PDF
ePub
Citation
Print
INTRODUCTION
The Seoul Neuropsychological Screening Battery (SNSB) is one of the most widely used neuropsychological screening batteries in South Korea.12 The current version of SNSB (SNSB-II) is composed of comprehensive cognitive tests that evaluate the level of cognitive functioning or impairment in five different cognitive domains: attention, language and related functions, visuospatial functions, memory, and frontal/executive functions.2 It also includes other related tests, such as clinical dementia rating (CDR), Barthel-activities of daily living (B-ADL), Korean-instrumental activities of daily living (K-IADL) and Geriatric Depression Scale (GDS). The estimated completion time of the whole battery is one and a half hours to two hours.
The brief version of the SNSB, named SNSB-Core (SNSB-C), has been developed to shorten completion time while ev-aluating core features of cognitive ability.3 The SNSB-C is composed of fourteen subtests derived from the SNSB-II:2 Vigilance Test, Digit Span Test,4 Comprehension Test, Repetition Test, short form of the Korean-Boston Naming Test,5 Ideomotor Apraxia Test (IAT), Rey Complex Figure Test,6 Seoul Verbal Learning Test-Elderly's version (SVLT-E), Contrasting Program, Go-No go Test, short form of the Korean-Color Word Stroop Test,7 Controlled Oral Word Association Test (COWAT),8 Korean-Trail Making Test-Elderly's version (K-TMT-E),9 and Digit Symbol Coding.10 Although each subtest score of the SNSB-C provides information on different features of broad cognitive functioning or impairment, a composite score depicting the overall level of cognitive ability is needed to identify the severity of global cognitive impairment. The present study aimed to develop and validate a composite score of the SNSB-C that would provide a normative based summary score of global cognitive functioning, especially for differentiating the elderly with cognitive impairment from the normal elderly.
METHODS
Subjects
We first used a normative sample to develop a composite score and then validated the score using different samples with cognitive impairment against a normal comparison group. For the normative sample, one thousand and sixty-seven elderly people, ranging in age from 45 to 90 years, were recruited across different regions in South Korea based on the Christensen's health screening criteria.11 The mean age was 68.1 years with a standard deviation of 11.8. The average years of education was 8.3 (SD=5.4) years, including 8.7% of illiterates. There were 600 females in the sample (56.2%).
The validation samples included different groups of patients with cognitive impairment: 41 patients with Alzheimer's disease (AD), 40 patients with vascular dementia (VD), 73 patients with amnestic multi-domain mild cognitive impairment (amMCI), 41 patients with vascular MCI (VaMCI), and 41 patients with Parkinson's disease with MCI (PD-MCI). All dementia patients met the clinical criteria for probable AD proposed by the NINCDS-ADRDA12 or for probable VD proposed by the NINDS-AIREN.13 Patients with amMCI, VaMCI, and PD-MCI met Petersen's criteria for MCI,14 American Heart Association/American Stroke Association's criteria for vascular cognitive impairment,15 and Movement Disorder Society's criteria for PD-MCI,16 respectively.
Additionally, 70 people without cognitive impairment were randomly selected from the normative sample for comparison with the patient groups. All of the subjects in the normative and the validation samples were administered all neuropsychological tests in the SNSB-II as well as the Korean-Mini Mental State Examination (K-MMSE)17 by trained clinical neuropsychologists. The CDR, B-ADL, K-IADL, and short version of the GDS (SGDS) were also administered to all subjects in the validation samples. Standard administration of the SNSB-II has been previously described in detail.2 Characteristics of the validation samples are presented in Table 1.
Average ages in the six groups were not statistically equivalent (F(5,300)=6.37, p<0.001), but post hoc comparisons using Tukey's HSD adjustment revealed that the normal group did not show a significant mean difference compared with any other groups. Average years of education were not significantly different across the six groups (F(5,300)=0.99, p=0.427). The validation groups differed in terms of mean SGDS (F(5,274)=6.68, p<0.001), where the normal group was significantly different from the other patient groups. The VD group showed a significantly lower mean value of the B-ADL compared with the other groups (F(5,300)=7.14, p<0.001). The average K-MMSE scores in the AD and the VD groups were significantly lower than those in the other groups, and the normal group was different from the amMCI group but not from the VaMCI and PD-MCI groups (F(5,300)=34.64, p<0.001).
Composite scores and statistical analysis
A composite score of the SNSB-C was developed using the normative sample. The subtest scores considered for the composite score are listed in Table 2 with their targeted cognitive domain. The COWAT score was calculated as sum of the semantic fluency score for the "animal" category and the phonemic fluency score for the phoneme "ㄱ." The K-TMT-E score in SNSB-C is the time in seconds, spent to complete the test with the limit of 300 seconds. Because the K-TMT-E score is negatively correlated with other subtest scores and distributed with high negative skewness, we used a log transformed K-TMT-E score [=loge(300/K-TMT-E)] to ensure that the transformed score has positive correlations with other subtest scores and less skewed distribution. These properties are desirable for constructing a composite score with other test scores. All three scores of immediate recall, delayed recall, and recognition on the SVLT-E were used in calculating the composite score.
Because the scales of subtests differ enormously (Table 2), standardized scores were used to construct a composite score. Two methods of standardization were explored: 1) a composite score that uses equal weight to each standardized subtest; 2) a composite score that uses equal weight to each standardized cognitive domain. To determine which composite score is better at representing overall cognitive ability, a confirmatory factor analysis (CFA) was conducted. The result of the general uni-factor model was compared to that of a higher-order factor model, where a first-order factor consists of subtests in the frontal domain and another first-order factor consists of subtests in the memory domain in terms of model fit indices.
Age, education, and gender effects on the general cognitive ability have been underscored in the literature.181920 Therefore, we developed a composite score corrected for age, education, and gender effects using the normative sample. Regression analysis was performed on the uncorrected composite score, where both the mean and the variance of the score were regressed on age, years of education, gender, and their plausible interactions. Details of the correction procedure can be found in the literature.21
Corrected and uncorrected composite scores were evaluated by receiver operating characteristic (ROC) curve analysis to differentiate patient groups with AD, VD, amMCI, VaMCI, or PD-MCI from the normal comparison group in the validation samples. Areas under the ROC curves for the composite scores were compared to those for the K-MMSE to ensure that they are comparable to K-MMSE in identifying general cognitive impairment.
RESULTS
Descriptive statistics and correlations of subtest scores are presented in Table 3. Overall, subtest scores were highly correlated with each other. On the other hand, scores of the other subtests not considered for the composite score, i.e., Vigilance Test, Comprehension Test, Repetition Test, Ideomotor Apraxia Test, Contrasting Program, and Go-No go Test, were not highly correlated with the ten subtest scores presented in Table 3 (all rs<0.41). In addition, most of the normative sample (65.3%, 95.3%) scored maximum points of the six subtest scores as expected.
The ten subtest scores, including the log transformed K-TMT-E score, were used to construct a composite score. CFA revealed that the general uni-factor model was not satisfactory [RMSEA=0.155, 95% confidence interval (CI)=(0.146, 0.164); CFI=0.849; NNFI=0.849; SRMR=0.065], but the higher-order factor model fitted the data well [RMSEA=0.054, 95% CI=(0.044, 0.065); CFI=0.983; NNFI=0.976; SRMR=0.026]. Accordingly, we adopted the composite score that uses equal weight to each standardized cognitive domain score. Frontal domain score was calculated by summing up the four standardized subtest scores in frontal domain (Table 2). Memory domain score was calculated as the sum of the three SVLT-E scores with the delayed recall score doubled so as to make scales of the three scores more similar to each other. Each of the visuospatial, language, and attention domains were composed of a single subtest score. The five domain scores were standardized or z-transformed and then summed up to form the composite score of the SNSB-C. The composite score itself was rescaled to have a mean of 50 and a standard deviation of 10 for interpretive purpose. Correlations of the composite score with its subtest scores were 0.65 or higher. The SNSB-C composite score was also highly correlated with the K-MMSE score [r=0.77, 95% CI=(0.74, 0.79)] in the normative sample (Table 3).
The effects of sex, log of years of education, illiteracy, and age on the uncorrected composite score were found to be significant (all |t|s ≥2.85, all ps<0.01). In addition, the effect of log of years of education interacted with sex (t=-2.91, p=0.004) and age (t=3.14, p=0.002). Variance of the composite score was also affected by age (z=3.66, p<0.001) and log of years of education (z=-5.66, p<0.001). The age-education-sex corrected composite score was obtained by subtracting the estimated mean from the uncorrected composite score and dividing it by the square root of the estimated variance, i.e., the estimated standard deviation. The corrected composite score was rescaled to have a mean of 50 and a standard deviation of 10.
The uncorrected and corrected composite scores of the SNSB-C were examined across the patient groups with different cognitive impairments as well as the normal group in the validation samples. Means and standard deviations of the composite scores for the validation samples are presented in Table 4. Group means of the uncorrected SNSB-C composite score (F(5,300)=23.36, p<0.001) and the corrected composite score (F(5,300)=35.80, p<0.001) were significantly different across the validation samples, but the corrected composite score showed more distinct mean differences across the groups. Although patients with MCI, i.e., the amMCI, VaMCI, and PD-MCI groups, did not show sizable group differences among them in both composite scores, all the groups were significantly different from the normal group in only the corrected SNSB-C score.
ROC analysis revealed that the SNSB-C composite scores were comparable to K-MMSE in differentiating normal people from the patients with cognitive impairment. Areas under the curve (AUCs) between the normal group and each patient group for the two composite scores and K-MMSE are presented in Table 5. The corrected composite score differentiated PD-MCI patients from the normal group significantly better than the K-MMSE total score [AUC difference=0.18, 95% CI=(0.06, 0.31)]. AUCs produced by the three scores were not significantly different from each other in all other discriminations.
DISCUSSION
This study was conducted to develop a composite score of the SNSB-C and to validate its usefulness as a screening measure for differentiating the MCI and dementia patients with various etiologies from the normal elderly. Confirmatory factor analyses showed that the composite score that uses equal weight to each standardized cognitive domain of SNSB-C is appropriate for indexing overall cognitive functioning. We developed two composite scores, uncorrected and age-education-sex corrected scores, with five cognitive domain scores derived from ten subtest scores of the SNSB-C.
The results showed that the corrected composite score could differentiate normal aging from MCI and MCI from dementia as well as normal aging from dementia. These findings suggest that the corrected composite score of the SNSB-C reflects the severity of cognitive dysfunction and can be used as a valid measure for tracking progression of a dementing process.
The results also demonstrated that the corrected and uncorrected composite scores yielded satisfactory size of the area under the ROC curve comparable to the K-MMSE. Both composite scores of the SNSB-C and K-MMSE total score showed quite similar discriminability of amMCI and VaMCI, as well as AD and VD. The corrected composite score in the PD-MCI group, however, showed better discriminability than the KMMSE total score. Along with memory, frontal/executive function is the most common cognitive domain affected in PDMCI.22 The SNSB-C includes more frontal/executive function subtests than the MMSE which insufficiently probes executive function.23 This result indicates that the corrected composite score of the SNSB-C is more sensitive to frontal/executive dysfunction than the K-MMSE score.
All these findings strongly support the claim that the composite scores of SNSB-C, especially the corrected score, provide an index of overall cognitive functioning, and they can be used as an alternative to K-MMSE for screening patients with cognitive impairment. We expect that the composite scores of SNSB-C will be widely used for the detection and progression monitoring of MCI and dementia. In addition, another benefit of using a composite score is that the score lends itself to data analysis and is useful for change over time analyses. Therefore, the composite scores of SNSB-C will be very useful in research relating to dementia, especially in longitudinal studies.
XML Download