Validation of a Latent Construct for Dementia in a Population-Wide Dataset from Singapore

Abstract

Background: The latent variable δ has been proposed as a proxy for dementia. Previous validation studies have been conducted using convenience samples. It is currently unknown how δ performs in population-wide data.

Objective: To validate δ in Singapore using population-wide epidemiological study data on persons aged 60 and above.

Methods: δ was constructed using items from the Community Screening Instrument for Dementia (CSI’D) and World Health Organization Disability Assessment Schedule (WHODAS II). Confirmatory factor analysis (CFA) was conducted to examine δ model fit. Convergent validity was examined with the Clinical Dementia Rating scale (CDR) and GMS-AGECAT dementia. Divergent validity was examined with GMS-AGECAT depression.

Results: The δ model demonstrated fit to the data, χ²(df) = 249.71(55), p < 0.001, CFI = 0.990, TLI = 0.997, RMSEA = 0.037. Latent variable δ was significantly associated with CDR and GMS-AGECAT dementia (range: β= 0.32 to 0.63), and was not associated with GMS-AGECAT depression. Compared to unadjusted models, δ model fit was poor when adjusted for age, gender, ethnicity, and education.

Conclusion: The study found some support for δ as a proxy for dementia in Singapore based on population data. Both convergent and divergent validity were established. In addition, the δ model structure appeared to be influenced by age, gender, ethnicity, and education covariates.

Keywords

Cognition dementia functional status Singapore validation studies

INTRODUCTION

Dementia is defined as a clinical syndrome caused by neurodegeneration, characterized by progressive deterioration in cognition in domains such as memory, learning, orientation, language, and judgment [1]. In older adults aged 60 years and above, the global prevalence of dementia was estimated to be between 5 to 7% based on the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV), the International Classification of Diseases, tenth edition (ICD-10), or similar clinical criteria [2, 3]. Importantly, it has been recognized that cognitive impairments in dementia are often accompanied with functional impairments and increasing need for care [3]. Given that deficits in cognition and functioning are both key characteristics of dementia, Royall et al. proposed a latent variable δ as a proxy for “cognitive correlates of functional status” [4]; specifically, δ indicates the dementia phenotype, and aims to tease apart general cognitive ability (g) [5] into domains associated (δ) or unassociated (g’) with functional status (f). This differentiation may be important for a more targeted identification of dementia caseness beyond the use of a general battery of cognitive assessments.

Latent variable modeling has been proposed to be an improved estimation method for dementia identification using a battery of cognitive tests [6]. Similarly, Royall et al. used bifactor latent modeling to construct δ, but with the use of both objective cognitive tests and measures of Activities of Daily Living (ADL), to estimate dementia caseness that is free from measurement error [4, 7]. In particular, the latent modeling approach to dementia is versatile; it allows δ to be constructed from various combinations of cognitive and functional status measures and can be tested in different populations and contexts [7, 8].

Many studies have demonstrated that δ can serve as a potential marker for dementia. Specifically, δ has been demonstrated to achieve good discrimination between older adults with and without clinically diagnosed dementia [7 –10]. Secondly, δ has shown convergent validity with dementia severity status and discriminant validity with global mental state and geriatric depression measures [7 , 11]. It has also been shown to mediate the association between executive function and ADL [12]. Thirdly, δ has been associated with biomarkers of dementia, including abnormal cerebrospinal fluid amyloid-β/tau ratio [13], serum protein biomarkers [14, 15], and regional gray matter atrophy in the default mode network [11]. Furthermore, δ has showed longitudinal associations with cognitive decline and dementia status[4 , 16].

Given the global prevalence of dementia and the potential clinical utility of δ in identifying dementia, it is thus important to investigate if δ is valid and applicable in various populations. This is especially significant for Singapore given its rapidly aging and multi-ethnic population composition. Singapore is a country in Southeast Asia with a resident population of 3.9 million of which 74.3% are Chinese, 13.3% Malays, 9.1% Indians, and 3.2% of other ethnicities [17]. Recent epidemiological data from the Well-being of the Singapore Elderly (WiSE) study estimated the prevalence of dementia in older adults in Singapore to be 4.6% based on the DSM-IV and 10% based on 10/66 dementia criteria [18].

The present study aims to validate δ in Singapore using the WiSE study data. The WiSE study included population-wide comprehensive data on cognition and functional status which can be used to construct δ, as well as dementia and depression outcomes which can be used to validate δ. Cognitive variables included a battery of objective tests from the Community Screening Instrument for Dementia (CSI’D) [19], while variables on functioning included self and informant-reported measures on ADL limitation. The use of CSI’D to construct δ is ideal as it is one of the key measures used in establishing 10/66 dementia in Singapore [18] and in low and middle income countries [20]. Given that previous studies on δ have used convenience samples [21], the validation of δ using the WiSE study data provides an opportunity to examine δ’s performance in population-wide epidemiological data.

METHODS

Participants and procedure

The present study used data from the WiSE study, a single-phase population-based survey on older adults in Singapore, conducted between August 2012 and November 2013. Participants were older adult Singapore residents aged 60 and above; older adults in day care centers, nursing homes, and institutional care were also included in the study. A total of 2,565 older adults and 2,421 informants participated in the study. Details of the study have been previously reported elsewhere [18]. Briefly, the 10/66 protocol [20] was adopted for the WiSE study. The 10/66 protocol includes a comprehensive set of cognitive and functional assessment questionnaires designed to assess dementia (measures as described below). Participants and informants completed a set of 10/66 questionnaires, which were administered by trained field interviewers. Regular field observations were conducted throughout the study period to monitor interview and data quality. Although English is the predominant language of communication in Singapore, other languages commonly used by older adults include Mandarin, Malay, Tamil, and three major dialects. The 10/66 questionnaires are available in English, Mandarin, and Tamil, while Malay and dialect versions were translated, cognitively tested, and adapted for local use. In the context of Singapore, 10/66 derived dementia has been validated against gold standard clinical assessment by experienced clinicians based on DSM-IV criteria for dementia [18].

Cognitive assessment

A cognitive test battery derived from the CSI’D [19, 22] was used to assess cognitive functioning across nine domains:

Memory recall was measured using the Consortium to Establish a Registry for Alzheimer’s Dementia word-list learning task (1. WLLT) and the East Boston Story (2. EBS) [19, 22]. The WLLT required participants to learn a list of 10 words over 3 trials; performance was assessed with delayed recall. The EBS required participants to recall as many details as they can after listening to a short story.

Language expression was measured by three CSI’D domains: abstract thinking (3. ABST), agnosia (4. AGNOS), and aphasia (5. APHAS) [19, 22]. Abstract thinking was measured based on participants’ ability to describe 4 items (bridge, hammer, church, pharmacy). Agnosia was measured based on the ability to name 7 objects displayed to them (e.g., pencil, watch, chair). Aphasia was measured based on the ability to repeat the phrase ‘no ifs, ands, or buts’ with clear enunciation.

Verbal fluency was measured by the Consortium to Establish a Registry for Alzheimer’s Dementia animal naming verbal fluency task (6. NVFT) [19, 22]; participants were to name as many different animals as they can in one minute.

Language comprehension (7. COMPRE) was measured by three CSI’D motor tasks [19, 22]; performance was assessed based on participants’ ability to respond to a set of instructions (e.g., nod your head, point first to the window and then to the door).

Constructional ability was measured using the CSI’D domain of apraxia (8. APRAX) [19, 22]; performance was assessed based on participants’ graphomotor ability to copy and draw two sets of overlapping shapes.

Orientation was measured by two CSI’D domains of orientation (9. ORIENT) to time and place [19, 22]. Orientation to time included ‘year’, ‘month’, ‘day’, and estimated time of day. Orientation to place included 5 items modified for local relevance (‘country’, ‘prime minister’, ‘streets’, ‘local market/store’, and ‘address’).

Functional assessment

Informant-reported functional status was measured based on informant-rated care needs (CARE) and ADL limitation. Care needs of the older adult were measured using a single item rated from 1 (completely independent) to 3 (needs care much of the time). ADL items were obtained from the CSI’D informant interview [19, 22]. Instrumental ADL items included the ability to perform household chores, and handle finances; items were rated from 0 (no impairment) to 2 (specific incapacity). Basic ADL items included the ability to eat, dress, and toilet; items were rated from 0 (no impairment) to 3 (specific incapacity).

Self-reported functional status was measured using items from the World Health Organization Disability Assessment Schedule (WHODAS II) [23]. Disability domains included the ability to ambulate, bathe, and dress, perform household chores, participate in community activities, and perform day-to-day activities; items were rated from 0 (no impairment) to 4 (specific incapacity).

Clinical assessment

Dementia was assessed using the CDR [24]. The CDR is based on informant-report of the older adults’ cognitive ability to function in 6 domains: memory, orientation, judgment and problem solving, community affairs, home and hobbies, and personal care. The scale is rated 0 (none), 0.5 (questionable), 1 (mild), 2 (moderate), and 3 (severe).

Dementia and depression was assessed using the Geriatric Mental State (GMS) package [25, 26]. The GMS is a semi-structured computer-assisted clinical interview. At stage 1, symptom scores for psychiatric diagnoses were generated based on a computerized algorithm (AGECAT). At stage 2, dementia and depression caseness were assigned based on a hierarchical algorithm. The GMS-AGECAT has been previously used in Singapore [27, 28] and elsewhere as part of the 10/66 dementia protocol [20]. The study used stage 2 GMS-AGECAT identified cases or subcases of dementia and depression respectively.

Statistical analyses

All estimates were weighted to ensure that findings were representative of the elderly population in Singapore during the study period. Analyses were performed using SAS 9.2 survey data analysisprocedures [29] and Mplus version 5.21 [30]. Three measurement models were tested (Fig. 1): a 2-factor (model 1), a bifactor model (model 2), and a 1-factor model (model 3). The 2-factor model included latent variables g and f. The bifactor model, as proposed by Royall et al. [7], included orthogonal latent variables δ, g’, and f’ (where factor f’ indicates variance in functional status unassociated with dementia). The 1-factor model included a single latent variable δ. Confirmatory factor analyses (CFA) were conducted to examine if the bifactor model provided better fit compared to the 2-factor and 1-factor models. Factor variances of the latent variables were fixed to 1 to allow all factor loadings to be estimated. Weighted least squares (WLSMV) with means and variance adjusted estimation was used as many of the variables were measured on a scale with binary and ordinal response categories and normal distribution was not assumed. Compared to maximum likelihood estimation, WLSMV estimation is a more robust approach for analyses with a combination of continuous, binary, and ordinal variables; WLSMV estimates the link between discrete variables and the continuous latent variable based on probit or ordered probit models [30 –32]. There was minimal missing data (5.6%) and this was addressed using pairwise deletion in accordance with Mplus WLSMVestimation.

Based on structural equation modeling, the final model was then validated against the CDR (scores≥1.0) and GMS-AGECAT dementia (yes/no) respectively for convergent validity; followed by GMS-AGECAT depression (yes/no) for divergent validity. δ was hypothesized to be positively associated with CDR and GMS-AGECAT dementia, but not associated with GMS-AGECAT depression. To explore if the final model and its associations with CDR and GMS-AGECAT dementia is influenced by demographic variables, unadjusted models were compared against models adjusted for age, gender, ethnicity, and education. Age (continuous variable), gender (categorical; male/female), ethnicity (categorical; Chinese, Malay, Indian, other ethnicities), and educational attainment (continuous variable; no education to tertiary education) were included as covariates in adjusted models for CDR, GMS-AGECAT dementia, and GMS-AGECAT depression, respectively.

Goodness-of-fit indices

Model fit was assessed using four indices: a nonsignificant χ², comparative fit index (CFI) of above 0.95, Tucker Lewis index (TLI) of above 0.95, and root-mean-square-error of approximation (RMSEA) of below 0.05 [33 –35]; notably, there is a tendency for χ² to be significant in large samples. All fit indices were considered when assessing modelgoodness-of-fit.

Ethical approval

The study protocol was approved by the National Healthcare Group of Singapore Domain Specific Review Board and the SingHealth Centralized Institutional Review Board, and conducted in accordance with the declaration of Helsinki. All participants provided written informed consent. If participants were unable to provide informed consent, written informed consent was obtained from a legally acceptable representative or next-of-kin.

RESULTS

Descriptive statistics from the WiSE study sample are presented in Table 1. The majority of the participants were aged 60 to 74 years, and of Chinese ethnicity; gender distribution was fairly even. Based on the CDR, 91.5% did not have dementia or had questionable dementia, and 8.6% had mild to severe dementia. Based on stage 2 GMS-AGECAT, 15.4% met criteria for dementia, and 11.9% met criteria for depression.

Table 2 presents results from CFA. The 2-factor model demonstrated fit to the data based on CFI and TLI indices but not for RMSEA, χ² = 751.17, df = 53, p < 0.001, CFI = 0.964, TLI = 0.990, RMSEA = 0.072. All cognitive variables loaded significantly on g (range: λ= 0.50 to 0.86, p < 0.001), and all variables on functioning loaded significantly on f (range: λ= 0.87 to 0.99, p < 0.001). In addition, g was strongly correlated with f (r = 0.81, p < 0.001).

The bifactor model demonstrated fit to the data based on CFI, TLI, and RMSEA indices, χ² = 249.71, df = 55, p < 0.001, CFI = 0.990, TLI = 0.997, RMSEA = 0.037. All cognitive variables loaded significantly on δ (range: λ= –0.48 to –0.82, p < 0.001). Compared to model 1, factor loadings on g’ (model 2) were reduced (range: λ= 0.03 to 0.72), and variable WLLT became marginally significant while APRAX became non-significant. Similarly, all variables on functioning loaded significantly on δ (range: λ= 0.70 to 0.97, p < 0.001). Compared to model 1, factor loadings on f’ (model 2) were also reduced (range: λ= 0.08 to 0.61) and ADL finances became non-significant.

Compared to 2-factor and bifactor models, the 1-factor model demonstrated poorer fit to the data on all indices, χ² = 1134.93, df = 49, p < 0.001, CFI = 0.944, TLI = 0.983, RMSEA = 0.093. Factor loadings on δ were significant for all variables (cognitive variables range: λ= –0.47 to –0.77, p < 0.001; variables on function range: λ= 0.86 to 0.995, p < 0.001).

The superior bifactor model was subsequently validated against the CDR, GMS-AGECAT dementia and depression outcomes, respectively (Fig. 2). Table 3 presents results from unadjusted and adjusted models. Convergent validity was demonstrated with both unadjusted and adjusted models on CDR and GMS-AGECAT dementia. Firstly, CDR dementia was associated with g’ (unadjusted β= –0.47; adjusted β= –0.44) and δ (unadjusted β= 0.53; adjusted β= 0.44). CDR was marginally associated with f’ in the adjusted model (adjusted β= –0.05) and not associated with f’ in the unadjusted model (unadjusted β non-significant). In addition, model fit with CDR was poorer for the adjusted model compared to unadjusted model. Importantly, g’ appeared to account for a substantial proportion of CDR variance similar to δ. This observation appeared to be inconsistent with previous studies as g’ was expected to account for relatively little CDR variance compared to δ. To clarify this finding, the bifactor model was regressed on observed CDR scores (0 to 3) instead of dichotomized CDR scores (≥1.0). Accordingly, results showed that δ was significantly and more strongly associated with CDR dementia (unadjusted β= 0.63; adjusted β= 0.54) compared to g’ (unadjusted β= –0.38; adjusted β= –0.34). Similarly, CDR was marginally associated with f’ (adjusted β= –0.08; unadjusted β non-significant) and model fit was poorer for the adjusted model compared to unadjusted model.

Secondly, GMS-AGECAT dementia was associated with all latent variables. Specifically, GMS-AGECAT dementia was more strongly associated with δ (unadjusted β= 0.56; adjusted β= 0.32) compared to g’ (unadjusted β= 0.06; adjusted β= 0.05) and f’ (unadjusted β= –0.19; adjusted β= –0.13). Similarly, model fit with GMS-AGECAT dementia was poorer for the adjusted model compared to unadjusted model.

Divergent validity was demonstrated with both unadjusted and adjusted models on GMS-AGECAT depression. Both δ and g’ were not associated with GMS-AGECAT depression (p > 0.05), while f’ showed weak associations with GMS-AGECAT depression in both unadjusted and adjusted models.

DISCUSSION

The present study aimed to construct and validate a latent model for dementia (δ), as proposed by Royall et al. [7], in Singapore. We constructed δ using objective cognitive tests and informant-reported functional status from the CSI’D, and also self-reported functional status from the WHODAS II. Results from this study demonstrated some support for the validity of the latent model δ in Singapore. To the best of our knowledge, this is the first study to apply δ to a population-wide epidemiological dataset (i.e., WiSE study; [18]); all previous studies have used institutional datasets such as the Texas Alzheimer’s Research and Care Consortium [7 , 36], University of Kansas Brain Aging Project [11], National Alzheimer’s Coordinating Center [9], or convenience samples from university memoryclinics and selected communities for the elderly [4, 10, 12, 13].

Consistent with previous research [7 –9], CFA from this study provided support for a bifactor δ model. Specifically, model fit for a bifactor δ model was improved or better compared to a 2-factor model (with latent variables for cognitive and functional status) and a 1-factor model (with latent variable δ). Firstly, this suggests that the bifactor δ model may be applicable to population-wide data, and in samples with demographics distinct from western countries. Secondly, the results also suggest that δ can be constructed using different cognitive and functional status measures. Previous studies have constructed δ using various (1) cognitive measures such as the Mini-Mental State Exam, Boston Naming Test, Weschsler’s intelligence/memory scales, clock-drawing, and trail making tasks, and (2) functional status measures such as the Functional Activities Questionnaire, Older Adults Resources Scale, the Alzheimer’s Disease Cooperative Study Activities of Daily Living Scale for Mild Cognitive Impairment, Physical Performance Test, and other ADL measures [4 , 7–12], while this study used the CSI’D and WHODAS II. Despite differences in measurement, the δ model achieved robust model fit. This supports the view that a latent modeling approach to dementia allows δ to be constructed from various combinations of cognitive and functional status measures [7, 8].

More importantly, δ demonstrated both convergent validity with clinical indicators of dementia and divergent validity with depression; δ was significantly associated with CDR and GMS-AGECAT dementia, and not associated with GMS-AGECAT depression. Moreover, δ was more strongly associated with CDR (observed scores) and GMS-AGECAT dementia compared to latent variables for cognition (g’) and function (f’). Given that cognitive impairments in dementia have been associated with functional impairments and increased need for care [3], the current finding on δ was consistent with previous studies which showed that the association between δ and dementia was stronger compared to the association between g’ and dementia [7–11 , 37]. Notably however, both g’ and δ accounted for a substantial amount variance in CDR when CDR was dichotomized based on scores≥1.0. This was unexpected and different from findings with observed CDR scores and GMS-AGECAT dementia, and with previous studies. The apparent inconsistent findings may be due to the way CDR was dichotomized (i.e., no/questionable dementia versus mild/moderate/severe dementia) such that there may be inaccuracies or difficulties in discriminating between no/questionable dementia and mild dementia in the population. In addition, previous studies on δ have had some challenges in discriminating mild cognitive impairment (MCI) with healthy controls [10, 15]. Thus, it may be worthwhile for future research to extend on the current findings and investigate δ’s performance in discriminating among dementia, MCI, and healthy cognition based on population data. Notwithstanding the limitation, the current study provided evidence in support of δ as an indicator of dementia based on both convergent and divergent validity.

In addition, δ did not demonstrate model invariance after accounting for age, gender, ethnicity, and education covariates. Although the bifactor δ model demonstrated fit to the data for unadjusted models on dementia (CDR and GMS-AGECAT dementia), model fit indicators were substantially weakened and/or reduced to non-significant when adjusted for demographic variables (age, gender, ethnicity, and education). This suggests that the δ model structure may vary across age, gender, ethnicity, and/or education. One reason may be that the δ model structure varies across ethnicity. In the validation study by Royall et al. [7], the δ model was adjusted for age, gender, and education, but not ethnicity. Further studies found potential moderating effects of ethnicity on δ and dementia indicators: (1) dMA (δ homolog) and CDR were more strongly associated in Non-Hispanic Whites compared to Hispanics [8]; (2) the association between d (δ homolog) and serum protein biomarkers of dementia was stronger in Non-Hispanic Whites compared to Hispanics [14]; (3) the effect of ethnicity persisted despite constructing δ for cross-ethnicity factor invariance (δ homolog “d(=)”); the association between d(=) and CDR/serum protein biomarkers of dementia remained stronger in Non-Hispanic Whites compared to Hispanics [15]. In the context of Singapore, previous research from the WiSE study data found that Indians had lower odds of dementiacompared to the Chinese, while no gender differences were observed [18]. Another reason may be that the δ model structure varies by gender. There may be gender role differences in terms of ADL measures such as housekeeping and managing finances. Accordingly, closer examination of the data showed that compared to men, women had significantly greater impairments in functioning related to housekeeping, managing finances, and community participation. While the examination of measurement invariance is beyond the scope of the current study, the findings suggest that further research is needed to test measurement invariance by comparing δ’s performance across subgroups by gender (men and women) or by ethnicity.

Certain limitations should be noted for this study. Firstly, although this study used two converging indicators for dementia, a clinical gold standard based on clinician diagnosis was not used as data was only available on a smaller sample subset; it is not known how δ would perform against clinical diagnosis. This however would have been impractical given the epidemiological study design. Secondly, differences within types of dementia were not examined in this study (e.g., Alzheimer’s disease versus vascular dementia); it is unclear if δ’s performance is similar in Alzheimer’s disease and all-cause dementia for Singapore’s population. Lastly, this study used a cross-sectional design; prospective longitudinal studies are needed to examine association between δ and cognitive decline in Singapore.

The validation of δ in a population-wide dataset has implications for epidemiological studies on dementia. Firstly, a latent variable approach to dementia case-finding may be a feasible and cost-effective method to screen for dementia in population-based samples, without the need for formal clinician assessment or the administration of a full battery of cognitive tests. Secondly, it is also a flexible method as it allows dementia to be estimated from various cognitive and functional status measures as available in a community or country. As demonstrated in this study, δ was constructed and validated using the items from the CSI’D and WHODAS II from an existing population-based dataset. Although we validated δ in a high-income country (i.e., Singapore), it may be possible to translate current findings to low- and middle-income countries. This is because the CSI’D has been previously validated in low and middle income countries [22, 38]. Our approach in constructing δ suggests a more efficient way to obtain preliminary estimates of dementia prevalence in underserved communities, where access to formal clinician diagnoses or biomarker facilities may be limited or unavailable. Importantly, further studies are needed to improve δ’s performance in detecting clinically significant dementia based on diverse population-based data.

To conclude, the present study validated a latent model for dementia in Singapore. The δ model demonstrated fit to a population-wide epidemiological data. δ demonstrated convergent validity with CDR and GMS-AGECAT dementia, and divergent validity with GMS-AGECAT depression.

Footnotes

ACKNOWLEDGMENTS

The WiSE study was funded by the Ministry of Health, Singapore, and the Singapore Millennium Foundation of the Temasek Trust. The funding organizations had no involvement in the study design, collection, analysis, or interpretation of data, writing the manuscript, and the decision to submit the manuscript for publication.

Authors’ disclosures available online ().

References

Alzheimer’s Disease International (2009) World Alzheimer Report 2009.

Prince

, Bryce

, Albanese

, Wimo

, Ribeiro

, Ferri

(2013) The global prevalence of dementia: A systematic review and metaanalysis. Alzheimers Dement 9, 63–75.e2.

Alzheimer’s Disease International (2015) World Alzheimer Report 2015: The Global Impact of Dementia - An analysis of prevalence, incidence, cost and trends.

Royall

, Palmer

(2012) Getting past g: Testing a new model of dementing processes in dementia. J Neuropsychiatry Clin Neurosci 24, 37–46.

Spearman

(1904) General Intelligence, objectively determined and measured. Am J Psychol 15, 201–292.

Shen

, Crane

, Gao

(2003) A latent variable model approach for assembling and scoring screening tests for dementia. J Alzheimers Dis 5, 399–407.

Royall

, Palmer

, O’Bryant

(2012) Validation of a latent variable representing the dementing process. J Alzheimers Dis 30, 639–649.

Royall

, Palmer

(2013) Validation of a latent construct for dementia case-finding in Mexican-Americans. J Alzheimers Dis 37, 89–97.

Gavett

, Vudy

, Jeffrey

, John

, Gurnani

, Adams

(2015) The δ latent dementia phenotype in the uniform data set: Cross-validation and extension. Neuropsychology 29, 344–352.

10.

Royall

, Palmer

, Matsuoka

, Kato

, Taniguchi

, Ogawa

, Fujimoto

, Okamura

, Shibata

, Nakamura

, Nakaaki

, Koumi

, Mimura

, Fukui

, Narumoto

(2015) δ scores are exportable across cultural and linguistic boundaries. J Alzheimers Dis 49, 561–570.

11.

Royall

DR1

, Palmer

, Vidoni

, Honea

, Burns

(2012) The default mode network and related right hemisphere structures may be the key substrates of dementia. J Alzheimers Dis 32, 467–478.

12.

Royall

, Palmer

(2014) Executive functions cannot be distinguished from general intelligence: Two variations on a single theme within a symphony of latent variance. Front Behav Neurosci 8, 369.

13.

Koppara

, Wolfsgruber

, Kleineidam

, Schmidtke

, Frolich

, Kurz

, Schulz

, Hampel

, Heuser

, Peters

, Reischies

, Jahn

, Luckhaus

, Hull

, Gertz

, Schroder

, Pantel

, Rienhoff

, Ruther

, Henn

, Wiltfang

, Maier

, Jessen

, Kornhuber

, Wagner

(2015) The latent dementia phenotype δ is associated with cerebrospinal fluid biomarkers of Alzheimer’s disease and predicts conversion to dementia in subjects with mild cognitive impairment. J Alzheimers Dis 49, 547–560.

14.

Royall

, Palmer

(2014) Does ethnicity moderate dementia’s biomarkers? Neurobiol Aging 35, 336–344.

15.

Royall

, Palmer

(2014) Ethnicity moderates dementia’s biomarkers. J Alzheimers Dis 43, 275–287.

16.

Palmer

, Royall

(2015) Future dementia severity is almost entirely explained by the latent variable δ’s intercept and slope. J Alzheimers Dis 49, 521–529.

17.

Department of Statistics Singapore (2015) Population Trends, 2015.

18.

Subramaniam

, Chong

, Vaingankar

, Abdin

, Chua

, Eng

, Heng

, Hia

, Huang

, Jeyagurunathana

, Kua

, Lee

, Mahendran

, Magadi

, Malladi

, McCrone

, Pang

, Picco

, Sagayadevan

, Sambasivam

, Seng

, Seow

, Shafie

, Shahwan

, Tan

, Yap

, Zhang

, Ng

, Prince

(2015) Prevalence of dementia in people aged 60 years and above: Results from the WiSE Study. J Alzheimers Dis 45, 1127–1138.

19.

Hall

, Hendrie

, Rodgers

, Osuntokun

, Brittain

, Norton

, Prince

, Pillay

, Blue

, Kaufert

, Nath

, Shelton

, Postl

(1993) The development of a dementia screening interview in two distinct languages. Int J Methods Psychiatry 3, 1–28.

20.

Prince

, Ferri

, Acosta

, Albanese

, Arizaga

, Dewey

, Gavrilova

, Guerra

, Huang

, Jacob

, Krishnamoorthy

, McKeigue

, Rodriguez

, Salas

, Sosa

, Sousa

RMM

, Stewart

, Uwakwe

(2007) The protocols for the 10/66 dementia research group population-based research programme. BMC Public Health 7, 165.

21.

Royall

(2015) Welcome back to your future: The assessment of dementia by the latent variable δ. J Alzheimers Dis 49, 515–519.

22.

Hall

, Ogunniyi

, Hendrie

, Osuntokun

, Hui

, Musick

, Rodenberg

, Unverzagt

, Guerje

, Baiyewu

(1996) A cross-cultural community based study of dementias: Methods and performance of the survey instrument Indianapolis, USA, and Ibadan. Nigeria Int J Methods Psychiatr Res 6, 129–142.

23.

World Health Organization (2000) World Health Organization disability assessment schedule: WHODAS II. Phase 2 field trials. Health services research, World Health Organization, Geneva/.

24.

Morris

(1997) Clinical dementia rating: A reliable and valid diagnostic and staging measure for dementia of the Alzheimer type. Int Psychogeriatr 9(Suppl 1), 173–178.

25.

Copeland

, Dewey

, Griffiths-Jones

(1986) A computerized psychiatric diagnostic system and case nomenclature for elderly subjects: GMS and AGECAT. Psychol Med 16, 89–99.

26.

Copeland

, Prince

, Wilson

KCM

, Dewey

, Payne

, Gurland

(2002) The Geriatric Mental State Examination in the 21st century. Int J Geriatr Psychiatry 17, 729–732.

27.

Kua

(1992) A community study of mental disorders in elderly singaporean chinese using the gms-agecat package. Aust N Z J Psychiatry 26, 502–506.

28.

Kua

, Ko

(1995) Prevalence of dementia among elderly Chinese and Malay residents of Singapore. Int Psychogeriatrics 7, 439–446.

29.

SAS Institute Inc (2008) SAS/STAT^® 9.2 User’s Guide., SAS Institute Inc., Cary, NC.

30.

Muthen

, Muthen

(2010) Mplus user’s guide, Muthen & Muthen, Los Angeles, CA.

31.

Brown

(2006) Confirmatory factor analysis for applied research. Guildford, New York.

32.

Muthen

(1984) A general structural equation model with dichotomous, ordered categorical, and continuous latent variable indicators. Psychometrika 49, 115–132.

33.

Bentler

(1990) Comparative fit indexes in structural models. Psychol Bull 107, 238–246.

34.

Browne

, Cudeck

(1992) Alternative ways of assessing model fit. Sociol Methods Res 21, 230–258.

35.

, Bentler

(1999) Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equ Model 6, 1–55.

36.

Royall

, Palmer

(2016) Thrombopoietin is associated with δ’s intercept, and only in Non-Hispanic Whites. Alzheimers Dement (Amst) 3, 35–42.

37.

Royall

, Palmer

, Matsuoka

, Kato

, Taniguchi

, Ogawa

, Fujimoto

, Okamura

, Shibata

, Nakamura

, Nakaaki

, Koumi

, Mimura

, Fukui

, Narumoto

(2015) Greater than the sum of its parts: Delta can be constructed from item level data. J Alzheimers Dis 49, 571–579.

38.

Hall

, Gao

, Emsley

, Ogunniyi

, Morgan

, Hendrie

HC.

(2000) Community screening interview for dementia (CSI “D”); performance in five disparate study sites. Int J Geriatr Psychiatry 15, 521–531.