Validation of a modified Chinese version of Mini‐Addenbrooke's Cognitive Examination for detecting mild cognitive impairment

Abstract Background For detecting mild cognitive impairment (MCI), brief cognitive screening tools are increasingly required for the advantage of time saving and no need for special equipment or trained raters. We aimed to develop a modified Chinese version of Mini‐Addenbrooke's Cognitive Examination (C‐MACE) and further evaluate its validation in detecting MCI. Methods A total of 716 individuals aged from 50 to 90 years old were recruited, including 431 cognitively normal controls (NC) and 285 individuals with MCI. The effect size of Cramer's V was used to explore which items in the Chinese version of Addenbrooke's Cognitive Examination‐III (ACE‐III‐CV) best associated with MCI and to form the C‐MACE. Receiver operating characteristic (ROC) analyses were carried out to explore the ability of C‐MACE, ACE‐III‐CV, Chinese version of Montreal Cognitive Assessment‐Basic (MoCA‐BC), and Mini‐Mental State Examination (MMSE) in discriminating MCI from NC. Results Five items with greatest effect sizes of Cramer's V were selected from ACE‐III‐CV to form the C‐MACE: Memory Immediate Recall, Memory Delayed Recall, Memory Recognition, Verbal Fluency Animal and Language Naming. With a total score of 38, the C‐MACE had a satisfactory classification accuracy in detecting MCI (area under the ROC curve, AUC = 0.892), superior to MMSE (AUC = 0.782) and comparable to ACE‐III‐CV (AUC = 0.901) and MoCA‐BC (AUC = 0.916). In the subgroup of Age > 70 years, Education ≤ 12 years, the C‐MACE got a highest classification accuracy (AUC = 0.958) for detecting MCI. Conclusion In the Chinese‐speaking population, C‐MACE derived from ACE‐III‐CV may identify MCI with a good classification accuracy, especially in aged people with low education.


INTRODUCTION
Mild cognitive impairment (MCI) is regarded as a transitional cognitive state between normal aging and dementia (Petersen, 2016). As most dementia can hardly be reversed, it is important to identify individuals with MCI for early intervention and to decrease the risk of further cognitive decline (Langa & Levine, 2014). Up to now, neuropsychological tests still play an important role in identifying MCI. Though comprehensive neuropsychological assessments performed well in the diagnosis of MCI (Jak et al. 2009), brief cognitive assessments are consistently required for the advantage of time saving and no need for special test equipment or trained raters, especially in primary care settings and large populations. Besides the widely used cognitive screening tests of Mini-Mental State Examination (MMSE) (Folstein et al., 1975) and Montreal Cognitive Assessment (MoCA) (Nasreddine et al., 2005), Addenbrooke's Cognitive Examination (ACE) is a designed for the detection and classification of dementia which includes subtests of orientation, attention, verbal fluency, memory, language, and visuospatial ability (Mathuranath et al., 2000). By covering the key cognitive domains, ACE showed better ability than MMSE and MoCA in predicting the development of dementia (Lischka et al., 2012). In order to increase the sensitivity and facilitate cross-cultural usage and translation, contents in memory, language, and visuospatial domains of ACE were modified to form the revised version of ACE (ACE-R) (Mioshi et al., 2006). By further addressing several weakness items in ACE-R, the latest third version of ACE (ACE-III) correlated well with the standardized neuropsychological tests and showed high diagnostic accuracy for both MCI and dementia (Bruno & Vignaga, 2019;Hsieh et al., 2013). However, as a comprehensive cognitive screening test, it takes about 15-20 min to complete the whole items of ACE-III, a time frame that may be beyond the scope of many busy clinical settings. For addressing this shortage, five items (time orientation, 7-element name and address immediate recall and delayed recall, animal verbal fluency, and clock drawing) derived from ACE-III were used to develop a short version of ACE-III named Mini-Addenbrooke's Cognitive Examination (MACE) with a total score of 30 and takes under 5 min to administer (Hsieh et al., 2015). For discriminating individuals with MCI from normal controls (NC), previous studies showed that the reliability and classification accuracy of MACE was superior to MMSE and equivalent to MoCA to a certain extent (Larner, 2017;Senda et al., 2020;Yang et al., 2019). However, latest systematic review indicated that this published version of MACE had a variable sensitivity and more variability in specificity according to the cutoff established in index study (Beishon et al., 2019). Furthermore, due to the linguistic and cultural diversities, previous study even showed a not ideal validation of the original MACE in discriminating MCI from control subjects (Miranda et al., 2018).
So far, ACE-R and ACE-III have been translated into different languages and the most culturally adapted items in ACE-III and its predecessors were anterograde memory, retrograde memory, and language (Mirza et al., 2017). Similarly, our previously published Chinese version of Addenbrooke's Cognitive Examination-III (ACE-III-CV) was translated and mainly culture adapted in the items of anterograde memory, verbal fluency, phrase repetition, and visuospatial perception (Pan et al., 2021). Though this Chinese version of ACE-III showed high ability to differentiate MCI from normal controls, the established cutoff scores and corresponding sensitivity and specificity differed significantly from other versions Peixoto et al., 2018).

Neuropsychological assessment and diagnostic criteria
Besides the cognitive screening tests of ACE-III-CV, MoCA, and MMSE, a battery of neuropsychological tests were carried out to diagnose MCI according to an actuarial neuropsychological method proposed by Jak and Bondi (Bondi et al., 2014). In this method, six standardized neuropsychological tests were included: Auditory Verbal Learning Test (AVLT) 30-min delayed recall and AVLT recognition for the measurements of memory (Zhao et al., 2015); Animal Verbal Fluency Test (AFT) (Zhao et al., 2013) and 30-item Boston Naming Test (BNT) (Guo et al., 2006) for the measurements of language; Shape Trail Test Part A and B (STT-A, STT-B) (Zhao et al., 2013) for the measurements of executive function. Each test has been widely used in China and was standardized using published normative data. Global functional status was also assessed by Functional Assessment Questionnaire (FAQ) (Pfeffer et al., 1982) according to informants. Participants met one of the following criteria were diagnosed as MCI: (1) impaired scores (defined as >1 standard deviation (SD) below the age-corrected normative mean) on two of the six neuropsychological indexes in the same cognitive domain (memory, language, or executive function); (2) impaired scores (defined as >1 SD below the age-corrected normative mean) in each of the three cognitive domains; (3) FAQ score ≥9. All the screening tests and standardized neuropsychological tests were administered on the same day by raters who were blind to the diagnosis.

Statistical analyses
All the items in ACE-III-CV were evaluated using Cramer's V, respectively, to determine which items best indicate a strong association with MCI. The effect size of Cramer's V ranges from 0 to 1 and greater than 0.5 is generally taken as a large value. The best items were supposed to have the largest effect sizes and would be selected as the items to form C-MACE.

Demographics and neuropsychological tests in the group of NC and MCI
The demographics, scores of standardized neuropsychological tests, global function tests, and cognitive screening tests for the groups of NC and MCI are shown in Table 1

ROC analyses of C-MACE, ACE-III-CV, MoCA-BC, and MMSE for discriminating MCI from NC
As shown in Table 2

Comparison of AUCs according to different age and education level
As the performance of screening tests was probably affected by age and education level, classification accuracies of the C-MACE, ACE-III-CV, MoCA-BC, and MMSE were further compared in different age and education level groups. Subjects in this study were divided into four subgroups according to the ages and formal education years: Age ≤ 70 years, Education ≤ 12 years; Age ≤ 70 years, Education > 12 years; Age > 70 years, Education ≤ 12 years; Age > 70 years, Education > 12 years. The education level was divided according to the current situation of Chinese education, that is, 12 years for educated from senior high school, and more than 12 years for the education in universities. The age cutoff of 70 years was chosen as dementia rises steeply in persons aged more than 70 years (Bachman et al., 1993).

DISCUSSION
This study developed a modified C-MACE with five items derived from the ACE-III-CV. The measurement accuracy of each item was confirmed by the significant associations with standardized neuropsychological tests. Besides the internal consistency verified by Cronbach's coefficient, the C-MACE has a good convergent validity with MoCA-BC. No obvious ceiling and floor effects were found. Validation study verified that this C-MACE had a satisfactory classification accuracy in detecting MCI from NC according to significantly higher AUC than MMSE and comparable to ACE-III-CV and MoCA-BC.
In the process of developing C-MACE, a statistically based method was used instead of clinical judgment or prior experience. As a result, with a total score of 38, this C-MACE mainly includes two cognitive domains: 19 points for verbal episodic memory assessed by learning and recall a 7-item name and address; 19 points for semantic memory assessed by verbal fluency of generating animals and naming of the pictures presented. This is consistent with the fact that episodic memory and semantic memory appeared to be substantially impaired in MCI and preclinical Alzheimer's disease (AD) (Mickes et al., 2007;Salmon, 2012). Previous studies showed that memory tests with free recall alone were more likely to generate high false positive and this shortage was overcome with the task of cued recall (Cullum et al., 1993;Kuslansky et al., 2002). The optimal cutoff score for detecting MCI and corresponding sensitivity and specificity of these tests were determined by the maximum Youden index. The sensitivity could be improved at the cost of specificity by using a higher cutoff score, and vice versa (Soreide, 2009). As early detection of MCI may help decrease the risk of further cognitive decline (Langa & Levine, 2014), a higher cutoff score with good sensitivity and minimized false negative rate should be adopted in a population with high prevalence of cognitive impairment such as individuals with subjective memory complaint (Kielb et al., 2017). However, there is no modifying-course therapy for MCI due to AD currently, and adverse psychological effects such as anxiety and depression may be incited once the prodromal phase of dementia was determined (Milne, 2010).
In view of this, lower cutoff scores prefer specificity and reduce the false positive rate should be selected in community population with an unusually high prevalence of cognitive impairment. It should be noted that in our study, the optimal cutoff scores of MACE and ACE-III-CV for detecting MCI were both lower than the pre-specified thresholds (82 and 88 for ACE-III, 21 and 25 for MACE) (Beishon et al., 2019). We attribute this phenomenon to several reasons. First, a comprehensive neuropsychological criterion which showed higher diagnostic accuracy than conventional criteria was used for the diagnosis of MCI in this study (Bondi et al., 2014). Second, item such as episodic memory in ACE-III-CV was modified for reducing the influence of implicit memory.
Third, the pre-specified thresholds were generated from case-control studies and thus may lead to a risk of bias.
Diagnostic accuracy of cognitive assessment tools were highly cor- Although cognitive screening instruments were widely used, the underdiagnosis rate of cognitive impairment is still high and partly due to the pressure on screening time (Iliffe et al., 2009). Besides the relatively lengthy administration time, not all the items in cognitive screening instruments were deemed essential and contributed to the classification accuracy (Braekhus et al., 1992). In our study, five items derived from the ACE-III-CV formed C-MACE. As a very brief cognitive screening instrument, neither specialist training nor specific test material is needed. However, this abbreviated version of ACE performed better than MMSE and was comparably accurate to the standard Chinese version of ACE-III and MoCA-BC in identifying MCI, especially in aged people with low education. Compared to the original version of MACE proposed by Hsieh et al. (2015), as the test of naming is more time saving than clock drawing, administration time of this C-MACE is obviously shorter and this would be more applicable in a large population or busy clinical settings. In addition, progressive decline in cognition demonstrated by longitudinal cognitive assessments provides more evidence for MCI due to AD (Albert et al., 2011). Raters may easily obtain the previous scores of the C-MACE for individuals assessed by ACE-III-CV in one's early stage and get a longitudinal cognitive comparation. However, as a brief screening tool, the C-MACE will not provide a formal diagnosis, individuals screened as cognitive impairment usually need further comprehensive assessments.
Several limitations in this study should be noted. First, the C-MACE was based on the administration of ACE-III-CV and not carried out as a unique test in this study. The standard administration procedure for this C-MACE has not been determined. Second, the time interval between learning and recall in the item of 7-item name and address came from ACE-III-CV, what would be an appropriate time interval in the C-MACE need further investigation. Third, the number of participants diagnosed as MCI in the group of Age ≤ 70 years, Education > 12 years was relatively low and sampling bias may exist in this group.
In conclusion, this present study developed a modified C-MACE with five items derived from ACE-III-CV. For detecting MCI, this C-MACE showed satisfactory classification accuracy with significantly higher AUC than MMSE and comparable to ACE-III-CV and MoCA-BC, especially in aged people with low education.

ACKNOWLEDGMENTS
This work was supported by the National Key R&D Program of China under grant number 2016YFC1306305.

CONFLICT OF INTEREST
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.