Discrepancy Between Patient and Caregiver Estimate of Apathy Predicts Dementia in Patients with Amnestic Mild Cognitive Impairment

Abstract

Background:

Apathy is a frequent behavioral symptom of Alzheimer’s disease (AD). The Apathy Evaluation Scale (AES) is a tool exploring the perception of apathy by both caregivers (CG-AES) and patients (PT-AES), and the discrepancy in their ratings is a proxy of patients’ disease unawareness.

Objective:

To assess in a cohort study of patients with amnesic mild cognitive impairment (aMCI) whether apathy and awareness of apathy predict progression to dementia and timing.

Methods:

From the global AES scores of 110 patients with aMCI and their caregivers, we obtained two principal indices for analysis: 1) ‘Apathy’, the mean of PT-AES and CG-AES, and 2) ‘Discrepancy’, obtained by subtracting CG-AES from PT-AES. Patients were followed with visits every six months for three years or until dementia. AES indices and the principal demographical/neuropsychological variables were filtered from multicollinearity. The most robust variables entered a logistic regression model and survival analyses (Cox regression, log-rank test of Kaplan-Meier curves) to estimate which predicted the risk and timing of progression, respectively.

Results:

Sixty patients (54.5%) developed dementia (57 AD) after 6.0–36.0 months, 22 (20%) remained in an MCI stage, and 28 (25.5%) dropped out. ‘Discrepancy’ was a robust and accurate predictor of the risk of progression (AUC = 0.73) and, after binarization according to a computed cutoff, of timing to dementia.

Conclusion:

A structured evaluation of apathy, both self-assessed and estimated by caregivers, can provide useful information on the risk and timing of progression from aMCI to dementia. The discrepancy between the two estimates is a fairly reliable index for prediction purposes as a proxy of disease unawareness.

Keywords

Alzheimer’s disease apathy awareness dementia discrepancy mild cognitive impairment

INTRODUCTION

Apathy is a complex phenomenon defined as diminished or loss of motivation regarding goal-directed behavior, cognitive ability, or emotion that causes significant impairment in daily life [1]. Three aspects of apathy with different clinical manifestations have been mostly recognized, namely 1) affective-emotional apathy, i.e., the inability to use emotional context to guide behavior, 2) behavioral apathy, i.e., the deficit in initiating and maintaining spontaneous physical activity that requires several prompts, and 3) cognitive apathy, i.e., the loss of motivation to participate in goal-directed behavior [1]. Apathy is one of the most frequently reported among the behavioral symptoms occurring in Alzheimer’s disease (AD) (pooled prevalence of 50%) [2] and is related to faster cognitive decline and higher caregiver burden [3 –5]. Moreover, apathy may precede cognitive symptoms by up to 5 years and is associated with an increased risk of progression from mild cognitive impairment (MCI) to AD dementia (ADD) [6, 7]. Indeed, apathy in AD has strong neurobiological roots. Abnormalities in frontal regions (associated with impairments in planning and decision making) and anterior cingulate cortex (related to emotional blunting and loss of motivation) have been associated with apathy in AD through structural (magnetic resonance imaging, MRI, computed tomography, CT) and functional (fMRI, positron emission tomography, PET, and single photon emission computed tomography, SPECT) neuroimaging techniques, as thoroughly reviewed by Stella et al. [8]. Studies on functional connectivity have highlighted the role of reduced connectivity between the left insula and the right superior parietal cortex in apathetic ADD patients [9]. From a pathological point of view, in a study of subjects with MCI a significant association between increased apathy and greater cortical amyloid-β (Aβ) deposition was highlighted [10]. Similar results were found in another study in ADD patients, revealing that prefrontal Aβ deposition correlates with apathy [11]. Apathy has also been associated with greater neurofibrillary tangle burden of the anterior cingulate cortex in moderate to severe ADD, increased cerebrospinal fluid total tau, and hyperphosphorylated tau in mild ADD [12, 13].

Since apathy is a frequent and early symptom and is related to the accumulation of AD pathology over time, it could be a good predictor of progression from the MCI to the dementia stage. The majority of data supporting this hypothesis have been obtained with the apathy subscale of the Neuropsychiatric Inventory [7 , 14–16], while few data are available using the specific Apathy Evaluation Scale (AES) [17]. Such a scale explores not only the informant but also the patient’s perception and focuses on each of the three dimensions of apathy [1]. In the only prospective study using the AES in individuals with MCI and cognitively normal elderly patients, the authors showed that self-reported AES may be less reliable in subjects with MCI and that the clinician-reported AES predicted progression from MCI to ADD [17]. However, that study was limited by the relatively small sample size and a mean follow-up time of approximately 1.4 years and did not investigate the possible relative weight of the three apathy dimensions in the prediction of progression from MCI to ADD. On the other hand, there is no research studying the impact of the discrepancy between self- and caregiver-ratings of apathy on the risk of conversion from amnestic MCI (aMCI) to dementia. This discrepancy is one of the principal methodologies to assess awareness in neurodegenerative pathologies. Previous studies indicated that apathy positively correlated with impaired awareness in AD, allowing discrimination among healthy controls, mild and moderate-to-moderately-severe AD patients [18 –20].

With the aim to assess the role of apathy and awareness of apathy in the risk of progression from MCI to dementia, we administered the AES to patients with aMCI and their caregivers in a prospective study with a follow-up time of three years or until progression to dementia.

MATERIAL AND METHODS

Patients

We prospectively recruited consecutive subjects complaining of memory deficit who accessed our memory clinic for the first time, according to the following main inclusion criteria: age range 50–85 years; onset of cognitive symptoms between 6 months and 3 years before assessment as estimated by the patient and/or the caregiver; a Mini-Mental State Examination (MMSE) score higher than or equal to 23 that ensured that all patients had preserved global cognition score as derived from the Italian elderly population normative [21]. The main exclusion criteria were: the presence of dementia according to the Clinical Dementia Rating and Instrumental Activities of Daily Living (IADL) scales [22]; illiteracy; cognitive impairment explained by systemic or iatrogenic causes; other neurological disorders; a diagnosis of an established psychiatric disorder according to the DSM-V. Mild-to-moderate depressive symptoms were not an exclusion criterion if the Geriatric Depression Scale with 15 items (GDS-15) scored below 10.

We performed MRI (or CT if MRI was contraindicated) and excluded those patients with cognitive impairment of probable vascular origin, according to Gorelick et al. criteria [23], or due to other brain diseases. The presence of white matter hyperintensities was not an exclusion criterion providing the Wahlund scale score was <2 in all regions [24].

All subjects underwent an extended neuropsychological test battery, which included the trail-making test, Stroop color and color word tests for visual attention, task switching, and executive functions, Rey auditory verbal learning and Babcock Story Recall tests for episodic and logical memory, categorical and phonological verbal fluency tests to assess language and freehand copying of drawings for visuospatial abilities. The references to the Italian versions of these tests and the corresponding normative data are reported in a previous paper [25].

We included 110 patients with either single or multi-domain aMCI diagnosis [26], as collegially established by the neurologist or the geriatrician who was in charge of the patient with an expert neurologist (FN) and neuropsychologist (NG) based on the substantial loss of autonomies (impairment in instrumental activities of daily living) and the results of neuropsychological tests (scores >1.5 SD below the normative mean on any test within a specific cognitive domain or >1.0 SD below the normative mean on at least two separate tests of one cognitive domain or in at least three cognitive domains [27]). As it was beyond the aim of our study, we excluded 70 patients with non-amnestic MCI referring to our center at the same time.

To assess the degree of apathy, the Apathy Evaluation Scale (AES) was administered to both patients (AES-S) and caregivers (AES-I) [1]. Briefly, the AES is composed of four sub-scales exploring the three main domains of apathy, namely behavioral apathy, cognitive apathy, emotional aspects of goal-directed behavior, and by a not further characterized ‘mixed’ domain (hereafter ‘other’) made by 3 items exploring initiative, motivation, and insight of own problems. These four subscales have different score ranges and altogether contribute to computing a global AES score, ranging from 18 (no apathy) to 72. The patient and caregiver versions are identical. The suggested cut-off scores of 41 for AES-I and 40 for AES-S, which are often used to define significant apathy, were not applied to the present data analysis. From the global AES scores, we obtained the two principal indices for analysis, namely 1) the ‘Apathy’, as the mean of the apathy scored by the patient (PT-AES) and the pertinent caregiver (CG-AES) computed as (PT-AES+ CG-AES)/2 and 2) the ‘Discrepancy’, obtained subtracting caregiver- from self-ratings (=PT-AES minus CG-AES) as done in previous research to assess patients’ impaired awareness of current deficits [18].

Therefore, higher values of ‘Apathy’ and lower values of ‘Discrepancy’ indicate greater impairment in terms of patients’ apathy and awareness, respectively. Additionally, the same two indices were computed for each of the four AES subscales.

According to clinical judgment, further biomarker investigations were performed to ascertain aMCI etiology. Based on the clinical-neuropsychological presentation, MRI/CT findings, and results of biomarkers, a diagnostic hypothesis was generated by the treating neurologist (FM, FN, or MP), and treatment started according to current options. These mainly included the control of systemic comorbidities and vascular risk factors as needed, the use of mild-to-moderate doses of antidepressant agents if indicated, and acetylcholinesterase inhibitors in those patients receiving a diagnosis of MCI due to AD according to the NIA-AA criteria [28].

Follow-up visits

All patients underwent 6-month based follow-up visits for at least 3 years. Twenty-eight patients dropped out because of several reasons, mainly the decision to avoid further visits (18 patients), the onset of other severe systemic pathologies (3 patients), referral to other centers (4 patients), or death (3 instances). Sixty patients (converters) developed dementia after 6.0–36.0 months (mean 22.3±9.7). Progression from MCI to distinct dementias was established during clinical follow-up by one of the physicians mentioned above, considering the pertinent diagnostic criteria and supported by biomarkers, when available, but blinded to AES score. ADD was diagnosed in 57 according to the National Institute of Aging-Alzheimer Association criteria [29] (28 with intermediate likelihood based on [¹⁸F]-Fluorodeoxyglucose-PET (FDG-PET) and 8 with high likelihood from an additional positive amyloid-PET). Behavioral variant frontotemporal dementia was diagnosed in 3 (all with consistent dysmetabolic findings at FDG-PET) [30]. Among the AD diagnosis, 3 patients developed frontal variant AD according to the proposed criteria [31] and one posterior cortical atrophy [32]. Twenty-two patients (non-converters) remained in an MCI stage (mean follow-up time 37.1±8.5 months, range 36.0–46.7).

All procedures performed were in accordance with the ethical standards laid down in the 1975 Declaration of Helsinki or comparable ethical standards.

Statistics

Comparison between converters and non-converters

We compared demographic, neuropsychological test, and AES scores, including ‘Apathy’ and ‘Discrepancy’ indices, between the subgroups of converters and non-converters (Wilcoxon rank-sum test with false discovery rate, FDR, method to account for multiple comparisons and Chi-squared test for categorical variables).

Prediction of progression to dementia

As our primary analysis method, we opted to use a multivariate generalized linear regression model (GLM), which enables to determine the relationship between many predictor variables and a binary outcome. The GLM in our study aimed to find the most robust predictors of progression from aMCI to dementia. We employed a feature selection approach in conjunction with the regression to extract the most important covariates, intending to reduce the number of variables to the minimum due to the small sample size.

Among the available information, we considered a priori the following variables of interest: sex, age, years of formal education, and scores of MMSE (at baseline), 15-item Geriatric Depression Scale (GDS-15), ‘Apathy’ and ‘Discrepancy’. Moreover, we identified as further potential predictors to be selected for the GLM those neuropsychological variables that resulted significantly different between converters and non-converters (p < 0.05, uncorrected for multiple comparisons). The possible multicollinearity among all these covariates was assessed through correlation analysis and the evaluation of the variance inflation factor (VIF) to assess whether some covariate could be incorporated into its proxy. We looked for the smallest significant model using the Elastic net approach (10-fold cross-validation) and then using the Akaike Information Criterion (AIC) to verify the minimal selection of covariates to be included in the GLM using the dementia progression state (i.e., yes/no, excluding the drop-out subgroup of patients) as the dependent variable. Finally, a receiver operating characteristic (ROC) analysis was performed for the GLM prediction output and the most significant predictor score using the Youden Index (YI) to compute the cutoffs.

As a supplementary analysis, we performed a principal component analysis (PCA) of the PT-AES and CG-AES subscales, normalized in a 0-to-1 scale to account for their different ranges. The PCA provides a linear data transformation and adapts well to exploratory data analysis. The resulting loads, namely the coefficients of the linear combination of the variables from which the principal components are constructed, are scalar products among vectors and allow a simple qualitative interpretation of the weight each variable has on the principal component. This was aimed to assess the principal components explaining the AES subscores (without a priori choice) and whether one or more subscales had greater weight than the others in patients or caregivers.

Survival analysis

The same process was used to filter and select the covariates to enter a Cox regression analysis of their predictive power with respect to the time-from-baseline, defined as the elapsed time from the first visit (i.e., when the diagnosis of MCI was made) to dementia. In this analysis, we could include the patients who dropped out of the study (regarded as right-censored). Testing the time-dependent covariates is equivalent to testing for a non-zero slope in a linear regression of the scaled Schoenfeld residuals on functions of time. We have verified that the slope was zero within 95% CI for all covariates, verifying the hypothesis of the proportional risk after running the Cox model. Furthermore, we binarized the covariates according to the relevant cutoff based on the YI value to run a Cox regression analysis with binary variables and a log-rank test on Kaplan-Maier curves. This allowed searching for the significant contribution and relative risk (RR) of the covariates about the time of progression to dementia.

As supplementary, we performed the same survival analyses (Cox regression before and after binarization of covariates, log-rank test on Kaplan-Maier curves) but considering all the variables of interest before the selection filter by the Elastic net and AIC methods, namely sex, age, years of education, MMSE, GDS-15, ‘Apathy’ and ‘Discrepancy,’ and those neuropsychological variables that resulted significantly different between converters and non-converters (Rey auditory verbal learning test, RAVLT - immediate (IR) and delayed recall (DR) Stroop Color Word (Stroop-CW), and Trail making test A (TMT-A). Additionally, we added to this analysis the prediction output of the GLM.

The threshold for significance was set at p < 0.05 in all the analyses.

For the statistical analysis, the MATLAB/Simulink software, version R2021b, was used (The MathWorks Inc., Natick, MA, USA).

RESULTS

Comparison between converters and non-converters

As shown in Table 1, non-converters performed better for MMSE (p = 0.005), RAVLT-IR (p = 0.004), RAVLT-DR (p = 0.032), TMT-A (p = 0.010), and Stroop-CW (p = 0.011). CG-AES (p = 0.013) and ‘Discrepancy’ (p = 0.0009) (both these latter two in the global, cognitive, and emotional subdomains) were less affected in the non-converters, while ‘Apathy’ (p = 0.266) did not significantly differ between converters and non-converters (see Fig. 1). Among them, only MMSE (q = 0.035), RAVLT-IR (q = 0.035), CG-AES cognitive subscale (q = 0.035) and ‘Discrepancy’ (global, q = 0.030, and cognitive, (q = 0.035)) remained statistically significant when correcting for multiple comparisons.

Table 1

Main demographic characteristics and neuropsychological test scores of patients

	All patients	Converters	Non-converters	Statistics (U)	p ^*	q^#	Drop-out
	(n = 110)	(n = 60, 54.5%)	(n = 22, 20%)				(n = 28, 25.5%)
Age	76.74±5.51	77.68±5.19	75.5±5.33	484.0	0.064	0.160	75.67±6.10
Sex (M:F)	47:63	22:38	13:9	399.5	0.069		12:16
Education (y)	9.76±4.34	9.51±4.11	10.0±4.42	618.5	0.741	0.728	10.10±4.85
MMSE	26.62±1.82	26.25±1.68	27.5±1.87	380.5	0.005	0.035	26.71±1.86
GDS-15	3.27±2.58	3.01±2.5	2.86±2.37	452.5	0.664	0.675	3.75±2.87
RAVLT- IR	18.95±7.2	17.45±6.7	21.30±4.95	553.0	0.004	0.035	20.23±9.01
RAVLT- DR	2.3±2.25	1.70±1.8	2.86±2.29	409.0	0.032	0.096	3.11±2.73
Babcock Story Recall	6.33±4.25	5.62±3.6	7.2±4.93	347.0	0.265	0.364	7.17±4.79
TMT-A	80.9±45.19	85.03±47.03	59.4±16.93	399.5	0.010	0.052	89.10±52.05
TMT-B	279.93±155.3	282.81±150.6	251.04±157.06	618.5	0.278	0.365	300.9±163.1
Stroop Color	31.62±8.36	30.7±7.8	33.36±8.133	520.5	0.344	0.417	32.07±9.59
Stroop CW	10.31±6.25	8.46±4.72	11.9±5.34	361.0	0.011	0.052	12.48±8.26
Phonological verbal Fluency	26.47±7.86	25.55±8.3	28.59±6.92	512.5	0.123	0.207	26.78±7.7
Semantic verbal Fluency	24.35±8.21	22.86±7.08	26.22±7.23	489.5	0.074	0.167	26.07±10.54
Figure Copy, simple	8.84±2.07	8.55±2.25	9.4±1.7	504.0	0.118	0.207	9.0±1.83
Patient AES	35.28±7.55	35.41±7.47	36.54±7.06	581.5	0.415	0.435	34.0±8.14
behavioral	9.38±2.46	9.36±2.11	9.72±2.52	637.0	0.811	0.774	9.14±3.07
cognitive	15.78±3.82	15.76±3.82	16.6±3.62	551.5	0.257	0.364	15.17±3.98
emotional	3.75±1.24	3.75±1.33	3.63±1.0	648.0	0.900	0.810	3.85±1.23
other	6.25±1.73	6.48±1.73	6.59±1.62	644.5	0.871	0.807	5.46±1.59
Caregiver AES	41.14±9.01	43.13±8.77	37.45±7.78	426.5	0.013	0.054	39.75±9.53
behavioral	10.69±2.89	11.15±2.97	9.91±2.74	510.0	0.115	0.207	10.32±2.7
cognitive	18.68±4.32	19.61±4.14	16.72±3.56	399.0	0.005	0.035	18.21±4.73
emotional	4.64±1.46	4.86±1.55	4.0±1.15	440.0	0.018	0.063	4.64±1.36
other	7.10±1.83	7.45±1.77	6.81±1.81	515.0	0.125	0.207	6.57±1.85
Apathy AES	38.21±6.92	39.28±6.89	37.00±6.64	553.0	0.266	0.364	36.88±7.07
behavioral	10.04±2.15	10.26±2.06	9.82±2.19	576.5	0.384	0.434	9.73±2.33
cognitive	17.23±3.14	17.69±3.14	16.66±3.17	567.5	0.336	0.417	16.70±3.04
emotional	4.20±1.11	4.31±1.20	3.82±0.87	501.5	0.095	0.199	4.25±1.07
other	6.67±1.49	6.97±1.43	6.70±1.54	577.0	0.388	0.434	6.02±1.43
Discrepancy AES	–5.85±9.22	–7.72±8.71	–0.91±6.70	351.0	0.0009	0.030	–5.75±10.71
behavioral	–1.31±3.21	–1.78±3.13	–0.18±2.94	485.5	0.066	0.160	–1.18±3.45
cognitive	–2.90±5.20	–3.85±4.91	–0.14±3.37	379.5	0.003	0.035	–3.04±6.29
emotional	–0.88±1.55	–1.12±1.64	–0.36±1.29	461.0	0.033	0.095	–0.79±1.50
other	–0.85±1.93	–0.97±2.03	–0.23±1.54	544.0	0.221	0.347	–1.11±1.95

MMSE, Mini-Mental State Examination; GDS-15, 15-item Geriatric Depression Scale, RAVLT, Rey Auditory Verbal Learning Test –immediate (IR) or delayed recall (IR); TMT, Trail Making Test; CW, color-word; AES, Apathy Evaluation Scale; U, test statistic. ^* p-values from Wilcoxon rank-sum test (significance threshold set at p < 0.05). No significance when correcting for multiple comparisons. ^# FDR adjusted p-value. Significant values are in bold.

Fig. 1

Box and whiskers plots comparing the scores of ‘Discrepancy’, ‘Apathy’ (A, B), and ‘Discrepancy’ for the subscales (C-F) in converter and non-converter subgroups. The center line denotes the median value, while the blue box contains the 25th to 75th percentiles of dataset. The black whiskers mark the 5th and 95th percentiles, and values beyond these upper and lower bounds are considered outliers, marked with light blue, empty, circles. The red line indicates the cutoff value based on the Youden Index. Lower values of ‘Discrepancy’ and higher values of ‘Apathy’ indicate greater impairment.

Prediction of progression to dementia

Therefore, the scores of RAVLT-IR, RAVLT-DR, TMT-A, and Stroop-CW tests were added as variables of interest in the correlation analysis along with those chosen a priori, namely sex, age, education, GDS-15, ‘Apathy’, and ‘Discrepancy’. We found that MMSE significantly correlated with all the neuropsychological test (RAVLT-IR, RAVLT-DR, TMT-A, and Stroop-CW) and education, which in turn correlated with TMT-A, additionally; age with TMT-A and Stroop-CW; ‘Apathy’ with ‘Discrepancy’ and GDS-15; RAVLT-IR with RAVLT-DR. The variance inflation factor (VIF) for all covariates was less than the 2.5 cutoff for collinearity, mostly used in the literature [33]. Although, VIF values were highest for RAVLT-DR (VIF = 1.94) and RAVLT-IR (VIF = 1.86), which were also highly correlated (p < 0.0001), suggesting mutual collinearity. Details are given in Supplementary Table 1.

The Elastic net and the AIC methods showed that the best GLM could be minimized to five regressors, namely MMSE, RAVLT-IR, TMT-A, Stroop-CW, and ‘Discrepancy’ scores, with the dementia progression state as the dependent variable. The GLM yielded ‘Discrepancy’ (p = 0.011), RAVLT-IR (p = 0.021), and MMSE (p = 0.044) as the most significant predictors of progression to dementia (Table 2).

The ROC analysis showed a fair discriminatory ability in terms of progression to dementia of the GLM (area under the curve, AUC = 0.86) and the most significant predictor, namely the ‘Discrepancy’ (AUC = 0.73) (Fig. 2).

Table 2

The generalized logistic regression model

Variables	Estimate	OR	CI 95%	p
MMSE	–0.41	0.66	0.45–0.99	0.044
RAVLT-IR	–0.13	0.88	0.79–0.98	0.021
TMT-A	0.02	1.02	1.00–1.05	0.103
Stroop-CW	–0.10	0.90	0.79–1.03	0.117
Discrepancy	–0.11	0.90	0.83–0.98	0.011

The generalized logistic regression model (GLM) of the resulting variables after filtering for collinearity and using the dementia progression state (i.e., yes/no, excluding the drop-out subgroup of patients) as the dependent variable is shown. The most significant predictors of progression from MCI to dementia (p < 0.05) are in bold. OR, odds ratio; CI, confidence interval. Other abbreviations as in Table 1.

Fig. 2

Discriminatory performance in the generalized linear regression model (GLM) prediction and ‘Discrepancy’ scores between converters and non-converters. The ROC analysis showed a fair-to-good discriminatory ability in terms of progression to dementia of the GLM (area under the curve, AUC = 0.86; sensitivity = 0.78, specificity = 0.82; cutoff = 0.66) and ‘Discrepancy’ (AUC = 0.73, sensitivity = 0.82, specificity = 0.65; cutoff = –4.00). Cutoffs based on the Youden Index are indicated with the dot.

The PCA analysis confirmed that ‘Apathy’ (i.e., the weighted average of PT-AES and CG-AES) and ‘Discrepancy’ (i.e., the contrast between PT-AES and CG-AES) are the principal components of the AES subscales (Supplementary Figure 1). None of the AES subscores alone resulted in having a strong effect on either ‘Apathy’ or ‘Discrepancy’ (load ≤0.5) (Supplementary Figure 1A, B). Nonetheless, when evaluating the ‘Discrepancy,’ the relevant contribution of each subscale had a similar ordinality in both patients and caregivers and was highest for the emotional and lowest for the behavioral ones (Supplementary Figure 1C).

Survival analysis

The Elastic net regression and the AIC methods showed that the best Cox regression analysis could be minimized to two regressors, namely education and ‘Discrepancy’ scores, with time-from-baseline as the dependent variable. The ‘Discrepancy’ resulted in a significant fitting of the survival curve in this model only when considering binary variables (RR = 1.72 in case of values ≤–4; p = 0.048) (Table 3).

Table 3

The Cox regression analysis with respect to time-from-baseline

Variables	β	HR	CI 95%	p
Education	–0.04	0.96	0.90–1.03	0.237
Discrepancy	–0.01	0.99	0.96–1.01	0.342
Binary variables^§	β	HR	CI 95%	p
Education (<11)	0.47	1.60	0.91–2.80	0.097
Discrepancy (< –4)	0.54	1.72	1.00–2.94	0.048

^§According to the relative cutoffs based on the Youden Index. The condition relatively to the cutoff (<or ≥) is set to obtain positive β values. HR, hazard ratio. Other abbreviations as in Tables 1 and 2.

The supplementary Cox regression analysis with unfiltered covariates yielded no significant results (Supplementary Table 2). However, TMT-A (RR = 2.36, p = 0.006) and ‘Apathy’ (RR = 2.27, p = 0.026) were significant predictors of time of progression to dementia when binarized according to the pertinent cutoff (Supplementary Table 3). Finally, the log-rank test showed that survival curves -i.e., time from baseline to dementia- of binarized variables were statistically different for RAVLT-DR (χ= 4.91, p = 0.027), TMT-A (χ= 9.26, p = 0.002), ‘Apathy’ (χ= 7.84, p = 0.005), ‘Discrepancy’ (χ= 4.27, p = 0.039) (Fig. 3).

Fig. 3

Kaplan-Meier curves. Survival curves for the covariates of interest: time from baseline to dementia (in months) is displayed in orange and green when the values of the variables were below (<) or above than/equal to (≥) the relevant cutoff computed through the Youden index (in brackets), with survival probability in the Y axis. The p-values of the log-rank test on the survival curves are shown for each variable (bold values denote significance at the <0.05 level). GLM, generalized linear regression model (GLM) prediction score, others as in Table 1.

DISCUSSION

In this study, we have analyzed the ability of apathy and awareness of apathy to predict the progression to dementia in patients with aMCI using the AES scale, specifically validated for AD [1]. By fixing a priori at least three years of follow-up, we minimized the risk of including late converter patients in the non-converter group. As expected by the inclusion of patients with aMCI, most converted to ADD, with an annual progression rate to ADD of 23.2%, which fits with the literature data [34, 35].

The ability of the AES to evaluate the patient apathy, both self-perceived and caregiver-assessed, proved particularly relevant. Their ‘Discrepancy’ significantly predicted conversion to dementia together with MMSE and RAVLT-IR in the most compact linear regression model we could build from several demographical and neuropsychological variables of interest. This traces the results of the comparisons between converters and non-converters, where such variables disclosed a significant difference even after the most rigorous FDR correction.

The finding of ‘Discrepancy’ as a reliable predictor highlights the meaningful role of the caregiver as an informant and that impaired self-awareness of cognitive and emotional deficits, albeit with different expression among patients [36], is a clinically relevant feature of MCI [37, 38] and AD [19]. Apathy may be difficult to identify since it does not strictly correlate with the performances on neuropsychological tests [39]. Previous studies using the apathy sub-score of the Neuropsychiatric Index (NPI) showed that patients with both aMCI and significant apathy on NPI had an almost sevenfold risk of progression to ADD compared to aMCI patients without apathy [7]. This risk increased with a higher NPI total score and a higher number of affected NPI items [40]. However, these studies did not analyze the relationship between the NPI apathy score in predicting the progression timing, nor did they evaluate the interaction between caregivers’ and patients’ scores. This latter was a reliable proxy of the patient’s lack of awareness of their condition, namely anosognosia [19]. Particularly, previous evidence pointed out that the assessment of the discrepancy between the ratings of the patient and the relative caregiver about several cognitive, mood/behavioral, and functional domains is a practical tool to assess anosognosia [18, 19]. Using the caregiver-patient discrepancy strategy, Jacus et al. [20] found apathy was the only awareness dimension allowing discrimination among healthy controls, mild and moderate-to-moderately-severe AD patients, whereas others, i.e., cognitive, autonomy in daily living, and behavioral, only distinguished patients from healthy controls. In this scenario, the caregiver more accurately detects the patient’s apathy, and the discrepancy with the patient self-rating— as an expression of the reduced insight into apathy by patients— proves to parallel AD course and cognitive decline. Unlike previous studies, ours is the first to have focused on the ability of the discrepant ratings of apathy at baseline to predict progression to dementia in the mid-term. As for the neurobiological roots, it has been shown that metabolism of the cingulate cortex directly correlates with awareness of cognitive deficit in MCI patients, independent of the severity of cognitive impairment [41]. In the same study, patients with lower awareness and metabolic levels in the cingulate cortex tended to convert more frequently to dementia in the short-to-medium term than those with preserved awareness and cingulate metabolism.

AES consists of subscales that evaluate the three main domains of apathy, namely affective-emotional, behavioral, and cognitive apathy. Our choice to focus on ‘Discrepancy’ as a reliable index of apathy is substantiated by the supplementary principal component analysis (PCA) confirming that both ‘Apathy’ (i.e., the weighted average of PT-AES and CG-AES) and ‘Discrepancy’ are the principal components of the AES subscales. The same PCA indicated that none of the AES subscores strongly affected either ‘Apathy’ or ‘Discrepancy’. This may be the consequence of the concept of apathy, which is a multifaceted and heterogeneous psychological-behavioral state in which the three sub-domains can vary greatly in weight from subject to subject. Therefore, it seems more likely that a comprehensive score succeeds in picking up the apathy phenomenon and its clinical correlates rather than somewhat artifactual and reductive segregation in sub-domains. However, in the PCA, the emotional subscale was found to have the highest relative weight on the total ‘Discrepancy’ assessment, whereas the behavioral had the lowest, and this ordinality was both in the rating by patients and caregivers. This indicates how suppression of emotions defines apathy [1] and, along with cognitive deficits, is more sensitively detected by caregivers recognizing the highest impairment in faster converters.

As for the timing of the progression to dementia from the first visit, the ‘Discrepancy’ and the years of formal education remained the principal amid different putative predictors after filtering for collinearity. However, only binarized ‘Discrepancy’ yielded a significant result, so patients with a score lower than < –4 are 72% more likely to progress to dementia, and this was consistent with the survival curve analysis where patients with a higher difference between patient and caregiver’s ratings had a shorter progression time (25.7 versus 30.2 months). As for the lack of significance of ‘Discrepancy’ as a continuous measure, the effect of the relatively small sample size and the estimation of time-from-baseline— that might be suboptimal when dealing with non-acute events (i.e., the onset of dementia)— needs to be considered. Furthermore, the dichotomization of variables can result in a loss of precision and greater difficulty in interpretation but helps assess the relative risk of different predictors. Finally, we acknowledge that the relationship between the Discrepancy in the AES measure and conversion to dementia in the Cox model may not be linear and that this could be another reason for the discrepancy in the findings.

In a supplementary Cox regression analysis that included several putative covariates without filtering for collinearity, only binarization of variables according to the relevant cutoff allowed finding of other significant predictors of progression time, namely ‘Apathy’ and TMT-A, whereas RAVLT-DR had a trend for significance. These variables, additionally, were significant in the survival curve analysis, associating with shorter time to dementia and approximately twice the risk of progression in the case the scores were in the range of abnormality. According to such an exploratory analysis, ‘Apathy’ can predict progression time similar to memory deficit (reflected by TMT-A and RATLV-DR scores) as proxies of the AD pathology that accumulates over the years. Noteworthy, the results of these supplementary survival analyses need to be carefully interpreted, considering that the variables were not filtered with a rigorous method for collinearity. With this premise, we may put forward that ‘Discrepancy’ is the most robust predictor of mid-term dementia amid the indices of apathy resulting in significant either when the variable time was considered (survival analyses) or not (generalized linear regression model).

The ‘Discrepancy’, either alone or as a component of the GLM, showed a fair-to-good discriminatory ability in progression to dementia (AUC = 0.73 and AUC = 0.86, respectively). Hence, we might propose that the ‘Discrepancy’ score could allow staging aMCI patients according to their risk of progression to dementia in research settings such as in clinical trials, and may express the highest prediction potential when included in a multimodal index along with other predictors, such as those derived from neuropsychological evaluation. However, the role of ‘Discrepancy’ as a clinical tool for the single-subject risk estimation would require much more rigorous validation in larger, independent cohorts before being deployed, given this small sample and the risk of false positive findings due to multiple statistical tests.

Strengths and limitations

Among the strengths of this work are the enrollment of patients in a real-world setting, the thorough statistical approach, and the follow-up time extended to at least three years or to progression to dementia. The main limitations are the choice not to use the AES version for the physician, mainly because different physicians managed the patients and the limited number of non-converters. Also, a characterization of caregivers in terms of age, sex, education, caregiver burden, and socio-economic status is lacking. Furthermore, the quality of the relationship between patient and caregiver might impact on an accurate estimation when using the patient-caregiver discrepancy method [18], albeit this factor is difficult to control and further reflects the real-world experience. Moreover, interpreting the results of the survival analyses needs caution in the absence of a precise date of onset of dementia that was inferred based on the closest clinical assessment, as routinely happens in the case of non-acute events, and especially after the dichotomization of predictor variables.

Finally, it is to note that the present results apply only to aMCI patients, who are, however the most numerous in a memory clinic setting and need to be confirmed in an independent larger population to be generalized.

Conclusions

A structured evaluation of apathy both self-assessed and estimated by caregivers can provide reliable information about the risk and timing of progression from aMCI to AD dementia. According to our results, the discrepancy between the two estimates may be considered for prediction as an index of awareness, and the AES might be an easy-to-use tool for the purpose. Further studies in larger validation cohorts are needed to propose the ‘Discrepancy’ index of the AES to enter statistical models of prediction of progression in the aMCI populations.

Footnotes

ACKNOWLEDGMENTS

The authors have no acknowledgments to report.

FUNDING

The authors have no funding to report.

CONFLICT OF INTEREST

Flavio Nobili has received fees for participation in boards by Biogen and by Roche, for consultation from Bial, and for teaching lecture by G.E. Healthcare.

Matteo Pardini has received research support from Novartis and received honoraria from Novartis, Merck, and Roche.

Fiammetta Monacelli, Flavio Nobili, and Nicola Girtler are Editorial Board Members of this journal but were not involved in the peer-review process nor had access to any information regarding its peer-review.

All other authors declare that they have no conflicts of interest.

DATA AVAILABILITY

The data supporting the findings of this study are available on request from the corresponding author.

The supplementary material is available in the electronic version of this article: .

References

Marin

(1991) Apathy: A neuropsychiatric syndrome. J Neuropsychiatry Clin Neurosci 3, 243–254.

Leung

DKY

, Chan

, Spector

, Wong

GHY

(2021) Prevalence of depression, anxiety, and apathy symptoms across dementia stages: A systematic review and meta-analysis. Int J Geriatr Psychiatry 36, 1330–1344.

Cerejeira

, Lagarto

, Mukaetova-Ladinska

(2012) Behavioral and psychological symptoms of dementia. Front Neurol 3, 73.

Lyketsos

, Carrillo

, Ryan

, Khachaturian

, Trzepacz

, Amatniek

, Cedarbaum

, Brashear

, Miller

(2011) Neuropsychiatric symptoms in Alzheimer’s disease. Alzheimers Dement 7, 532–539.

Vergallo

, Giampietri

, Pagni

, Giorgi

, Nicoletti

, Miccoli

, Libertini

, Petrozzi

, Bonuccelli

, Tognoni

(2019) Association between CSF beta-amyloid and apathy in early-stage Alzheimer disease. J Geriatr Psychiatry Neurol 32, 164–169.

Lanctôt

, Agüera-Ortiz

, Brodaty

, Francis

, Geda

, Ismail

, Marshall

, Mortby

, Onyike

, Padala

, Politis

, Rosenberg

, Siegel

, Sultzer

, Abraham

(2017) Apathy associated with neurocognitive disorders: Recent progress and future directions. Alzheimers Dement 13, 84–100.

Palmer

, di Iulio

, Varsi

, Gianni

, Sancesario

, Caltagirone

, Spalletta

(2010) Neuropsychiatric predictors of progression from amnestic-mild cognitive impairment to Alzheimer’s disease: The role of depression and apathy. J Alzheimers Dis 20, 175–183.

Stella

, Radanovic

, Aprahamian

, Canineu

, de

Andrade LP

, Forlenza

(2014) Neurobiological correlates of apathy in Alzheimer’s disease and mild cognitive impairment: A critical review. J Alzheimers Dis 39, 633–648.

Jones

, de Marco

, Manca

, Bell

, Blackburn

, Wilkinson

, Soininen

, Venneri

(2019) Altered frontal and insular functional connectivity as pivotal mechanisms for apathy in Alzheimer’s disease. Cortex 119, 100–110.

10.

Marshall

, Donovan

, Lorius

, Gidicsin

, Maye

, Pepin

, Becker

, Amariglio

, Rentz

, Sperling

, Johnson

(2013) Apathy is associated with increased amyloid burden in mild cognitive impairment. J Neuropsychiatry Clin Neurosci 25, 302–307.

11.

Mori

, Shimada

, Shinotoh

, Hirano

, Eguchi

, Yamada

, Fukuhara

, Tanimukai

, Zhang

M-R

, Kuwabara

, Ueno

S -i.

, Suhara

(2014) Apathy correlates with prefrontal amyloid deposition in Alzheimer’s disease. J Neurol Neurosurg Psychiatry 85, 449–455.

12.

Marshall

, Fairbanks

, Tekin

, Vinters

, Cummings

(2006) Neuropathologic correlates of apathy in Alzheimer’s disease. Dement Geriatr Cogn Disord 21, 144–147.

13.

Skogseth

, Mulugeta

, Ballard

, Rongve

, Nore

, Alves

, Aarsland

(2008) Neuropsychiatric correlates of cerebrospinal fluid biomarkers in Alzheimer’s disease. Dement Geriatr Cogn Disord 25, 559–563.

14.

Lanctôt

, Amatniek

, Ancoli-Israel

, Arnold

, Ballard

, Cohen-Mansfield

, Ismail

, Lyketsos

, Miller

, Musiek

, Osorio

, Rosenberg

, Satlin

, Steffens

, Tariot

, Bain

, Carrillo

, Hendrix

, Jurgens

, Boot

(2017) Neuropsychiatric signs and symptoms of Alzheimer’s disease: New treatment paradigms. Alzheimers Dement (N Y) 3, 440–449.

15.

Leoutsakos

J-MS

, Forrester

, Lyketsos Constantine

, Smith

(2015) Latent classes of neuropsychiatric symptoms in NACC controls and conversion to mild cognitive impairment or dementia. J Alzheimers Dis 48, 483–493.

16.

Pink

, Stokin

, Bartley

, Roberts

, Sochor

, Machulda

, Krell-Roesch

, Knopman

, Acosta

, Christianson

, Pankratz

, Mielke

, Petersen

, Geda

(2015) Neuropsychiatric symptoms, APOE 4, and the risk of incident dementia: A population-based study. Neurology 84, 935–943.

17.

Guercio

, Donovan

, Munro

, Aghjayan

, Wigman

, Locascio

, Amariglio

, Rentz

, Johnson

, Sperling

, Marshall

(2015) The Apathy Evaluation Scale: A comparison of subject, informant, and clinician report in cognitively normal elderly and mild cognitive impairment. J Alzheimers Dis 47, 421–432.

18.

Starkstein

(2014) Anosognosia in Alzheimer’s disease: Diagnosis, frequency, mechanism and clinical correlates. Cortex 61, 64–73.

19.

Azocar

, Rapaport

, Burton

, Meisel

, Orgeta

(2022) Risk factors for apathy in Alzheimer’s disease: A systematic review of longitudinal evidence. Ageing Res Rev 79, 101672.

20.

Jacus

J-P

, Voltzenlogel

, Antoine

, Cuervo-Lombard

C-V

(2022) Apathy is the best dimension to consider for awareness assessment in Alzheimer’s disease. J Alzheimers Dis 87, 149–154.

21.

Magni

, Binetti

, Bianchetti

, Rozzini

, Trabucchi

(1996) Mini-Mental State Examination: A normative study in Italian elderly population. Eur J Neurol 3, 198–202.

22.

Lawton

, Brody

(1969) Assessment of older people: Self-maintaining and instrumental activities of daily living. Gerontologist 9, 179–186.

23.

Gorelick

, Scuteri

, Black

, DeCarli

, Greenberg

, Iadecola

, Launer

, Laurent

, Lopez

, Nyenhuis

, Petersen

, Schneider

, Tzourio

, Arnett

, Bennett

, Chui

, Higashida

, Lindquist

, Nilsson

, Roman

, Sellke

, Seshadri

(2011) Vascular contributions to cognitive impairment and dementia. Stroke 42, 2672–2713.

24.

Wahlund

, Barkhof

, Fazekas

, Bronge

, Augustin

, Sjögren

, Wallin

, Ader

, Leys

, Pantoni

, Pasquier

, Erkinjuntti

, Scheltens

(2001) A new rating scale for age-related white matter changes applicable to MRI and CT. Stroke 32, 1318–1322.

25.

Mattioli

, Pardini

, Famà

, Girtler

, Brugnolo

, Orso

, Meli

, Filippi

, Grisanti

, Massa

, Bauckneht

, Miceli

, Terzaghi

, Morbelli

, Nobili

, Arnaldi

(2021) Cuneus/precuneus as a central hub for brain functional connectivity of mild cognitive impairment in idiopathic REM sleep behavior patients. Eur J Nucl Med Mol Imaging 48, 2834–2845.

26.

American Psychiatric Association (2013) Diagnostic and Statistical Manual of Mental Disorders.

27.

Jessen

, Amariglio

, Buckley

, van der Flier

, Han

, Molinuevo

, Rabin

, Rentz

, Rodriguez-Gomez

, Saykin

, Sikkes

SAM

, Smart

, Wolfsgruber

, Wagner

(2020) The characterisation of subjective cognitive decline. Lancet Neurol 19, 271–278.

28.

Albert

, DeKosky

, Dickson

, Dubois

, Feldman

, Fox

, Gamst

, Holtzman

, Jagust

, Petersen

, Snyder

, Carrillo

, Thies

, Phelps

(2011) The diagnosis of mild cognitive impairment 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 7, 270–279.

29.

McKhann

, Knopman

, Chertkow

, Hyman

, Jack

, Kawas

, Klunk

, Koroshetz

, Manly

, Mayeux

, Mohs

, Morris

, Rossor

, Scheltens

, Carrillo

, Thies

, Weintraub

, Phelps

(2011) 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 7, 263–269.

30.

Rascovsky

, Hodges

, Knopman

, Mendez

, Kramer

, Neuhaus

, van Swieten

, Seelaar

, Dopper

EGP

, Onyike

, Hillis

, Josephs

, Boeve

, Kertesz

, Seeley

, Rankin

, Johnson

, Gorno-Tempini

M-L

, Rosen

, Prioleau-Latham

, Lee

, Kipps

, Lillo

, Piguet

, Rohrer

, Rossor

, Warren

, Fox

, Galasko

, Salmon

, Black

, Mesulam

, Weintraub

, Dickerson

, Diehl-Schmid

, Pasquier

, Deramecourt

, Lebert

, Pijnenburg

, Chow

, Manes

, Grafman

, Cappa

, Freedman

, Grossman

, Miller

(2011) Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134, 2456–2477.

31.

Ossenkoppele

, Pijnenburg

YAL

, Perry

, Cohn-Sheehy

, Scheltens

NME

, Vogel

, Kramer

, van der Vlies

, Joie

R la

, Rosen

, van der Flier

, Grinberg

, Rozemuller

, Huang

, van Berckel

BNM

, Miller

, Barkhof

, Jagust

, Scheltens

, Seeley

, Rabinovici

(2015) The behavioural/dysexecutive variant of Alzheimer’s disease: Clinical, neuroimaging and pathological features. Brain 138, 2732–2749.

32.

Crutch

, Schott

, Rabinovici

, Murray

, Snowden

, Flier

, Dickerson

, Vandenberghe

, Ahmed

, Bak

, Boeve

, Butler

, Cappa

, Ceccaldi

, Souza

, Dubois

, Felician

, Galasko

, Graff-Radford

, Hof

, Krolak-Salmon

, Lehmann

, Magnin

, Mendez

, Nestor

, Onyike

, Pelak

, Pijnenburg

, Primativo

, Rossor

, Ryan

, Scheltens

, Shakespeare

, Suárez González

, Tang-Wai

, Yong

KXX

, Carrillo

, Fox

(2017) Consensus classification of posterior cortical atrophy. AlzheimersDement 13, 870–884.

33.

Johnston

, Jones

, Manley

(2018) Confounding and collinearity in regression analysis: A cautionary tale and an alternative procedure, illustrated by studies of British voting behaviour. Qual Quant 52, 1957–1976.

34.

Ding

, Zhao

, Guo

, Liang

, Luo

, Yu

, Zheng

, Hong

(2016) Progression and predictors of mild cognitive impairment in Chinese elderly: A prospective follow-up in the Shanghai Aging Study. Alzheimers Dement (Amst) 4, 28–36.

35.

Schmidtke

, Hermeneit

(2008) High rate of conversion to Alzheimer’s disease in a cohort of amnestic MCI patients. Int Psychogeriatr 20, 96–108.

36.

Chudoba

, Schmitter-Edgecombe

(2020) Insight into memory and functional abilities in individuals with amnestic mild cognitive impairment. J Clin Exp Neuropsychol 42, 822–833.

37.

Fragkiadaki

, Kontaxopoulou

, Beratis

, Andronas

, Economou

, Yannis

, Papanicolaou

, Papageorgiou

(2016) Self-awareness of cognitive efficiency: Differences between healthy elderly and patients with mild cognitive impairment (MCI). J Clin Exp Neuropsychol 38, 1144–1157.

38.

Roberts

, Clare

, Woods

(2009) Subjective memory complaints and awareness of memory functioning in mild cognitive impairment: A systematic review. Dement Geriatr Cogn Disord 28, 95–109.

39.

Zanetti

, Vallotti

, Frisoni

, Geroldi

, Bianchetti

, Pasqualetti

, Trabucchi

(1999) Insight in dementia: When does it occur? Evidence for a nonlinear relationship between insight and cognitive status, J Gerontol B Psychol Sci Soc Sci 54B, P100–P106.

40.

Somme

, Fernandez-Martínez

, Molano

, Jose Zarranz

(2013) Neuropsychiatric symptoms in amnestic mild cognitive impairment: Increased risk and faster progression to dementia. Curr Alzheimer Res 10, 86–94.

41.

Nobili

, Mazzei

, Dessi

, Morbelli

, Brugnolo

, Barbieri

, Girtler

, Sambuceti

, Rodriguez

, Pagani

(2010) Unawareness of memory deficit in amnestic MCI: FDG-PET findings. J Alzheimers Dis 22, 993–1003.