Participants in this study included 93 patients with AD and 43 patients with amnestic MCI (aMCI) who were recruited from the Memory Disorder Clinic at the Samsung Medical Center in Seoul, Korea from November 2005 to January 2007. All patients with AD met the criteria for probable AD proposed by the National Institute of Neurological and Communicative Diseases and Stroke and Alzheimer's disease and Related Disorders Association (NINCDS-ADRDA) (3). The aMCI patients were diagnosed according to the criteria proposed by Peterson et al. (4): 1) subjective memory complaint as described by the patient and/or caregiver, 2) normal general cognitive function, as defined by a score of 24 or greater on the Korean version of Mini-Mental State Examination (MMSE), 3) ability to participate in normal activities of daily living (ADL), judged clinically and by an ADL scale, 4) objective memory decline below the 16th percentile on neuropsychological tests, and 5) non-conformance to clinical criteria for diagnosis of dementia.

All patients underwent a comprehensive evaluation consisting of a detailed medical history, neurological examinations, and a neuropsychological evaluation (SNSB). Additionally, laboratory tests were used to confirm that there were no secondary causes for dementia or cognitive impairment. Magnetic Resonance Imaging (MRI) was performed on all patients, and all patients with structural brain lesions or severe white matter ischemia (caps or band >10 mm and deep white lesion >25 mm) were excluded. Patients who were illiterate were also excluded, regardless of formal education status.

The normal control (NC) group consisted of 77 healthy spouses or caregivers of patients from the memory disorder clinic. All controls were screened for neurological and psychiatric illnesses, and those who were identified to have any of these illnesses were excluded from the study. All subjects in the NC group met the criteria for healthy controls proposed by Christensen et al. (5) and did not have dementia as defined by the score below 8 points on the Korean Dementia Screening Questionnaire (KDSQ) (6) as well as by an ADL score less than 8 on Seoul Instrumental Activities of Daily Living (S-IADL) (7).

The three groups included in this study did not differ significantly in age, education level, and gender. In terms of Clinical Dementia Rating (CDR) scores (8), all participants in the NC group had a CDR score of 0; all patients in the MCI group had a CDR score of 0.5; and of the participants in the AD group 35 had a CDR score of 0.5 (mild stage, 38%), 42 had a CDR score of 1 (mild to moderate stage, 45%), and 16 had a CDR score of 2 (moderate stage, 17%). Among patients with AD, the CDR groups did not differ in age, education level, and gender. We obtained informed consents from all the patients and controls, and this study was approved by the Institutional Review Board of Samsung Medical Center (2005-02-008).

Like the original SNSB, the SNSB-D consists of sub-domains assessing attention, language and related function, visuospatial function, memory, and frontal/executive function (Table 1). We devised a GCF score drawn from the results of the SNSB-D that includes a maximum of 300 points, which is the sum of each subdomain with 17/300 (6%) from attention, 27/300 (9%) from language and related function, 36/300 (12%) from visuospatial function, 150/300 (50%) from memory, and 70/300 (23%) from frontal/executive function.

Table 1 summarizes the differences between the SNSB-D and the original SNSB, and describes how the SNSB-D was modified from the original SNSB. The attention domain score was derived from the sum of the raw scores for forward and backward digit span. Of the many items that assess language and related functions, only naming and calculation tests were included. Specifically, a shortened version (15 item version, Form A) of the Korean version of the Boston Naming Test (K-BNT) and 12 trials of written calculations (three trials each for addition, subtraction, multiplication, and division) were included. The copying test from the Rey Complex Figure Test (RCFT) was used to assess visuospatial function. Several domains of memory function were assessed, including orientation, verbal memory, and visual memory. Orientation was assessed using four measures assessing orientation to time (year, month, date, day of the week) and two items assessing orientation to place ("What is this place for?" and "What's the name of this place?"). The verbal memory score was derived from the sum score of three recall trials, a delayed recall task, and a recognition task from the Seoul Verbal Learning Test (SVLT). The visual memory score was composed of the sum of the scores from the immediate recall, delayed recall, and recognition scores of the RCFT. The frontal/executive function subdomain was assessed using motor impersistence, contrasting program, go-no-go test, fist-edge-palm task, and the Luria loop task, each of which were rated on a 0 to 3 scale. Additionally, a word fluency and Stroop color-reading test were administered. For the word fluency test, a category word generation task (animal) and a phonemic word generation task ('ㄱ'/g/) were given, with a maximum score of 20 for the category word generation task and 15 for the phonemic word generation task. Thus, if the number of appropriate words generated was greater than 20 or 15, respectively, the measure was scored as 20 and 15 points. Stroop test score was converted to a maximum score of 20 by dividing the number of correct responses of color reading test by 5 and dropping the digits after the decimal points. Because of the shortening of several tests from the original SNSB as describ. ed in the above, it takes about 40-50 min to administer the SNSB-D compared with 60-80 min to complete the original SNSB.

To assess test-retest reliability, the SNSB-D was readministered one to three months (mean=51 days) following the initial test in 10 patients with AD, three patients with MCI, and eight participants from the NC group.

Data were analyzed using SPSS 13.0. ANOVA was used to compare age and education levels across groups, and a chi-square test was used to compare gender across groups. To verify the discriminant validity of the SNSB-D, GCF scores were compared among all groups. To examine discriminant validity based on severity of dementia, all subjects were classified into four groups (CDR 0, 0.5, 1, 2), and GCF scores were compared among these four groups. To examine the convergent validity of the SNSB-D, Pearson correlation coefficients were calculated for the SNSB-D and the MMSE. The reliability of the SNSB-D was evaluated by assessing internal consistency and test-retest reliability. Cronbach's alpha was used to test for internal consistency, and Pearson correlation coefficients were used to test for test-retest reliability among each of the groups.

To evaluate the diagnostic utility of the SNSB-D, we examined the sensitivity and specificity of SNSB-D, using a receiver operating characteristics (ROC) curve. Discrimination ability of GCF score derived from the SNSB-D was compared with the results of MMSE using area under the curve (AUC) analyses.

GCF scores differed significantly among the groups (NC, MCI and AD; P<0.001; Table 2). Tukey's post hoc analyses were used to examine differences in GCF scores between each of the groups. GCF scores were significantly lower for the AD group than for the MCI group (P<0.001), and GCF scores were significantly lower for the MCI group than for the NC group (P<0.001). In contrast, a post hoc analysis of MMSE scores revealed that there were no statistically significant differences between the NC and MCI groups (P=0.105), and that the only significant difference in MMSE scores was between the MCI and AD groups (P<0.001).

With respect to each of the cognitive domain scores, the three groups differed in each of the domains assessed: attention, language and related function, visuospatial function, memory, and frontal/executive function (all P<0.001). Tukey's post hoc analyses of the domain scores revealed that there were significant differences between the NC and MCI groups, as well as between the MCI and AD groups in language and related function, memory, and frontal/executive function. However, in attention and visuospatial function, differences were only noted between the MCI and AD groups, but not between the NC and MCI groups.

The same analyses were performed to examine each of the sub-test scores individually. Overall group differences were present in all tests at the significance level of P<0.001, with the exception of the motor impersistence test (P=0.072) from the frontal/executive function sub-domain. Post hoc analyses revealed that there were significant differences between the NC and MCI groups, as well as between the MCI and AD groups on the KBNT; the immediate recall, delayed recall, and the recognition of the SVLT; the immediate and delayed recall, as well as the recognition of the RCFT; and the category word generation test. However, in the remaining sub-tests, including forward digit span, backward digit span, the calculation task, the RCFT copy and orientation task, contrasting program, go-no-go test, fistedge-palm test, the Luria loop task, and the word fluency test, there were statistically significant differences between the MCI and AD groups, but not between the NC and MCI groups.

Participants were also classified according to the CDR score. As the CDR stage increased, the GCF score decreased: CDR 0 (211.0±24.3), 0.5 (146.1±36.7), 1 (93.6±27.1), 2 (55.2±26.6), and GCF scores differed significantly between the four CDR groups. There were also significant differences between each of the consecutive pairs of CDR groups: CDR 0 vs. 0.5, CDR 0.5 vs. 1, and CDR 1 vs. 2 (P<0.001).

To further examine the utility of the SNSB-D to discriminate according to dementia severity rating, participants were reclassified into five groups according to CDR scores. Specifically, the NC group was assigned a CDR score of 0 (NC-0), MCI patients were assigned a CDR score of 0.5 (MCI-0.5), and AD patients were assigned a CDR score of 0.5 (AD-0.5), 1 (AD-1), or 2 (AD-2), depending on the patient's severity of dementia. According to post hoc analyses, the five CDR groups differed significantly with respect to GCF scores. Specifically, the GCF score decreas. ed significantly as the CDR increased (Table 2). We were particularly interested in comparing the MCI-0.5 and AD-0.5 groups. Post hoc analyses revealed that the two groups differed significantly across all cognitive domains except attention.

In examining convergent validity, it was determined that SNSB-D GCF scores and MMSE scores were highly correlated (r=0.876). Additionally, each of the cognitive sub-domain scores of the SNSB-D was significantly correlated with the overall GCF score, as well as with the MMSE score. Specifically, it was found that the memory domain score had the highest correlation with the GCF among the five cognitive sub-domains, with the other correlations as follows: attention (0.629), language and related function (0.848), visuospatial function (0.779), memory (0.945), frontal/executive function (0.919).

Cronbach's alphas for the SNSB-D were 0.797 in the AD group, 0.710 in the MCI group, 0.637 in the NC group, and 0.871 for the full sample, suggesting a high level internal consistency. The test-retest reliability of the GCF score was assessed using Pearson correlation coefficients. Test-retest reliability was high, with a correlation coefficient of 0.978 for the whole sample, 0.918 for the AD group, 0.999 for the MCI group, and 0.960 for the NC group.

We used the ROC curve analysis and calculated the area under the curve (AUC), as a measure of diagnostic utility for the SNSB-D. First, in discriminating the MCI group from the NC group, the AUC of the SNSB-D was 0.883 and the SNSB-D had a sensitivity of 0.844 and a specificity of 0.814, with a cut-off score of 188.25 (Table 3, Fig. 1A). The AUC of the SNSB-D GCF score was significantly superior to the MMSE (AUC=0.716, P=0.001). Second, in discriminating the MCI group from the AD group, the AUC of the SNSB-D was 0.939 and the appropriate SNSB-D cut-off score was 134.25 with a sensitivity of 0.860 and a specificity of 0.860 (Table 3, Fig. 1B). The AUC of the MMSE (0.942) was comparable to that of the SNSB-D (P=0.842).