Professional Documents
Culture Documents
Abstract Background: Communication problems are common in Alzheimer’s disease (AD) patients, but in-
struments to assess these symptoms are limited. Our objective was to create a new scale, based on
the language subscale of the Severe Impairment Battery (SIB), as a sensitive and reliable measurement
of treatment effects on language performance.
Methods: All 24 items of the SIB language subscale were chosen for analysis. Baseline scores of
1320 moderate-to-severe patients (Mini-Mental State Examination [MMSE] score, ,15), from a com-
bined AD database of four Memantine clinical trials (Study Codes: IE-2101, MEM-MD-01, MEM-
MD-02, and MRZ-9605), were used for item reduction according to a standard principal components
factor analysis. All items with loadings .0.5 on the identified factors were selected for inclusion in the
new language scale. Correlations with existing AD scales were examined.
Results: The analysis indicated six factors, with 21 of 24 items showing loadings .0.5. The resulting
21-item SIB Language (SIB-L) scale exhibited high internal consistency (Cronbach’s a 5 0.809). The
maximal SIB-L score was 41 points, with a measurement error of 3.7 points. The stratification of base-
line SIB-L scores (mean, 31.7; SD, 8.4) by MMSE scores (mean, 9.7; SD, 3.3) showed a high variance
in SIB-L scores. This confirms that patients with a low MMSE score can possess preserved language
abilities. The SIB-L scale did not exhibit substantial floor-and-ceiling effects.
Conclusions: The new SIB-L is a fast (,15 minutes) and easily administered scale with favorable
psychometric characteristics for assessing language impairment and treatment effects on the language
performance of patients with moderate to severe AD.
Ó 2009 The Alzheimer’s Association. All rights reserved.
Keywords: Alzheimer’s disease; Clinical trial; Severe Impairment Battery; Language; Scale
substantially enhanced quality of life for both patients and the Mini-Mental State Examination [19], the Functional As-
their relatives/caregivers. sessment Staging (FAST) instrument [20], the Neuropsychi-
Despite the immense importance of language perfor- atric Inventory (NPI) [21], and the 19-item severe AD version
mance, only a few rating scales include evaluations of symp- of the Alzheimer’s Disease Cooperative Study Activities of
toms specifically related to it. This may be because most Daily Living Inventory (ADCS-ADL19) [22]. Only patients
scales were designed to focus on cognitive outcomes in with an MMSE score of ,15, designated as the group of
mild to moderate patients [7,8]. Consequently, widely used moderate-to-severe AD patients, were included in these
scales such as the Mini-Mental State Examination (MMSE) four clinical trials and therefore in the overall database, which
and Alzheimer’s Disease Assessment Scale-cognitive sub- was used for statistical analyses in this study.
scale (ADAS-cog) include only a few items related to lan-
guage and communication, and they are not sensitive 2.2. Scale construction
enough for patients in the later stages of AD. Scales designed
to evaluate language and communication, such as the Refer- A factor analysis, based on the principal-components
ential Communication Task [9] or the Functional Linguistic method using an orthogonal rotational procedure (Varimax),
Communication Inventory [10], are difficult to administer, was performed on baseline scores of the 24-item SIB lan-
and are suitable only for patients in earlier stages of AD. guage subscale. All analyses were performed using SAS soft-
An exception is the Severe Impairment Battery (SIB), ware, version 9.1.3 (SAS, Inc., Cary, NC). Initial factors were
which contains a language subscale with items related to selected on the basis of Eigen values .1. To decide whether
naming, reading, writing, and repetition. The SIB was devel- scale items should be retained or drop out of the new lan-
oped in 1990 to assess cognition specifically in later AD guage scale, a threshold of 0.5 was chosen for the single-
stages, and is now frequently administered in several lan- item loading of each identified factor. Items with a factor
guages [11–14]. Its sensitivity, reliability, and validity were loading of ,0.5 were deleted.
established in several studies. The current version consists
of 51 items in nine ‘‘cognitive’’ domains (social interaction, 2.3. Statistical analyses
memory, orientation, language, attention, praxis, visuospatial
abilities, constructional abilities, and orientation to name), To determine the internal consistency of the resulting
has a maximal score of 100, and requires approximately instrument, Cronbach’s a was calculated from baseline SI-
30 minutes to administer. B language subscale data. The measurement error of the
Recently, a short form of the SIB was developed by means SIB language subscale was calculated using the formula:
of exploratory factor analyses, based on data from two patient
Standard deviation Oð12 Cronbach’s aÞ;
groups [12]. In this shortened SIB, the original nine-domain
structure was maintained and 25 items were dropped, yield- and Pearson correlation coefficients were calculated with
ing a 26-item scale with a maximal score of 50 and reduced baseline scores of the complete SIB, nonlanguage SIB items,
administration time. MMSE, FAST, NPI, and ADCS-ADL19.
The goal of the present study was to establish a language
scale specifically for patients with moderate to severe AD. To
identify items for the new instrument, we constructed a com- 3. Results
bined AD clinical trial database and performed a factor anal- 3.1. Development of the SIB-L scale
ysis on the SIB language subscale, using the respective
baseline scores. A database was constructed with baseline data from 1320
patients from four previously conducted clinical trials. The
characteristics of the safety populations (all patients who
2. Methods took at least one dose of study medication) of each trial
and the combined total sample are provided in Table 1. At
2.1. Overall database
baseline, the mean age was 75.8 (SD, 8.7) years, the mean
A combined database was constructed with data from MMSE score was 9.7 (SD, 3.3), and the mean total SIB score
1320 patients in four randomized, placebo-controlled, was 73.9 (SD, 18.6).
double-blind Memantine clinical trials with study codes: The SIB language subscale consists of seven categories
IE-2101 [15], MEM-MD-01 [16], MEM-MD-02 [17], and (writing, reading, comprehension, verbal fluency, naming,
MRZ-9605 [18]. These studies were performed to assess repetition, and discourse), comprising the 24 items shown
the effects of Memantine versus placebo treatment in patients in Table 2. The internal consistency of this subscale was cal-
with moderate to severe AD. In the study MEM-MD-02, culated using data from the combined database. The resulting
Memantine or placebo was administered to patients on stable value for Cronbach’s a was 0.570, indicating only moderate
donepezil treatment. In all studies, the following clinical out- internal consistency of the SIB language subscale.
come measures were applied at baseline and during the A factor analysis was also performed on the SIB language
course of the study: the Severe Impairment Battery [11], subscale, using the respective baseline scores of the
S. Ferris et al. / Alzheimer’s & Dementia 5 (2009) 375–379 377
Table 1
Patient baseline characteristics in database
IE-2101 MEM-MD-01 MEM-MD-02 MRZ-9605 Total
n 315 350 403 252 1320
Age (years) 73.3 (9.4) 78.2 (7.9) 75.5 (8.6) 76.1 (8.1) 75.8 (8.7)
Female (%) 70.8 71.4 65.0 67.5 68.6
Education (years) 10.0 (2.9) 12.5 (3.1) 11.1 (3.2)
BMI (kg/m2) 22.0 (3.1) 25.1 (4.3) 25.5 (4.5) 24.7 (4.4) 24.4 (4.4)
MMSE 10.1 (3.0) 10.1 (3.0) 10.1 (3.1) 7.9 (3.6) 9.7 (3.3)
SIB 71.0 (17.9) 75.8 (18.6) 78.7 (14.9) 67.1 (21.7) 73.9 (18.6)
NOTE. Data represent means, with standard deviations in parentheses.
Abbreviations: n, number of patients; BMI, body mass index; MMSE, Mini-Mental State Examination; SIB, Severe Impairment Battery.
combined database. Of 24 items on the SIB language sub- Two items from the category ‘‘naming’’ (‘‘forced choice
scale, six factors were identified, for a total of 21 items naming—cup’’ and ‘‘shape identification—square’’) and
with factor loadings R0.5 (Table 2). The resulting factor 1 the sole item of ‘‘discourse’’ (‘‘free discourse’’) were not
comprises the categories ‘‘writing,’’ ‘‘reading,’’ and ‘‘com- included, because their factor loading was ,0.5.
prehension,’’ and includes five items, whereas factor 6 con- Compared with the SIB language subscale, the resulting
sists of two items from the category ‘‘repetition.’’ The new SIB-Language (SIB-L) scale exhibited a much-im-
remaining four factors comprise three or four items from proved Cronbach’s a (0.809), indicating its high internal con-
the category ‘‘naming.’’ One of these, factor 2, also includes sistency. The maximal score of the SIB-L is 41 points, with
the ‘‘verbal fluency’’ item. a measurement error of 3.7 points, calculated on the basis
of the standard deviation and Cronbach’s a as an indicator
of the scale’s reliability [23].
Table 2
Items of SIB language subscale used for SIB-L construction*
3.2. Relationship of SIB-L to MMSE and other instruments
SIB language subscale Factor analysis
Item Category Factor Loading The SIB-L baseline scores were stratified into three de-
mentia severity groups, based on MMSE score: %5, 6–9,
4a: Write name Writing Factor 1 0.767
4b: Copy name 0.771
and 10–14 (Fig. 1). According to the data, patients from all
9a: Reading comprehension Reading 0.651 MMSE groups showed a wide range of SIB-L scores. Inter-
9c: Reading 0.569 estingly, patients with very low MMSE scores (%5) also ex-
9b: Verbal comprehension Comprehension 0.622 hibitws a broad range of SIB-L scores, including scores .30
13: Fluency Verbal fluency Factor 2 0.649 points. Therefore, even patients who are severely demented,
6: Months of year Naming 0.516
8a: Responsive naming—cup 0.674
as based on MMSE score, can exhibit fairly well-preserved
8b: Responsive naming—spoon 0.719 language abilities.
30a: Color naming—red Factor 3 0.645 Moreover, the high variation in SIB-L scores indicates an
30b: Color naming—green 0.640 absence of substantial floor-and-ceiling effects, particularly
34a: Shape identification—circle 0.648 for patients with an MMSE score of ,10. For patients with
34b: Shape identification—triangle 0.560
20: Confrontation naming—spoon Factor 4 0.747
an MMSE score R10, a small ceiling effect might occur:
22: Object naming—spoon 0.793 in this study, only 48 (6.8%) patients of the group with
24: Forced 0.720
choice naming—spoon
15: Confrontation naming—cup Factor 5 0.685
17: Object naming—cup 0.691
26: Color naming—blue 0.592
y
19: Forced (0.482)
choice naming—cup
y
30c: Shape identification—square (0.466)
11a: People Repetition Factor 6 0.580
spend money
11b: Baby 0.805
y
40: Free discourse Discourse (0.410)
Cronbach’s a 0.570 0.809
*The SIB-L scale can be obtained upon request from the corresponding
author.
y
Items with a factor loading of ,0.5 were omitted. Fig. 1. Boxplot of SIB-L scores, stratified by MMSE groups.
378 S. Ferris et al. / Alzheimer’s & Dementia 5 (2009) 375–379
MMSE scores of 10–14 achieved the maximum SIB-L value not necessarily reflect the clinical relevance of this subdo-
of 41 points. main, the inclusion and testing of further items related to lan-
Consistent with the observed high variance of SIB-L guage and communication may be appropriate, especially for
scores within the three MMSE severity groups, the Pearson a more comprehensive assessment of the communicative and
correlation coefficients between SIB-L and MMSE scores social-interactive behavior of AD patients.
were low for the group with MMSE scores %5 (r 5 The value of Cronbach’s a demonstrated that the SIB-L
0.203), and moderate for the other two groups (MMSE has high internal consistency. Although the SIB-L is not ex-
6–9, r 5 0.299; MMSE 10–14, r 5 0.392). pected to exhibit substantially lower reliability than the SIB,
Moreover, the Pearson correlation of SIB-L with SIB was this needs to be confirmed in a study designed to assess psy-
high (r 5 0.943), indicating that the SIB-L scale maintains chometric variables, such as interrater and test-retest reliabil-
the sensitivity of the complete SIB scale. As expected, the ity, and specifically to assess floor-and-ceiling effects.
correlation was lower between the nonlanguage SIB items One notable finding of this study was the broad range of
and SIB-L items (r 5 0.826). A Pearson correlation coeffi- baseline SIB-L scores throughout the assessed population,
cient close to zero was found between the total scores of especially for patients with very low MMSE scores (%5).
SIB-L and NPI (r 5 20.096), confirming that these two This clearly supports the idea of Livingston et al. [8] that pa-
scales measure different symptom domains. In contrast, there tients with severe AD comprise a heterogeneous population
were moderate correlations between SIB-L and level of func- with a varying spectrum of abilities and disabilities. As
tioning, as measured by the FAST (r 5 20.379) and ADCS- shown here, patients with MMSE scores of 5 or even lower
ADL19 (r 5 0.513), respectively. may still retain considerable language abilities.
Moreover, only a few patients with MMSE scores of ,10
4. Discussion exhibited very low or very high SIB-L scores. This indicates
the absence of substantial floor-and-ceiling effects for this
Although AD includes symptoms of impaired communi- population. A small ceiling effect was observed for patients
cation [6], limited information exists about language perfor- with MMSE scores of 10 to 14. More than 90% of these pa-
mance in the later stages of AD, partly because of the lack of tients showed lower-than-maximum SIB-L scores, and thus
appropriate assessment tools. We aimed to fill this gap by language impairment was detectable in these patients. Also,
choosing the language subscale of the SIB as the basis for de- clinically relevant changes in SIB-L score, defined as
veloping a specific AD language tool. The SIB can be admin- a change of 3.7 points, was measurable in half of the patients
istered by using simple commands and gestures, which with an MMSE score R10. Hence, we conclude that SIB-L
makes it suitable even for patients in very late stages of can be applied to all patients with MMSE scores of ,15.
AD. The language domain of SIB comprises tasks mainly re- However, for moderate AD patients (MMSE 10–14 group)
lated to naming various items in everyday life, reading and whose SIB-L scores were .37.3, i.e., for whom clinically rel-
writing, verbal comprehension, and repetition. evant changes in language impairment cannot be measured,
A recently developed short version of the SIB maintains a more demanding language scale should be administered.
the nine-domain structure of the complete SIB, but consists For AD, as for many other diseases, no ‘‘gold standard’’
of only 26 out of 51 SIB items [12]. In a convincing argument exists for determining clinically relevant differences. There-
for the short SIB, the authors cited the reduced administration fore, a change of at least the magnitude of the measurement
time, which accommodates the restricted attention span of error (i.e., 3.7 SIB-L points) might be used to indicate clinical
patients with more severe AD. The SIB-L scale presented relevance [23]. If language performance over time is to be
here shares this advantage of relatively short administration evaluated, no improvement can be measured for patients
time: the scale consists of 21 out of 51 SIB items, correspond- with maximum or near-maximum SIB-L scores (and a poten-
ing to a reduction in administration time from approximately tial ceiling effect may mask worsening). Therefore, only pa-
30 minutes to a maximum of 15 minutes. Given its applica- tients with a measurable language deficit should be included
bility in more severe stages of AD and its derivation from in studies to evaluate treatment effects on language. Because
the SIB, which is a widely used tool in the AD research com- the measurement error of SIB-L was calculated to be 3.7
munity, the SIB-L scale is likely to be acceptable as an instru- points, evaluation over time is desirable only for patients
ment to focus on the language abilities of patients with with a baseline SIB-L score %37.3 (maximal score, 41
moderate to severe AD. points; measurement error, 3.7 points).
We performed a principal-components factor analysis on In conclusion, the removal of three items from the SIB lan-
the SIB language subscale, leading to the deletion of three guage subscale yielded the SIB-L scale, an efficient and reli-
items. The resulting 21-item SIB-L scale consists of 21 items able tool for language assessment in moderate-to-severe AD
distributed over six factors, with a maximum score of 41 patients. This patient population displays high variance in
points. The only SIB item assessing the interactive compo- language performance. The SIB-L scale seems ideal for the
nent of language (‘‘discourse’’) was not retained after our assessment of basic language abilities and the evaluation of
analysis, because its factor loading was ,0.5. Because this treatment effects on language abilities in patients with mod-
decision was based on a mathematical algorithm and did erate to severe AD.
S. Ferris et al. / Alzheimer’s & Dementia 5 (2009) 375–379 379