Utility of Self-Rated vs. Informant-Rated Ascertain Dementia-8 for Detection of Early Cognitive Impairment: Experience of a “Real-World” Memory Clinic
Article information
Abstract
Background
The Ascertain Dementia 8-item Questionnaire (AD8) is a validated informant-based interview for early dementia detection. Research suggests the utility of self-rated AD8 to identify milder dementia forms in research settings. This study compares the factor structure, reliability, and diagnostic performance between AD8-Self and AD8-Informant for early cognitive impairment (ECI) in a clinical setting.
Methods
Five hundreds fifteen patient-informant dyads (43 cognitively intact and 472 ECI) from a tertiary memory clinic completed both self and informant-reported AD8. We conducted exploratory factor analysis to determine the factor structure, Cronbach’s alpha for internal consistency, and receiver operating characteristic (ROC) curve analysis for ECI, including a subgroup analysis for mild cognitive impairment (MCI).
Results
The mean age and education of ECI participants were 75.61 years (range, 51–95) and 5.6 years (range, 0–20), respectively, and 72.6 years (range, 51–89) and 6.8 years (range, 0–16) in the MCI subgroup. Unlike AD8-Informant’s one-factor structure, AD8-Self had a two-factor structure corresponding to memory and non-memory domains. AD8-Self demonstrated lower reliability (Cronbach’s alpha: ECI 0.666 vs. 0.764; MCI 0.663 vs. 0.709). In ECI, AD8-Informant (cutoff score ≥3) showed better diagnostic performance (sensitivity 89%, specificity 79%) than AD8-Self (cutoff score ≥4; sensitivity 27.1%, specificity 95.3%) (AUC 0.915 vs. 0.593; p<0.001). Similar results were found in MCI (sensitivity 64.7% vs. 26.5%; specificity 79.1% vs. 95.3%; AUC 0.745 vs. 0.600; p=0.002).
Conclusion
AD8-Self has a distinct factor structure, lower reliability, and inferior diagnostic performance compared to AD8-Informant for ECI/MCI detection. Our result do not support AD8-Self as a standalone tool for detecting ECI or MCI.
INTRODUCTION
More than 55 million people currently live with dementia worldwide, with over 60% residing in low- and middle-income countries.1) Nearly 10 million new cases are diagnosed annually, making dementia a major global public health priority. In line with the global trend of increasing dementia prevalence due to population aging, the prevalence of dementia in Singapore was reported to be 8.8% among older persons aged 60 and above, according to the second Well-Being of the Singapore Elderly (WISE) study (2023).2)
Aligned with the Global Action Plan on the Public Health Response to Dementia 2017–2025, Singapore’s National Dementia Strategy emphasizes timely diagnosis and post-diagnostic support for persons living with dementia and their caregivers. As the diagnosis of dementia remains clinical in nature, identifying intra-individual cognitive and functional decline from baseline—based on input from reliable informants and objective cognitive testing—remains the cornerstone for early diagnosis and disease staging.3) This underscores the need for assessment tools with strong diagnostic performance and reliability that are also easy to administer and feasible for use in clinical practice.
The Washington University AD8 (Ascertain Dementia 8-item Informant Questionnaire) is a brief tool that captures subjective cognitive symptoms via informants. Initially designed to distinguish early dementia from normal cognition, it assesses intra-individual changes in memory (e.g., consistent memory problems, repetition, forgetting appointments), temporal orientation, judgment (e.g., decision-making, managing finances), and function (e.g., reduced interest in activities, difficulty using appliances).4) It has been translated and validated across multiple languages and regions, including Asia.5-12) Though originally an informant-based screening tool, the AD8 was later evaluated as a self-rated questionnaire, showing utility in differentiating at-risk individuals from those without dementia, particularly in mild cases.12,13) While informant-based assessments offer a longitudinal estimate of cognitive change, self-rated assessments may be particularly useful when informants are unavailable or when older adults prefer to maintain their privacy of not disclosing the assessment outcome to their family members.14,15) However, prior studies indicate a performance gap between the informant-reported (AD8-Informant) and self-reported (AD8-Self) versions,16,17) with AD8-Informant generally providing better screening accuracy at a cutoff score of ≥2.
While many studies support the utility of either AD8-Informant or AD8-Self as sensitive and reliable tools to differentiate non-demented from demented individuals,7,8,13,14,18) some report that AD8-Informant offers more accurate results.16,17,19) Locally, findings have been mixed. One study reported AD8-Self effective in identifying mild cognitive impairment (MCI) and early dementia at a memory clinic,12) whereas others emphasized stronger performance of AD8-Informant in distinguishing cognitive impairment and dementia.17,20,21) Differences in study setting (clinic vs. community), participants’ characteristics, or the expertise of test administrator may account for these inconsistencies. Moreover, inconsistent cutoff scores across different versions and settings hinder consensus on the optimal threshold for identifying cognitively impaired individuals (Table 1).22-27) Most studies focus on the utility of AD8-Informant and AD8-Self to distinguish cognitively impaired from unimpaired individuals. While Wang et al.5) demonstrated a one-factor model for AD8-Informant, the factor structure of the informant and self-rated versions has not yet been compared.
Summary of studies on AD8 performance and factor structure in Asian population
Therefore, this study aimed to compare the diagnostic performance and reliability of AD8-Self and AD8-Informant in identifying early cognitive impairment (ECI)—MCI and mild dementia—with optimal cutoff scores for each version in a predominantly Chinese Asian population of healthy older persons attending a tertiary memory clinic. Additionally, we compared the factor structures of AD8-Self and AD8-Informant, which may help explain differences in diagnostic performance.
MATERIALS AND METHODS
Study Design and Participants
This cross-sectional study involved 515 participant–informant dyads of community-dwelling older adults with ECI, defined as MCI or dementia with a Clinical Dementia Rating (CDR) score of 0.5 to 1.28) Participants were recruited from the Memory Clinic at the Department of Geriatric Medicine, Tan Tock Seng Hospital, Singapore, between January 2011 and December 2013. Clinicians at the clinic were trained in using informant-based tools, including the CDR and AD8 questionnaire, for ECI evaluation.
We studied participant–informant dyads in which participants were aged 55 years and older and were diagnosed with MCI, very mild or mild dementia (CDR global score 0.5–1). Individuals with a CDR score of ≥2 were excluded, as individuals with moderate to severe dementia typically have limited insight and are unlikely to provide reliable self-assessment of their cognitive status. This study is a secondary analysis of a previously approved “Memory Clinic Service Standing Database” (SDB-2022-0056-TTSH-01), maintained by the Cognition and Memory Disorder Service. Ethics approval was waived by the National Healthcare Group Domain Specific Review Board for use of fully anonymized data. The database, collected between 2009 and 2017, includes participants’ demographic data, comorbidities, medication history, cognitive and functional assessments, neuropsychological and neuropsychiatric data, neuroimaging results, informant characteristics, and caregiver burden scores.
Assessment
All participants underwent a comprehensive assessment by a nurse clinician and a doctor, followed by blood tests and neuroimaging. Participants with subtle cognitive complaints, but not yet fulfilling diagnostic criteria for dementia syndrome received standardized, locally validated neuropsychological assessments by neuropsychologists; 67.8% of study participants completed these assessments.
Cognitive diagnoses, including severity and etiology, were determined through consensus meetings involving geriatricians, neuropsychologists, and trained nurse clinicians. The clinical diagnosis of dementia was based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria and it is to note that diagnosis of dementia was established purely based on clinical history. Classification of etiology was made using international published criteria for dementia:
(1) Alzheimer’s disease: National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) criteria,29)
(2) Vascular and mixed dementia: National Institute of Neurological Disorders and Stroke–Association Internationale pour la Recherche et l’Enseignement en Neurosciences (NINDS-AIREN),30)
(3) Frontotemporal lobar degeneration: consensus clinical criteria,31) and
(4) Parkinson’s disease dementia: established clinical diagnostic criteria.32)
MCI was diagnosed using the revised Petersen criteria.33) Cognitive impairment severity was rated using the locally validated CDR scale,34) where 0 indicated normal cognition; 0.5, MCI or very mild dementia; and 1, 2, and 3 represented mild, moderate, and severe dementia, respectively. This study included participants with CDR scores of 0–1 and excluded those with scores ≥2.
Measurements and Instruments
Baseline demographics for both participants and informants were collected, including age, sex, education, ethnicity, and relationship to the participant. Trained nurse clinicians or geriatricians administered both self-reported and informant-reported AD8 questionnaires. Participants underwent locally validated cognitive screening assessment: Abbreviated Mental Test (AMT), a 10-question cognitive screening instrument which assesses the orientation, memory (immediate memory with delayed recall) and attention,35) modified Chinese Mini-Mental State Examination (m-CMMSE),35) and modified Chinese Frontal Lobe Assessment Battery (m-CFAB).36) The participants were also assessed for their function via modified Barthel Index for basic activities of daily living (BADL)37) and the Lawton and Brody’s instrumental activities of daily living scale (IDAL),38) mood disorders using the DSM-IV criteria for major depressive disorder and neuropsychiatric symptoms using the Neuropsychiatric Inventory-Questionnaire (NPI-Q).39) Where applicable, participants received standardized, locally validated neuropsychological assessment battery, including word list memory (immediate recall, delayed recall and recognition), verbal fluency (animal category), constructional praxis test, the modified Boston Naming Test, the Welchsler Adult Intelligence Scale-Revised (WAIS-R) subtest of block design and object assembly which measures visuospatial and non-verbal reasoning abilities.40)
Statistical Analysis
Descriptive and inferential statistical analyses were conducted using IBM SPSS version 27.0 (IBM, Armonk, NY, USA). Chi-square tests were used for categorical variables, and one-way analysis of variance (ANOVA) with Bonferroni correction for post hoc comparisons of continuous variables. All tests were two-sided, with a significance level set at p<0.05.
Receiver operating characteristic (ROC) curves were used to compare the diagnostic performance of AD8-Self and AD8-Informant, using clinical diagnosis (normal/MCI/ECI) as the reference standard. Areas under the ROC curves (AUCs) were compared using the DeLong method to demonstrate that the difference in AUCs of AD8-Self and AD8-Informant were statistically significant.41) Optimal cutoff values were determined using the Youden index (sensitivity+specificity–1),42) and diagnostic performance was evaluated by calculating sensitivity, specificity, positive predictive value, and negative predictive value. Internal consistency was assessed using Cronbach’s alpha.
Exploratory factor analysis was performed to examine the factor structures of AD8-Self and AD8-Informant. Sampling adequacy was evaluated with the Kaiser-Meyer-Olkin (KMO) statistic, and the Bartlett test of sphericity was used to determine the suitability of performing factor analysis. Principal component analysis with varimax rotation was used to extract factors, and parallel analysis to determine the optimal number of factors to retain. Items with factor loadings below 0.5 were excluded.43)
RESULTS
Demographic and Clinical Characteristics
This study included 515 participant–informant dyads, predominantly of Chinese ethnicity. Among these, 43 were cognitively normal, while 472 participants had ECI, comprising 102 participants with MCI (CDR=0.5) and 370 with mild dementia (CDR 0.5–1.0). The mean age of the total cohort was 75.08±7.16 years, and 58.8% were female. Participants with normal cognition were slightly younger (mean age: 69.28±7.87 years) and had more years of formal education (9.58±4.83 years) compared to those with ECI (75.61±6.9 years and 5.6±4.9 years, respectively) (Table 2).
Baseline characteristics of study cohort (n=515)
As expected, cognitively normal participants scored higher on cognitive and functional assessments. Among participants with ECI, the mean AMT and m-CMMSE scores were 7.0±2.2 and 19.1±4.8, respectively. The mean AD8-Self scores were 1.74±1.2 in the normal cognition group and 2.4±1.8 in the ECI group. Corresponding AD8-Informant scores were 1.67±1.25 and 5.29±2.04, respectively. Participants with ECI exhibited relatively preserved BADLs, with a Barthel Index score of 96.45±7.1, while experiencing some degree of impairment in IADLs, with a Lawton and Brody’s IADL score of 15.1±4.8 (Table 2).
Diagnostic Performance of AD8-Self and AD8-Informant with Cutoff Scores
ROC analysis was used to compare the diagnostic performance of AD8-Self and AD8-Informant for detecting ECI. The AD8-Informant demonstrated significantly superior diagnostic performance, with an optimal cutoff score of ≥3, yielding a sensitivity of 89.0%, specificity of 79.1%, and AUC of 0.915. In contrast, AD8-Self demonstrated a significantly lower performance, with an optimal cutoff score ≥4, sensitivity of 27%, specificity of 95.3%, and an AUC of 0.593 (p<0.001) (Table 3, Fig. 1A).
Reliability and diagnostic performance of AD8-Self vs. AD8-Informant
Comparison of AUC of AD8-Self versus AD8-Informant between two groups: (A) early cognitive impairment, (B) mild cognitive impairment. AUC, area under the curve; AD8-Self, Ascertain Dementia 8-item Questionnaire-Self reported; AD8-Informant, Ascertain Dementia 8-item Questionnaire-Informant reported.
Similar findings were observed in the MCI subgroup. AD8-Informant achieved a sensitivity of 64.7%, specificity of 79.1%, and AUC of 0.745 (cutoff score ≥3), whereas AD8-self had a sensitivity of 26.5%, specificity of 95.3%, and AUC of 0.600 (cutoff score ≥4) (p=0.002) (Table 3, Fig. 1B).
Reliability of AD8-Self and AD8-Informant
The internal consistency of AD8-Informant was higher than that of AD8-Self, as measured by Cronbach’s alpha. For the overall ECI group, Cronbach’s alpha was 0.764 for AD8-Informant and 0.666 for AD8-Self. In the MCI subgroup, values were 0.709 and 0.663, respectively (Table 3).
Factor Analysis of AD8-Self and AD8-Informant
Factor analysis was deemed appropriate for both AD8-Self and AD8-Informant, with KMO statistics of 0.742 and 0.85, respectively, and significant Bartlett’s test of sphericity (χ²=457.8 and 764.39, respectively; p<0.0001) (Table 4).
Factor structure of AD8-Self vs. AD8-Informant
For AD8-Self, principal component analysis and parallel analysis supported a two-factor structure, accounting for 44.69% of the total variance. The first factor (30.11% of variance) included three items related to non-memory domains: less interest in hobbies/activities, trouble learning how to use tools or gadgets, and difficulty handling complicated financial affairs. The second factor (14.58% of variance) encompassed memory-related items: daily problems with thinking and/or memory, trouble remembering appointments, and repeating the same thing over and over. Two items, “forgetting the correct week, month, or year” and “problems with judgment,” were excluded due to low factor loadings (0.447 and 0.487, respectively).
In contrast, parallel analysis recommended a one-factor structure for AD8-Informant, which accounted for 38.58% of the total variance. All items showed strong loadings on this single factor.
DISCUSSION
This study compared the diagnostic performance of AD8-Self with AD8-Informant to evaluate the utility of AD8-Self as a standalone screening tool in detecting ECI in clinical settings. Although AD8-Self has shown potential in previous research, there are studies emphasizing better diagnostic performance of AD8-Informant. Although AD8 has been validated in several Asian countries, most studies focused on the performance of AD8-Self or AD8-Informant, with only a few studies comparing the performance of both tools.16,17,22) Our study comprehensively investigated the diagnostic performance of both tools; optimal cutoff scores, and reliability in differentiating early cognitively impaired older individuals (with a subgroup of MCI) from cognitively unimpaired individuals in a tertiary memory clinic. The added strength of our study is to explore the factor structures of AD8-Self and AD8-Informant. To our knowledge, this is the first study to analyse the factor structures of both assessment. Our finding demonstrate that AD8-Self has significantly poorer discriminant ability in identifying both ECI and MCI, with AUCs of 0.593 and 0.600, respectively. AD8-Informant performed better across all metrics and showed a one-factor structure, while AD8-Self demonstrated a two-factor structure.
First, the optimal cutoff score in our study was ≥3 for AD8-Informant and ≥4 for AD8-Self to identify individuals with ECI, higher than previous reports—e.g., ≥2 and ≥1, respectively, by Galvin et al.4,13) and a local study by Chin et al.12) These discrepancies may reflect differences in study settings (community vs. clinic), demographic characteristics, such as age, educational level, and the severity of cognitive impairment. Our sample comprised older individuals with lower educational attainment and included only MCI and mild dementia cases, while other studies included ECI as well as moderate to severe dementia. The variation raises question about whether factors like age and education influence the performance of AD8. While Galvin et al.18) indicated that AD8-Informant was not likely to be influenced by age, education, or cultural differences, this was not the case in a study by Cai et al.,9) which showed education affected AD8-Self in a community dwelling older Chinese populations.
Second, our results highlighted AD8-Self’s reduced discriminant ability more pronounced in ECI (AUC 0.915 vs. 0.593) than in MCI (AUC 0.745 vs. 0.600), which is likely attributable to anosognosia, a neurological symptom characterized by limited awareness of one’s cognitive and functional deficits, commonly seen in individuals with cognitive impairment. This lack of insight into their changes in memory and cognition can cause underreporting of their symptoms.44,45) Therefore, in the absence of reliable informants, clinicians should consider supplementing self-reported AD8 with other objective cognitive assessments or performance-based tools, including neuropsychological testing, as recommended by Lim et al.46) Still, the high specificity of AD8-Self may make it useful in the very early stage of cognitive impairment; a positive result should prompt further evaluation and follow-up, in line with Gao et al.’s recommendation.47)
Third, although both AD8 versions showed strong performance, the performance in the ECI group was much higher than in the MCI group. This can be attributed to the subtle nature of cognitive changes in MCI, which may not be easily noticeable to informants. In contrast, individuals with dementia typically exhibit more noticeable cognitive symptoms, which informants or caregivers can identify. Informants may also underestimate the severity of cognitive symptoms or may not be aware of certain cognitive deficits in individuals with subtle changes in MCI group. Additionally, informant characteristics—such as gender, education, and contact frequency—can influence reporting accuracy which may impact the diagnostic performance of informant-based assessment tools. Studies have shown that female caregivers are often more likely to assume primary caregiving role and attuned to behavioural, cognitive and functional changes.48,49) These sociodemographic factors may partially explain the weaker performance of AD8-Informant in the MCI subgroup, although our study did not explore these directly.
Fourth, our findings align with those of Li et al.11) and Dong et al.17) showing that AD8-Informant has stronger reliability with higher Cronbach’ alpha values. Due to the lack of insight, which is commonly seen in individuals with cognitive impairment, the reliability of AD8-Self may be questionable. Despite its lower reliability, AD8-Self still showed acceptable internal consistency (Cronbach’s alpha: 0.666 for ECI and 0.663 for MCI), suggesting that individuals with ECI can still recognize their cognitive and/or functional decline to a certain extent, although their insight may be limited. Another important observation is that the Cronbach’s alpha for AD8-Informant in our study was lower than in Galvin’s original study18) and other Asian studies,6,7,16) possibly reflecting differences in clinical setting, cultural difference between Asian and Western populations and test administrator of the tools. In our study, assessments were administered by memory clinic physicians and nurse clinicians.
Additionally, our study provides insight into the distinct factor structures of AD8-Self versus AD8-Informant: AD8-Informant showed a unidimensional structure, consistent with prior work,50) while AD8-self displayed two factors reflecting memory and non-memory domains. One possible explanation for this could be that individuals with ECI may not associate cognitive decline with functional issues or may attribute cognitive changes to aging and functional changes to other physical health conditions, leading to a dissociation between cognitive and functional symptoms. Prior studies on the factor structure of AD8-Informant are limited and this is the first study, to our knowledge, to highlight structural differences between self- and informant-rated AD8s (two vs. one factor).
Our study has several limitations. Firstly, AD8 was originally designed for community screening, our study was, however, conducted in a tertiary memory clinic, resulting in a relatively smaller number of cognitively normal participants compared to the ECI group. It could introduce spectrum bias and limit generalizability to community-dwelling populations. Secondly, we recognise that our data was collected a decade ago, and findings should be further validated in newer cohorts. Finally, despite identifying a distinct factor structure for AD8-Self, we did not analyse correlations between AD8-Self’s factor structure and cognitive severity or assess the impact of participant demographics on test performance.
In conclusion, AD8-Self demonstrated inferior sensitivity and reliability compared to AD8-Informant in detecting ECI. Our findings suggest that AD8-Self alone may be insufficient for clinical screening without collateral input or additional objective cognitive testing. In clinical settings, collateral history from a reliable informant remains an integral role in confirming the degree of cognitive decline and functional impairment. In situations where informants are unavailable, we recommend that clinicians base their evaluations and judgement on objective cognitive assessments, incorporating neuropsychological testing when clinically indicated. The differing factor structures of the two versions underscore the need for further research to explore how these structural variations relate to dementia severity.
Notes
We would like to thank the doctors, nurse clinicians of the Cognition and Memory Disorders Service, Department of Geriatric Medicine, Tan Tock Seng Hospital and research assistants from the Institute of Geriatrics and Active Aging, Tan Tock Seng Hospital, for their substantial contribution in data acquisition.
CONFLICT OF INTEREST
The researchers claim no conflicts of interest.
FUNDING
None.
AUTHOR CONTRIBUTIONS
Conceptualization, WSL, KKW; Data curation, WSL; Methodology, WSL, KKW; Supervision, WSL; Writing-original draft, KKW; Writing-editing & review, WSL, JC, JPL, EH, NA, MC.