Evaluating the Beijing Version of Montreal Cognitive Assessment for Identification of Cognitive Impairment in Monolingual Chinese American Older Adults

Abstract

Objective:

This study aims to evaluate the performance of a Chinese version of the Montreal Cognitive Assessment (MoCA) as a screener to detect mild cognitive impairment (MCI) and dementia from normal cognition in the monolingual Chinese-speaking immigrant population.

Method:

A cohort of 176 Chinese-speaking older adults from the National Alzheimer’s Coordinating Center Uniform Data Set is used for analysis. We explore the impact of demographic variables on MoCA performance and calculate the optimal cutoffs for the detection of MCI and dementia from normal cognition with appropriate demographic adjustment.

Results:

MoCA performance is predicted by age and education independent of clinical diagnoses, but not by sex, years of living in the U.S., or primary Chinese dialect spoken (i.e., Mandarin vs. Cantonese). With adjustment and stratification for education and age, we identify optimal cutoff scores to detect MCI and dementia, respectively, in this population. These optimal cutoff scores are different from the established scores for non-Chinese-speaking populations residing in the U.S.

Conclusions:

Our findings suggest that the Chinese version of MoCA is a valid screener to detect cognitive decline in older Chinese-speaking immigrants in the U.S. They also highlight the need for population-based cutoff scores with appropriate considerations for demographic variables.

The Montreal Cognitive Assessment (MoCA) is developed as a screening tool to detect mild cognitive impairments (MCI)¹. In the original sample collected in Canada consisting primarily White participants with a combined average of approximately 12 years of education, this brief measure has demonstrated strong psychometric properties ¹. MoCA has since been widely applied in a variety of populations as a tool to detect not only mild but also more profound cognitive impairments ^2,3. A cultural neuropsychology perspective emphasizes that test validation in the intended cultural group is essential prior to any research or clinical application of the instrument ⁴. Indeed, studies across the globe have documented a variety of cultural and linguistic adaptations of the MoCA, as well as a wide range of normative scores and psychometric properties ^5,6. A set of demographic variables such as education ^1,7,8, age ^7,8, and race/ethnicity ⁹ are also found to impact MoCA performance.

MoCA is one of the most extensively researched performance-based neuropsychological tools in Chinese-speaking populations. In fact, at least 9 versions of Chinese translations/adaptations for the MoCA version 7.1 have been made available, though not every translation/adaptation has been empirically validated ^10–12. It can be speculated that the strong interest in MoCA derives from 1) its relatively low level of culture-specific stimuli and thus relative ease of translation and adaptation, and 2) its brevity and ease of administration combined with the potential for strong diagnostic performance. These features allow MoCA, as a neuropsychological tool born in a Western centric context, to bear the potential of becoming a valuable clinical tool for an enormous Chinese-speaking population of 1.3 billion across the globe, as a screener to detect cognitive decline. Studies on the Chinese MoCA have, to a certain extent, illustrated the high level of heterogeneity within this population, as optimal cutoff scores for detection of cognitive impairments ranged from 13 to 26 when tested in different Chinese-speaking groups (e.g., mainland China ^13–15, Taiwan ¹⁶, Hong Kong ¹⁷, Singapore ¹⁸). These studies also note impact of demographic variables that are not necessarily consistent across different Chinese-speaking groups. For instance, age ^13,18 and education ^13,18,19 are commonly found to be associated with the MoCA performance, while the effects of sex and place of residence (i.e., urban vs. rural) are reported to be significant in only one study ¹³.

Despite the rather abundant literature on MoCA in various Chinese populations, the Chinese diaspora is a group of around 40 million worldwide that is largely under investigated. Approximately 5 million Americans are of Chinese descent as of 2015 ²⁰ and nearly 3 million of them identify a Chinese dialect as their primary language, with Mandarin and Cantonese being the most common ones ²¹. Assessment of immigration populations requires special consideration of acculturation, which has been repeatedly shown to impact performance on both verbal and nonverbal cognitive tests ^22,23. To our best knowledge, only two studies have examined the neuropsychological performance of Chinese immigrants residing in the U.S. and Australia, respectively ^24,25. These two studies are import initial work in highlighting the importance of cognitive test validation for Chinese immigrants and exploring the impact of demographic variables; though the samples were rather small (n = 71 and n = 145, respectively) for the comprehensive batteries implemented. However, the MoCA was not included in these two studies and has not been validated in the Chinese immigrant population. In the current study, we aim to evaluate the performance of the MoCA as a screening tool to detect early cognitive decline (i.e., MCI) and dementia in Chinese-speaking older adults residing in the U.S. Specifically, we aim to explore the impact of demographic variables that are unique in this population, and to determine the optimal cutoff scores to detect MCI and dementia from normal cognition (NC) with appropriate demographic adjustments.

Methods

Data Source

The current study utilized data of Chinese speakers recruited to the Alzheimer’s Disease Research Centers (ADRC) at Icahn School of Medicine at Mount Sinai and University of California San Francisco. These two NIH funded ADRCs contain the largest percentile of Chinese-speaking participants in their cohorts as of February 2020. They contributed data to the National Alzheimer’s Coordinating Center (NACC) using instruments in the Unform Data Set (UDS), which consisted of a standardized clinical and neuropsychological assessment ²⁶. The MoCA was implemented as a part of the standardized protocol (UDS v3.0) in March 2015. In 2018, the UDS was translated into Chinese, with plans for use in Alzheimer’s research in China and the United States. As of February 2020, two centers had participants who completed the MoCA in Chinese. Data came from these two ADRC sites with English-Chinese bilingual staff and Chinese-speaking older adults in their cohort. Staff were trained to follow a structured protocol for administration and were supervised by a neuropsychologist at each site. The ADRC at Icahn School of Medicine at Mount Sinai also administered the Beijing version of the MOCA as part of the standard dementia evaluation prior to the official implementation of the Chinese UDS v3.0 in 2018. Diagnosis of NC, MCI, and dementia was recorded on the UDS form D1 “Clinical Diagnosis,” which was completed by a clinician or by a consensus diagnosis ^26,27. In addition to the UDS, Icahn School of Medicine at Mount Sinai also collected demographic information including years of living in the U.S. and birth country for participants who were immigrants as part of the local data. Data collection at all ADRCs was overseen by the local institutional IRBs.

The sample for this study was selected from NACC data for UDS visits between March 2015 and February 2020, with a data extraction date of December 2020. Participants who completed MoCA at their baseline visit in Chinese with core demographic information (i.e., age, sex, and education) were included in these analyses.

Variable Construction

Participants completed a Chinese version of the MoCA (MoCA Beijing) ¹¹. Compared to the original English version, MoCA Beijing made the following adaptations. First, the letters in the trail making test (ABCDE) were replaced with Chinese nominal sequential characters () because Latin alphabet is not an inherent part of the Chinese language and can thus be unfamiliar to many Chinese-speaking individuals. Second, the letters in the cancellation task were replaced with Arabic numerals for the same rationale. Third, the phonemic fluency task was replaced with a category fluency task (i.e., animal naming) because the phonemic knowledge structure is much less prominent in logographic languages such as Chinese. All other instructions and verbal stimuli were directly translated from English to Chinese. The same written form in simplified Chinese was used for all Chinese-speaking participants, while the language for oral administration of the test was matched with participants’ primary language (i.e., Mandarin or Cantonese). The total score of MoCA (0-30) at baseline were used for analyses (n =176).

We included participants who were diagnosed as having NC, MCI, or dementia following the approach of Milani et al. ⁹. Participants coded as impaired, not MCI, were not included due to the ambiguity over their diagnosis. For demographic variables, sex and primary language (i.e., Mandarin and Cantonese, no other Chinese dialects were endorsed as the primary language in this cohort) of the participants were entered as categorical variables, and age and education were entered as continuous variables. Acculturation was proxied by years living in the U.S. as a continuous variable, which was only available for a subset of total sample (n = 162).

Based on the results of the demographic impact, we created adjustments for age and education. Specifically, we created three education groups (low, ≤9; medium, 10-12; high, >12) and three age bands (≤65, 66-75, >75). Minimum cell size for stratification was determined by the a priori power analysis (AUC = .80, significance level = .05, power = .80). We also created adjusted total score based on the established adjustment method (i.e., one extra point for individuals with ≤12 years of education) ¹ and the total MoCA scores of the NC participants within the current sample. Stratification and total score adjustment were used in the ensuing cutoff analyses.

Statistical Analyses

Statistical analyses were performed with R ²⁸. Descriptive statistics were calculated for population characteristics including age, sex, education, and years living in the U.S. One-way analysis of variance and Chi-square analysis were used to perform between-group comparisons (NC, MCI, and dementia) on demographic characteristics and MoCA performance. For these analyses, p ≤.05 was set as the threshold for statistical significance. Rejection of these one-way analysis of variance or Chi-square analysis were followed by post hoc multiple comparisons. A Bonferroni correction was used for post hoc analyses with an adjustment of p ≤.05/15 based on 15 univariant analyses within the sample. Multiple linear regression models were used to evaluate the relation between total MoCA score and demographic variables. The R package cutpointr ²⁹ was used to calculate optimal cutoffs to detect MCI and dementia from NC, using the Youden index ³⁰ as the metric. Bootstrapped confidence intervals for areas under the ROC curves (AUC) and cutoffs were calculated using 5,000 iterations.

Results

Sample Characterization

The sample included 176 Chinese-speaking adults aged between 52 and 94 years (M = 72.7, SD = 7.3), with a mean of 12.9 years of education (SD = 4.2), and a mean of 33.0 years of living in the U.S. 63.1% of the sample identified as female, and 56.8% of the sample reported Mandarin as their primary language. Participants from the Icahn School of Medicine at Mount Sinai (n = 163) were significantly older than those from the University of California San Francisco (n = 13; t = −2.52, p = .03). Participants from the two sites were otherwise statistically comparable on MoCA total score (t = −.62, p = .55), years of education (t = 1.34, p = .20), and sex composition (odds ratio = .93), and were combined for further analysis. Table 1 showed detailed demographics of the total sample and each diagnostic groups (i.e., NC, MCI, dementia). The three diagnostic groups were statistically comparable in age, years of living in the U.S, and primary dialect composition. The NC group had a slightly higher education level and the dementia group had a higher proportion of male than the other two groups, but these differences were not found significant in the post hoc analyses. The mean MoCA scores were significantly different across the three groups, with the highest being in the NC group and the lowest being in the dementia group.

Table 1.

Characteristics of study participants, by clinical diagnosis (n = 176)

Characteristics	Overall (n = 176)	NC (n = 101; 57.39%)	MCI (n = 48; 27.27%)	DM (n = 27; 15.34%)	P value
Mean age (SD)	72.65 (7.29)	71.65 (7.06)	73.50 (6.84)	74.85 (8.43)	.08
Sex
Male	65 (36.93%)	30 (29.70%)	19 (39.58%)	16 (59.26%)
Female	111 (63.07%)	71 (70.30%)	29 (60.42%)	11 (40.74%)	.02*
Mean years of education (SD)	12.92 (4.21)	13.67 (3.58)	12.06 (5.00)	11.63 (4.43)	.02*
Years in US (SD; n = 162)	32.97 (14.93)	34.14 (15.08)	33.34 (13.86)	27.45 (15.72)	.15
Primary Language
Mandarin	100 (56.82%)	59 (58.42%)	29 (60.42%)	12 (44.44%)
Cantonese	76 (43.18%)	42 (41.58%)	19 (39.58%)	15 (55.56%)	.36
MoCA total score	21.36 (5.86)	24.41 (3.19)	19.92 (4.29)	12.56 (6.22)	<.001**

Abbreviations: DM, dementia; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; NC, normal cognition; SD, standard deviation.

^*p < .05.

^**p < .001.

Demographic Impact on MoCA Total Score

Correlation between MoCA total score and continuous demographic variables were evaluated in the NC group. MoCA total score was found to be correlated to age (r = −.28, p = .005) and education (r = .44, p < .001) but not with years living in the US. Multiple linear regression analysis with all demographic variables (age, years of education, years of living in the U.S., sex, and primary languages) and clinical diagnosis entered as predictors found that MoCA total score was associated with age and education independent of clinical diagnosis (Table 2). Sex, years of living in the U.S., and primary language were not found to have significant predictive power for MoCA total score. Alternative models with interaction terms did not improve the model fit.

Table 2.

Regression analysis summary for demographic variables predicting MoCA total score

Variable	b (SE)	t	P
Education	.39 (.07)	5.78	<.001*
Age	−.14 (.04)	−3.50	<.001*
Years in US	.02 (.02)	1.02	.31
Sex (Male)	.49 (.58)	.84	.40
Primary Language (Mandarin)	.37 (.61)	.61	.55
Clinical Diagnosis (MCI)	−4.07 (.63)	−6.45	<.001*
Clinical Diagnosis (DM)	−9.66 (.84)	−11.47	<.001*
Adjusted R2	.61
F for change in R2	38.23 (P < .001***)

Note. Degree of freedom = 154.

^*p < .001.

Based on these results, we performed stratification and total score correction for education and age in the ensuing cutoff score analyses. Multiple regression analyses conducted in each of the education/age subgroups showed that education/age was no longer a significant predictor of the MoCA total score within each subgroup.

Optimal Cutoff Scores for MCI Detection from NC

The optimal MoCA cutoff scores, defined by the highest Youden index ³⁰, for detecting MCI from NC were calculated in the overall sample and in each age-/education-stratified group (Table 3). In the overall sample, unadjusted MoCA total score yielded an optimal cutoff score of 23, with sensitivity of .79, specificity of .67, and accuracy of .71. The two education adjustments (adjustment 1 added one extra point to individuals with ≤12 years of education based on the established adjustment method ¹; adjustment 2 added one extra point to individuals with 10 to 12 years of education and four extra points to those with ≤9 years of education based on the mean MoCA scores of the NC participants within the current sample) and the age adjustment (two extra points were added to individuals over 75 years old based on the mean MoCA scores of the NC participants within the current sample) all resulted in the same optimal cutoff score of 23 and largely comparable AUC estimates. Education adjustment 2 yielded the highest diagnostic accuracy (.77) and Youden index (.53).

Table 3.

Optimal MoCA cutoffs to distinguish between normal cognition and MCI

	NC		MCI
	n	MoCA M (SD)	n	MoCA M (SD)	Optimal Cutoff *	Sen	Spe	Youden*	Acc*	AUC*
Total Score Adjustment
No Adjustment	101	24.4 (3.19)	48	19.9 (4.29)	≤23 (20.6-23.8)	.79	.67	.46 (.32-.61)	.71 (.66-.83)	.80 (.72-.87)
Edu Adj 1+	101	24.8 (3.03)	48	20.5 (4.02)	≤23 (21.0-24.1)	.77	.74	.51 (.29-.63)	.75 (.66-.83)	.81 (.73-.88)
Edu Adj 2+	101	25.3 (2.84)	48	21.4 (3.59)	≤23 (22.3-24.4)	.75	.78	.53 (.34-.67)	.77 (.69-.84)	.81 (.73-.88)
Age Adj+	101	24.9 (3.10)	48	20.8 (4.11)	≤23 (21.3-23.9)	.73	.74	.47 (.30-.60)	.74 (.68-.83)	.79 (.71-.86)
Education Stratification
≤9	17	21.4 (3.39)	15	16.7 (3.10)	≤18 (16.7-20.1)	.73	.88	.62 (.31-.85)	.81 (.66-.94)	.84 (.68-.97)
9-12	26	24.2 (3.43)	11	18.9 (4.46)	≤22 (18.4-23.3)	.91	.73	.64 (.11-.80)	.78 (.60-.89)	.83 (.68-.94)
>12	58	25.4 (2.39)	22	22.6 (3.16)	≤23 (21.6-24.9)	.59	.83	.42 (.18-.64)	.76 (.63-.86)	.76 (.64-.88)
Age Stratification
52-65	19	24.7 (2.47)	6	19.3 (5.99)	--	--	--	--	--	--
66-75	55	24.9 (2.82)	21	21.5 (3.71)	≤23 (20.6-24.7)	.71	.73	.44 (.16-.64)	.72 (.62-.84)	.78 (.65-.88)
76-94	27	23.1 (4.01)	21	18.5 (3.97)	≤21 (18.2-23.1)	.71	.74	.46 (.20-.70)	.73 (.60-.85)	.79 (.65-.91)

Abbreviations: Acc, accuracy; AUC, area under the ROC curve; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; NC, normal cognition; SD, standard deviation; Sen, sensitivity; Spe, specificity.

Boldface values are recommended cutoff values to use in clinical practice.

^*Parentheses contain 95% confidence intervals with bootstrapping (5,000 iterations; random number seed: 35)

⁺Edu Adj 1: +1 for education ≤12; Edu Adj 2: +1 for education 10-12, +4 for education ≤9; Age Adj: +2 for age > 75.

To better understand the demographic effects, we also performed cutoff score analyses using unadjusted MoCA total score in each education- or age-stratified group (Table 3). We found that optimal cutoff scores of 18 and 22 in the low (≤9) and medium (9-12) education groups, respectively, yielded better diagnostic performance (low education: Youden index = .62, accuracy = .81; medium education: Youden index = .64, accuracy = .78) than using adjusted total score in the aggregated sample. In the high education group, however, an optimal cutoff score of 23 yielded poorer diagnostic performance with a high specificity (.83) yet low sensitivity (.59; Youden index = .42). In terms of age stratification, the youngest age band (52-65) was not included in the analyses due to an insufficient sample size determined by the a priori power analysis. In the other two age bands, optimal cutoff scores of 21 and 23 using unadjusted MoCA total score did not yield better diagnostic performance than the adjusted total score in the aggregated sample.

Optimal Cutoff Scores for Dementia Detection from NC

The optimal MoCA cutoff scores for detecting dementia from controls were also calculated in the overall sample and in each age-/education-stratified group (Table 4). In the overall sample, unadjusted MoCA total score demonstrated excellent diagnostic properties (sensitivity = 1.00, specificity = .83, accuracy = .87). The education adjustments and age adjustment yielded comparable AUCs and Youden indices.

Table 4.

Optimal MoCA cutoffs to distinguish between normal cognition and dementia

	NC		DM
	n	MoCA M (SD)	n	MoCA M (SD)	Optimal Cutoff *	Sen	Spe	Youden*	Acc*	AUC*
Total Score Adjustment
No Adjustment	101	24.4 (3.19)	27	12.6 (6.22)	≤21 (18.0-20.9)	1.00	.83	.83 (.67-.92)	.87 (.84-.97)	.97 (.94-.99)
Edu Adj 1+	101	24.8 (3.03)	27	13.2 (6.19)	≤19 (18.4-21.7)	.85	.95	.80 (.66-.92)	.93 (.84-.97)	.97 (.94-.99)
Edu Adj 2+	101	25.3 (2.84)	27	14.3 (5.84)	≤22 (19.4-22.0)	1.00	.84	.84 (.63-.90)	.88 (.85-.95)	.97 (.94-.99)
Age Adj+	101	24.9 (3.10)	27	13.6 (6.24)	≤21 (18.8-21.8)	.96	.85	.81 (.68-.90)	.88 (.84-.95)	.97 (.94-.99)
Education Stratification
≤9	17	21.4 (3.39)	10	9.5 (6.62)	≤18 (9.7-18.0)	1.00	.88	.88 (.38-.94)	.93 (.70-.96)	.96 (.86-1.00)
9-12	26	24.2 (3.43)	8	15.4 (4.98)	≤21 (14.9-21.0)	1.00	.73	.73 (.21-.88)	.79 (.68-.97)	.93 (.82-1.00)
>12	58	25.4 (2.39)	9	13.4 (5.83)	≤20 (15.8-20.0)	1.00	.97	.97 (.50-.98)	.97 (.91-.99)	.99 (.97-1.00)
Age Stratification
52-65	19	24.7 (2.47)	3	18.0 (2.00)	≤20 (17.2-20.0)	1.00	.89	.89 (.17-1.00)	.91 (.73-1.00)	.96 (.88-1.00)
66-75	55	24.9 (2.82)	10	11.4 (7.71)	≤21 (14.1-21.0)	1.00	.85	.85 (.47-.92)	.88 (.85-.99)	.98 (.93-1.00)
76-94	27	23.1 (4.01)	14	12.2 (5.25)	≤18 (14.7-19.8)	.93	.85	.78 (.50-.90)	.88 (.76-.95)	.95 (.87-1.00)

Abbreviations: Acc, accuracy; AUC, area under the ROC curve; DM, dementia; MoCA, Montreal Cognitive Assessment; NC, normal cognition; SD, standard deviation; Sen, sensitivity; Spe, specificity.

Boldface values are recommended cutoff values to use in clinical practice.

^*Parentheses contain 95% confidence intervals with bootstrapping (5,000 iterations; random number seed: 35)

⁺Edu Adj 1: +1 for education ≤12; Edu Adj 2: +1 for education 10-12, +4 for education ≤9; Age Adj: +2 for age > 75.

In the education- or age-stratified subgroups, unadjusted MoCA score also demonstrated strong diagnostic properties. Specifically, in the education-stratified subgroups, the optimal cutoff scores (low education, 18; medium education, 21; high education, 20) all yielded maximized sensitivity (1.00). Compared to the specificity of unadjusted MoCA score in the aggregated sample (.83), specificity in the low and high education groups were higher (.88 and .97, respectively) while that in the medium education group was lower (.73). In age-stratified groups (52-65, 66-75, 76-94), the optimal cutoff scores of 20, 21, 18, respectively, all demonstrated strong diagnostic performance (sensitivity, .93 to 1.00; specificity, .85 to .89; accuracy, .88 to .91). The results of the youngest age band (52-65) should be treated with caution, though, given the small cell size.

Discussion

The current study was the first study to examine the utility of MoCA as a screener to detect cognitive decline (MCI and dementia) in monolingual Chinese-speaking older adults residing in the U.S. Our study contributed to the rather limited body of literature for the Chinese immigrant population by employing a comparably large sample and examining the effects of demographic variables unique to this population (i.e., participants’ primary dialect and their length of immigration history). Our findings noted that the MoCA total score was impacted by age and education, and not by sex, years of living in the U.S., or primary dialect. Applying appropriate demographic adjustment, we found that MoCA demonstrated excellent diagnostic performance to detect dementia from NC, whereas its performance to detect MCI from NC was found adequate in individuals with low and medium education but not those with high education. Results suggested that, though some patterns were similar, the Chinese immigrant population in the U.S. demonstrated distinct performance on MoCA compared to other Chinese-speaking groups and ultimately required their own validation study. Main findings of several validation studies in Chinese-speaking groups ^13–18, along with a study of non-Asian ethnic minorities residing in the U.S. ⁹, were summarized in Table 5 for comparison.

Table 5.

Optimal MoCA cutoff scores in diverse populations

	Population	Diagnoses	Demographic Stratification		Cutoff	Sensitivity	Specificity	AUC
Mellor et al., 2016	Chinese in mainland China	NC = 701 MCI = 265 AD = 49	Overall (MCI/AD)		22.5/19.5	.87/1.00	.73/.87	.89/.98
			Edu (MCI/AD)	≤6	16.5/14.5	.83/.94	.81/.90	.88/.96
				7-9	22.5/19.5	.74/1.00	.81/.95	.85/.99
				≥10	24.5/19.5	.89/1.00	.80/.96	.87/.87
			Age (MCI/AD)	57-64	24.5/--	.97/--	.73/--	.89/--
				65-69	23.5/--	.88/--	.80/--	.87/--
				70-76	19.5/18.5	.73/.89	.86/.89	.86/.97
				77-80	17.5/19.5	.82/.1.00	.81/.73	.88/.93
				81-97	16.5/11.0	.74/.83	.81/.95	.82/.82
			Gender (MCI/AD)	Male	24.5/19.5	.95/1.00	.63/.91	.89/.98
				Female	22.5/19.5	.89/1.00	.70/.83	.89/.98
Tsai et al., 2016	Chinese in Taiwan	NC = 26 MCI = 59 DM = 57	Overall (MCI/DM)		24/20	.88/.79	.74/.80	.81/.80
Tan et al., 2014	Chinese in mainland China	NC = 4150 MCI = 2311 DM = 984	Age (MCI/DM)	60-79++	25/24	.86/.91	.85/.80	.92/.91
				80-89++	24/21	.85/.82	.96/.82	.94/.90
				≥90++	23/19	.90/.83	.97/.75	.94/.86
Yeung et al., 2014	Chinese in Hong Kong	NC = 49 MCI = 93 DM = 130	Overall (MCI/DM)		21/18	.83/.92	.74/.92	.84/.97
Ng et al., 2013	Primarily Chinese in Singapore	NC = 103 MCI = 49 Mild AD = 60	Edu (MCI/AD)	≤10	26/24	.96/.85	.30/.81	--
				>10	27/25	.94/.90	.19/.70	--
Lu et al., 2011	Chinese in mainland China	NC = 6283 MCI = 1687 DM = 441	Education (ACI)	0	13	.81	.83	--
				1-6	19	.84	.83	--
				≥7	24	.90	.82	--
Milani et al., 2018	Non-Hispanic Blacks in US	NC = 345 MCI = 158	Overall (MCI/DM*)		23/16	.72/.78	.72/.89	.77/.90
			Edu (MCI/DM*)	≤12+	19/13	.56/.78	.84/.96	.72/.92
				13-16	23/17	.73/.78	.75/.93	.77/.87
				>16	23/19	.65/.90	.82/.84	.79/.94
	Hispanics in US	NC = 72 MCI = 45	Overall (MCI/DM*)		24/16	.84/.72	.56/.98	.73/.89
			Edu (MCI/DM*)	≤12+	23/15	.92/.91	.30/1.00	.60/.97
				13-16	24/16	.83/.75	.62/1.00	.77/.86
				>16	24/19	.80/.78	.93/.73	.93/.73

Abbreviations: ACI, all cognitive impairment; AD, Alzheimer’s disease; DM, dementia; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; NC, normal cognition.

⁺Education corrected total score (one extra point for individuals with ≤12 years of education).

⁺⁺Education corrected total score (one extra point for individuals with >6 and ≤12 years of education, two extra point for individuals with ≤6 years of education).

^*Comparison was between MCI and dementia rather than normal control and dementia.

For MCI detection in aggregated sample, MoCA total score with education adjustment based on the current sample (i.e., 1 extra point for individuals with 10 to 12 years of education and four extra points for individuals with ≤9 years of education) yielded the best diagnostic performance with an optimal cutoff score of 23. This score was roughly comparable with samples collected in mainland China ^14,15 and the Hispanic and non-Hispanic Black samples collected in the U.S. ⁹, but was notably lower than a primarily Chinese sample from Singapore ¹⁸ and the commonly used cutoff score of 26 established in samples that were primarily Whites ¹. A closer examination within each education-stratified group showed that the diagnostic performance of MoCA was considerably better in the low (≤9) and medium (9-12) education groups than in the high education (>12) group, where an optimal cutoff score of 23 yielded adequate specificity but low sensitivity. This raised concerns for MoCA’s utility as a screener for MCI in Chinese-speaking individuals with a high level of education, potentially resulting in a high number of false negatives where individuals with true cognitive impairments were screened out and deprived of the opportunity for follow-up assessment. Age stratification for individuals over 65 years old did not demonstrate concerning diagnostic properties within each subgroup; however, it should be noted that we did not have enough sample size to perform an optimal cutoff score analysis in individuals younger than 65 years, rendering cautions to be exercised when drawing conclusion about this age band. In sum, current results could not support the use of MoCA to detect MCI in U.S.-residing Chinese-speaking adults who were under 65 years old and/or with >12 years of education. However, for those who were over 65 years old and/or with ≤12 years of education, MoCA appeared to be a valid screener for MCI. Use of education stratified cutoff scores (18 for ≤9 years of education, 22 for 9-12 years of education) is recommended.

For dementia detection, unadjusted MoCA total score in aggregated sample yielded excellent diagnostic performance with a cutoff score of 21. Education and age adjustment to total score did not improve diagnostic performance. This cutoff was lower than a few Chinese samples ^15,18, but higher than the others ^14,16. With education- or age-stratification, MoCA continued to demonstrate strong diagnostic properties in each subgroup. However, the small sample size for individuals who are under 65 years old is still worth noting. Altogether, MoCA appeared to be a valid screening tool to detect dementia in U.S.-residing Chinese-speaking older adults who were over 65 years old. Use of unstratified and unadjusted MoCA total score with a cutoff of 21 appeared to be sufficient.

An additional aim of this study was to explore the impact of demographic variables on MoCA performance, especially those variables that were unique to the immigrant population. We did not find participants’ primary Chinese dialect (Mandarin vs. Cantonese) and their length of living in the U.S. to be significant predictors of the MoCA performance. However, we acknowledge that the self-reported primary dialect and the length of living in the U.S. were both rather rudimentary proxies for the immigration experience and acculturation. Additional metrics such as acculturation questionnaires and objective assessment of bilingualism should be performed before drawing any conclusion about the role of acculturation on MoCA performance in Chinese-speaking immigrants.

Several other limitations and directions for future research should be considered. First, the current sample size precludes several analyses, including machine learning-based classification and simultaneous stratification for age and education. We plan to continue collecting data and to perform these analyses with adequate sample size. Second, to fully understand the ecological validity of MoCA in the Chinese-speaking population, we plan to include other outcome variables such as longitudinal disease outcomes and neuroimaging markers. Third, the current sample, though a part of a national effort, is comprised of participants who are primarily urban living. Future research should work to collect samples that are more representative of the heterogeneity of the Chinese-speaking population residing in the U.S. and the changing makeup of this population. Randomized data collection that are designed to represent the racial and ethnic composition of the U.S. would greatly benefit this line of research. Relatedly, testing material of both simplified and traditional Chinese should be made available in future studies in order to best approximate participants’ cultural background.