The Impact of Amyloid Burden and APOE on Rates of Cognitive Impairment in Late Life Depression

INTRODUCTION

Late life depression (LLD) is a highly prevalent and disabling psychiatric disorder in older adults. Up to 5% of community-dwelling older adults meet criteria for major depressive disorder (MDD) and 8–16% report clinically significant depressive symptoms [1, 2]. LLD is associated with a host of negative outcomes in aging, including increased rates of cognitive impairment (CI), disability, incident dementia, suicide, and all-cause mortality [3–6]. CI has been identified as a key feature of LLD that contributes to increased functional disability and mental health care costs, poor treatment response and compliance, and higher rates of relapse [7–10]. Cognitive deficits in LLD have been reported in multiple domains, including executive functioning, information processing speed, episodic memory, language, and visuospatial skills [11]. However, determining the incidence of clinical CI in LLD is complicated by significant variability in methods of normative comparison and the complex relations between LLD and neurodegenerative diseases of aging, which both feature CI as a primary symptom. It remains unclear to what extend CI in LLD is related to underlying concurrent neurodegenerative pathologies.

While previous studies have explored the pattern, course, and neural substrates of cognitive dys-function in LLD, few have reported these findings using clinically meaningful thresholds for impair-ment or demographically corrected normative comparisons. Commonly cited estimates suggest that roughly 25–60% of individuals with LLD demonstrate impaired performance in at least one domain of cognitive functioning [12–15]. However, others have reported much lower estimates in the range of 0–7% [16, 17]. This variability in the literature may be accounted for, in part, by several methodological factors. There is a relative dearth of published studies that report rates of impairment using clinically meaningful thresholds. While most, if not all, widely cited manuscripts on cognitive dysfunction in LLD use the term “impairment,” most do not apply clinical standards for interpreting cognitive test performance as impaired based on normative comparison standards. Of the few studies that do utilize normative comparisons, there is significant variability in the threshold for clinical impairment, ranging from 1-2 standard deviations below the reference group mean [14–18]. Further, published impairment rates are based on relative comparison to healthy control samples, which are often small, poorly characterized, and free of common medical comorbidities that increase risk for CI. To our knowledge, there are no published rates of CI in LLD that utilize demographically corrected normative comparisons beyond age-adjusted scores, which are commonly used in clinical practice to account for variability in cognitive performance due to factors such as age, race/ethnicity, gender, and education. Additionally, no published studies of an LLD sample have applied regression-based norming procedures, which allow for more precise correction across multiple demographic factors relative to traditional norms [19].

Cognitive dysfunction in LLD often persists despite treatment and remission of other depressive symptoms, suggesting that observed impairments are more than simply state-dependent effects of depression, but rather reflect multiple neurobiologi-cal pathways including cerebrovascular, inflammatory, and neurodegenerative etiologies [14, 20–22]. Prior research examining determinants of cognitive dysfunction in LLD has focused largely on vascular [23] and inflammatory pathways [24], while relatively less is known about the contribution of APOE ɛ4 genotype and cerebral amyloid-β (Aβ) deposition, two widely recognized Alzheimer’s disease (AD) biomarkers that are associated with increased CI and decline in non-demented older adults [25, 26]. APOE ɛ4 genotype and Aβ positivity are estimated to be present in up to 30% [27] and 30–43% [28, 29] of cognitively normal older adults depending on age, respectively, and the relationships among APOE, Aβ burden, and cognitive dysfunction may not be apparent early in the course of AD pathological changes. Thus, a significant proportion of cognitively normal older adults with and without LLD likely have undetected elevated risk for CI and future decline due to AD risk factors, which are not routinely accounted for in studies of cognitive functioning in LLD.

The full impact of APOE and Aβ on CI in LLD is further obscured by the use of cognitively normal nondepressed (ND) comparison groups in the existing literature [30–32]. Given that up to 48% of LLD samples meet criteria for mild cognitive impairment (MCI), typically based on the presence of memory impairment, mild functional difficulties, and preserved global cognition [18], this approach limits our understanding of how AD risk and other age-related neurologic changes (e.g., white matter hyperintensities, cerebral atrophy) may contribute to CI in LLD samples. In the absence of comparison groups with equivalent levels of global cognitive functioning and MCI diagnosis, it is difficult to determine the degree to which previously described patterns of CI in LLD could be accounted for by AD pathology. Instead, comparison of individuals with LLD to a ND normal control sample with equivalent rates of MCI and levels of global cognitive functioning allows for more precise assessment of factors that contribute to CI in LLD and ND older adults.

The current study was conducted with two primary aims. The first aim was to assess the specific profile and rates of neuropsychological impairment in LLD relative to ND controls using regression-based demographically corrected norms and a clinically meaningful cutoff for impairment. We hypothesized that, in a sample matched on key demographic (age, sex, education) and clinical characteristics (MCI status, global cognition, and Aβ positivity), LLD participants would demonstrate a greater proportion of impairment overall and higher rates of impaired performance on tasks of executive functioning, information processing speed, and episodic memory relative to ND participants. The second aim was to assess the relative contributions of Aβ burden and APOE ɛ4 genotype on cognitive performance using a well-characterized sample of LLD and participants without LLD matched on key demographic and clinical characteristics, including MCI status and global cognitive functioning. We hypothesized that higher Aβ burden and APOE ɛ4 genotype would be associated with poorer performance on measures of global cognition and episodic memory in the overall sample.

METHODS

RESULTS

Rates of impairment using normative comparison

When CI was examined as a categorical variable (i.e., percentage of individuals falling at or below the 7th percentile for each measure), 56.4% of the LLD group was impaired on at least one cognitive domain compared to 42.9% of the ND group. The distribution of CI differed between groups, with a greater proportion of the LLD group impaired on more domains compared to the ND control group (H = 7.13, p = 0.008). A significantly greater proportion of the LLD group was impaired relative to ND controls on AVLT Learning (21.8% versus 11.2%, p = 0.006), AVLT Immediate Recall (19.5% versus 8.4%, p = 0.003), and AVLT Delayed Recall (17.8% versus 9.6%, p = 0.023, see Fig. 1). Rates of impairment did not differ between groups on any other measures (all ps > 0.05).

OPEN IN VIEWER

Fig. 1

Proportion of participants with clinically significant impairments on neuropsychological test performance at –1.5 standard deviations. LLD, late life depression; ND, non-depressed; ADAS, Alzheimer’s Disease Assessment Scale; AVLT, Auditory Verbal Learning Test; LM, Logical Memory; BNT, Boston Naming Test; TMT, Trail Making Test; ^*p; < 0.05; ^**p; < 0.01; ^***p; < 0.001.

DISCUSSION

The present study was conducted to determine the incidence of CI in LLD using regression-based demographically corrected norms, as well as to assess the impact of Aβ burden and APOE genotype on cognitive functioning in a sample of individuals with LLD and ND controls matched on key clinical and demographic variables. Our results showed that, after matching for AD risk factors and global cognitive status, 54.6% of the LLD sample and 42.9% of the ND control sample demonstrated impairment in at least one cognitive domain. We found that LLD was associated with focal deficits in list learning and memory relative to nondepressed controls, and that increased Aβ burden and APOE ɛ4 carrier status were independently associated with poorer cognitive functioning on tests of story learning and memory and rapid set shifting in the sample as a whole, but these relationships did not differ for LLD and ND participants. Each of these results will be discussed below.

Our finding that 56.4% of the LLD group demonstrated CI is ostensibly consistent with the extant literature, though it should be noted that our threshold for impairment was more conservative (i.e., performance < 7th percentile) than that used in widely-cited estimates (< 10th percentile) [12, 14]. This result is important for several reasons. First, to our knowledge, this is the first study to use regression based, demographically corrected norms to classify CI in LLD. While others have established impairment in LLD participants via relative comparisons to standard scores derived from performance of healthy control participants [12, 14–17], this approach is vulnerable to confounding effects of uncontrolled or unknown demographic and clinical factors in the comparison sample. Thus, while the rate of impairment we report is consistent with previous studies, our findings suggest that using traditional metrics of control group performance as the standard for relative comparison may result in artificially inflated impairment rates. Second, our results demonstrate that, when matched on key demographic and clinical variables, over 40% of the ND control group also showed clinically meaningful CI in at least one domain. This is likely driven, in part, by AD biomarkers, as nearly 35% of our ND control group met criteria for MCI, 20% were Aβ positive, and 30% were APOE positive. Taken together, these findings illustrate how defining CI relative to a limited sample of healthy control participants may misrepresent the true incidence of CI in LLD, especially when neurodegenerative disease burden is unknown.

When comparing rates of impairment across neuropsychological tasks using regression-based norms we found that individuals with LLD showed greater rates of impaired performance on learning, immediate recall, and delayed recall trials for a word list episodic memory task (AVLT Trials 1–5, Immediate Recall & Delayed Recall) compared to ND participants. For each of these metrics the rate of impairment was approximately twice as high in the LLD group relative to the ND group even after matching for MCI status and overall cognitive functioning. These results highlight a particularly salient dysexecutive learning process in LLD characterized by inefficient learning that negatively impacts performance on subsequent memory trials. In contrast, when comparing cognitive performance using raw scores without regression-based normative correction, individuals with LLD showed worse performance on learning and immediate memory only (AVLT Trials 1–5 & Immediate Recall), suggesting that reliance on scores that are not corrected for demographic variables to characterize cognitive dysfunction may obscure clinically meaningful memory impairment unique to LLD. Contrary to our hypothesis, we did not find evidence for reduced performance or increased CI in the LLD group on standalone measures of executive functioning or information processing speed, which are thought to represent core features of cognitive dysfunction in LLD [12, 53], though there was indirect evidence of a dysexecutive memory profile. Given that our sample was uniquely matched on multiple AD risk factors, these more focal findings suggest that early effects of AD pathology may account for some previously described multi-domain cognitive deficits in LLD.

With regard to the contribution of AD biomarkers to cognitive performance in the whole sample, we found that Aβ SUVR predicted worse performance on measures of story learning and memory (LM I & II) and rapid set-shifting (TMT-B). APOE ɛ4 carrier status was significantly associated with worse performance solely on story learning (LM I). These findings are consistent with existing literature showing associations between Aβ, APOE, and measures of episodic memory in nondemented older adults [26, 27] and highlight the importance of accounting for the impact of AD biomarkers when evaluating aging populations, even in samples that are largely cognitively intact. Given that APOE ɛ4 genotype is present in 30% of nondemented older adults depending on age and Aβ status [27] and Aβ burden increases with age, with up to 60% of adults over the age of 85 meeting criteria for Aβ positivity [54], there is potential for AD to function as a significant confounding factor in geriatric psychiatry research samples. Though the relation of Aβ SUVR and TMT-B performance was unexpected in a sample that was largely cognitively normal, it is consistent with the executive dysfunction commonly present in clinical AD and may reflect early cognitive changes secondary to elevated AD pathology. We also report that the relationship of Aβ and cognitive dysfunction did not differ between depressed and non-depressed older adults. This importantly suggests that elevated rates of CI in LLD are independent of Aβ deposition.

Participants in our analyses were matched on global measures of cognition and proportion of MCI with the intent to make the most conservative comparison of specific CI and relationship of Aβ to cognition. While we recognize that matching the groups on MCI impacts our ability to detect memory differences between these groups, particularly on the Logical Memory subtest, we did report more impairment in LLD on a test of list learning. We interpret these results to largely reflect an impaired learning process in earlier trials demonstrated by the LLD group that may be more evident in list learning paradigms than for story learning. However, we acknowledge that based on our matching criteria a similar deficit in Logical Memory may have been obscured. Additionally, we found that in the ND sample, clinician diagnosed MCI was more likely to be rated as amnestic than in the ADNI-D sample despite identical criteria for MCI diagnosis. In the ADNI, the MCI participants are almost always rated as amnestic, in part because the ADNI was designed to capture memory impairment characteristic of early AD, as opposed to other cognitive phenotypes (e.g., dysexecutive, multi-domain). However, as our results show, multi-domain and non-amnestic MCI were present in the ADNI ND sample but not captured by clinician rating. Even in participants categorized as cognitively normal by ADNI criteria, there is still significant CI in non-memory domains. These findings highlight the high degree of variability in clinician ratings of MCI, even when following established diagnostic criteria, which has the potential to negatively impact studies of aging and cognition. Given evidence from prior studies showing improved sensitivity, specificity, validity, and reliability of more comprehensive neuropsychological diagnostic methods with actuarial decision-making relative to ADNI MCI criteria [55–57], future research is warranted to examine the potential impact of MCI diagnostic criteria on geriatric psychiatry research.

Though these results are robust, there are some limitations to state. The cross-sectional nature of the data makes it impossible to speculate about longitudinal change. For example, this highly educated sample may have already experienced cognitive decline but remains within a normative range thus appearing unimpaired. The cross-sectional design also is uninformative with respect to future cognitive decline in either group. Further, it is possible that the relative contribution of AD risk to CI in this sample will increase over time as underlying neurodegenerative processes progress. Thus, it will be important for future research to examine longitudinal trajectories of cognitive change in LLD. Although matching on key demographic and clinical variables is considered a strength of this study, matching on MMSE and MCI status, which is based in part on memory performance, may have obscured or minimized memory impairments secondary to LLD. Finally, our study sample and normative reference sample were both highly educated and predominantly Caucasian, which limits the generalizability of our findings to racial/ethnic minorities and individuals with lower levels of education. Ideally, a normative reference sample would include equal proportions of representation across all levels and categories of variability; however, we were restricted to broader, less specific categories (e.g., white/non-white) due to limited representation in low education and racial minority groups.

To our knowledge this is the first study to assess rates of neuropsychological impairment in LLD relative to ND controls using regression-based demographically corrected norms and first to investigate the impact of in vivo Aβ burden and APOE on cognitive functioning in a large sample of clinically depressed and ND older adults matched on key demographic and clinical variables, including AD biomarkers. When controlling for AD biomarkers, LLD was associated with a ∼30% higher rate of CI relative to ND participants. Though the overall rate of CI in LLD was consistent with previous findings, our results suggest that the profile of LLD neurocognitive dysfunction may be more focal than previously reported. We found increased rates of impairment on measures of list learning and memory that support a pattern of memory impairment in LLD primarily characterized by a dysexecutive, retrieval-based memory profile. However, we acknowledge that controlling for MCI and controlling for multiple comparisons may have limited the results to the most prominent differences.

Our results also highlight the confounding role of AD processes in cognitive functioning in LLD, which have not been well characterized in prior studies and may account for more diffuse patterns of impairment reported in the extant literature. Our findings underscore the potential for AD risk to impact results of clinical trials in LLD and other geriatric psychiatry research examining cognition as an outcome. Collectively, these findings offer important clarification on the relative contributions of depression, Aβ burden, and APOE genotype to CI in nondemented older adults and serve as an important step in disambiguating the role of AD as a confound in the assessment of CI in LLD.