosis (RR, 1.45; 95% CI, 1.01-2.06; P = .04). The presence of mild cognitive impairment prior to stroke tended to be significant (RR, 1.05; 95% CI, 0.99-1.11; P = .06). The MTLA was not an independent predictor of PSD.

Of the 23 patients with MTLA who had dementia within 3 years, the presumed cause of dementia was AD in 10 patients (43.5%), VaD in 10 patients (43.5%), and remained unknown in 3 patients (13%). Of the 11 patients without MTLA, the presumed cause of dementia was VaD in 9 patients (81.8%) and remained unknown in 2 patients (18.2%). Among the 10 patients who developed dementia and fulfilled criteria for AD, all had MTLA on the CT scan performed at admission; 9 (90%)of them had prestroke cognitive impairment without dementia. Among the 19 patients who developed dementia and fulfilled criteria for VaD, 10 (52.6%) had MTLA on the CT scan performed at admission, and 14 had prestroke cognitive impairment without dementia. In 5 of these 19 patients, a degenerative factor was suspected because of a progressive time-course without evidence of new stroke: 3 of them had MTLA and memory disturbances before stroke.

This study revealed that after a strict exclusion of patients having prestroke dementia, MTLA was present in approximately half of patients and that the risk of PSD was 2.3-fold increased in patients having MTLA. Nevertheless, MTLA was not an independent risk factor in multivariate analysis after adjustment for age, diabetes mellitus, severity of the neurological deficits, and leukoaraiosis.

The risk of PSD was twice as high in patients with MTLA compared with patients without MTLA. Medial temporal lobe atrophy has previously been found to be a predictor of dementia in patients with mild cognitive impairment.^{29 - 30} De Léon et al³⁰ followed up subjects with MTLA and mild cognitive impairment and found MTLA to be highly predictive of future AD. More recently,³¹ 80 patients with mild cognitive impairment³² had an MRI measurement of both hippocampi and were followed up for 32 months: the proportion of patients who developed AD was 37% (25/67) for those with MTLA and 15% (3/13) for those without MTLA. Our results are similar: 32% of patients with an initial MTLA developed dementia within 3 years after stroke onset, whereas 15% without MTLA developed dementia within the same time span. Nevertheless, in our study, MTLA was not an independent predictor of PSD after adjustment for age, diabetes mellitus, severity of clinical deficits, and leukoaraiosis. This could partly be explained by the median age of our population. Scheltens et al³³ have suggested that MTLA is strongly associated with the diagnosis of AD in subjects younger than 75 years, while MTLA in older subjects is more an age-related atrophy than an indicator of preclinical AD. Golomb et al³⁴ evaluated hippocampal atrophy in 154 cognitively healthy subjects aged 55 to 88 years: 33% had hippocampal atrophy and the prevalence of atrophy significantly increased with age up to 56% in subjects older than 76 years. Another explanation for the lack of independent relationship between MTLA and PSD may be that we did not consider the severity of MTLA. However, in a previous study³¹ conducted in patients with mild cognitive impairment, among the 27 patients who had dementia, 19 patients (70.4%) had severe MTLA.

We evaluated MTLA on CT scans performed at the acute stage of stroke scan-temporal lobe–oriented plane that enables imaging of the long axis of the hippocampal region and permits a good view of the anteroposterior extent of the perihippocampal region. The method is simple, quick, and routinely applicable, and a cut-off value of 11.5 mm provides good diagnostic accuracy for a diagnosis of AD.^{10 ,29 ,35} Indeed, MRI may offer more accurate information owing to a best spatial resolution, less volume-averaging effect, and bone-hardening artifact, but requires a dedicated software and is not always feasible at the acute stage of stroke. We could have assessed MTLA on delayed MRI but this procedure would have introduced several of the following biases: (1) exclusion of patients who died early, who had contraindications to MRI, or who were medically unstable; and (2) delayed structural changes in the medial temporal lobes in patients with infarcts in the posterior cerebral artery territory. Therefore, we considered that the evaluation of MTLA on the first CT scan was probably the best reflection of what this area was before stroke.

Our findings concerning the influence of age, diabetes mellitus, and severity of the neurological deficit on PSD are similar to those of the few previous published studies.^{1 - 2 ,4 - 5 ,13} Other demographic characteristics or vascular risk factors seem irrelevant. We also found leukoaraiosis as an independent predictor of PSD: this finding contradicts the results of the Framingham study cohort,³⁶ but agrees with the results of the previous studies conducted in patients hospitalized for stroke.^{2 ,5}

In most cases, the diagnosis of dementia was made at month 6, suggesting that stroke itself has an important role in the occurence of dementia. Stroke may, of course, in many cases be the major cause of cognitive decline. However, about 30% of patients were diagnosed as having AD and not VaD, always at the month 6 visit, supporting the hypothesis that the vascular lesion might also reveal an underlying asymptomatic AD.

We found that all patients who developed AD after stroke had MTLA, and about 50% of patients who developed VaD had MTLA. The main limitation of our study is, of course, that we did not perform any neuropathological evaluation of the cause of dementia in patients who died: clinicians may have preconceived notions that slowly progressive decline is more likely to signify AD. However, the diagnosis of AD was based not only on the course of dementia, but also on the type of neuropsychological profile, and on evolution of CT data, with new CT or MRI performed in case of neuropsychological worsening. The recent development of symptomatic treatments has emphasized the need for an early and accurate diagnosis of AD, explaining that many studies have evaluated the utility of MTLA in differentiating healthy brain structures from the atrophy associated with AD.^{9 - 11} However, MTLA does not seem to be specific of AD; it is also observed in VaD.^{35 ,37 - 39} Indeed, MTLA was found in patients having VaD diagnosed according to Diagnostic and Statistical Manual of Mental Disorders, Revised Third Edition criteria.^{35 - 37} However, as these criteria do not require any temporal relationship between stroke and dementia, VaD may have been diagnosed in patients having AD associated with stroke. However, Pasquier et al²⁹ also found MTLA in patients having VaD diagnosed according to the NINDS-AIREN criteria.²⁷ Recently, Fein et al,³⁸ comparing MTLA in normal aging, AD, and subcortical VaD, found MTLA in patients with VaD and suggested that a combination of degenerative and ischemic pathologic conditions could explain this finding. However, neuropathological examination performed in 3 patients with VaD did not find any significant AD pathology in the neocortex. The authors suggested that cortical atrophy and MTLA in VaD might result from secondary axonal and transsynaptic degeneration following subcortical injury. And last, O'Brien et al³⁹ found no differences in medial temporal lobe width among AD-affected brains, VaD, and dementia with Lewy bodies, MTLA being useful only to distinguish between depression and dementia. The authors concluded that the measures of medial temporal lobe could be more helpful at an early stage of the disease, as suggested by Jobst et al³⁵ : MTLA in non-AD dementia may be a late feature in contrast to AD, where it occurs early. In our study, patients with MTLA had an increased risk of dementia, with about half of the patients having AD and half of the patients having VaD, suggesting that MTLA is not specific of AD. However, in PSD, global cerebral atrophy (with white-matter lesions and ventricular dilatation) might be a reflection of the magnitude of the overall ischemic process, whereas MTLA reflects a different pathogenic process, more often AD. Among our patients without MTLA who had dementia within 3 years, none was diagnosed as having AD, suggesting that MTLA could be more frequent in patients with AD than in patients with VaD.

Nevertheless, we cannot exclude that a larger following could have shown a closer relationship between MTLA and further dementia in stroke patients. A new study with initial MRI volumetric evaluation, using medial temporal lobe thickness and change in neuropsychological testing as continuous rather than dichotomous variables, with a longer follow-up and neuropathological confirmation might provide other important findings.

Corresponding author and reprints: Hilde Hénon, MD, PhD, Department of Neurology, Stroke Unit, Hôpital Roger Salengro, F-59037 Lille, France (e-mail: hhenon@chru-lille.fr).

Accepted for publication September 19, 2002.

Author contributions: Study concept and design (Drs Cordoliani-Mackowiak, Hénon, Pruvo, Pasquier, and Leys); acquisition of data (Drs Cordoliani-Mackowiak and Hénon); analysis and interpretation of data (Drs Cordoliani-Mackowiak, Hénon, Pasquier, and Leys); drafting of the manuscript (Drs Cordoliani-Mackowiak and Hénon); critical revision of the manuscript for important intellectual content (Drs Cordoliani-Mackowiak, Hénon, Pruvo, Pasquier, and Leys); statistical expertise (Drs Cordoliani-Mackowiak and Hénon); obtained funding (Dr Hénon); administrative, technical, and material support (Dr Pasquier); study supervision (Drs Hénon, Pruvo, Pasquier, and Leys).

This study was supported by grant 9306 from the Centre Hospitalier et Universitaire de Lille, Lille; Association pour la Recherche et l'enseignement en Pathologie Neurovasculaire (APREPAN), Lille; Association d'Etude et de Recherche sur la Maladie d'Alzheimer (ADERMA), Lille; Association pour le Développement, la Recherche, l'Innovation du Nord Pas de Calais (ADRINORD), Lille; a grant from the Ministere de l'Education National, de la Recherche et de la Technologie (MENRT), Paris, France; and grant EA 2691 from Research and Technology for the Research Group on Cognition in Degenerative and Vascular Disorders, Lille.

We thank Jocelyn Campos for her technical assistance and Sylvie Pollet and Carole Mouly for neuropsychological assistance.