Improved Accuracy of the Addenbrooke’s Cognitive Examination–Revised in the Diagnosis of Mild Cognitive Impairment,Mild Dementia Due to Alzheimer’s Disease and Behavioral Variant Frontotemporal Dementia Using Mokken Scale Analysis

Abstract

Background:

The Addenbrooke’s Cognitive Examination-Revised (ACE-R) is an accessible cognitive tool that supports the early detection of mild cognitive impairment (MCI), Alzheimer’s disease (AD), and behavioral variant frontotemporal dementia (bvFTD).

Objective:

To investigate the diagnostic efficacy of the ACE-R in MCI, AD, and bvFTD through the identification of novel coefficients for differentiation between these diseases.

Methods:

We assessed 387 individuals: 102 mild AD, 37 mild bvFTD, 87 with amnestic MCI patients, and 161 cognitively unimpaired controls. The Mokken scaling technique facilitated the extraction out of the 26 ACE-R items that exhibited a common latent trait, thereby generating the Mokken scales for the AD group and the MCI group. Subsequently, we performed logistic regression, integrating each Mokken scales with sociodemographic factors, to differentiate between AD and bvFTD, as well as between AD or MCI and control groups. Ultimately, the Receiver Operating Characteristic curve analysis was employed to assess the efficacy of the coefficient’s discrimination.

Results:

The AD-specific Mokken scale (AD-MokACE-R) versus bvFTD exhibited an Area Under the Curve (AUC) of 0.922 (88% sensitivity and specificity). The AD-MokACE-R versus controls achieved an AUC of 0.968 (93% sensitivity, 94% specificity). The MCI-specific scale (MCI-MokACE-R) versus controls demonstrated an AUC of 0.859 (78% sensitivity, 79% specificity).

Conclusions:

The ACE-R’s capacity is enhanced through statistical methods and demographic integration, allowing for accurate differentiation between AD and bvFTD, as well as between MCI and controls. This new method not only reinforces its clinical value in early diagnosis but also surpasses traditional approaches noted in prior studies.

Keywords

Aging Alzheimer’s disease cognition dementia frontotemporal dementia mild cognitive impairment neuropsychological tests

INTRODUCTION

Cognitive assessment through objective tests is frequently valuable for tracking cognitive impairment or dementia. This is of paramount importance in clinical practice. 1 However, identifying cognitive decline and establishing the diagnosis of neurodegenerative diseases can be challenging due to the overlap of symptoms among different types of dementia. 2 Despite the remarkable advances in neuroimaging and biomarker analyses, these resources are not available worldwide, given the high cost and need of specialized professionals. This fact underscores the importance of accessible clinical methods.

The Addenbrooke’s Cognitive Examination–Revised (ACE-R)3,4, 3,4 is a widely established cognitive assessment battery used in clinical practice in various countries. The instrument provides a comprehensive and quantitative assessment of cognitive functions, such as short-term and long-term memory, attention, language, visual-spatial abilities, visual-perceptual abilities, and executive functions, among other cognitive aspects that can be qualitatively analyzed. Therefore, ACE-R is particularly valuable for cognitive screening and assisting in profiling the patient’s cognitive status in a relatively short period (∼20 min). Additionally, ACE-R is a versatile tool that can be applied in various clinical settings, from outpatient clinics to hospitals.5 –16

Some authors have recognized the utility of ACE-R in evaluating mild cognitive impairment (MCI).4,17,18 , 4,17,18 In the very first study with the ACE-R, MCI patients showed intermediate scores between dementia patients and controls. 4 However, there is contradictory evidence regarding its effectiveness in identifying MCI. 19 In preliminary analyses of our group comparing MCI versus controls based solely on raw scores (univariate discriminative analysis), ACE-R was inefficient in discriminating between the groups (AUC <0.754). 20 There are indications that ACE-R may not be accurate in identifying MCI when using cutoff scores. 21 Hence, the inquiry persists as to whether the instrument lacks efficacy in identifying this group of patients or if the employed type of analysis lacks the sensitivity to demonstrate its effectiveness.

On the other hand, ACE-R has been helpful in aiding the differentiation between Alzheimer’s disease (AD) and behavioral variant frontotemporal dementia (bvFTD).4,20–23 , 4,20–23 The classic clinical presentation of AD predominantly involves episodic memory decline, 24 while bvFTD presents with behavioral changes and executive function impairment. 25 As an attempt to differentiate between AD and FTD with ACE-R, in 2006, authors developed the VLOM ratio ((Verbal fluency (Max score: 14 points) + Language (Max score: 26 points) / Orientation (Max score: 10 points) + Episodic memory/recall (Max score: 7 points)). 4 However, the effectiveness of VLOM remains controversial. While there are English language studies that did not find favorable results for its use,19,26, 19,26 the original study and studies from South America confirmed its value.4,20,27 , 4,20,27

The value of having a broad scope of diagnostic methods that assist in diagnostic reasoning to detect the risk of dementia and its early stages is unquestionable. Although there are specific biomarkers for AD, none exist for FTD, making clinical analysis paramount.28,29, 28,29 Furthermore, with the advent of new modifying therapies for AD, cognitive assessment can help in the early screening of patients who may benefit from treatment and rule out possible negative cases, as it is not possible to indiscriminately request biomarker tests. 2

It is worth mentioning that in 2015, among more than 40 dementia screening tests covered in an international systematic review, ACE-R was indicated as one of the two best available alternatives, 15 along with the Mini-Cog. 30 Hence, more sophisticated statistical analyses, aimed at improving the diagnostic efficacy of instruments such as ACE-R, can offer advantages, particularly for clinicians grappling with the intricate task of precise differential diagnosis.

Previously, researchers employed Mokken analysis to examine the hierarchical structure of the 26 ACE-R items in AD and other dementias. 31 However, they did not proceed to investigate the diagnostic value of ACE-R based on the Mokken scaling analysis. Thus, the aim of this study was to investigate the discriminative ability of ACE-R and each of its items, as potential new coefficients for the differential diagnosis between AD and bvFTD, and MCI and controls.

METHODS

This study included patients diagnosed with probable AD 24 and patients with bvFTD, 25 all with mild dementia according to the criteria of the Clinical Dementia Rating Scale (CDR).32,33, 32,33 Participants who met diagnostic criteria for amnestic MCI of multiple domains 34 were also selected. The groups were compared with cognitively healthy controls, who were extracted from a sample previously examined in a standardization study conducted by the same authors. 35

All individuals should be aged 45 years or older and have four or more years of schooling. Patients achieved results equivalent to or higher than the 10th percentile in the Mattis Dementia Rating Scale (DRS).36,37, 36,37 Individuals with atypical subtypes of AD or FTD, those with associated relevant depressive symptomatology classified as moderate intensity (score 2 or 3) on the “Depression” item of the Neuropsychiatric Inventory,38,39, 38,39 or moderate cerebrovascular disease (Fazekas ≥2), 40 or those using benzodiazepine medications were not included.

The diagnostic groups were determined by an experienced medical team at the Hospital das Clínicas, Universidade Federal de Minas Gerais, Belo Horizonte, Brazil, or at the Hospital das Clínicas, Universidade de São Paulo, São Paulo, Brazil. These public institutions are part of the Brazilian free-of-charge social healthcare system. To broaden the inclusion criteria to individuals with higher educational and socioeconomic levels, invitations were extended to patients under the care of private clinics to participate in the study. The study received approval from the ethics committees of both institutions (Ethics Committee Approval Numbers: 1.975.023 and 1.956.759, respectively), and all participants provided written informed consent. Procedures involving experiments on human subjects were also done in accord with the Helsinki Declaration of 1975.

Mokken scaling plus multivariate discriminative analyses

The 26 items of the ACE-R underwent Mokken scaling analyzes to assess parameters which might confirm the conformity of the data to the generated Mokken scale. These parameters include scalability (H), which measures the strength of an item in discriminating the assumed latent variable related to these items. Scalability expresses the power to reliably order respondents at the latent trait level through their total score sum and indicates the Mokken scale quality. The item scalability coefficient (Hi) indicates item discrimination ability, and values below 0.3 suggest poor discrimination and potential exclusion from the Mokken scale. Monotonicity ensures that higher latent trait values increase the probability of a correct response, while non-intersection means item step response functions do not intersect, maintaining an invariant order of items (IIO), which is vital. IIO signifies that items adhere to a hierarchical scale with the same order of difficulty or attractiveness for all respondents. This property aids in detecting unexpected scoring patterns and characterizing differences between subgroups and various forms of dementia. 41

The second part of this statistical investigation of the ACE-R proceeded with multivariate discriminative analyses. The Variance Inflation Factor (VIF) was assessed, as it is an important preliminary analysis for the creation of the regression model. Only variables that respected the threshold of a value less than 2.50 were selected for the next step.

Next, logistic regression with cross-validation (leave-one-out) was used to identify which of the independent variables (as age, education, sex, Mokken scale generated for AD or MCI, and the remaining ACE-R) could be jointly used in the diagnostic differentiation between groups, in pairs. The outcome of this analysis is a proposed logistic model for each pair (AD versus bvFTD, AD versus controls, MCI versus controls). The quality of this model was investigated through its residual deviance, normality of the residue, dispersion of each subscale from predicted values, and complementary Pearson residual. Outliers were checked by the Bonferroni test.

To maximize the sensitivity and specificity of these results, the best cutoff point (τ) on the probability scale and its equivalent on the odds scale (logit τ) were chosen. By calculating this model’s cutoff point, its statistical validity in the database of this study was verified. Finally, the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the model in distinguishing groups in this sample were measured.

Statistical analyses were performed using IBM SPSS Statistics 22 and R software version 3.6.1, all for Windows.

RESULTS

The total sample in this study comprised 387 subjects. The study included 102 patients with mild probable dementia due to AD, 37 patients with mild bvFTD, 87 individuals with MCI, and 161 cognitively healthy controls. The age ranged from 45 to 89 years in the patient groups and from 50 to 93 years among controls. In both groups, education ranged from 4 to 24 years.

Sex distribution showed no significant difference among groups (p = 0.170). Significant differences were observed for age and education, DRS and ACE-R scores between the groups (p < 0.001) (Table 1).

Table 1

Demographic data and cognitive performance in the AD, bvFTD and control groups

		Controls	MCI	AD	bvFTD	p
	N	161	87	102	37
Sex	N (F/M)	92F/69M	45F/42M	61F/41M	9F/28M	0.17
	% Female	57%	51%	59%	24%
Age	Mean	69.4	74.8	75.6	64.6	*
	SD	10.9	6.2	7.9	8.4
	Min	50	60	53	45
	Max	93	86	89	79
^aControls < AD^; Controls > bvFTD (p = 0.022); Controls < MCI^;
AD > bvFTD^; AD = MCI (p = 1.000); bvFTD < MCI^
Schooling	Mean	10.2	13.8	11.8	12.2	*
	SD	5.2	4.8	5	6.2
	Min	4	4	4	4
	Max	24	23	22	24
^aControls = AD (p = 0.104); Controls = bvFTD (p = 0.188); Controls < MCI^*;
AD = bvFTD (p = 1.000); AD < MCI (p = 0.034); bvFTD = MCI (p = 0.655)
DRS	Mean	136.1	130.5	120.2	123.5	*
	SD	5.4	7.7	10.2	11.5
	Min	123	104	100	100
	Max	144	144	139	144
^aControls > AD^; Controls > bvFTD^; Controls > MCI^*;
AD = bvFTD (p = 0.258); AD > MCI^; bvFTD > MCI^
ACE-R	Mean	85.1	79.8	70.2	72.2	*
	SD	8.2	7.8	10.8	11.1
	Min	62	57	42	51
	Max	98	95	91	94
^aControls > AD^; Controls > bvFTD^; Controls > MCI^*;
AD = bvFTD (p = 1.000); AD > MCI^; bvFTD > MCI^
VLOM	Mean	2.54	3.13	3.66	2.52	*
	SD	0.43	0.73	1.07	0.7
	Min	2	2	2	2
	Max	4	6	7	5
^aControls < AD^; Controls = bvFTD (p = 1.000); Controls < MCI^;
AD > bvFTD^; AD > MCI^; bvFTD < MCI^*

All comparisons were made using ANOVA Test, except for the sex variable (Chi-square). ^a Post hoc analysis was performed using the Bonferroni test. MCI, mild cognitive impairment; AD, mild dementia due to Alzheimer’s disease; bvFTD, behavioral variant frontotemporal dementia; DRS, Dementia Rating Scale; SD, standard deviation; p, significance value; ^* p < 0.001.

Mokken analyses of ACE-R in AD (AD-MokACE-R) and multivariate discriminative analyses

Among the groups with AD and bvFTD, the AD group was chosen to generate the Mokken scale, given its larger sample size. Of the 26 items in the ACE-R, the Mokken scale of ACE-R in AD (AD-MokACE-R) identified 12 items measuring the same latent concept (Table 2). All these items adhered to the item scalability coefficients (Hi >0.3) and showed no violations of monotonicity or invariant item ordering without intersection (Table 2). ACE-R items with Hi <0.3 were discarded.

Table 2

AD-MokACE-R and MCI-MokACE-R: scalability and violations analysis

	ACE-R items	Hi	Violation’s analysis
			M	NI	IIO
			#vi^*	#vi^*	#vi^*
AD-MokACE-R (H: 0.39)	Memory: Retrograde memory (Max 4)	0.36	0	2	0
	Verbal Fluency: letter P (Max 7)	0.35	0	3	0
	Verbal Fluency: Animals (Max 7)	0.39	0	3	0
	Language: Comprehension “Close your eyes” (Max 1)	0.60	0	0	0
	Language: Repetition of words (Max 2)	0.31	0	3	0
	Language - Repetition “Above, beyond and below” (Max 1)	0.46	0	0	0
	Language: Naming of 10 figures (Max 10)	0.42	0	5	0
	Language: Comprehension (Max 4)	0.42	0	4	0
	Language: Reading of words (Max 1)	0.38	0	0	0
	Visuospatial abilities: Wire cube (Max 2)	0.43	0	2	0
	Visuospatial abilities: Clock drawing (Max 5)	0.36	0	2	0
	Perceptual abilities: Identifying letters (Max 4)	0.48	0	0	0
	AD-MokACE-R Maximum score: 48 points.
MCI-MokACE-R (H: 0.37)	Memory - Anterogrady Memory (Max 7)	0.32	0	0	0
	Memory - Retrograde memory (Max 4)	0.45	0	0	0
	Verbal Fluency - Animals (Max 7)	0.37	0	0	0
	Language - writing (Max 1)	0.59	0	0	0
	Language - Repetition of words (Max 2)	0.35	0	0	0
	Language - naming (Max 10)	0.37	0	0	0
	Language - comprehension (Max 4)	0.42	0	0	0
	Language - reading words (Max 1)	0.40	0	0	0
	Visuospatial abilities – overlapping pentagons (Max 1)	0.37	0	0	0
	Memory – recall (Max 7)	0.34	0	0	0
	MCI-MokACE-R Maximum score: 44 points.

^* p < 0.001. AD-MokACE-R, the AD-specific Mokken scale; MCI-MokACE-R, the MCI-specific scale; H, whole scalability coefficient; Hi, item scalability coefficient; #vi, number of violations; M, monotonicity; NI, non-intersecting; IIO, Invariant item ordering.

AD versus bvFTD

In this sample, all variables adhered to the maximum VIF value of 2.50. The logistic regression method with cross-validation identified that for bvFTD versus AD, AD-MokACE-R, being male, age, Orientation in time, Orientation in space, and Memory Recall were selected as significant variables (p < 0.05; Table 3). In Table 3, the β (beta) value indicates that each additional point in scores on Orientation in time, Orientation in space, Memory Recall, and being male increase the chance of the individual being in the bvFTD group in this sample. Similarly, each additional point in the AD-MokACE-R score and each additional year on age decreases the chance of bvFTD. The exponential value of β (eβ) is interpreted as the effect size of the variable on the chance of the individual being in the bvFTD group. For example, each additional point in the Orientation in space, on average, increases the chance of the subject being bvFTD compared to AD (18.651, i.e., 18x) or if the individual is male, it becomes more likely for him to be bvFTD (5.370 increases by 5.3x). Following this reasoning, the proposed logarithm for this sample is:

Table 3

Logistic regression model with cross-validation (leave-one-out) of AD versus bvFTD, AD versus controls and MCI versus Controls

		β	eβ	Standard Error	p
^*AD versus bvFTD	(Intercept)	–4.164	–	5.956	0.484
	AD-MokACE-R	–0.194	0.824	0.058	<0.001
	Male	1.681	5.370	0.725	<0.05
	Age	–0.182	0.833	0.048	<0.001
	Orientation in time	1.516	4.555	0.528	<0.001
	Orientation in location	2.926	18.651	1.209	<0.05
	Memory (name and address recall)	0.493	1.637	0.209	<0.05
^**AD versus Controls	(Intercept)	18.413	–	3.684	<0.001
	AD-MokACE-R	–0.183	0.832	0.054	<0.001
	Schooling	0.315	1.370	0.069	<0.001
	Orientation in time	–3.019	0.049	0.623	<0.001
	Memory (name and address recall)	–0.795	0.451	0.167	<0.001
^***MCI versus Controls	(Intercept)	8.714	–	3.148	<0.001
	MCI-MokACE-R	–0.214	0.807	0.044	<0.001
	Age	0.036	1.037	0.02	0.061
	Schooling	0.288	1.334	0.046	<0.001
	Orientation in time	–1.415	0.243	0.519	<0.001
	Memory (recognition)	–0.310	0.733	0.206	0.131

The result of logarithmic function seeks the identification of cases of ^*bvFTD, ^**AD, ^***MCI. AD-MokACE-R, the AD-specific Mokken scale; MCI-MokACE-R, the MCI-specific scale; MCI, mild cognitive impairment; AD, mild dementia due to Alzheimer’s disease; bvFTD, behavioral variant frontotemporal dementia; β, the beta value; eβ exponential value of effect on chance; p, significance value.

Log = Beta value of the Intercept (–4.164) + ((–0.194×score obtained on AD-MokACE-R items) + (1.681×1 if male) + (–0.182×age) + (1.516×score on Orientation in time) + (2.926×score on Orientation in space) + (0.493×score on Memory Recall)).

When the residual deviation of these logarithmic models was investigated, there was no significant difference from the perfect model that would fit each data, p-value (χ ² 93.95 df = 129) = 0.999 for AD versus bvFTD. Three outliers were identified and excluded from the construction of the logarithmic model, confirmed by Bonferroni test (p < 0.05).

In the cross-validation analysis, the AUC of the model for AD versus bvFTD was 0.922 (Fig. 1, Table 4), therefore it was classified as highly satisfactory and superior to ACE-R Total score (AUC = 0.496), Mini-Mental State Examination (MMSE) (AUC = 0.692) and VLOM (AUC = 0.829), with a cutoff point on the chance scale for logit τ= –0.82. In this analysis, the sensitivity and specificity of the model were 88%. The PPV and NPV were 71% and 96%, respectively (Table 4).

Fig. 1

Comparison of ROC Curves in Group Discrimination for Proposed Models Derived from Multivariate Logistic Regression utilizing Mokken Scales versus ACE-R, MMSE and VLOM. AD, mild dementia due to Alzheimer’s disease; bvFTD, behavioral variant frontotemporal dementia; MCI, mild cognitive impairment; Multivariate logistic regression, Logistic regression model with cross-validation (leave-one-out); VLOM, VLOM ratio; MMSE, Mini-Mental State Examination.

Table 4

Multivariate discriminative analyses for each group comparison (considering ACE-R subscores and sociodemographic variables)

	Chance logarithm	AUC	Sens.	Spec.	PPV	NPV
	(logit τ)
AD versus bvFTD	τ= –0.82	0.922	88%	88%	71%	96%
AD versus Controls	τ= –0.45	0.968	93%	94%	90%	95%
MCI versus Controls	τ= –0,64	0.859	78%	79%	66%	86%

MCI, mild cognitive impairment; AD, mild dementia due to Alzheimer’s disease; bvFTD, behavioral variant frontotemporal dementia; Sens., sensitivity; Spec., specificity; PPV, positive predictive value; NPV, negative predictive value.

Lastly, the model’s validity within the data itself was examined. For AD versus bvFTD, among the 102 AD patients, the model correctly identified 90 cases and misdiagnosed the remaining 12 cases. In 34 bvFTD (3 subjects were removed after being identified as outliers), 30 subjects were correctly identified as bvFTD, and 4 cases were incorrectly considered AD. Therefore, based on the presented data, the ability of ACE-R to discriminate AD from bvFTD using the proposed model, which includes AD-MokACE-R, demonstrates strong discriminative power.

AD versus controls

The AD-MokACE-R, Years of Education, Orientation in time, and Memory Recall were the variables that showed statistical significance (p < 0.05; Table 3). The β (beta) value suggests that higher education and lower scores on AD-MokACE-R, Orientation in time, and Memory Recall increase the chance of an individual having AD in this sample. The proposed logarithm for this sample is:

Log = Beta value of the Intercept (18.413) + ((–0.183×score on MokACE-R-AD) + (0.315×years of study) + (–3.019×score on Orientation in time) + (–0.795×score on Memory Recall))

For AD versus controls, when investigating the residual deviation of these logarithmic models, there is no significant difference from the perfect model that would fit each data, p-value (χ ² 98.99 df = 258) = 0.999. No outliers were identified, confirmed by Bonferroni test (p < 0.05).

In the cross-validation analysis for AD versus controls, the AUC of the model was 0.968 (Fig. 1, Table 4), therefore classified as highly satisfactory and superior to ACE-R Total score (AUC = 0.856), MMSE (AUC = 0.835), and VLOM (AUC = 0.859), with a chosen cutoff of logit τ= –0.45. In this analysis, the sensitivity and specificity of the model were 93% and 94%, respectively. The PPV and NPV were 90% and 95%, respectively (Table 4).

Finally, the model’s validity within the data itself was examined. For AD versus controls, among the 102 AD patients, the model correctly identified 95 cases and misdiagnosed the remaining 7 cases. Among 161 controls, 151 subjects were correctly identified, and 10 were incorrectly considered AD. Therefore, based on the presented data, the ability of ACE-R to discriminate AD from controls using the proposed model, which includes AD-MokACE-R, is highly efficacious.

MCI-MokACE-R and multivariate discriminative analyses

The 10 subitems of ACE-R that measure the same latent concept in this group are listed in Table 2.

MCI versus controls

In this sample, all variables adhered to the maximum limit of 2.50 for the VIF value. Through logistic regression with cross-validation, MCI-MokACE-R, Age, Education, Orientation in time, and Memory Recognition were selected as independent variables that jointly contributed significantly to diagnostic differentiation between MCI and controls (p < 0.05; Table 3). The proposed logarithm for this sample is:

Log = Intercept Beta value (8.714) + ((–0.214×MCI-MokACE-R score) + (0.036×age) + (0.288×years of education) + (–1.415×Orientation in time) + (–0.310×Memory Recognition score))

Investigating the residual deviation of this model revealed no significant difference compared to the perfect model that would fit each data point (p-value χ ² 98.99 GL = 258 = 0.999). No outliers were identified, confirmed by Bonferroni test (p < 0.05).

In cross-validation analysis, the model’s AUC was 0.859 (Fig. 1, Table 4), superior to ACE-R Total score (AUC = 0.714), MMSE (AUC = 0.569), and VLOM (AUC = 0.774). The determined cutoff point was logit τ= –0.64 on the chance scale. Validating the model on the dataset itself, among 161 controls, the model correctly identified 127 cases and misdiagnosed the 34 remaining cases. For 87 cases with MCI, 68 subjects were correctly identified, and 19 cases were incorrectly considered controls. In this analysis, sensitivity and specificity were 78% and 79%, respectively. The PPV and NPV were 66% and 86%, respectively (Table 4).

Therefore, based on the presented data, the ACE-R capacity to discriminate MCI from controls, using the proposed model with MCI-MokACE-R, is highly satisfactory.

DISCUSSION

We investigated the diagnostic value of the ACE-R in AD, bvFTD, and MCI by extracting Mokken scales for AD and MCI and employing logistic regression analysis along with ROC curves. Our results highlight that by appropriately combining selected ACE-R items with demographic factors affecting cognitive performance, the ACE-R demonstrated substantial capacity in distinguishing AD from bvFTD, and AD from controls. Additionally, it revealed high efficiency in differentiating MCI from controls.

Although Item Response Theory (IRT) and the selection and formation of items are insightful approaches, they have not been widely used in the past.42 –44 In a 2015 study using IRT and Mokken analysis, the authors also examined all ACE-R items for hierarchical order analysis and latent trait identification in AD and other dementias. 31 Similarly, we believe that important information might be overlooked when considering total scores. Therefore, we applied the Mokken scaling technique, enabling the extraction of items which share the same latent trait to generate the Mokken scale for AD (AD-MokACE-R) and MCI groups (MCI-MokACE-R).

The AD-MokACE-R in our sample comprised 12 items. However, compared to the 12 items composing the Australian AD Mokken scale, only results in Retrograde Memory, Clock Drawing, and Language – Comprehension converged. In that study, various AD phenotypes were grouped in the same category, such as logopenic variant of primary progressive aphasia. 31 Differences between samples may explain such distinct findings. Nevertheless, a significantly larger sample might provide further insights into these differences.

Contrary to McGrory et al. study, 31 in formulating the Mokken scale for our sample, we assessed the relationship between AD-MokACE-R and demographic factors (sex, age, education) through logistic regression analysis and the discriminative ability of ACE-R using this statistical model through ROC curves.

In the comparison between AD and bvFTD, relevant factors for improved group discrimination included being Male, Age, Orientation in time, Orientation in space, and Memory recall, along with AD-MokACE-R. Obviously, these items (Orientation in time, Orientation in space, and Memory recall) were not selected by the Mokken analysis for AD-MokACE-R. The logit τ= –0.82 cutoff on the probability scale achieved a 92% discriminative power. This result surpasses the AUC of 0.865 published recently using a logarithmic model with ACE-R subscales (Attention and Orientation, Fluency, and Language) plus Age. 22 Furthermore, using this new method, discriminative ability is higher than the VLOM (AUC = 0.829) found in our previous study, 22 with higher sensitivity and specificity than the ones reported by Argentinian authors (81%, and 79%, respectively). 27 Besides, even though the VLOM ratio is a simple formula, there are studies where the developers of ACE-R question its discriminatory power, suggesting the need for other tests with better sensitivity for FTD. 21

ACE-R also proved to be useful to differentiate AD from controls. The relevant factors along with AD-MokACE-R were Education and the remaining ACE-R items Orientation in time and Memory – recall. The model with the cutoff logit τ= –0.45 reached 96% accuracy, with equally remarkable sensitivity, specificity, PPV, and NPV. As this is an unprecedented approach in the literature applied to ACE-R (or any other ACE version), there are no directly comparable data. Nevertheless, we can infer that these represent the accuracy data with the highest and most consistent parameters reported to date when assessing the discriminative capacity between AD and controls.15,19,20,45 , 15,19,20,45 Considering that AD is the most common form of dementia, responsible for 60–70% of cases, and over 60% of them live in low- and middle-income countries, 46 ACE-R may be an accessible tool to assist the diagnostic reasoning of physicians.

Finally, considering the success of this combination of statistical methods, the Mokken scale was generated for the MCI group, which was compared to controls by logistic regression and ROC curves. Like the others, the discriminative ability of the statistical model reached 85%. The model indicated that the relationship between MCI-MokACE-R, Age, Education, Orientation in time, and Memory – recognition better discriminated between the groups. Using the cutoff logit τ= –0.64 on the probability scale, in this analysis, the model achieved good sensitivity and specificity, and demonstrated high reliability by indicating that someone does not have MCI (NPV).

While using only cutoff scores from the raw totals of ACE-R, as done in various studies 21 and preliminary analyses, 20 ACE-R had proven ineffective for this population with cognitive impairment without dementia. Therefore, the new findings presented here strongly indicate that the analysis method used in the literature so far was inadequate or insufficient. Besides offering a wealth of information on cognition in a short time, it becomes evident that ACE-R becomes a more powerful instrument when a set of statistical methods that leverage the most relevant items for differential diagnosis is employed.

Although the ACE-R results proposed in this article have very good accuracy, it is important to note that, in the very early stages of neurocognitive conditions, the method of assessing an individual’s performance through a cross-sectional approach may not be as sensitive and specific as longitudinal analysis. Future studies with two or more successive assessments using the ACE-R may provide a “time index” with quantification of severity over time and enable a more accurate understanding of the disease progression, as first proposed by Ashford and colleagues. 47

We acknowledge that our study has limitations. It is crucial to note that the AD and bvFTD groups demonstrated comparable levels of education. This similarity in educational backgrounds may account for the absence of education detection by the logarithmic model in the analysis comparing bvFTD to AD. As highlighted in a systematic review, 48 education typically influences performance in cognitive tests, including ACE-R.16,35, 16,35 Therefore, new logarithmic model analyses with heterogeneous groups in terms of education, age, and sex may enhance the robustness of our findings. However, for an average education of approximately 11 years, we have sufficient evidence to affirm the quality of the model and the ACE-R.

While the research was carried out in two Brazilian Cognitive Neurology units, it is important to emphasize that potential biases were actively addressed. One might question the variability in test application and examiner differences due to the dual-unit setup. However, it is crucial to note that physicians and neuropsychologists, despite not being blinded to clinical diagnoses, possessed extensive experience, ensuring a high level of expertise in their assessments. Furthermore, strict adherence to recommended administration and scoring modes was maintained, aiming to minimize any subjectivity interference and enhance the overall reliability of the study.

Moreover, it is evident that researchers are increasingly transitioning from traditional paper-and-pencil cognitive tests to computerized assessments, which can even be self-administered. We acknowledge the significant potential of these methods to precisely measure memory and other cognitive functions. They offer accurate and practical diagnostic approaches and can be time-efficient in terms of application and scoring. 49 However, we express concern that the digital format may still present challenges and potentially act as a barrier for the older adult population (with or without dementia risk) or those with lower educational levels, particularly in low- and middle-income countries such as Brazil and many others, where the prevalence of dementia is higher compared to high-income countries. 46 Furthermore, the self-administration of cognitive tests in individuals with suspected bvFTD can be problematic, as it requires cooperation and sustained engagement, which may be compromised in the early stages of the disease. In this context, we think that interaction with an examiner may enhance the likelihood of a successful assessment.

In conclusion, our research suggests that the ACE-R is an effective tool for the assessment of dementia, and it also demonstrates a high potential for identifying individuals at risk. The findings of the present study highlight the need for the use of statistical approaches which address the relationship of cognitive and demographic factors with diagnostic outcomes. Therefore, our study examines the diagnostic value of the ACE-R taking into account the intricate relationship between different variables in distinguishing AD, bvFTD, and MCI. Each of our logarithmic models has been subjected to a thorough validation process, ensuring that they effectively distinguish between groups and provide consistent, reliable data. Moreover, the final ROC curves underscore the enhanced discriminative power of these methods, particularly the Mokken scales for AD (AD-MokACE-R) and MCI (MCI-MokACE-R), outperforming traditional methods documented in previous literature.20 –22 Given its accessibility, the ACE-R may be considered a useful tool for clinicians, especially in regions with limited resources but a high occurrence of dementia. Our statistical methodologies emphasize the importance of further exploration and collaboration in larger-scale endeavors in refining our understanding of the ACE-R utility across diverse diagnostic scenarios. We invite practitioners and researchers to use our online platform (https://cogneuroufmg.org/) for free ACE-R score calculations, including the methods used in this paper.

AUTHORS CONTRIBUTIONS

Viviane Amaral-Carvalho (Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Resources; Validation; Visualization; Writing – original draft; Writing – review & editing); Thais Bento Lima-Silva (Data curation; Investigation; Resources; Writing – review & editing); Luciano Inácio Mariano (Data curation; Investigation; Resources; Writing – review & editing); Leonardo Cruz de Souza (Methodology; Resources; Supervision; Writing – review & editing); Henrique Cerqueira Guimarães (Resources; Writing – review & editing); Valéria Santoro Bahia (Resources; Supervision; Writing – review & editing); Ricardo Nitrini (Resources; Supervision; Writing – review & editing); Maira Tonidandel Barbosa (Resources; Supervision; Writing – review & editing); Mônica Sanches Yassuda (Methodology; Resources; Supervision; Writing – review & editing); Paulo Caramelli (Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Resources; Supervision; Validation; Visualization; Writing – original draft; Writing – review & editing).

Footnotes

ACKNOWLEDGMENTS

We are very thankful to Prof. Eneida Mioshi and Prof. John Hodges for their kind support to this study, since the first step. We also thank Drs. Elisa França de Paula Resende, João Carlos Machado and Rachael Brant Machado Rodrigues for referring some patients to this study.

FUNDING

The authors have no funding to report.

CONFLICT OF INTEREST

Paulo Caramelli is an Editorial Board Member of this journal but was not involved in the peer-review process of this article nor had access to any information regarding its peer-review.

All other authors have no conflict of interest to report.

DATA AVAILABILITY

The data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

References

Foster

, Bondi

, Das

, et al. Quality improvement in neurology: Mild cognitive impairment quality measurement set. Neurology 2019; 93: 705–713.

Dubois

, von Arnim

CAF

, Burnie

, et al. Biomarkers in Alzheimer’s disease: role in early and differential diagnosis and recognition of atypical variants. Alzheimers Res Ther 2023; 15: 1–13.

Amaral-Carvalho

and Caramelli

. Brazilian adaptation of the Addenbrooke’s Cognitive Examination-Revised (ACE-R). Dement Neuropsychol 2007; 1: 212–216.

Mioshi

, Dawson

, Mitchell

, et al. The Addenbrooke’s Cognitive Examination Revised (ACE-R): a brief cognitive test battery for dementia screening. Int J Geriatr Psychiatry 2006; 21: 1078–1085.

de Almeida

, Dalpubel

, Ribeiro

, et al. Subjective cognitive impairment, cognitive disorders and self-perceived health: The importance of the informant. Dement Neuropsychol 2019; 13: 335–342.

Beckert

, Loureiro

, Menta

, et al. Performance of low-educated elders with depression on Addenbrooke’s Cognitive Examination-Revised (ACE-R) test. Dement Neuropsychol 2016; 10: 19–25.

Bento

, Ottaviani

, Brigola

, et al. Use of digital game therapy among elderly persons undergoing dialytic treatment: cognitive aspects and depressive symptoms. Rev Bras Geriatr Gerontol 2018; 21: 447–455.

da Silveira

, Portuguez

. Effects of computer use on cognition, emotional state, quality of life and motor skills of the elderly. Psicol Teor Pesquis 2019; 35: 1–10.

Fukushima

RLM

, Micali

, Carmo

EG do

, et al. Cognitive abilities and physical activity in chronic kidney disease patients undergoing hemodialysis. Dement Neuropsychol 2019; 13: 329–334.

10.

Gesualdo

, Duarte

, Zazzetta

, et al. Cognitive impairment of patients with chronic renal disease on hemodialysis and its relationship with sociodemographic and clinical characteristics. Dement Neuropsychol 2017; 11: 221–226.

11.

Lima-Silva

, Bahia

, Amaral-Carvalho

, et al. Translation, cross-cultural adaptation and applicability of the Brazilian version of the Frontotemporal Dementia Rating Scale (FTD-FRS). Dement Neuropsychol 2013; 7: 387–396.

12.

Ordonez

, Borges

, Kanashiro

, et al. Actively station: Effects on global cognition of mature adults and healthy elderly program using eletronic games. Dement Neuropsychol 2017; 11: 186–197.

13.

Rocha

MSG

, Bassetti

, Oliveira

, et al. Addenbrooke’s Cognitive Examination-Revised is accurate for detecting dementia in Parkinson’s disease patients with low educational level. Dement Neuropsychol 2014; 8: 20–25.

14.

Sobreira

, Pena-Pereira

, Eckeli

, et al. Screening of cognitive impairment in patients with Parkinson’s disease: diagnostic validity of the Brazilian versions of the Montreal Cognitive Assessment and the Addenbrooke’s Cognitive Examination-Revised. Arq Neuropsiquiatr 2015; 73: 929–933.

15.

Tsoi

KKF

, Chan

JYC

, Hirai

, et al. Cognitive tests to detect dementia. JAMA Intern Med 2015; 175: 1450–1458.

16.

César

, Yassuda

, Porto

FHG

, et al. Addenbrooke’s cognitive examination-revised: Normative and accuracy data for seniors with heterogeneous educational level in Brazil. Int Psychogeriatr 2017; 29: 1345–1353.

17.

Alexopoulos

, Ebert

, Richter-Schmidinger

, et al. Validation of the German Revised Addenbrooke’s Cognitive Examination for detecting mild cognitive impairment, mild dementia in Alzheimer’s disease and frontotemporal lobar degeneration. Dement Geriatr Cogn Disord 2010; 29: 448–456.

18.

Lonie

, Parra-Rodriguez

, Tierney

, et al. Predicting outcome in mild cognitive impairment: 4-Year follow-up study. BJPsych 2010; 197: 135–140.

19.

Crawford

, Whitnall

, Robertson

, et al. A systematic review of the accuracy and clinical utility of the Addenbrooke’s Cognitive Examination and the Addenbrooke’s Cognitive Examination—Revised in the diagnosis of dementia. Int J Geriatr Psychiatry 2012; 27: 659–669.

20.

Amaral-Carvalho

, Lima-Silva

, Mariano

, et al. Brazilian Version of Addenbrooke’s Cognitive Examination - Revised in the differential diagnosis of Alzheimer’s disease and behavioral variant frontotemporal dementia. Arch Clin Neuropsychol 2022; 37: 437–448.

21.

Hodges

and Larner

. Addenbrooke’s Cognitive Examinations: ACE, ACE-R, ACE-III, ACEapp, and M-ACE. In: Larner

(eds) Cognitive Screening Instruments, Cham: Springer, 2017, pp. 109–137.

22.

Amaral-Carvalho

, Bento Lima-Silva

, Inácio Mariano

, et al. Predicting dementia due to Alzheimer’s disease and behavioral variant frontotemporal dementia using algorithms with the Addenbrooke’s Cognitive Examination-Revised subscores combined with sociodemographic factors. Curr Alzheimer Res 2023; 20: 341–349.

23.

Mathuranath

, Nestor

, Berrios

, et al. A brief cognitive test battery to differentiate Alzheimer’s disease and frontotemporal dementia. Neurology 2000; 55: 1613–1620.

24.

McKhann

, Knopman

, Chertkow

, et al. The diagnosis of dementia due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 2011; 7: 263–269.

25.

Rascovsky

, Hodges

, Knopman

, et al. Sensitivity of revised diagnostic criteria for the behavioral variant of frontotemporal dementia. Brain 2011; 134: 2456–2477.

26.

Bier

, Ventura

, Donckels

, et al. Is the Addenbrooke’s Cognitive Examination effective to detect frontotemporal dementia? J Neurol 2004; 251: 428–431.

27.

Torralva

, Roca

, Gleichgerrcht

, et al. Validation of the spanish version of the addenbrooke’s cognitive examination-revised (ACE-R). Neurología (English Edition) 2011; 26: 351–356.

28.

Olfati

, Shoeibi

and Litvan

. Clinical spectrum of tauopathies. Front Neurol 2022; 13: 1–25.

29.

Santos

, Cabreira

, Rocha

, et al. Blood biomarkers for the diagnosis of neurodegenerative dementia: a systematic review. J Geriatr Psychiatry Neurol 2023; 36: 267–281.

30.

Borson

, Scanlan

, Brush

, et al. The Mini-cog: A cognitive ‘vital signs’ measure for dementia screening in multi-lingual elderly. Int J Geriatr Psychiatry 2000; 15: 1021–1027.

31.

McGrory

, Starr

, Shenkin

, et al. Does the order of item difficulty of the Addenbrooke’s Cognitive Examination add anything to subdomain scores in the clinical assessment of dementia? Dement Geriatr Cogn Dis Extra 2015; 5: 155–169.

32.

Chaves

MLF

, Camozzato

, Godinho

, et al. Validity of the Clinical Dementia Rating Scale for the detection and staging of dementia in Brazilian patients. Alzheimer Dis Assoc Disord 2007; 21: 210–217.

33.

Morris

. Clinical Dementia Rating: a reliable and valid diagnostic and staging measure for dementia of the Alzheimer type. Int Psychogeriatr 1997; 9: 173–176.

34.

Petersen

. Mild cognitive impairment. N Engl J Med 2011; 364: 2227–2234.

35.

Amaral-Carvalho

, Caramelli

. Normative data for healthy middle-aged and elderly performance on the Addenbrooke Cognitive Examination-Revised. Cogn Behav Neurol 2012; 25: 72–76.

36.

Foss

, Amaral-Carvalho

, Machado

, et al. Mattis Dementia Rating Scale (DRS): Normative data for the Brazilian middle-age and elderly populations. Dement Neuropsychol 2013; 7: 374–379.

37.

Mattis

. Dementia Rating Scale: professional manual. Florida: Psychological Assessment Resources, 1988.

38.

Camozzato

, Kochhann

, Simeoni

, et al. Reliability of the Brazilian Portuguese version of the Neuropsychiatric Inventory (NPI) for patients with Alzheimer’s disease and their caregivers. Int Psychogeriatr 2008; 20: 383–393.

39.

Cummings

, Mega

, Gray

, et al. The Neuropsychiatric Inventory: Comprehensive assessment of psychopathology in dementia. Neurology 1994; 44: 2308–2308.

40.

Fazekas

, Chawluk

, Alavi

, et al. MR signal abnormalities at 1.5 T in Alzheimer’s dementia and normal aging. AJR Am J Roentgenol 1987; 149: 351–356.

41.

Mokken

. A Theory and Procedure of Scale Analysis: With Applications in Political Research. Berlin, New York: De Gruyter Mouton; 1971.

42.

Scanlan

and Borson

. The Mini-Cog: Receiver operating characteristics with expert and naïve raters. Int J Geriatr Psychiatry 2001; 16: 216–222.

43.

Ashford

, Kolm

, Colliver

, et al. Alzheimer patient evaluation and the Mini-Mental State: item characteristic curve analysis. J Gerontol 1989; 44: 139–146.

44.

Mendiondo

, Ashford

, Kryscio

, et al. Designing a brief Alzheimer screen (BAS). J Alzheimers Dis 2003; 5: 391–398.

45.

Larner

, Mitchell

. A meta-analysis of the accuracy of the Addenbrooke’s Cognitive Examination (ACE) and the Addenbrooke’s Cognitive Examination-Revised (ACE-R) in the detection of dementia. Int Psychogeriatr 2014; 26: 555–563.

46.

World Health Organization. Fact sheet: dementia. https://www.who.int/news-room/fact-sheets/detail/dementia (2023, accessed 27 April 2024).

47.

Ashford

, Shan

, Butler

, et al. Temporal quantification of Alzheimer’s disease severity: ‘Time Index’ Model. Dement Geriatr Cogn Disord 1995; 6: 269–280.

48.

Seblova

, Berggren

and Lövdén

. Education and age-related decline in cognitive performance: systematic review and meta-analysis of longitudinal cohort studies. Ageing Res Rev 2019; 58: 101005.

49.

Ashford

, Schmitt

, Bergeron

, et al. Now is the time to improve cognitive screening and assessment for clinical and research advancement. J Alzheimers Dis 2022; 87: 305–315.