Author’s personal copy (e-offprint) Zeitschrift fr Neuropsychologie, 25 (1), 2014, 49 – 63

Original Article

Prevalence of Mild Cognitive Impairment Subtypes in Patients with Parkinson’s Disease – Comparison of two Modes of Classification Johann Lehrner, Heidemarie Zach, Doris Moser, Andreas Gleiß, Eduard Auff, Walter Pirker, and Gisela Pusswald Department of Neurology, Medical University of Vienna

Abstract: Early detection of dementia in Parkinsons disease is becoming increasingly important. The goal of this study was to establish prevalence of mild cognitive impairment subtypes in Parkinsons disease using two different modes of mild cognitive impairment classification. Categorizing patients into mild cognitive impairment subtypes according to the minimum mode of mild cognitive impairment classification revealed the following results: three patients (2.5 %) were categorized as cognitively healthy, whereas 117 patients (97.5 %) met the criteria for mild cognitive impairment. When categorizing patients according to the mean mode of mild cognitive impairment classification, 41.7 % of the patients were categorized as cognitively healthy, whereas 58.3 % met the criteria for mild cognitive impairment. Frequency of mild cognitive impairment varies substantially, depending on how impairment is defined. Keywords: Parkinsons disease, cognition, mild cognitive impairment subtypes, neuropsychological testing

Prvalenz verschiedener Subtypen der leichten kognitiven Stçrung bei Morbus Parkinson – Vergleich zweier Klassifizierungsmethoden Zusammenfassung: Das frhzeitige Erkennen einer Demenz im Rahmen der Parkinsonkrankheit wird immer wichtiger. Das Ziel der Studie war die Erfassung der Prvalenz der leichten kognitiven Stçrung mittels zweier Klassifizierungsmethoden. Wurden die Patienten anhand der Minimum Mode Methode in Subtypen klassifiziert kam es zu folgenden Ergebnissen: Drei Patienten (2.5 %) wurden als kognitiv gesund kategorisiert, 117 Patienten (97.5 %) erfllten die Kriterien fr eine leichte kognitive Stçrung. Wurden die Patienten anhand der Mean Mode Methode in Subtypen klassifiziert kam es zu folgenden Ergebnissen: 41.7 % der Patienten wurden als kognitiv gesund kategorisiert whrend 58.3 % der Patienten erfllten die Kriterien fr eine leichte kognitiven Stçrung Die Hufigkeit der Subtypen leichten kognitiven Stçrung bei der Parkinsonkrankheit variiert betrchtlich in Abhngigkeit von der Klassifizierungsmethode. Schlsselwçrter: Parkinsonkrankheit, Subtypen der leichten kognitive Stçrung, neuropsychologische Testung

Introduction During the past decade, the concept of mild cognitive impairment (MCI), an early transition stage from normal cognition to dementia, has gained wide acceptance in its association with Parkinsons disease (PD; Fernandez, Crucian, Okun, Price & Bowers, 2005). Mild cognitive impairment in Parkinsons disease (PD-MCI) is associated with an increased risk for dementia (Kehagia, Barker & DOI 10.1024/1016-264X/a000116

Robbins, 2010). The prevalence rate of PD dementia (PDD) ranges from 10 – 40 %, depending on the sample population and criteria used (Aarsland, Andersen, Larsen, Lolk & Krogh-Sorensen, 2003). Longitudinal studies show that the risk of developing dementia is four to six times higher in PD than in subjects without PD (Aarsland et al., 2001; Hobson & Meara, 2004).The cumulative prevalence of PD dementia in patients with long-term survival is at least 75 % (Aarsland et al. , 2003; Hely, Reid, Adena,

Z. Neuropsychol. 25 (1)  2014 Verlag Hans Huber, Hogrefe AG, Bern

Author’s personal copy (e-offprint) 50

J. Lehrner et al.: MCI and Parkinsons Disease (PD)

Halliday & Morris, 2008).The reason for the growing interest in the concept of PD-MCI is based on the idea that identifying patients with an increased risk for dementia will enable early treatment. Characterization of PD-MCI and its subtypes in patients with PD has become an important research goal (Caviness et al., 2007; Janvin, Larsen, Aarsland & Hugdahl, 2006). Different MCI subtypes are defined primarily through neuropsychological criteria. Since there is no standard protocol for cognitive assessment in PD, recent studies have operationally defined PD-MCI criteria in different ways (Dalrymple-Alford et al., 2011; Litvan et al., 2011). The determination of abnormality on a neuropsychological test battery depends on a number of parameters, including the nature of the psychometric measures used, the number of tests in the cognitive domain being investigated, the reliability of individual measures as compared with composite measures, the quality of the normative data, and the statistical threshold used to indicate impairment (Dalrymple-Alford et al., 2011; Litvan et al., 2011). Currently, different procedures with varying numbers of tests for assessing cognitive functions are used. Some authors evaluated screening instruments for the detection of PD-MCI (Hoops et al., 2009), while others used neuropsychological assessment evaluating different cognitive domains. For example, Liepelt-Scarfone et al. (2011) used a test battery of 19 tests for 6 cognitive domains known to be affected in PD-MCI. Their goal was to compare different criteria of PD-MCI classification: they defined standard (z) scores for -1, -1.5 and -2 as cut off values for the definition of PD-MCI and differentiated whether a subject scored low in at least 1 or 2 tests per cognitive domain. They found that varying cut off values for the definition of PD-MCI affected the frequency of PD-MCI subjects, ranging from 9.9 % to 92.1 % (Liepelt-Scarfone et al., 2011). In another study, Dalrymple-Alford et al. (2011) compared the influence of different PD-MCI criteria on PD-MCI frequency. A prevalence of 14 % was found when using a cut-off of at least 2 standard deviations (SD) below normative scores in 2 tests in 1 cognitive domain with none of the controls scoring in the MCI range. The frequency of PD-MCI went up to 89 % when the cut-off was set at 1 SD below normative scores in one test in one cognitive domain. However, when using this definition, 70 % of the control subjects were categorized as having MCI (DalrympleAlford et al., 2011). Varying outcomes regarding the frequency of MCI have also been encountered in studies assessing MCI prevalence in non-parkinsonian patients with reported cognitive problems and attending a memory clinic (Hughes, Snitz & Ganguli, 2011). For instance, our group recently reported MCI prevalence in an outpatient sample of 673 independently living patients who reported having cognitive problems. Using the presence of impairment on a mean composite score of a certain domain (mean mode of MCI classification) versus a single cognitive measure (minimum mode of MCI classification), prevalence varied from 39.5 % to 89.4 %, respectively (Pusswald et al., 2013).

To further explore suitable methods for defining PDMCI, the goals of the present study were to establish a well characterized, cognitively intact control group and to group tests into domains (domain structure) using empirically validated methods. By examining the distribution of impaired patients across neuropsychological measures using norms of this control group, we investigate two different approaches of PD-MCI classification for PD-MCI subtyping, namely minimum mode of PD-MCI classification and mean mode of PD-MCI classification. To the best of our knowledge, these two modes of classification have not been directly compared yet in PD studies. We expected that using the minimum mode of PD-MCI classification would lead to a higher prevalence rate than using the mean mode of PD-MCI classification. We further compared the two modes of PD-MCI classification regarding PD-MCI subtype prevalence in our cohort and the relationship to age and sex. We also compared whether PD-MCI subtypes emerging from these analyses differ regarding cognitive performance across single tests.

Patients and Methods Subjects and procedure The current data are part of a larger research project, the Vienna Mild Cognitive Impairment and Cognitive Decline in Parkinsons Disease Study (VMCI-CD-PD Study). The VMCI-CD-PD Study is a prospective cohort study including consecutive, community-dwelling PD patients who attend the movement disorder clinic for assessment of their parkinsonism. The primary goal of the study is to determine the prevalence of four clinical PD-MCI subtypes. The secondary goal is to evaluate conversion rates from PDMCI to PDD and to investigate whether the risk for conversion to PDD varies according to PD-MCI subtype. For the purpose of this paper, we only present neuropsychological data regarding PD-MCI subtypes at the baseline. The study protocol was in accordance with the Helsinki Declaration and approved by the Ethical Committee of the Medical University of Vienna. All PD patients underwent a clinical examination and neuropsychological testing. The clinical assessment encompassed a complete medical history, a detailed history of PD, which was obtained using a standardized interview, and a complete neurological examination including the motor section of the Unified Parkinsons Disease Rating Scale (UPDRS-III; Fahn & Elton 1987) and the modified Hoehn and Yahr scale (Fahn & Elton, 1987). Clinical examination and neuropsychological testing were performed during the “on” state. Standardized questionnaires were used to obtain information from the patients relatives or caregivers. Patients who had never undergone computed tomography (CT) or magnetic resonance imaging (MRI) during the course of PD and patients showing clinical features incompatible with previous imaging results were referred to structural imaging. Both neuroimaging and

Z. Neuropsychol. 25 (1)  2014 Verlag Hans Huber, Hogrefe AG, Bern

Author’s personal copy (e-offprint) J. Lehrner et al.: MCI and Parkinsons Disease (PD)

clinical features were used to determine significant cerebrovascular disease or other co-morbid conditions with a potential impact on cognitive outcomes. Inclusion and exclusion criteria were similar to those used in other studies. All PD patients had to fulfill UK Parkinsons Disease Society Brain Bank criteria (Gibb & Lees, 1988) for probable PD. Patients were excluded from the study if any of the following conditions applied: (a) evidence of having had a stroke as determined by neuroradiologic and clinical examination, (b) history of severe head injury, (c) current psychiatric diagnosis according to ICD-10 (Dilling, Mombour & Schmidt, 2000) with the exception of patients with (sub)depressive symptoms, (d) any medical condition that can lead to cognitive deterioration including renal, respiratory, cardiac and hepatic disease, or (e) a diagnosis of dementia according to DSM IV (Saß, Wittchen, Zaudig & Houben, 2003). Patients were assessed on their regular medication and were required to have a Mini-mental State Examination (MMSE) score of  23.

Parkinson’s disease Patients One-hundred twenty consecutive patients fulfilled the inclusion criteria and were included in the study. Patients were either referred by physicians or were self-referrals. The median age of PD patients was 69 (range 52 – 86) years. Altogether, 62.5 % of the patients were male and 37.5 % of the patients were female. The mean years of formal education were 10.6  3.4. The median MMSE performance of patients was 28 (range 23 – 30). The median PD onset was 62 (range 39 – 77) years. The median disease duration was 7 (range 0.25 – 19) years. The mean UPDRSIII Motor score was 28.4  13.3.

Cognitively healthy control subjects Great care was taken to enroll a sufficient number of cognitively healthy control subjects living independently at home. Control subjects were recruited by means of advertisements. They underwent a rigorous screening evaluation using standardized clinical interview and cognitive screening. Imaging procedures, neurological examination, standard laboratory blood tests and informant reports were not included in the evaluation. Controls were assessed as being in good health. Criteria for healthy function were identified as being similar to those in the Mayo research studies (Greenaway, Smith, Tangalos, Geda & Ivnik, 2009): (a) no active neurological or psychiatric disease, (b) no psychotropic medications, and (c) the subjects may have medical disorders but neither they nor their treatment compromises cognitive function. Cognitive status was given special attention and cognitively healthy control subjects were screened for intact cognition. They were required to have an MMSE score greater than or equal to 27 and an MOCA score (adjusted for education) greater than or

51

equal to 26. Control subjects did not report any cognitive problems. Adequate normative data using cognitively healthy subjects (N = 250) for the neuropsychological measures were thus available. This normative sample will be used to estimate the dependence of neuropsychological test outcomes on age, education and sex as a basis of z-score computation (see below; Pusswald, et al., 2013).

Neuropsychological Measures All participants were subjected to the Neuropsychological Test Battery Vienna (NTBV) (Lehrner, Maly, Gleiß, Auff & Dal-Bianco, 2007). The NTBV includes tests for attention, executive functioning, language and memory according to suggestions for the neuropsychological evaluation of patients with PD using a balanced mixture of “speed” and “power” tests (Ringendahl et al., 2000; Litvan et al., 2012). The Alters-Konzentrations-Test (AKT; Gatterer, 2008), a geriatric cancellation test, the digit symbol subtest of the German WAIS-R (Tewes, 1994), the symbol counting task from the cerebral insufficiency test (C.I.) (Lehrl & Fischer, 1997), the Trail Making Test B (Reitan, 1979), and the score difference of the Trail Making Tests A and B (Reitan, 1979) were applied to assess attention. Executive functions were investigated using the Trail Making Test A (Reitan, 1979), the Five-Point Test (Regard, Strauss & Knapp, 1982), the Maze Test from the NAI Test Battery (Oswald & Fleischmann, 1997), the Stroop Test from the NAI Test Battery (Oswald & Fleischmann, 1997), and the interference test from the C.I. (Lehrl & Fischer, 1997). Phonemic verbal fluency was assessed by instructing participants to name as many words as possible beginning with the letters b, f, and l (one minute time was allowed for each letter condition). In order to test language functions, we used verbal fluency tasks and a confrontation naming task (Goodglass & Kaplan, 1983). Naming as many animals, supermarket items and tools that came to mind within one minute for each task was used to assess semantic verbal fluency. The modified Boston Naming Test (mBNT) (Morris et al., 1989) was used for assessing naming capabilities. Episodic memory was tested using the Verbal Selective Reminding Test (VSRT; Lehrner, Gleiß, Maly, Auff & Dal-Bianco, 2006) with the subtests of immediate recall, total recall, delayed recall and recognition. The ability of the NTBV subtests to detect dementia has been established in a previous study, and we recently published results for sensitivity, specificity, positive predicted value, and negative predicted value, and found very good discrimination power for the NTBV in detecting dementia. Specifically, comparing the predictive accuracy using results from receiver operating characteristic curve analyses (area under the curve), we found very good discriminative power for single tests with an area under the curve ranging from 0.79 for the modified Boston Naming Test to 0.99 for the Verbal Selective Reminding Test–delayed recall for patients with dementia versus cognitively healthy control subjects (Lehrner et al., 2007).

Z. Neuropsychol. 25 (1)  2014 Verlag Hans Huber, Hogrefe AG, Bern

Author’s personal copy (e-offprint) 52

J. Lehrner et al.: MCI and Parkinsons Disease (PD)

Cognitive testing lasted approximately 45 minutes for each patient. Testing was performed within one test session. Cognitive function tests were selected to assess a broad range of cognitive abilities commonly affected by PDD and other forms of dementia.

PD-MCI classification procedure In order to characterize the neuropsychological profile of PD-MCI patients, a z-score was calculated for each variable, indicating in SD units the degree of impairment relatively to cognitively healthy subjects, thereby allowing direct comparison across different cognitive tests. Since age, education and sex effects on cognitive variables have been reported in the literature (Pusswald, et al., 2013) zscores were estimated for each neuropsychological variable as depending on these demographic variables based on the sample of 250 cognitively healthy controls. For this purpose the flexible GAMLSS (Generalized Additive Models for Location, Scale and Shape) model class was used (Stasinopoulus & Rigby, 2007). Such a model comprises two ingredients: first, a distribution function (e. g., normal, log-normal, Box-Cox power exponential) is selected which, for each given set of values for the independent variables (age, years of formal schooling, sex), is able to describe the distribution of the observed values. Second, the dependence of the parameters of the selected distribution (corresponding to measures of location, scale, skewness and kurtosis) on the independent variables is described using appropriate functions (linear, polynomial, cubic splines). Adequate distribution functions, corresponding link functions as well as adequate functions for the parameters dependence on age and years of formal schooling were determined for each sex and each neuropsychological variable individually based on statistical information criteria (Akaike Information Criterion, Schwarz-Bayes Criterion; Pusswald et al., 2013; Stasinopoulus & Rigby, 2007). An Excel-based program to compute z-scores is available from the authors of the present paper. The domain structure of the NTBV was investigated empirically by means of cluster analysis based on the cognitively healthy subjects results. The clustering into domains reflects the correlation structure of the z-scores as implemented in “proc varclus” of the software package SAS, version 9.2. The variable clustering procedure revealed a 6-domain solution defining 6 domains for the 30 cognitive variables. The naming of the domains is not meant to be absolute and is oriented toward the most strongly correlated variables. See Table 1 for the 6 domains with cognitive variables and corresponding R-square with its own domain and R-square with the next closest domain, respectively (Pusswald et al., 2013). Neurological examination and standard laboratory blood tests were performed approximately two weeks prior to neuropsychological testing. The cognitive status of PD-MCI subtypes was determined according to the Peterson criteria, and the cut off score used was 1.5. SD below

age and education corrected norms using the cognitively healthy controls (Pusswald et al., 2013).

Modes of PD-MCI classification for subtyping PD-MCI: Minimum mode and mean mode For the minimum mode of PD-MCI classification, PD-MCI patients were divided into five groups of patients based on cognitive features as follows: cognitively healthy patients (z-scores of each single test were greater than -1.5 SD), amnestic PD-MCI single domain patients (the z-score of at least one memory test was below -1.5 SD, all other z-scores were greater than -1.5 SD), amnestic PD-MCI multiple domain (the z-score of at least one memory test was below -1.5 SD, and at least one other z-score of the remaining tests was below -1.5 SD), non-amnestic PD-MCI single domain patients (there is exactly one domain other than memory in which the z-score of at least one test was below -1.5 SD), and non-amnestic PD-MCI multiple domain patients (at least two tests from different domains other than memory lead to z-scores below -1.5 SD). Thus, each domain was basically assessed according to the minimum of the z-scores of all constituent tests (Pusswald et al., 2013). For the mean mode of PD-MCI classification, PD-MCI patients were divided into five groups of patients based on cognitive features as follows: cognitively healthy patients (mean z-scores of each domain were greater than -1.5 SD), amnestic PD-MCI single domain patients (mean z-score of the memory domain was below -1.5 SD, all other mean zscores of the remaining domains were greater than -1.5 SD), amnestic PD-MCI multiple domain (the mean z-score of the memory domain was below -1.5 SD and at least one of the mean z-scores of the remaining domains was below -1.5 SD), non-amnestic PD-MCI single domain patients (there is exactly one domain other than memory in which the mean of the z-scores was below -1.5 SD), and non-amnestic PD-MCI multiple domain patients (mean z-scores of at least two domains other than memory were below -1.5 SD). Thus, each domain was assessed according to the mean over the z-scores of all constituent tests (Pusswald et al., 2013).

Statistical methods Years of formal education and UPDRS-III Motor score are described by means and standard deviations. Age, age at PD onset, disease duration and MMSE scores are presented as median and range due to the skewed distribution of these variables. Z-scores of the neuropsychological test variables are described by means and standard deviations. In order to compare z-scores of neuropsychological variables between subtypes, one-way ANOVAs were computed. Uncorrected p-values are given and significance according to the method of Bonferroni-Holm for multi-

Z. Neuropsychol. 25 (1)  2014 Verlag Hans Huber, Hogrefe AG, Bern

Author’s personal copy (e-offprint) J. Lehrner et al.: MCI and Parkinsons Disease (PD)

53

Table 1 A variable clustering procedure based on 250 cognitively healthy subjects resulted in a 6 domain solution Domain / Neuropsychological variable

R-squared with own domain

R-squared with next closest

AKT time

0.70

0.14

AKT total/time

0.69

0.14

Trail Making Test – TMTB

0.58

0.18

Digit-Symbol -Test (WAIS-R)

0.44

0.24

TMTB – TMTA difference

0.43

0.07

Symbols counting (C.I.)

0.38

0.19

Phonemic Verbal Fluency PWT total words

0.98

0.10

Phonemic Verbal Fluency PWT l-words

0.72

0.08

Phonemic Verbal Fluency PWT f-words

0.66

0.07

Phonemic Verbal Fluency PWT b-words

0.64

0.09

Stroop color words

0.87

0.17

Stroop total/time

0.86

0.16

Interference (C.I.) time

0.68

0.30

Interference (C.I.) total/time

0.67

0.26

Stroop color words – colors

0.55

0.05

Stroop colors

0.55

0.24

0.66

0.22

Domain 1 Attention

Domain 2 Executive function – phonemic verbal fluency

Domain 3 Executive function – interference

Domain 4 Language Semantic verbal fluency SWT total words Semantic verbal fluency SWT supermarket items

0.65

0.08

Semantic verbal fluency SWT animals

0.65

0.09

Semantic verbal fluency SWT tools

0.63

0.10

Boston Naming Test (mBNT)

0.05

0.02

Verbal memory total recall (VSRT)

0.78

0.08

Verbal memory immediate recall (VSRT)

0.72

0.03

Verbal memory delayed recall (VSRT)

0.63

0.07

Verbal memory recognition (VSRT)

0.21

0.02

Planning Maze Test – NAI time

0.82

0.13

Planning Maze Test – NAI total/time

0.77

0.11

Nonverbal Fluency Five Point Test – total correct

0.45

0.14

Trail Making Test – TMTA

0.43

0.19

Nonverbal Fluency Five Point Test – perseverations

0.38

0.10

Domain 5 Memory

Domain 6 Executive Function – planning and nonverbal fluency

Abbreviations: AKT, Alters-Konzentrations-Test; WAIS-R, Wechsler Adult Intelligence Scale – Revised; TMTA, Trail Making Test Version A; TMTB, Trail Making Test Version B; PWT, Phonematische Wortflssigkeit; NAI, Nrnberger Alters Inventar; C.I., Cerebral Insufficiency Test; SWT, Semantische Wortflssigkeit; VSRT, Verbal Selective Reminding Test; mBNT, modified Boston Naming Test.

plicity correction (for 30 neuropsychological variables) is indicated. Post-hoc pairwise comparisons were adjusted using Tukeys method.

In order to test for a potential influence of age, sex and years of formal schooling on the proportion of patients classified as PD-MCI, a multi-variable logistic regression

Z. Neuropsychol. 25 (1)  2014 Verlag Hans Huber, Hogrefe AG, Bern

Author’s personal copy (e-offprint) 54

J. Lehrner et al.: MCI and Parkinsons Disease (PD)

model was used (with age and schooling treated as scale variables). The reported p-values are the results of two-sided tests. P-values  0.05 were considered to be statistically significant. All computations were performed using SAS software Version 9.2 (SAS Institute Inc., Cary, NC, USA, 2002 – 2008), except GAMLSS estimation, which was done using R 2.11.1 (Development Core Team, Vienna, Austria, 2010).

Results Z-score values of the cognitive variables are presented as mean and standard deviation in Table 2 for the total PDMCI patient sample. For each cognitive variable, the frequency of patients showing impaired performance (defined as performing at least 1.5 SD below the medians of healthy controls matched for age, sex and schooling) are also reported in Table 2. Categorizing PD-MCI patients into PD-MCI subtypes according to the minimum mode of PD-MCI classification revealed the following results. Three patients (2.5 %) were categorized as cognitively healthy, whereas 117 patients (97.5 %) met the criteria for PD-MCI. PD-MCI patients were subtyped as amnestic PD-MCI single domain (1 patient; 0.8 %), amnestic PD-MCI multiple domain (60 patients; 50.0 %), non-amnestic PD-MCI single domain (16 patients; 13.3 %), and non-amnestic PD-MCI multiple domain (40 patients; 33.2 %). Figure 1 A displays corresponding results for mean domain z–scores across PD-MCI subtypes. Categorizing PD-MCI patients into PD-MCI subtypes according to the mean mode of PD-MCI classification revealed the following results. Fifty patients (41.7 %) were categorized as cognitively healthy, whereas 70 patients (58.3 %) met the criteria for PD-MCI. PD-MCI patients were subtyped as amnestic PD-MCI single domain (3 patients; 2.5 %), amnestic PD-MCI multiple domain (19 patients; 15.8 %), non-amnestic PD-MCI single domain (18 patients; 15.0 %), and non-amnestic PD-MCI multiple domain (30 patients, 25.0 %). Figure 1B displays corresponding results for mean domain z–scores across PD-MCI subtypes. For the minimum mode of PD-MCI classification, statistical analyses for z-scores of all neuropsychological variables revealed significant subtype differences after Bonferroni-Holm correction (all corrected p-values