The current study aimed to investigate whether four cognitive tests used to screen for dementia, namely the Mini Addenbrooke’s Cognitive Examination (M-ACE [51]), Cross-Linguistic Naming Test (CLNT [52]), the Rowland Universal Dementia Assessment Scale (RUDAS [34]), the Visual Short-Term Memory Binding Test (VSTMBT [53, 54]), reveal any bias in performance associated to participants’ ethnicity. Additionally, the effects of education, acculturation and language proficiency, as potential drivers of performance bias, were also investigated [55], along with self-rated English proficiency scores. A key methodological characteristic of the study is that we did not use interpreters. We predicted that, since a limited language involvement is required for the VSTMBT, performance on this test would be insensitive to the ethnic background of those assessed, a pattern that would not hold for the other screening tests.

To be eligible, participants had to be aged 60 or over, resident in the UK and needed to meet the following group-specific criteria:

1. EM group: a) identify as belonging to a South Asian EM group, b) speak English as an additional (non-native) language, c) have sufficient mastery of English to understand test instructions and give informed consent. 2. Ethnic majority group: a) identify as belonging to the UK ethnic majority group, b) speak English as a first language. Participants were excluded if they met any of the following criteria: Diagnosis of dementia or any other neurological or health conditions (e.g., stroke, psychiatric illness) that may affect cognitive functioning; disabilities that could interfere with testing or would require special accommodations (e.g., uncorrected vision or hearing problems). No subjects who met the inclusion criteria were excluded due to the above criteria.

To determine whether participants included in each group were cognitively healthy, authors employed a combination of self-report and established inclusion/exclusion criteria. Self-report: Participants were asked to self-report their medical history, current health status, and any cognitive concerns at the start of the assessment. They were asked questions related to memory, attention, language abilities, and overall cognitive functioning. Self-report allows researchers to gather initial information about participants’ cognitive health directly from the individuals themselves.

The final sample consisted of 43 subjects. The EM group comprises South Asian participants (N = 23; but see footnote of Table 1 for missing data), mainly from India and Pakistan, as they represent the largest EM in Scotland at 2.7% of the total population [56] and 2.3% in the UK [57]. Ethnic majority participants (N = 20) were recruited through a combination of convenience, snowball, and purposive sampling.

Education—RQF is a score between 1–8 based on the UK’s Regulated Qualifications Framework (2019). Education Level was calculated by dividing three levels RQF [level 1 ≤ GCSE (General Certificate of Secondary Education); level 2 > GCSE; level 3 is a bachelor’s degree or higher]. Years in the UK, the percentage years in the UK and English proficiency were retrieved from the Multi-Ethnic Acculturation Scale (MAS [55]). English proficiency was recorded by a self-reported scale between excellent and poor from the MAS [55], the only acculturation scale designed for use with individuals from diverse ethnic groups [84]. This scale uses simple, commonly used vocabulary, suitable for linguistic minorities.

Data from neuropsychological tests were compared across groups using either independent-samples t-tests or Mann-Whitney U-tests (see Table S1 for normality checks and Table S2 for non-parametric tests). We reported on the outcomes from parametric tests here and those of non-parametric analysis are shown in Supplementary material. This methodological decision aims to demonstrate that the significance of the results observed was independent of the type of statistical analysis used. Variables from the VSTMBT were compared using a parametric 3-way mixed ANOVA model with Group (EM vs. ethnic majority) as the between-subjects factors, and Condition (shape vs. binding) and Set Size (2 vs. 3) as the within-subjects factors. A Group × Condition interaction will suggest potential cultural bias and this would be followed up with appropriate tests. If proved valid, we calculated the Binding Cost [7] which is the variable considered a marker for Alzheimer’s disease [27, 53, 54]. Finally, stepwise regression models were developed to explore underpinnings of the sensitivity of the screening tests investigated to sociocultural factors. The data were analysed using IBM SPSS Statistics (version 28).

The data collected in this manuscript were obtained in compliance with the Declaration of Helsinki. The study received ethical approval by the University of Edinburgh’s School of Health in Social Science (CLIN623). Every participant provided written consent. The average assessment time was approximately forty minutes.

There was a significant effect of Group [F (1,40) = 6.12, P = 0.018, η² = 0.133]. Both the effect of Condition [F (1,40) = 66.89, P < 0.001, η² = 0.63] and Set Size [F (1,40) = 33.31, P < 0.001, η² = 0.45] were significant. Interestingly, none of the interactions reached the significance threshold [Group × Condition: F (1,40) = 2.43, P = 0.127, η² = 0.06; Group × Set Size: F (1,40) = 0.41, P = 0.524, η² = 0.01; Set Size × Condition: F (1,40) = 0.06, P = 0.805, η² = 0.002; Group × Condition × Set Size: F (1,40) = 0.62, P = 0.805, η² = 0.002] (Figure 1). We, therefore, collapsed performance across Set Size, and used such scores to calculate the Binding Cost [Binding Cost = (Performance on Shape-Only – Performance on Shape-Colour Binding)/Performance on Shape-Only * 100, [85]] which entered further analyses.

Among the neuropsychological tests, the CLNT [t (20.27) = –2.19, P = 0.04], M-ACE [t (29.52) = –3.42, P < 0.001] and RUDAS [t (27.82) = –2.16, P = 0.04] revealed significant differences across Group, while the VSTMBT, particularly the cost of binding [t (40) = 1.16, P = 0.27], did not (Table 1).

To explore the potential underpinnings of the sociocultural sensitivity shown by the neuropsychological screening tests, we ran regression models (Table 2). For the stepwise regression analysis, Group was first included as the predictor of interest (Model 1). A subsequent model included education level, years in the UK, percentage years in the UK, English proficiency, and the MAS as additional regressors (Model 2). Such an approach allowed assessment of whether variability on the cognitive screening tests could be accounted for by the participants’ ethnic background before (Model 1), and after (Model 2) controlling for sociocultural factors.

Ethnicity was a significant predictor of CLNT. This predictive value remained significant after controlling for other sociocultural factors, with percentage of years in the UK and scores on the MAS also retained as significant predictors. Ethnicity significantly predicted performance on the M-ACE and RUDAS. However, after controlling for other sociocultural factors, ethnicity was no longer a significant predictor. The MAS was retained as a predictor of performance on the M-ACE whereas education level, years in the UK, and percentage of years in the UK were retained as predictors of performance on the RUDAS. Neither ethnicity nor other sociocultural factors yielded models that significantly predicted the cost of binding (see Table 2).

Performance on the CLNT seems to be influenced by the percentage of years in the UK. The words found within the CLNT are derived from the Swadesh word list [58], a set of words considered to be universally found in most of the spoken languages. However, major criticisms have arisen about the degree to which the Swadesh words are universal [86]. The principal critique is that the Swadesh word list presumes that there are direct word equivalents between English and the many incorporated languages in the list. For example, the English word “you”, found in the Swadesh vocabulary, does not have a direct one-to-one word equivalent in other languages such as Urdu, a primarily South Asian language spoken by many throughout India and Pakistan [87].

Level of acculturation was a significant predictor of the M-ACE. The significant moderate correlation between acculturation and M-ACE suggests that higher adoption of the dominant culture is related to better test performance above age and education [84]. As culture represents adaptation to live in a specific context [88], what is relevant and worth learning for one individual in a particular cultural context may be less relevant for someone from another cultural background [89].

The M-ACE has been validated in different translated language versions [63–65], but it was not designed as a cross-cultural dementia assessment [51]. Thus, the results for the M-ACE test between the two cultural groups are not particularly surprising. These findings indicate that using the M-ACE in an EM group when administered in English may increase the risk in some ethnic groups performing below clinical cut-off scores, resulting in false diagnoses of cognitive deficit [90].

RUDAS seems affected by percentage of years in the UK and English proficiency. These results do not replicate the findings reported in the literature [34, 36, 67, 70, 71, 91, 92] which indicated that RUDAS scores were not significantly affected by culturally and linguistically diverse status. This discrepancy between the current and earlier studies may be explained by the use of interpreters in earlier studies [20, 37]. Cognitive scores on the RUDAS between the South Asian and British groups differed from previous studies [71]. The RUDAS was originally designed as a cross-cultural assessment for dementia and scores for both groups were expected to be similar [36]. One potential drawback of the RUDAS is the number of action tasks requiring constant feedback from both the tester and subjects (Praxis and Body Orientation Task) [36]. In fact, whereas CLNT and Mini-ACE only had one significant predictor of group differences, RUDAS had three (i.e., percentage of years in the UK, English proficiency and education level).

The VSTMBT has proved to be insensitive to the individual’s ethnicity. The VSTMBT indexes low level of visual functions and makes it language-independent, hence a high level of English proficiency is not required [32, 83]. The task relies on a simple set of instructions easy to understand and follow by people with little formal education and assessed in a language they are not proficient [33, 83]. It is also not affected by prior knowledge, experience, or nor practice effects. Previous studies showed that level of education does not affect the performance which indicates that it may be appropriate to test patients from various educational and socio-cultural backgrounds [32, 54], however, none of these earlier studies considered EMs. Here we have further demonstrated that even the most taxing version of this task (i.e., Set Size 3 [83]) did not render it more challenging for EMs. This is important because it was recently suggested that while the version with three items can aid dementia’s risk screening in asymptomatic stages, the version with two items would help detect Alzheimer’s disease dementia in its mild stage [83]. Our results suggest that EMs in the UK can benefit from such an assessment from the asymptomatic stages to the early clinical stages of the disease.

Our results suggest that a one-size-fits-all approach would not help identify the vulnerabilities of cognitive and neuropsychological assessment to sociocultural factors. Mungas et al. (2009) [93] suggested that such task properties would not only influence our interpretation of the association between sociocultural factors and behaviours, but also our interpretation of the association between brain pathology and behavioural outcomes. Complex influences seem to drive the sensitivity of traditional neuropsychological tasks to sociocultural as demonstrated by regression models. Different tests showed different vulnerabilities to different sociocultural factors, an observation that is in line with [93]. Taken together these results suggest that assessments for dementia risk in EM groups ought to consider a wide range of meaningful sociocultural factors when it comes to the selection of appropriate tools. For example, linguistic barriers significantly influence the accuracy and validity of the cognitive evaluation, despite the presence of a professional interpreter [94]. Our study is innovative because we investigated the effect of ethnicity on neuropsychological test results when administered in a language other than the participant’s native language without the use of interpreters. This allowed us to observe what can happen in the real clinical testing setting where interpreters are neither always available nor reliable [4, 12, 13].

Our study had some limitations. The small sample size may have led to some bias [95] considering that South Asians in the UK are not a homogeneous ethnic group, but include Indians 2.3%, Pakistanis 1.9%, Bangladeshis 0.7% and other Asians [57]. Although the South Asians and British were matched for age, education and gender; occupations between the two groups were not perfectly equivalent. In terms of predictor variables, English proficiency was a self-rated measure of language competence. As such, this proxy variable may not necessarily be reflective of the participants’ true level of English literacy in which self-knowledge bias may play a role [96]. EM differ from the majority group more than language. A range of variables are likely to contribute to the difference in test performance, such as acculturation [85], education [97] or socio-economic status [98]. The contributions of the latter variable cannot be fully disentangled by the present study because it was not collected for ethical reasons. Education was measured based on stratification of years of education. However, it is unlikely that quantitative variations in education reflect qualitative differences in education across countries with different education systems [99].