Development of Apathy, Anxiety, and Depression in Cognitively Unimpaired Older Adults: Effects of Alzheimer’s Disease Pathology and Cognitive Decline

BACKGROUND
The impact of Alzheimer's disease (AD) pathology and cognitive deficits on longitudinal neuropsychiatric symptoms is unclear, especially in early disease stages.


METHODS
Cognitively unimpaired older adults (N = 356) enrolled in the prospective Swedish BioFINDER study were examined. Neuropsychiatric assessments encompassed the Apathy Evaluation Scale and the Hospital Anxiety and Depression Scale, performed biennially (together with tests of global cognition) for up to 8 years. Biomarkers were measured in cerebrospinal fluid or plasma at baseline. Magnetic resonance imaging quantified white matter lesions. We used linear mixed-effect models to test associations between baseline AD biomarkers (for amyloid-β [Aβ], tau, and neurodegeneration) and white matter lesions with longitudinal neuropsychiatric symptoms (apathy, anxiety, and depressive symptoms). We also tested associations between changes in cognition and changes in neuropsychiatric symptoms. Finally, we tested if change in cognition mediated the effects of different brain pathologies on neuropsychiatric symptoms.


RESULTS
Aβ pathology at baseline was associated with increasing levels of apathy (β = -0.284, p = .005) and anxiety (β = -0.060, p = .011) longitudinally. More rapid decline of cognition over time was related to increasing levels of apathy. The effects of baseline Aβ pathology on longitudinal apathy were partly mediated by changes in cognitive performance (proportion mediated 23%).


CONCLUSIONS
Aβ pathology may drive the development of both apathy and anxiety in very early stages of AD, largely independent of cognitive change. The effect of Aβ on apathy is only partially conveyed by worse cognition. Together, these findings highlight certain neuropsychiatric symptoms as early manifestations of AD.

amyloid-b (Ab) plaques, neurofibrillary tangles of hyperphosphorylated tau, and neurodegeneration, as well as clinical manifestations including both cognitive deficits and neuropsychiatric symptoms (NPSs) (e.g., apathy, depression, and anxiety) (1,2). According to the National Institute on Aging and Alzheimer's Association criteria, AD is defined as a neurobiological construct related to core AD pathologies, where its clinical progression is staged according to the level of cognitive deterioration (2). This view is supported by a robust relationship between AD pathology and future cognitive decline (1). Although the criteria do not highlight NPSs, it is known that the frequency and severity of NPSs increase with worsening cognition (3,4). This suggests that NPSs and cognitive deficits can develop in parallel and that NPSs may constitute early manifestations of AD (5). In support, cross-sectional studies in early disease stages have shown associations between AD pathology and NPSs (6)(7)(8)(9)(10). Other studies demonstrate NPSs as predictors of future cognitive decline and dementia already in preclinical AD (6,11,12). Moreover, anxiety and Ab are reported to interact, resulting in accelerated cognitive decline (6,13). In line, the novel concept of mild behavioral impairment emphasizes that NPSs can develop before, in concert with, or somewhat after mild cognitive impairment (MCI) due to neurodegenerative disease (14). However, only a few studies have tested effects of both neuropathology and cognition on the development of NPSs (8). Therefore, the exact temporal and causal relationships between pathology, cognition, and NPSs in AD remain unclear. Here, we investigated how biomarkers of AD pathology, white matter lesions (WMLs), and cognitive deficits potentially drive the development of apathy, anxiety, and depressive symptoms in cognitively unimpaired (CU) older adults. We also tested if cognitive change mediates the effect of brain pathologies on longitudinal NPSs.

METHODS AND MATERIALS Study Sample
CU participants (n = 359) in the prospective Swedish Bio-FINDER study (Clinical Trial No. NCT01208675) were recruited. However, only participants with at least one NPS rating during the biennial follow-up of up to 8 years were included (N = 356). Of note, not all had completed the 6-and 8-year visits at the time of data extraction. In short, the CU participants were eligible for inclusion in the BioFINDER study if they 1) were $60 years old, 2) had a Mini-Mental State Examination (MMSE) score of 28 to 30 at the screening visit (allowed MMSE 27-30 at the baseline visit), 3) were not in need of a Swedish interpreter, 4) had absence of cognitive symptoms, and 5) did not fulfill criteria of MCI or dementia. Details on design are provided in the Supplement and reported previously (6). Additional information is found at http://www.biofinder.se.

Standard Protocols, Registrations, and Patient Consents
The Regional Ethical Review Board in Lund, Sweden, approved the study. All participants gave their written informed consent.

Clinical Assessments
Clinical assessments were administered biennially for up to 8 years.
Apathy was assessed by the Swedish Apathy Evaluation Scale, self-rated (AES-S) and informant-rated (AES-I) (15). AES is a well-studied tool consisting of 18 items rated at a 4-point scale (not at all, slightly, somewhat, or a lot) (15). Higher scores indicate a higher level of apathy (range, 18-72).
Global cognition was measured using the MMSE (17) and a modified Preclinical Alzheimer's Cognitive Composite (mPACC5) (18). The color/form task in A Quick Test (AQT-CF) (19,20) assessed executive functioning and was further used in the mPACC5 composite [the executive test has also previously varied in different PACC5 publications (21)(22)(23)], as well as used separately in a post hoc analysis (outlined below). A more detailed description of AQT-CF and the computation of mPACC5 is provided in the Supplement.
AES was incorporated in the BioFINDER study after study start. Hence, some participants lacked AES at baseline or the 2-year follow-up. Given an administrative error at year 2, HADS was not distributed to some participants at this visit. The number of available assessments at each visit are presented in Figures S1 and S2.

Fluid Biomarkers
Cerebrospinal fluid (CSF) and blood samples were collected close in time after the baseline NPS examination (mean = 1.4 [SD = 0.1] months) and handled according to structured preanalytic protocols (24,25). Levels of CSF Ab 42 , Ab 40 , and neurofilament light (NfL) were measured on an Elecsys platform according to the manufacturer's instructions (Roche Diagnostics International Ltd.) (25,26). CSF Ab 42 , and Ab 40 were combined into a CSF Ab 42 /Ab 40 ratio, with high specificity for AD-related amyloidopathy (27). CSF NfL was used as a marker for cortical and subcortical axonal degeneration (28). Plasma phosphorylated tau (P-tau217) was analyzed, as previously described in detail (25), using immunoassay on a Mesoscale Discovery platform developed by Lilly Research Laboratories. There were missing data at baseline (CSF Ab 42 / Ab 40 , n = 33; CSF NfL, n = 35; plasma P-tau217, n = 36).

Magnetic Imaging Acquisition and Processing
High-resolution T1-weighted and T2-weighted FLAIR images were acquired on a Siemens Tim Trio 3T MR scanner (Siemens Medical Solutions) (mean = 0.6 [SD = 0.1] months from baseline). WML volumes were generated by an automated segmentation process, using the lesion prediction algorithm in the LST toolbox (http://www.statisticalmodelling.de/lst.html) for SPM (29). Eleven subjects lacked magnetic resonance imaging data.

Statistical Analyses
First, individual change per year (slope) for the cognitive measures MMSE, mPACC5, and AQT-CF were calculated using individual univariate linear regression models with cognitive scores as dependent variables and time as an independent variable.
Second, we tested associations between longitudinal NPSs (as dependent variables) and different predictors in primary linear mixed-effect (LME) models. Baseline measures of continuous CSF Ab 42 /Ab 40 , plasma P-tau217, CSF NfL, and WML volumes, individually, were entered as zero-centered predictors interacting with time (biomarker 3 time). In similar LME models, baseline values or slopes of MMSE or mPACC5 (extracted from linear regression in the first step of the analyses) were used as predictors interacting with time (cognition 3 time). All models included age, sex, and education as covariates, as well as random slopes and intercepts. The number of participants and NPS observations per model are presented in Table S1. We report on the interacting effects; main effects are provided in Tables S2 and S3. To reduce the risk of type I error, Bonferroni corrections were made sectionwise for each dependent variable in each table (in total: 32 models, 64 p values, 4 p values per correction).
We also conducted sensitivity analyses where the primary models were refitted when removing participants with only one NPS measure. Given missing AES data at baseline, we also reran the apathy models when including only the 2-year to 8year follow-up data. A survival bias analysis was further conducted using logistic regressions, where missingness of data at the 2-, 4-, or 6-year visit were predicted by neuropathology (one model per biomarker and visit). Age, sex, and education constituted covariates. Here, false discovery rate correction adjusted for multiple comparisons. Additional sensitivity analyses were conducted in which antidepressants at any visit (dichotomous variable) was added as a covariate to the primary models.
In post hoc analyses, associations between longitudinal apathy and baseline executive function or executive slopes (assessed by AQT-CF) were examined using similar models as in the primary LME analyses. Third, we conducted mediation analyses to test if the cognitive slopes for MMSE or mPACC5 over time mediated the effects of neuropathology on longitudinal NPSs. Analyses were restricted to models in which longitudinal NPSs had significant associations with both neuropathology and global cognitive decline also after Bonferroni correction. A bootstrap procedure (n = 1000 iterations) calculated 95% CI for the mediated effects. A detailed description of the model setups is provided in the Supplement. The number of participants and NPS observations for each mediation analysis are also presented in Table S4.
For all statistical tests, a significance threshold of p , .05 (two-sided) was used. Regression model assumptions were assessed by evaluating normality and homoscedasticity of residuals with probability plots and plots of residuals versus fitted values. Statistical analyses were performed using R version 3.6.1 with the packages "lme4," "lmerTest," "visdat," and "ggeffects" and IBM SPSS Statistics version 25 (IBM Corp.).

Demographics and Clinical Characteristics
Demographics and clinical characteristics are presented in Table 1. Mean age was 73.8 (SD = 5.1) years, 9.8% of the participants used antidepressants at any visit, and 28.5% were APOE ε4 carriers.

Effects of Pathology on Longitudinal NPSs
First, we tested associations between individual baseline biomarkers interacting with time and longitudinal NPS scores ( Table 2). Longitudinal increase in AES-I was greater in participants with lower (i.e., more abnormal) CSF Ab 42 /Ab 40 (b = 20.284, p = .005) or higher (i.e., more abnormal) plasma P-tau217 (b = 20.253, p = .015). Lower CSF Ab 42 /Ab 40 (b = 20.060, p = .011) or higher (i.e., more abnormal) CSF NfL (b = 0.054, p = .024) were also associated with higher longitudinal HADS-A scores. A high (i.e., more abnormal) WML volume (b = 0.136, p = .016) was associated with increased longitudinal AES-S scores. Only the effect of CSF Ab 42 /Ab 40 over time on AES-I and HADS-A remained significant after correction for multiple comparisons. Figure 1 demonstrates these associations for Ab-negative versus Ab-positive individuals and displays that those with the highest level of pathology show the steepest increases in NPS scores. None of the pathologies was associated with longitudinal HADS-D.

Effects of Cognition and Cognitive Slopes on Longitudinal NPSs
Next, we tested associations between baseline cognition or cognitive slopes over time and longitudinal NPS scores (Table 3). Over time, there was an effect on longitudinal AES-S by baseline mPACC5 (b = 20.126, p = .033), but this did not hold for Bonferroni correction.
As shown in Figure (Table S5).

Sensitivity Analyses
As a sensitivity analysis, the primary LME analyses were rerun on a restricted sample, removing participants with NPS data for only one visit (n removed: AES-S = 36, AES-I = 111, HADS-A = 53, and HADS-D = 53). Effects and p values were found to be similar to the primary analyses (Table S6), with the exceptions that WML volumes were now associated with change in AES-I and that the association between MMSE slope and longitudinal AES-S was lost.
Rerunning the primary apathy models including only 2-year to 8-year data also gave results consistent with the primary models (Table S7). Corroborating this, our survival bias analysis in general did not find associations between pathology and missing follow-up data (Table S8). The exceptions were that plasma P-tau217 strongly predicted the presence of missing AES-S data at the 2-year follow-up (odds ratio = 41.4, 95% CI = 6.0-286.4, p-adj = .002) and that CSF NfL predicted missing AES-I data at the 4-year visit (odds ratio = 1.003, 95% CI = 1.000-1.007, p-adj = .048).
We further controlled the primary models for the use of antidepressants at any visit, which did not change the results (Table S9).

Cognition as a Mediator of Pathology on NPSs
Finally, we tested whether some associations between neuropathology and longitudinal NPSs were statistically mediated via cognitive slopes. The association between baseline CSF Ab 42 /Ab 40 interacting with time and longitudinal AES-I was partly mediated by mPACC5 slopes with 23% mediation ( Figure 3A). The effect of CSF Ab 42 /Ab 40 over time on longitudinal AES-I remained significant also after controlling for mPACC5 slopes, indicating a remaining statistically direct effect of Ab independent from cognitive change. A similar result was obtained using MMSE ( Figure 3B).

DISCUSSION
This study explored associations between longitudinal NPSs and AD-related pathologies, WMLs, and cognition in CU individuals. Our main finding was that Ab exerted a weak to moderate effect over time on the trajectories of apathy and anxiety, and this was mainly independent from cognition. Longitudinal anxiety and cognitive decline associated merely on a trend level, and cognitive change only partially mediated the effect of Ab on longitudinal apathy.

Associations Between Ab and Longitudinal NPSs
Scores on repeated measures of informant-rated apathy increased in participants with signs of Ab pathology at study Development of Apathy, Anxiety, and Depression start. This finding conforms with most cross-sectional studies (6,(30)(31)(32) but also points to the direction of the relationship, where Ab to some extent could be accountable for the subsequent development of apathy. We are aware of a similar study on CU individuals from the Harvard Aging Brain Study, which did not find an association between Ab interacting with time and the development of apathy-anhedonia cluster items derived from the self-rated Geriatric Depression Scale (GDS)

AES-S, median (IQR)
Follow-up 0 years 28 (7) 18 to 43 Follow-up 2 years 27 (7) 18 to 53 Follow-up 4 years 28 (10) . This does not agree with our findings on informant-rated apathy but is well in line with our self-rated findings, which were not affected by Ab over time. Therefore, the discrepancy between our studies could reflect the critical challenges in rater source selection (as further discussed in Differences Between AES-S and AES-I in Their Relation to Neuropathology). The same Harvard Aging Brain Study instead displays associations between Ab interacting with time and the development of self-rated anxiety-concentration cluster item scores. In addition to earlier cross-sectional findings on nondemented samples (6,9,10,13,32,34), this aligns with our results where an effect over time by Ab on longitudinal self-rated anxiety was found. In contrast, longitudinal data from the PREVENT-AD cohort on CU individuals at increased risk of AD (due to a family history of sporadic AD) displayed a lack of such an association (32). Instead, they displayed a cross-sectional association between Ab and some latent behavioral factors, including, e.g., neuroticism, anxiety, and apathy (the latter two informant-rated). Potentially, the disagreement in longitudinal results is best explained by the somewhat shorter follow-up in the PREVENT-AD study, where participants on a group level might not have had time to progress in their anxiousness. Nevertheless, their cross-sectional finding implies the useful sensitivity of informant ratings even in early AD.
The relationship between Ab and depressive manifestations has remained unsettled (35). This study supports several previous studies that have reported a lack of such a relationship (6,34,(36)(37)(38)(39). However, other studies have displayed an Linear mixed-effect models to investigate the effects of different biomarkers for neuropathology over time (pathology 3 time interaction) on the development of NPSs in CU participants. Longitudinal NPS measures of apathy, anxiety, and depressive symptoms were entered as the dependent variable in separate models. Biomarker measures at baseline were one by one entered as fixed effects interacting with time (biomarker 3 time). Fixed effects were zero centered. All models were corrected for age, sex, and education and included random slopes and intercepts. Main effects are reported in Table S2. The significance threshold was set at p , .050. Bonferroni corrections were run sectionwise for each dependent variable. Overall, 35 participants lacked CSF data for Ab 42 /Ab 40 and NfL, 36 participants lacked data for plasma P-tau217, and 11 participants lacked WML volume data.
Ab  association (33,(40)(41)(42)(43)(44). There are many possible reasons for these divergent findings. One is that the definition of depression or the assessment of its severity varies considerably. Many subsyndromal depression studies that report a relationship have assessed depressive symptoms using the GDS.
In the Harvard Aging Brain Study, the authors displayed steeper rates of total GDS scores over time for participants with higher levels of Ab deposition (33). However, according to their subanalysis, in which the three item clusters of the GDS scale (dysphoria, anxiety-concentration, and apathyanhedonia) were analyzed, the average dysphoria item cluster score was shown lower than the other item clusters. Moreover, change in dysphoria was not linked to Ab. Similar findings on the GDS are reported from the Australian Imaging Biomarkers and Lifestyle Study (38). Together, these studies suggest that GDS total scores primarily reflect on anxiety or apathy rather than dysphoria. Because dysphoria could be argued central in the concept of major depression, this questions the validity of the GDS in samples where apathy and anxiety are prevalent, as in older adults at risk of neurodegenerative disease (3).

Associations Between Tau, Neurodegeneration, WMLs, and Longitudinal NPSs
We further report associations between longitudinal informantrated apathy and baseline plasma P-tau217 interacting with time and longitudinal anxiety and baseline CSF NfL interacting with time. Although this suggests links between tau or neurodegeneration and the trajectories of some NPSs, these findings did not hold for Bonferroni correction and are only suggestive findings. As for Ab, the literature on tau and neurodegeneration in relation to NPSs in CU individuals displays both mixed results and methodologies (7,8,32,34,40,45,46). In favor of an association, we earlier demonstrated crosssectional associations between tau and mild behavioral impairment among CU Ab-positive individuals (partly overlapping with this sample) (7). However, future longitudinal studies are needed to determine the role of these pathologies in the development of NPSs. We also demonstrate WMLs to be initially associated with longitudinal self-rated apathy (informant-rated was near the significance threshold). Yet, this association did not hold for correction for multiple comparisons. This was unexpected given that cross-sectional studies consistently have demonstrated strong associations between apathy and WMLs already in preclinical stages (6,47). However, another longitudinal study on apathy could also not report an effect of WMLs over time (48).

Differences Between AES-S and AES-I in Their Relation to Neuropathology
Our results diverge regarding self-and informant-rated apathy. For instance, AES-I, but not AES-S, was related to baseline Ab. We have also previously reported diverging results for these rater sources in relation to Ab (6). In the earlier study, using a mixed sample of CU individuals and individuals with MCI, the median scores for AES-I and AES-S among CU participants were similar. But in participants with MCI, the median for AES-S was less increased compared with AES-I. This suggests that participants with MCI tend to underreport, resulting in less steep slopes for AES-S than AES-I as participants progress from CU to MCI. This further agrees with a study comparing the repeated measures for the different versions in CU individuals and individuals with MCI (49). Hypothetically, individuals with AD could underreport NPSs due to lost insight in a similar fashion because this loss has been reported to affect assessments of memory complaints (50).

Association Between Cognition and Longitudinal NPSs
The connection between NPSs and cognition is emphasized by findings showing that the worse the cognitive status is in a sample, the higher is the frequency and severity of NPSs (4,51). Several reports have also shown that certain NPSs Linear mixed-effect models to investigate effects of cognition over time (cognition by time interaction) on the development of NPSs in CU participants. Longitudinal NPS measures of apathy, anxiety, and depressive symptoms were entered as the dependent variable in separate models. Cognitive measures at baseline, as well as cognitive slopes, were one by one entered as fixed effects interacting with time (cognition 3 time). Fixed effects were zero centered. Individual change per year (slope) in MMSE and mPACC5 score were generated using individual linear regression models in which longitudinal MMSE and mPACC5 were predicted by time. All models were corrected for age, sex, and education and included random slopes and intercepts. Main effects are reported in Table S3. The significance threshold was set at p , .050. Bonferroni corrections were run sectionwise for each dependent variable.
AES constitute risk markers for more rapid cognitive decline or conversion to dementia (6,11,12). For instance, in an earlier study, we demonstrated that apathy and anxiety predicted cognitive decline in CU individuals and individuals with MCI (6).
Yet, only few studies have addressed the impact of cognition on longitudinal NPSs. In a longitudinal study on CU individuals by the AIBL Research Group, neither cognitive test performance nor retrospective informant ratings of cognitive change exerted effects over time on apathy (48). Here, conversely, we show a strong statistical effect by longitudinal cognitive test performance on longitudinal apathy. However, similar to the prior study, no effect over time by baseline cognitive test performance was found. Perhaps cognitive tests at baseline are not affected to such an extent that associations with longitudinal NPSs can be detected, or the tests are not sensitive enough to capture more subtle deficits. Given that baseline NPSs seem to predict future cognitive decline (6,13), and not vice versa, this highlights the potential clinical utility of early monitoring of NPSs as prognostic markers for disease progression. However, these findings partly rest upon studies using mixed samples of CU individuals and individuals with MCI, which limits the interpretation (6,13).
We found stronger associations between apathy and cognitive slopes than with pathology. Maybe this is explained by resilience factors against both cognitive deficits and NPSs. Hence, those who develop clinical symptoms due to pathology show both cognitive deficits and NPSs. Hypothetically, this could inflate the statistical relationship.
We further show that elevated trajectories of anxiety are associated with mPACC5 slopes, but this association did not survive Bonferroni correction. Hence, longitudinal anxiety The direct effect of baseline CSF Ab 42 /Ab 40 on the mediator mPACC5 or MMSE is designated a and was obtained using linear regression modeling. The direct effect of the mediator mPACC5/MMSE on the development of AES-I is designated b. Models were corrected for age, sex, and education. All fixed and random effects, as well as covariates, were zero centered. Linear mixed-effect models included random slopes and intercepts. Confidence intervals (CIs) for mediation effects were obtained using bootstrapping with 1000 iterations. prop., proportion.
Development of Apathy, Anxiety, and Depression seems less related to cognitive decline compared with apathy. Individuals who debut with cognitive deficits could likely become anxious over its functional impact or worry over having a progressive neurocognitive disorder. However, anxiety does not inevitably accelerate owing to progressive cognitive deterioration. Instead, the anxiety or its increase over time due to cognitive change could remain stable, as indicated in Figure S4.

The Mediating Effects of Cognitive Decline for AD Pathology on Longitudinal NPSs
The association between Ab and longitudinal apathy was only partly (23%) mediated by cognitive slopes. This indicates that Ab mainly conveys its effect on apathy development through direct mechanisms somewhat independent from cognitive decline. In AD, Ab is known to accumulate early in the parietal and frontal cortices with effects on neuronal connectivity in the default mode network and the frontoparietal control network (52). Even if these networks serve several purposes, the default mode network is considered important for cognitive task performance, while the frontoparietal control network predominately relates to goal-related behavior (53). In support, apathy has been shown to be associated with interrupted connectivity in the frontoparietal control network but not in any other network (53). Aligning with our finding that longitudinal anxiety is associated with Ab but merely on a trend level with cognitive change, certain NPSs and cognitive decline could hypothetically share common anatomical locations of neuropathology but arise from dysfunction in separate functional brain networks.
Yet, our findings also support an indirect less prominent pathway to apathy, where Ab may act through cognitive decline. The mechanism behind this mediation needs further exploration. However, diagnostic criteria for apathy emphasize change in goal-directed cognitive activities as an essential part of the construct (54), and associations between apathy and executive functioning have been reported (55,56). Even so, our post hoc analyses only support a role of executive dysfunction in the development of self-rated, but not informant-rated, apathy. Hence, if these associations arise due to an overlap in the theoretical frameworks of these manifestations (e.g., the ability to take initiative or complete tasks) (54,56) or if they are given by a shared common functional network disruption needs further exploration. Perhaps the divergency between the ratings is attributed to the self-rated version's potential to register the internal experience of a reduced executive function or goal-directed cognition, whereas the informant-rated version is limited to observations of external goal-directed behavior.
All in all, with previous studies demonstrating a strong association between Ab positivity and future cognitive decline (57), our findings strengthen the proposed idea that cognitive deficits and NPSs can develop independent of, yet parallel to, each other, given a common underlying neuropathology. However, they also seem to reinforce one another, even if only to a limited extent (5).

Limitations and Strengths
The strength of this study is its well-characterized sample and its repeated measures of both NPSs and cognition. However, there are limitations. First, there are missing NPS data.
However, LME models are known to be advantageous in dealing with missing values, and our sensitivity and survival bias analyses argue against such a strong effect. Second, the NPS data rest on assessments, not clinical diagnoses, and major psychiatric illness at baseline constituted an exclusion criterion. This limits the generalizability toward CU individuals with only subsyndromal NPSs or good mental health. Third, findings are not controlled for a history of psychiatric illness, although we did control for antidepressants during study follow-up (data on other psychopharmacologic treatments were not available). Fourth, tau and neurodegeneration are believed to develop somewhat later than Ab in AD. As expected, levels of P-tau217 and NfL in this study on CU individuals are therefore low, which may reduce the power to detect very early associations between tau or neurodegeneration with longitudinal NPSs. Fifth, the corrections for multiple comparisons increase the risk of type II error. Nonetheless, we do display associations between NPSs, Ab, and cognition, and uncorrected p values are also provided. Finally, neuropathologies other than those studied here could have contributed to the evolution of NPSs.

Conclusions
Early Ab pathology may be a significant driver behind the development of both apathy and anxiety in CU older adults. The association between Ab pathology and longitudinal apathy is only partly conveyed by cognitive decline; hence, Ab pathology may influence apathy directly and somewhat independent of cognitive changes.