Journal of Alzheimer’s Disease 41 (2014) 887–901 DOI 10.3233/JAD-132186 IOS Press
887
Cognitive and Neuroimaging Profiles in Mild Cognitive Impairment and Alzheimer’s Disease: Data from the Spanish Multicenter Normative Studies (NEURONORMA Project) Gonzalo S´anchez-Benavidesa , Jordi Pe˜na-Casanovaa,b,∗ , Marta Casals-Colla , Nina Gramuntc , Jos´e L. Molinuevod , Beatriz G´omez-Ans´one , Miguel Aguilarf , Alfredo Roblesg , Carmen Ant´unezh , Carlos Mart´ınez-Parrai , Anna Frank-Garc´ıaj , Manuel Fern´andez-Mart´ınezk and Rafael Blesal for the NEURONORMA Study Team1 a Hospital
del Mar Research Institute, Barcelona, Spain of Behavioral Neurology and Dementias, Hospital del Mar, Parc Salut Mar, Barcelona, Spain c BarcelonaBeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain d Department of Neurology, Hospital Cl´ınic i Provincial, Barcelona, Spain e Department of Neuroradiology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain f Department of Neurology, Hospital M´ utua de Terrassa, Terrassa, Spain g Department of Neurology, Hospital Cl´ınico Universitario, Santiago de Compostela, Spain h Department of Neurology, Hospital Virgen Arrixaca, Murcia, Spain i Department of Neurology, Hospital Virgen Macarena, Sevilla, Spain j Department of Neurology, Hospital Universitario La Paz, Madrid, Spain k Department of Neurology, Hospital de Cruces, Barakaldo, Spain l Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain b Section
Handling Associate Editor: Montse Alegret Accepted 2 March 2014
Abstract. The aim of this study was to characterize the neuropsychological and neuroimaging profiles of mild cognitive impairment (MCI) and Alzheimer’s disease (AD) patients, and to study the magnitude of the differences by comparing both outcomes with healthy subjects in a cross-sectional manner. Five hundred and thirty-five subjects (356 cognitively normal adults (CONT), 79 MCI, and 100 AD) were assessed with the NEURONORMA neuropsychological battery. Thirty CONT, 23 MCI, and 23 AD subjects from this sample were included in the neuroimaging substudy. Patients’ raw cognitive scores were converted to age and education-adjusted scaled ones (range 2–18) using co-normed reference values. Medians were plotted to examine the cognitive profile. MRIs were processed by means of FreeSurfer. Effect size indices (Cohen’s d) were calculated in order to compare the standardized differences between patients and healthy subjects. Graphically, the observed cognitive profiles for MCI and AD groups produced near to parallel lines. Verbal and visual memories were the most impaired domains in both groups, ∗ Correspondence
to: Jordi Pe˜na-Casanova, Behavioral Neurology and Dementia Section, Neurology Department, Hospital del Mar, Passeig Maritim, 25-29, 08003 Barcelona, Spain. Tel.: +34 93 3160778; Fax: +34 93 3160723; E-mail: JPcasanova@ parcdesalutmar.cat.
ISSN 1387-2877/14/$27.50 © 2014 – IOS Press and the authors. All rights reserved
888
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
followed by executive functions and linguistic/semantic ones. The largest effect size between AD and cognitively normal subjects was found for the FCSRT (d = 4.05, AD versus CONT), which doubled the value obtained by the best MRI measure, the right hippocampus (d = 1.65, AD versus CONT). Our results support the notion of a continuum in cognitive profile between MCI and AD. Neuropsychological outcomes, in particular the FCSRT, are better than neuroimaging ones at detecting differences among subjects. Keywords: Alzheimer’s disease, magnetic resonance imaging, mild cognitive impairment, neuropsychological tests, neuropsychology
INTRODUCTION Neuropsychological assessment is crucial in the diagnosis and characterization of mild cognitive impairment (MCI) and Alzheimer’s disease (AD). Diagnostic guidelines for AD and MCI require proof of lower performance in cognitive tasks that is greater than would be expected for the subject’s age and educational background. In some cases, medical historytaking and bedside mental state examination are sufficient for diagnosis. In non-conclusive cases, formal neuropsychological assessment is needed to document cognitive impairment and quantify its severity [1, 2]. There are two main approaches in neuropsychological assessment: the flexible approach, in which the neuropsychologist selects the tests that best fit individual clinical questions and patients’ complaints; and the fixed one, in which the same array of predefined tests grouped in a neuropsychological battery are always employed. Both approaches are valid and the selection of one or the other depends upon the clinician’s background and the testing objectives. However, the use of a fixed, comprehensive, and standardized neuropsychological test battery presents some considerable advantages, such as the possibility of constructing a cognitive profile that can be compared across subjects and pathologies.
[5], and executive [6, 7] functions are also impaired in MCI subjects who subsequently develop dementia. Indeed, a report from a large, recent, longitudinal study states that single-domain amnestic MCI is rarely diagnosed when subjects undergo a comprehensive neuropsychological assessment [8], corroborating previous studies [9]. As claimed by some authors, the use of extensive batteries seems necessary to adequately describe MCI subjects [10]. Neuroimaging profiles In recent decades, neuroimaging techniques have undergone an increasing range of applications in AD. Imaging has moved from an exclusionary role to a recognized position in diagnosis and research. A variety of imaging modalities, including structural and functional ones, has shown distinctive changes in the brains of patients within the AD spectrum [11]. Overt dementia presents marked atrophy in the medial temporal lobe structures [12, 13] along with progressive thinning of the parietotemporal and frontal cortices [14, 15]. The pattern observed in MCI is similar, but to a lesser degree, to that of AD. Atrophy in medial temporal lobe structures is the most salient finding although volume and cortical thickness loss have been observed in other regions, such as the medial parietal and frontal areas [16, 17].
Cognitive profiles Cognition versus biomarkers A cognitive profile can be simply defined as the pattern of preserved and impaired cognitive functions. It is useful in differential diagnosis, identification of subtypes, and monitoring of disease progression [3]. Interpretation of the cognitive profile can be used in dementia to track the cognitive decline. Accurate analyses of dissociations between preserved/impaired domains can be applied, for example, in the design of strategies aimed at minimizing the impact of disability on daily living. The benefits of the cognitive profile approach may be especially valuable in prognosis at predementia stages. It has been demonstrated that, in addition to memory, attentional [4], visuoperceptual
New diagnostic trends in AD spectrum pathologies include biomarkers, such as cerebrospinal fluid (CSF) and neuroimaging in order to improve specificity [1, 2, 18]. Nonetheless, there is as yet no consensus on which measure (i.e., CSF, MRI, PET, or neuropsychological assessment), or combination of them, is best to discriminate MCI and AD individuals from normal ones. While a number of authors claim that imaging is superior to cognitive testing for AD diagnosis [19], other meta-analytic studies have found the opposite [20]. In agreement with the latter, the conversion from MCI to AD seems better predicted by neuropsycholog-
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
ical performance [21, 22]. Clearly, prediction accuracy increases when models that combine cognition, MRI, and/or CSF are employed [23–25]. Nevertheless, the use of CSF-based biomarkers, although of great interest in research, is presently far from suitable for clinical use. With regard to neuroimaging, MRI, and even PET, while widely available are expensive and do not constitute a core feature of diagnosis. At present, neuropsychological assessment is the most feasible, economical, and effective way of characterizing subjects in the AD pathology spectrum. The present study The present cross-sectional study provides data regarding the neuropsychological and MRI profiles of MCI and AD individuals. It compares the magnitude of the differences with normal aging in order to identify the best measures to differentiate patients from controls. The study is framed within the Spanish Multicenter Normative Studies (NEURONORMA project). In the NEURONORMA project, a group of MCI and AD subjects were recruited with the purpose of studying cognitive profiles and developing diagnostic norms. Our group has presented normative data from a Spanish population to provide valid and appropriate norms for commonly used neuropsychological tests. The use of a co-normed battery allows the direct comparison of test scores and facilitates the clinical interpretation of neuropsychological test profiles. The control sample and methodology have been previously described [26], and the norms have been published in a series of papers [26–31]. The specific objectives of this paper are: 1) to describe the characteristic neuropsychological profiles of MCI and AD subjects by means of the co-normed NEURONORMA battery and its age- and educationadjusted Spanish normative data; 2) to present results of brain atrophy in a subsample of MCI and AD; and 3) to study the magnitude of the differences amongst groups for both neuroimaging and neuropsychological outcomes. MATERIAL AND METHODS Participants A total sample of 535 subjects was studied. Participants included 356 cognitively normal adults (CONT), 79 patients diagnosed with MCI, and 100 patients with AD (70 with mild and 30 with moderate dementia). All subjects were Caucasian. The project was reviewed
889
and approved by the Research Ethics Committee of the Hospital del Mar Research Institute (Spain), and was conducted in accordance with the Declaration of Helsinki and its subsequent amendments, and the European Union regulations concerning medical research. All participants were required to have an informant who could answer questions about their cognition, function, and health. Patients were recruited from dementia consultations at nine hospitals in different regions of Spain between 2004 and 2007. Diagnosis was based on a clinical interview with the patient and the informant, and a physical and neurological examination. A series of previously validated instruments (i.e., cognitive, psychiatric, medical, and functional scales) was administered as part of the diagnostic process and classification of participants (see [26] for details). These scales included the Mini-Mental State Examination (MMSE) in its validated Spanish version [32], the Spanish version of the Interview for Deterioration of Daily living activities in Dementia (IDDD) [33], the Modified Hachinski Ischemia Score, and the Hamilton Depression Rating Scale. Laboratory (blood) tests and brain imaging (CT or MRI) were performed according to local site procedures. A subsample composed of 60 CONT, 25 MCI, and 25 AD from two centers (Hospital del Mar and Hospital Clinic de Barcelona) were included in the neuroimaging substudy. The final sample was made up of 30 CONT, 23 MCI, and 23 AD (19 with mild and 4 with moderate dementia). In addition, in a genetic sub-study (n = 226) the presence of APOE 4 alleles was determined. CSF samples, although of great interest, were not collected in this study for two main reasons. Firstly, the participation of nine hospitals complicated the harmonization of collection and processing of the samples. Secondly, because when this project was planned such studies were not yet implemented in almost any setting. A brief description of the main features of the sample is as follows: 1) CONT definition: Subjects were aged 50 to 85 years. Recruitment procedures and socio-demographic characteristics of the normative sample have been previously reported [26]; 2) MCI diagnostic criteria: Patients were diagnosed as having amnestic MCI according to the following criteria: (a) Presence of subjective memory complaints; (b) Essentially preserved functional abilities (confirmed by an informant); (c) A minimal evolution of 6 months; (d) Evidence of objective memory impairment, defined as a score below percentile 15 in the age and education-matched reference group on a test of verbal delayed recall (Barcelona Test paragraph recall) [34], but without any other significant deficits in the
890
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
Barcelona Test battery; (e) Not conforming to criteria for probable AD according to the National Institute of Neurological and Communication Disorders and Stroke/Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) [35]; and (f) a Global Deterioration Scale (GDS) score of 3; 3) AD diagnostic criteria: Diagnosis of dementia was made following the Diagnostic and Statistical Manual for Mental Disorders, 4th edition revised [36]; Diagnosis of AD was carried out according to the NINCDS/ADRDA criteria [35]. A minimum MMSE score of 14 points were required to be enrolled. Patients were staged as having mild to moderate dementia (GDS 4-5). Neuropsychological methods The neuropsychological protocol included the following tests: Verbal span (Digit Span forward and backward) [34]; Visuospatial Span (Corsi’s Test) from the WAIS-R-NI [37]; Letter-Number Sequencing (WAIS-III) [38]; Trail Making Test [39]; Symbol Digit Modalities Test [40]; Boston Naming Test [41, 42]; Token Test [43]; Selected test of the Visual Object and Space Perception Battery [44]; Judgment of Line Orientation [45]; Rey-Osterrieth Complex Figure [46, 47]; Free and Cued Selective Reminding Test (FCSRT) [48] (Copyright, 1996–2000. Albert Einstein College of Medicine of Yeshiva University. New York); Verbal fluency, including three semantic fluency tasks (animals, fruit and vegetables, and kitchen tools), three formal phonemic tasks (words beginning with p, m, and r), and three excluded letter fluency task (excluded a, e, and s); Stroop Color-Word Interference Test [49]; Tower of London Drexel University version [50]. All tests were administered and scored according to standardized procedures published in each test manual. For more details on administration procedures and specific variables, see [26–31]. In order to avoid an overload of information and redundancies in the cognitive profile graph, some variables such as Letter-Number Sequencing (WAIS-III) and most of the fluency tasks were excluded. Finally, 33 cognitive variables were studied (see Fig. 1 and Table 2 for a list of the included variables). MRI acquisition and analysis Subjects were imaged in a Signa LX 1.5 Tesla magnet (General Electric, Milwaukee, USA) using a specific T1 3D SPGR protocol (TE = Min Full; TR = 12; TI = 450; angle = 15; 256 x 192; FOV = 25; 1NEX; Thickness = 1.5; 128 locs). Voxel size was
0.97 × 1.5 × 0.97 mm). An initial visual assessment was performed to discard any other relevant brain pathology and to ensure appropriate study quality. Standard DICOM images were first anonymized and then processed by means of FreeSurfer v4.0.2 (http://surfer.nmr.mgh.harvard.edu/). One volumetric variable (total gray matter volume) was calculated by means of the mri segstats command from the version 5.0 of this software. In addition to volumetric quantification of predefined ROIs, FreeSurfer’s processing provides white matter and cerebral cortex models to produce representations of cortical thickness. Cortical thickness is calculated as the closest distance from the gray/white boundary to the gray/CSF boundary at each vertex on the tessellated surface [51]. A parcellation of the cerebral cortex into units based on gyral and sulcal structure [52, 53] was also carried out. FreeSurfer has been implemented in the cluster of the Port d’Informaci´o Cient´ıfica, at the Universitat Aut`onoma de Barcelona. It is composed of 170 HP blades with 2 quad-core CPU (Hewlett Packard, USA), each with 16 GB of RAM, running on Scientific Linux 4 (https://www.scientificlinux.org/). Subjects were processed in parallel and launched through a batch system. The PIC-Neuroimaging Center (PICNIC) remote web-based system was used to perform the processing. Finally, 95 MRI-based variables were studied (29 volumetric ROIs and 66 cortical thickness ROIs, see Results for details). In one third of the images, a limited number of manual corrections was performed to ensure measurement quality. Data management and statistical analysis Analyses and graphs were obtained using R statistical software (v2.7). A detailed description of the statistical procedure for obtaining the normative data has been described elsewhere [26]. In brief, the process of converting the subjects’ raw scores to age-adjusted scaled ones includes a transformation to a 2 to 18 range scale based on percentile ranges, and a subsequent education adjustment based on the coefficients obtained from linear regressions. This approach is similar to that employed by the authors of the Mayo’s Older Americans Normative studies [54]. Basic descriptive statistics included count, percentage of total, mean, and standard deviation. ANOVA and pairwise post-hoc comparisons (Tukey) were applied to test mean differences among groups for sociodemographic and eligibility variables. Since patients were older and less educated than control subjects, mean comparisons among groups for neuropsychological
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
891
Fig. 1. Cognitive profile of MCI and AD groups in the NEURONORMA battery. Medians of age- and education-adjusted scores are shown.
892
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
raw scores were carried out by applying ANCOVA (with age and years of education as covariables) and pairwise post-hoc comparisons (Tukey). Chi-square tests were applied for nominal and ordinal variables. Median values of the scaled scores (range 2–18) of each cognitive variable were used to construct graphical demographic age- and education-adjusted cognitive profiles for the MCI and AD samples. Neuroimaging data analysis was twofold. First, ANCOVA analyses were performed for ROI-based outcomes. Age was included as a covariable for cortical thickness variables; age and intracranial volume were included for volumetric ones. Second, a vertex-based analysis was performed by means of the QDEC tool provided in the FreeSurfer package. A Gaussian kernel of FWHM of 20 mm was applied to smooth images before the analysis. This analysis was corrected for multiple comparisons by the application of a False Discovery Rate of 5%. For all the other statistical procedures significance threshold was set to 0.05. The magnitude of the observed differences between both pathological groups and cognitively normal subjects was explored by calculating Cohen’s d. Cohen’s d values from a group of selected MRI-derived variables (i.e., global gray matter volume, global white matter volume, total CSF volume, entorhinal cortex thickness, hippocampal and amygdalar volumes, and cortical volumes from temporal, frontal, parietal, and occipital lobes) were plotted to illustrate the neuroimaging profile of MCI and AD groups. RESULTS Sociodemographic, neuropsychological, and APOE status data Table 1 summarizes demographic information and eligibility criteria of the three groups (CONT, MCI, and AD). Of the 535 participants, 322 were females (60.2%). CONT subjects were younger and more educated than patients (p < 0.001), there were, however, no significant differences in these variables between the MCI and AD groups (age, p = 0.309; years of education, p = 0.859). The IDDD scale, which measures impairment in activities of daily living, revealed a preserved performance of basic activities in MCI subjects (p = 0.939) and mild impairment in AD (p < 0.001). Complex activities were slightly affected in MCI (p < 0.001) and moderately impaired in AD (p < 0.001). While CONT and MCI subjects did not show differences in depression symptoms (p = 0.797), AD patients did (p < 0.05). The APOE study showed an increased
percentage of subjects carrying at least one 4 allele in both MCI and AD samples with respect to CONT (p < 0.05). However, MCI and AD percentages did not significantly differ (p = 0.285). A summary of the raw neuropsychological scores obtained from each group, and results of mean comparisons, are presented in Table 2. Average cognitive profiles of AD and MCI subjects, expressed as the medians of scaled scores adjusted for age and education, are presented in Fig. 1. This figure shows the relative impairment in each neuropsychological measure against the normative group. The most marked impairment was found in memory measures, in which the AD group scores were below the scaled score 2 (equivalent to percentile two) in both verbal and visual tasks. MRI data The demographic and inclusion variables of the MRI subsample were as follows (means and [SD] are given): CONT (n = 30, age = 72.02 [4.68], years of education = 11.37 [6.05], MMSE = 28.93 [1.36], IDDD = 33.30 [0.67]); MCI (n = 23, age = 72.91 [7.38], years of education = 8.26 [4.70], MMSE = 26.32 [2.08], IDDD = 36.51 [2.79]); AD (n = 23, age = 75.82 [6.30], years of education = 6.00 [3.26], MMSE = 20.08 [3.75], IDDD = 44.78 [8.10]). While groups did not significantly differ in age (p = 0.08), they did in education. In the post hoc test, the only comparison that reached significance was between CONT and AD (p < 0.001). Groups showed statistical differences in MMSE and IDDD scores. MRI and cognitive samples differ in number of subjects and the mean age of the control sample which was closer to that of the pathological samples in the neuroimaging subgroup. Nevertheless, age was included as a covariate in MRI analyses. We found significant differences in most of the MRI-based outcomes among the groups. Supplementary Table 1 shows the results of MRI analyses. To illustrate the neuroimaging profile of MCI and AD effect size, results for selected variables are presented graphically in Fig. 2. The results of the cortical thickness vertex-wise analyses can be seen in Supplementary Fig. 1. Effect size results Cohen’s d calculations were used to explore differences in neuroimaging-based measures and calculated for cognitive variables for comparison purposes. Effect size results can be partially seen in Table 3 (comprehensive data of the 128 variables [33 cognitive and 95
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
893
Table 1 General characteristics of study sample: eligibility and classification criteria Female, n (%) Age (years) Education (years) MMSE MMSE adjusted∗ IDDD Basic activities IDDD Complex activities IDDD Total† Hamilton DRS, M (SD) Ischemia score, M (SD) APOE 4 sub-sample, n At least one 4 allele of APOE, n (%)
CONT
MCI
AD
212 (59.6) 64.9 (9.3) 10.4 (5.4) 28.7 (1.5) 29.1 (1.4) 16 (0.0) 17.1 (0.5) 33.1 (0.6) 2.1 (2.5) 0.1 (0.3) 147 40 (27.2)
45 (57) 72.8 (6.5) 8.0 (4.7) 25.7 (2.2) 26.8 (2.2) 16.1 (0.3) 19.9 (2.4) 36.0 (2.5) 1.9 (2.3) 0.3 (0.5) 35 19 (54.3)
65 (65) 74.7 (7.5) 7.6 (4.6) 20.2 (4.0) 21.3 (3.8) 17.9 (4.0) 30.9 (7.0) 48.9 (10.0) 2.9 (3.3) 0.4 (0.7) 44 27 (61.4)
p value ns
887
Cognitive and Neuroimaging Profiles in Mild Cognitive Impairment and Alzheimer’s Disease: Data from the Spanish Multicenter Normative Studies (NEURONORMA Project) Gonzalo S´anchez-Benavidesa , Jordi Pe˜na-Casanovaa,b,∗ , Marta Casals-Colla , Nina Gramuntc , Jos´e L. Molinuevod , Beatriz G´omez-Ans´one , Miguel Aguilarf , Alfredo Roblesg , Carmen Ant´unezh , Carlos Mart´ınez-Parrai , Anna Frank-Garc´ıaj , Manuel Fern´andez-Mart´ınezk and Rafael Blesal for the NEURONORMA Study Team1 a Hospital
del Mar Research Institute, Barcelona, Spain of Behavioral Neurology and Dementias, Hospital del Mar, Parc Salut Mar, Barcelona, Spain c BarcelonaBeta Brain Research Center, Pasqual Maragall Foundation, Barcelona, Spain d Department of Neurology, Hospital Cl´ınic i Provincial, Barcelona, Spain e Department of Neuroradiology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain f Department of Neurology, Hospital M´ utua de Terrassa, Terrassa, Spain g Department of Neurology, Hospital Cl´ınico Universitario, Santiago de Compostela, Spain h Department of Neurology, Hospital Virgen Arrixaca, Murcia, Spain i Department of Neurology, Hospital Virgen Macarena, Sevilla, Spain j Department of Neurology, Hospital Universitario La Paz, Madrid, Spain k Department of Neurology, Hospital de Cruces, Barakaldo, Spain l Department of Neurology, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain b Section
Handling Associate Editor: Montse Alegret Accepted 2 March 2014
Abstract. The aim of this study was to characterize the neuropsychological and neuroimaging profiles of mild cognitive impairment (MCI) and Alzheimer’s disease (AD) patients, and to study the magnitude of the differences by comparing both outcomes with healthy subjects in a cross-sectional manner. Five hundred and thirty-five subjects (356 cognitively normal adults (CONT), 79 MCI, and 100 AD) were assessed with the NEURONORMA neuropsychological battery. Thirty CONT, 23 MCI, and 23 AD subjects from this sample were included in the neuroimaging substudy. Patients’ raw cognitive scores were converted to age and education-adjusted scaled ones (range 2–18) using co-normed reference values. Medians were plotted to examine the cognitive profile. MRIs were processed by means of FreeSurfer. Effect size indices (Cohen’s d) were calculated in order to compare the standardized differences between patients and healthy subjects. Graphically, the observed cognitive profiles for MCI and AD groups produced near to parallel lines. Verbal and visual memories were the most impaired domains in both groups, ∗ Correspondence
to: Jordi Pe˜na-Casanova, Behavioral Neurology and Dementia Section, Neurology Department, Hospital del Mar, Passeig Maritim, 25-29, 08003 Barcelona, Spain. Tel.: +34 93 3160778; Fax: +34 93 3160723; E-mail: JPcasanova@ parcdesalutmar.cat.
ISSN 1387-2877/14/$27.50 © 2014 – IOS Press and the authors. All rights reserved
888
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
followed by executive functions and linguistic/semantic ones. The largest effect size between AD and cognitively normal subjects was found for the FCSRT (d = 4.05, AD versus CONT), which doubled the value obtained by the best MRI measure, the right hippocampus (d = 1.65, AD versus CONT). Our results support the notion of a continuum in cognitive profile between MCI and AD. Neuropsychological outcomes, in particular the FCSRT, are better than neuroimaging ones at detecting differences among subjects. Keywords: Alzheimer’s disease, magnetic resonance imaging, mild cognitive impairment, neuropsychological tests, neuropsychology
INTRODUCTION Neuropsychological assessment is crucial in the diagnosis and characterization of mild cognitive impairment (MCI) and Alzheimer’s disease (AD). Diagnostic guidelines for AD and MCI require proof of lower performance in cognitive tasks that is greater than would be expected for the subject’s age and educational background. In some cases, medical historytaking and bedside mental state examination are sufficient for diagnosis. In non-conclusive cases, formal neuropsychological assessment is needed to document cognitive impairment and quantify its severity [1, 2]. There are two main approaches in neuropsychological assessment: the flexible approach, in which the neuropsychologist selects the tests that best fit individual clinical questions and patients’ complaints; and the fixed one, in which the same array of predefined tests grouped in a neuropsychological battery are always employed. Both approaches are valid and the selection of one or the other depends upon the clinician’s background and the testing objectives. However, the use of a fixed, comprehensive, and standardized neuropsychological test battery presents some considerable advantages, such as the possibility of constructing a cognitive profile that can be compared across subjects and pathologies.
[5], and executive [6, 7] functions are also impaired in MCI subjects who subsequently develop dementia. Indeed, a report from a large, recent, longitudinal study states that single-domain amnestic MCI is rarely diagnosed when subjects undergo a comprehensive neuropsychological assessment [8], corroborating previous studies [9]. As claimed by some authors, the use of extensive batteries seems necessary to adequately describe MCI subjects [10]. Neuroimaging profiles In recent decades, neuroimaging techniques have undergone an increasing range of applications in AD. Imaging has moved from an exclusionary role to a recognized position in diagnosis and research. A variety of imaging modalities, including structural and functional ones, has shown distinctive changes in the brains of patients within the AD spectrum [11]. Overt dementia presents marked atrophy in the medial temporal lobe structures [12, 13] along with progressive thinning of the parietotemporal and frontal cortices [14, 15]. The pattern observed in MCI is similar, but to a lesser degree, to that of AD. Atrophy in medial temporal lobe structures is the most salient finding although volume and cortical thickness loss have been observed in other regions, such as the medial parietal and frontal areas [16, 17].
Cognitive profiles Cognition versus biomarkers A cognitive profile can be simply defined as the pattern of preserved and impaired cognitive functions. It is useful in differential diagnosis, identification of subtypes, and monitoring of disease progression [3]. Interpretation of the cognitive profile can be used in dementia to track the cognitive decline. Accurate analyses of dissociations between preserved/impaired domains can be applied, for example, in the design of strategies aimed at minimizing the impact of disability on daily living. The benefits of the cognitive profile approach may be especially valuable in prognosis at predementia stages. It has been demonstrated that, in addition to memory, attentional [4], visuoperceptual
New diagnostic trends in AD spectrum pathologies include biomarkers, such as cerebrospinal fluid (CSF) and neuroimaging in order to improve specificity [1, 2, 18]. Nonetheless, there is as yet no consensus on which measure (i.e., CSF, MRI, PET, or neuropsychological assessment), or combination of them, is best to discriminate MCI and AD individuals from normal ones. While a number of authors claim that imaging is superior to cognitive testing for AD diagnosis [19], other meta-analytic studies have found the opposite [20]. In agreement with the latter, the conversion from MCI to AD seems better predicted by neuropsycholog-
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
ical performance [21, 22]. Clearly, prediction accuracy increases when models that combine cognition, MRI, and/or CSF are employed [23–25]. Nevertheless, the use of CSF-based biomarkers, although of great interest in research, is presently far from suitable for clinical use. With regard to neuroimaging, MRI, and even PET, while widely available are expensive and do not constitute a core feature of diagnosis. At present, neuropsychological assessment is the most feasible, economical, and effective way of characterizing subjects in the AD pathology spectrum. The present study The present cross-sectional study provides data regarding the neuropsychological and MRI profiles of MCI and AD individuals. It compares the magnitude of the differences with normal aging in order to identify the best measures to differentiate patients from controls. The study is framed within the Spanish Multicenter Normative Studies (NEURONORMA project). In the NEURONORMA project, a group of MCI and AD subjects were recruited with the purpose of studying cognitive profiles and developing diagnostic norms. Our group has presented normative data from a Spanish population to provide valid and appropriate norms for commonly used neuropsychological tests. The use of a co-normed battery allows the direct comparison of test scores and facilitates the clinical interpretation of neuropsychological test profiles. The control sample and methodology have been previously described [26], and the norms have been published in a series of papers [26–31]. The specific objectives of this paper are: 1) to describe the characteristic neuropsychological profiles of MCI and AD subjects by means of the co-normed NEURONORMA battery and its age- and educationadjusted Spanish normative data; 2) to present results of brain atrophy in a subsample of MCI and AD; and 3) to study the magnitude of the differences amongst groups for both neuroimaging and neuropsychological outcomes. MATERIAL AND METHODS Participants A total sample of 535 subjects was studied. Participants included 356 cognitively normal adults (CONT), 79 patients diagnosed with MCI, and 100 patients with AD (70 with mild and 30 with moderate dementia). All subjects were Caucasian. The project was reviewed
889
and approved by the Research Ethics Committee of the Hospital del Mar Research Institute (Spain), and was conducted in accordance with the Declaration of Helsinki and its subsequent amendments, and the European Union regulations concerning medical research. All participants were required to have an informant who could answer questions about their cognition, function, and health. Patients were recruited from dementia consultations at nine hospitals in different regions of Spain between 2004 and 2007. Diagnosis was based on a clinical interview with the patient and the informant, and a physical and neurological examination. A series of previously validated instruments (i.e., cognitive, psychiatric, medical, and functional scales) was administered as part of the diagnostic process and classification of participants (see [26] for details). These scales included the Mini-Mental State Examination (MMSE) in its validated Spanish version [32], the Spanish version of the Interview for Deterioration of Daily living activities in Dementia (IDDD) [33], the Modified Hachinski Ischemia Score, and the Hamilton Depression Rating Scale. Laboratory (blood) tests and brain imaging (CT or MRI) were performed according to local site procedures. A subsample composed of 60 CONT, 25 MCI, and 25 AD from two centers (Hospital del Mar and Hospital Clinic de Barcelona) were included in the neuroimaging substudy. The final sample was made up of 30 CONT, 23 MCI, and 23 AD (19 with mild and 4 with moderate dementia). In addition, in a genetic sub-study (n = 226) the presence of APOE 4 alleles was determined. CSF samples, although of great interest, were not collected in this study for two main reasons. Firstly, the participation of nine hospitals complicated the harmonization of collection and processing of the samples. Secondly, because when this project was planned such studies were not yet implemented in almost any setting. A brief description of the main features of the sample is as follows: 1) CONT definition: Subjects were aged 50 to 85 years. Recruitment procedures and socio-demographic characteristics of the normative sample have been previously reported [26]; 2) MCI diagnostic criteria: Patients were diagnosed as having amnestic MCI according to the following criteria: (a) Presence of subjective memory complaints; (b) Essentially preserved functional abilities (confirmed by an informant); (c) A minimal evolution of 6 months; (d) Evidence of objective memory impairment, defined as a score below percentile 15 in the age and education-matched reference group on a test of verbal delayed recall (Barcelona Test paragraph recall) [34], but without any other significant deficits in the
890
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
Barcelona Test battery; (e) Not conforming to criteria for probable AD according to the National Institute of Neurological and Communication Disorders and Stroke/Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) [35]; and (f) a Global Deterioration Scale (GDS) score of 3; 3) AD diagnostic criteria: Diagnosis of dementia was made following the Diagnostic and Statistical Manual for Mental Disorders, 4th edition revised [36]; Diagnosis of AD was carried out according to the NINCDS/ADRDA criteria [35]. A minimum MMSE score of 14 points were required to be enrolled. Patients were staged as having mild to moderate dementia (GDS 4-5). Neuropsychological methods The neuropsychological protocol included the following tests: Verbal span (Digit Span forward and backward) [34]; Visuospatial Span (Corsi’s Test) from the WAIS-R-NI [37]; Letter-Number Sequencing (WAIS-III) [38]; Trail Making Test [39]; Symbol Digit Modalities Test [40]; Boston Naming Test [41, 42]; Token Test [43]; Selected test of the Visual Object and Space Perception Battery [44]; Judgment of Line Orientation [45]; Rey-Osterrieth Complex Figure [46, 47]; Free and Cued Selective Reminding Test (FCSRT) [48] (Copyright, 1996–2000. Albert Einstein College of Medicine of Yeshiva University. New York); Verbal fluency, including three semantic fluency tasks (animals, fruit and vegetables, and kitchen tools), three formal phonemic tasks (words beginning with p, m, and r), and three excluded letter fluency task (excluded a, e, and s); Stroop Color-Word Interference Test [49]; Tower of London Drexel University version [50]. All tests were administered and scored according to standardized procedures published in each test manual. For more details on administration procedures and specific variables, see [26–31]. In order to avoid an overload of information and redundancies in the cognitive profile graph, some variables such as Letter-Number Sequencing (WAIS-III) and most of the fluency tasks were excluded. Finally, 33 cognitive variables were studied (see Fig. 1 and Table 2 for a list of the included variables). MRI acquisition and analysis Subjects were imaged in a Signa LX 1.5 Tesla magnet (General Electric, Milwaukee, USA) using a specific T1 3D SPGR protocol (TE = Min Full; TR = 12; TI = 450; angle = 15; 256 x 192; FOV = 25; 1NEX; Thickness = 1.5; 128 locs). Voxel size was
0.97 × 1.5 × 0.97 mm). An initial visual assessment was performed to discard any other relevant brain pathology and to ensure appropriate study quality. Standard DICOM images were first anonymized and then processed by means of FreeSurfer v4.0.2 (http://surfer.nmr.mgh.harvard.edu/). One volumetric variable (total gray matter volume) was calculated by means of the mri segstats command from the version 5.0 of this software. In addition to volumetric quantification of predefined ROIs, FreeSurfer’s processing provides white matter and cerebral cortex models to produce representations of cortical thickness. Cortical thickness is calculated as the closest distance from the gray/white boundary to the gray/CSF boundary at each vertex on the tessellated surface [51]. A parcellation of the cerebral cortex into units based on gyral and sulcal structure [52, 53] was also carried out. FreeSurfer has been implemented in the cluster of the Port d’Informaci´o Cient´ıfica, at the Universitat Aut`onoma de Barcelona. It is composed of 170 HP blades with 2 quad-core CPU (Hewlett Packard, USA), each with 16 GB of RAM, running on Scientific Linux 4 (https://www.scientificlinux.org/). Subjects were processed in parallel and launched through a batch system. The PIC-Neuroimaging Center (PICNIC) remote web-based system was used to perform the processing. Finally, 95 MRI-based variables were studied (29 volumetric ROIs and 66 cortical thickness ROIs, see Results for details). In one third of the images, a limited number of manual corrections was performed to ensure measurement quality. Data management and statistical analysis Analyses and graphs were obtained using R statistical software (v2.7). A detailed description of the statistical procedure for obtaining the normative data has been described elsewhere [26]. In brief, the process of converting the subjects’ raw scores to age-adjusted scaled ones includes a transformation to a 2 to 18 range scale based on percentile ranges, and a subsequent education adjustment based on the coefficients obtained from linear regressions. This approach is similar to that employed by the authors of the Mayo’s Older Americans Normative studies [54]. Basic descriptive statistics included count, percentage of total, mean, and standard deviation. ANOVA and pairwise post-hoc comparisons (Tukey) were applied to test mean differences among groups for sociodemographic and eligibility variables. Since patients were older and less educated than control subjects, mean comparisons among groups for neuropsychological
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
891
Fig. 1. Cognitive profile of MCI and AD groups in the NEURONORMA battery. Medians of age- and education-adjusted scores are shown.
892
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
raw scores were carried out by applying ANCOVA (with age and years of education as covariables) and pairwise post-hoc comparisons (Tukey). Chi-square tests were applied for nominal and ordinal variables. Median values of the scaled scores (range 2–18) of each cognitive variable were used to construct graphical demographic age- and education-adjusted cognitive profiles for the MCI and AD samples. Neuroimaging data analysis was twofold. First, ANCOVA analyses were performed for ROI-based outcomes. Age was included as a covariable for cortical thickness variables; age and intracranial volume were included for volumetric ones. Second, a vertex-based analysis was performed by means of the QDEC tool provided in the FreeSurfer package. A Gaussian kernel of FWHM of 20 mm was applied to smooth images before the analysis. This analysis was corrected for multiple comparisons by the application of a False Discovery Rate of 5%. For all the other statistical procedures significance threshold was set to 0.05. The magnitude of the observed differences between both pathological groups and cognitively normal subjects was explored by calculating Cohen’s d. Cohen’s d values from a group of selected MRI-derived variables (i.e., global gray matter volume, global white matter volume, total CSF volume, entorhinal cortex thickness, hippocampal and amygdalar volumes, and cortical volumes from temporal, frontal, parietal, and occipital lobes) were plotted to illustrate the neuroimaging profile of MCI and AD groups. RESULTS Sociodemographic, neuropsychological, and APOE status data Table 1 summarizes demographic information and eligibility criteria of the three groups (CONT, MCI, and AD). Of the 535 participants, 322 were females (60.2%). CONT subjects were younger and more educated than patients (p < 0.001), there were, however, no significant differences in these variables between the MCI and AD groups (age, p = 0.309; years of education, p = 0.859). The IDDD scale, which measures impairment in activities of daily living, revealed a preserved performance of basic activities in MCI subjects (p = 0.939) and mild impairment in AD (p < 0.001). Complex activities were slightly affected in MCI (p < 0.001) and moderately impaired in AD (p < 0.001). While CONT and MCI subjects did not show differences in depression symptoms (p = 0.797), AD patients did (p < 0.05). The APOE study showed an increased
percentage of subjects carrying at least one 4 allele in both MCI and AD samples with respect to CONT (p < 0.05). However, MCI and AD percentages did not significantly differ (p = 0.285). A summary of the raw neuropsychological scores obtained from each group, and results of mean comparisons, are presented in Table 2. Average cognitive profiles of AD and MCI subjects, expressed as the medians of scaled scores adjusted for age and education, are presented in Fig. 1. This figure shows the relative impairment in each neuropsychological measure against the normative group. The most marked impairment was found in memory measures, in which the AD group scores were below the scaled score 2 (equivalent to percentile two) in both verbal and visual tasks. MRI data The demographic and inclusion variables of the MRI subsample were as follows (means and [SD] are given): CONT (n = 30, age = 72.02 [4.68], years of education = 11.37 [6.05], MMSE = 28.93 [1.36], IDDD = 33.30 [0.67]); MCI (n = 23, age = 72.91 [7.38], years of education = 8.26 [4.70], MMSE = 26.32 [2.08], IDDD = 36.51 [2.79]); AD (n = 23, age = 75.82 [6.30], years of education = 6.00 [3.26], MMSE = 20.08 [3.75], IDDD = 44.78 [8.10]). While groups did not significantly differ in age (p = 0.08), they did in education. In the post hoc test, the only comparison that reached significance was between CONT and AD (p < 0.001). Groups showed statistical differences in MMSE and IDDD scores. MRI and cognitive samples differ in number of subjects and the mean age of the control sample which was closer to that of the pathological samples in the neuroimaging subgroup. Nevertheless, age was included as a covariate in MRI analyses. We found significant differences in most of the MRI-based outcomes among the groups. Supplementary Table 1 shows the results of MRI analyses. To illustrate the neuroimaging profile of MCI and AD effect size, results for selected variables are presented graphically in Fig. 2. The results of the cortical thickness vertex-wise analyses can be seen in Supplementary Fig. 1. Effect size results Cohen’s d calculations were used to explore differences in neuroimaging-based measures and calculated for cognitive variables for comparison purposes. Effect size results can be partially seen in Table 3 (comprehensive data of the 128 variables [33 cognitive and 95
G. S´anchez-Benavides et al. / Profiles in MCI and AD: NEURONORMA
893
Table 1 General characteristics of study sample: eligibility and classification criteria Female, n (%) Age (years) Education (years) MMSE MMSE adjusted∗ IDDD Basic activities IDDD Complex activities IDDD Total† Hamilton DRS, M (SD) Ischemia score, M (SD) APOE 4 sub-sample, n At least one 4 allele of APOE, n (%)
CONT
MCI
AD
212 (59.6) 64.9 (9.3) 10.4 (5.4) 28.7 (1.5) 29.1 (1.4) 16 (0.0) 17.1 (0.5) 33.1 (0.6) 2.1 (2.5) 0.1 (0.3) 147 40 (27.2)
45 (57) 72.8 (6.5) 8.0 (4.7) 25.7 (2.2) 26.8 (2.2) 16.1 (0.3) 19.9 (2.4) 36.0 (2.5) 1.9 (2.3) 0.3 (0.5) 35 19 (54.3)
65 (65) 74.7 (7.5) 7.6 (4.6) 20.2 (4.0) 21.3 (3.8) 17.9 (4.0) 30.9 (7.0) 48.9 (10.0) 2.9 (3.3) 0.4 (0.7) 44 27 (61.4)
p value ns
