We assessed whether the Korean version of modified Mini-Mental State Examination (K-mMMSE) has improved performance as a screening test for cognitive impairment or dementia in a general population compared with the Korean Mini-Mental State Examination (K-MMSE).
Screening interviews were conducted with people aged 65 and over in Noam-dong, Namwon-city, Jeonbuk province. There were 522 community participants, of whom 235 underwent clinical and neuropsychological examination for diagnosis of dementia and Cognitive Impairment No Dementia (CIND). Sensitivity, specificity and areas under the receiver operating characteristic (ROC) curves for the K-mMMSE and the K-MMSE were the main outcome measures.
Cronbach's alpha for the K-mMMSE was 0.91, compared with 0.84 for the K-MMSE. The areas under the ROC curves in identifying all levels of CIND or dementia were 0.91 for the K-mMMSE and 0.89 for the K-MMSE (P < 0.05). For the K-mMMSE, the optimal cut-off score for a diagnosis of CIND was 69/70, which had a sensitivity of 0.86 and a specificity of 0.79, while, for a diagnosis of dementia, the optimal cut-off score of 59/60 had a sensitivity of 0.91 and a specificity of 0.78. The K-mMMSE also had a high test-retest reliability (r = 0.89).
Our findings indicate that the K-mMMSE is more reliable and valid than the K-MMSE as a cognitive screen in a population based study of dementia. Considering the test characteristics, the K-MMSE and modified version are expected to be optimally used in clinical and epidemiologic fields.
The Mini-Mental State Examination (MMSE) is a brief screening test that quantitatively assesses the cognitive status of elderly people [1,2]. It is easy to administer and has shown good reliability. Although its validity as a screening test is acceptable for clinical samples, it has been shown to have difficulty in discriminating between demented and non-demented individuals in community-based samples . The MMSE has been found to be influenced largely by pre-morbid ability and is less sensitive to focal brain dysfunction  or mild dementia .
These limitations led to the development of the Modified Mini-Mental State Examination (3MS) in 1987 , which expanded the MMSE from 30 to 100 points to provide finer discrimination. In addition, the 3MS added four items: personal information, including date and place of birth; verbal fluency; abstract reasoning; and a second delayed recall trial. The 3MS also graded temporal orientation and broadened the delayed recall measures, which included cued and recognition formats. While retaining the brevity and ease of administration of the MMSE, the 3MS improved the validity and reliability of identifying individuals with dementia [6,7], and in predicting functional outcomes in patients with stroke .
In 2002, a group from the Cache County Study modified the 3MS for use as a cognitive screen in an epidemiologic study of dementia . These modifications substituted the recall of personal demographic information with the recall of current and past prominent politicians. The main reason for this modification was the difficulty the researchers had in verifying personal demographic information. In addition, the scaling of the items in the time orientation and writing parts of the test was changed, and the time allotted for animal naming was shortened. This revised form of the 3MS (3MS-R) demonstrated good sensitivity in detecting dementia in a general population and providing age- and education-specific normative data and cut-off values at the 7th and 10th percentiles .
In Korea, the MMSE was translated into two versions, the Korean version of the Mini-Mental State Examination (MMSE-K) and the Korean Mini-Mental State Examination (K-MMSE) [10,11]. Both Korean versions of the MMSE were tested for validity and efficacy in clinical settings [11,12] and partly in epidemiologic research . Both were somewhat modified to adjust better to the cultural background in Korea, but both shared all the limitations of the original MMSE. Recently the Korean Modified Mini-Mental State Examination (K-3MS) was introduced and validated in a clinical setting . Although the K-3MS was found to be a reliable cognitive screening measure, there was no significant difference between the K-3MS and extracted MMSE for detecting individuals with dementia. In addition, components of the K-3MS could not be compared with items extracted from the K-MMSE and MMSE-K. We have therefore introduced the Korean version of modified Mini-Mental State Examination (K-mMMSE), and we have determined whether its validity is superior to that of the K-MMSE as a screen for cognitive impairment or dementia in a community setting.
Potential participants for this study were recruited from all inhabitants of Noam-dong, Namwon-city, Jeonbuk-province, South Korea, aged 65 and over in 2003, as recorded in national residents registration lists. The area surveyed covered 8.93 km2 and had an estimated population of 6,883, of whom about 7% were aged 65 or over. All participants gave informed consent, and the study was conducted in accordance with the guidelines in The Declaration of Helsinki and approved by the appropriate research ethics committee.
Among the 522 eligible subjects aged 65 and over identified from registration lists, 235 (45%) completed clinical examinations after the interview and formed the study sample for principal analysis. Of the remaining subjects, we were unable to establish contact with 162 (31%), 75 (14%) refused to participate, 18 (3%) did not complete the survey, 7 (1%) had severe pre-morbid illness including blindness and deafness, 4 (0.8%) had changed address, and 3 (0.6%) had died before the visit. The principal apparent reason for the difficulties in establishing contact was that the person was in a regular daily activity or away from home, visiting family members living elsewhere. The 18 individuals (3%) who did not complete the survey questionnaire or were not examined clinically had a mean age of 74.9 ± 10.3 years; 13 (72%) were females, and 4 (22%) were educated. Of all the eligible subjects, participants and non-participants did not differ in age (73.5 ± 6.8 y vs. 74.6 ± 7.8 y, respectively) or gender (66% and 62%, all P values > 0.1).
Assessment and measurements
Interviewers received a seven-day training session on administering the screening instruments and were supervised throughout by the project neurologist. Cognitive status was classified in two stages. In the first stage, interviewers carried out home-based interviews for data on cognitive function, past medical history, and demographic characteristics. All participants were contacted for cognitive screening using a formulated battery, from which the K-mMMSE and K-MMSE were extracted. And they were rated by a knowledgeable informant using the Short form of Samsung Dementia Questionnaire (S-SDQ) and Korean Instrumental Activities of Daily Living (K-IADL) [15,16]. The S-SDQ is a Korean version of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE), with 15 items and scores ranging from 0 to 30 . The K-IADL is composed of 11 items that grade functional abilities, with scores calculated as the sum of points over the number of applicable questions and ranging from 0.0 to 3.0 . High scores on the S-SDQ and K-IADL indicate poor performance. Both tests were found to be uncontaminated by pre-morbid ability, including education or age [15,16].
At the second interview, physicians who were blinded to the cognitive scores performed a clinical examination and neuropsychiatric inventory on participants who completed the first survey questionnaire. The clinical examination verified the presence of cognitive impairment. The diagnostic criteria for dementia were based on those of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV)  and were subdivided according to the guidelines of the National Institutes of Neurological and Communicative Disorders and Stroke and the Alzheimer's disease and Related Disorders Association (NINCDS-ADRDA) . Severity of dementia was staged using the Korean version of the Expanded Clinical Dementia Rating (CDR) scale . The physician and neurologist made independent diagnoses and CDR scoring and subsequently held a case conference to reach a consensus diagnosis, classifying the person as either cognitively normal, cognitively impaired with no dementia (CIND), or having dementia. A diagnosis of CIND represents an attempt to classify people with recognizable cognitive decline who did not meet the criteria for dementia. This group included people who complained of cognitive decline and showed impaired memory function, but did not have any non-cognitive alterations, including intact activities of daily living. At both stages, home visits were repeated on at least two occasions if no contact was made.
We translated the original 3MS-R into Korean according to the guidelines recommended by the modifiers . The item regarding political figures, which asked the participants to name the current president, vice president, and state governors, was replaced with questions about current and previous presidents, because there is no vice president in the government of South Korea. We assessed temporal orientation according to three methods used to calculate year and time in Korea: the solar, lunar and Tangun era. We replaced the words "shirt," "nickel" and "honesty" in the memory task with the words "airplane," "pine tree" and "sincerity." In Korean, the first two words coincided with the K-MMSE items and ended with vowel sounds ("bee-haeng-gi," and "so-na-mu," respectively); while "seong-sil" in Korean, which means sincerity, was substituted for "jeong-jik," which is equivalent to the English word honesty, inasmuch as a pilot study found that "jeong-jik" was more difficult to hear or perceive than "seong-sil", possibly because the latter ended with a voiced sound and could be heard more comfortably.
In explaining the appropriate questions and answers, we presented a simple example prior to asking the first question, specifically, "the eyes and nose are different, but they are similar in being part of our face." In a pilot study, most elderly subjects could not understand the concept of similarity without this example, and they became intolerant to our interview unless an example was provided. After providing the example, however, most subjects were more cooperative and tried to answer properly. We asked subjects to write a spontaneous sentence and scored whether it was legible and correct, with or without prompting. The total possible score was 100, and a K-MMSE score could be generated from it.
For demographic factors, mean or median values and proportions were calculated according to the cognitive impairment strata. Cronbach's alpha was used to quantify internal consistencies of the K-mMMSE and K-MMSE . We assessed the stability and test-retest reliability with 30 subjects who took the K-mMMSE twice. Receiver operating characteristic (ROC) curves were used to determine the validity of the two screening tests graphically and statistically. The areas under the ROC curves (AUC) were calculated and described with standard errors (SE) using the trapezoidal rule , and comparison between AUCs was made by the algorithm with an estimated covariance matrix . Cut-off points were chosen to optimize the trade-off between false-negative and false-positive rates. The choice of whether to judge the screening tests by their ability to identify CIND or dementia was addressed practically by assessing both. The validity of the K-mMMSE and K-MMSE were compared, first between those with CIND or dementia and those who were normal, and then between those with dementia and those with CIND or normal. All statistical calculations were performed using Stata 8.2 software (Stata, College Station, TX).
Descriptive statistics are displayed in Table 1 for the total sample and according to the cognitive impairment strata. Of the 235 participants, 46 (19.6%) were classified as having dementia and 54 (22.9%) as having CIND. Overall levels of education were very low, in that 118 participants (50.2%) had no formal education, of whom 83 (70.3%) were illiterate.
Severity of cognitive impairment was directly related to mean age and inversely related to number of years of education. Women had higher rates of being cognitively impaired or demented than men. The median scores on the K-mMMSE and K-MMSE decreased with severity of cognitive impairment. Median scores on the informant questionnaires of the S-SDQ and K-IADL increased with poorer cognitive status. The inter-quartile ranges (IQR) also steadily increased, reflecting increasing variability in cognitive and functional status. Four respondents scored perfectly on the K-MMSE, whereas none scored perfectly on the K-mMMSE, suggesting that the K-mMMSE might be less prone to the ceiling effect in this population.
Table 1. Subject characteristics
The estimated Cronbach's alpha was 0.91 for the K-mMMSE and 0.84 for the K-MMSE. Relative to each cognitive impairment stratum (normal, CIND, and dementia), the alphas were 0.84, 0.81 and 0.81, respectively, on the K-mMMSE and 0.74, 0.72, and 0.63, respectively, on the K-MMSE. Neither age nor gender had any substantial impact on internal consistency. Stratum-specific alphas of the K-mMMSE for the different subgroups ranged from 0.86 to 0.91 in men and 0.89 to 0.90 in women.
The retest of the K-mMMSE was assessed in 30 subjects (mean interval, 26 days; range, 19–32 days). The correlation coefficients for the total scores were 0.89 on the K-mMMSE and 0.85 on the K-MMSE. The coefficients of the 15 items of the K-mMMSE were all significant, ranging from 0.37 for similarities to 0.83 for time orientation. The re-tested subjects were representative of the entire study population, in that the sociodemographic characteristics of the 30 retested subjects were similar to the other participants in mean age (74.0 ± 7.4 y vs. 73.5 ± 6.7 y), educational years (3.5 ± 3.2 y vs. 3.4 ± 3.9 y), K-mMMSE scores (63.8 ± 21.0 vs. 63.1 ± 20.4), and proportion of women (80.0% vs. 66.4%; P = 0.132).
Validity of the K-mMMSE and K-MMSE
Construct validity data between the K-mMMSE and other cognitive or functional measures are shown in Table 2. K-mMMSE was found to be significantly correlated with all measures, including CDR, Sum of Boxes of CDR (CDR-SB), and informant questionnaires such as the S-SDQ and K-IADL. The correlation coefficient between K-mMMSE and K-MMSE scores was 0.94. According to the CDR scores, the median values of the K-mMMSE and K-MMSE changed significantly (Table 3).
Table 2. Correlations between K-mMMSE, K-MMSE and cognitive or functional measures (CDR, KIADL, and S-SDQ)
Table 3. K-mMMSE and K-MMSE scores for each CDR group
Identification of combined CIND and dementia
The performance of the K-mMMSE (AUC ± SE, 0.91 ± 0.02) was significantly superior to that of the K-MMSE (0.89 ± 0.02; P = 0.041). The ROC curves plotted in the same graph suggested that the performance of the K-mMMSE was superior to that of the K-MMSE at almost all cut-off points (Figure 1). At a cut-off of 69/70 for CIND, the K-mMMSE had a sensitivity of 0.86 (95% Confidence Intervals, 0.78–0.92), a specificity of 0.79 (0.71–0.86), a positive likelihood ratio (LR) of 4.15, and a negative LR of 0.18. In comparison, at a cut-off of 20/21 for CIND, the K-MMSE had a sensitivity of 0.82 (0.73–0.89), a specificity of 0.79 (0.71–0.86), a positive LR of 3.95 and a negative LR of 0.23.
Figure 1. Receiver operating characteristic (ROC) curves of the K-mMMSE and the K-MMSE for combined CIND and dementia. K-mMMSE (light blue), K-MMSE (brown), and diagonal line. Area under ROC Curves (AUC): K-mMMSE = 0.91, K-MMSE = 0.89.
Identification of dementia
Both the K-mMMSE and the K-MMSE could properly discriminate demented from normal individuals, but there was no significant difference between them (AUC, 0.92 vs 0.91; P = NS). At the cut-off of 59/60, the K-mMMSE had a sensitivity of 0.91 (0.79–0.98), a specificity of 0.78 (0.72–0.84), a positive LR of 4.21 and a negative LR of 0.11. At a cut-off point of 18/19, the K-MMSE had a sensitivity of 0.91 (0.79–0.98), a specificity of 0.76 (0.69–0.82), a positive LR of 3.82 and a negative LR of 0.11.