Researchers are mapping the earliest end of the Alzheimer’s disease (AD) continuum to identify patients in a critical window for therapeutic intervention (Dubois et al., 2016). While in vivo detection of AD pathologies using biomarkers is central to this process (Sperling et al., 2011), it is not sufficient given the imperfect association between neuropathology and clinical manifestation of disease (Negash et al., 2013). Indeed, at least a third of cognitively normal older adults have evidence of pathological AD on autopsy (Negash et al., 2013) or amyloid imaging (Chételat et al., 2013), and the pathological definition of AD continues to be debated (de la Torre, 2004; Castellani and Smith, 2011; Castellani and Perry, 2014). The ongoing questions and controversies surrounding clinical-pathological correlations in AD (Castellani and Smith, 2011; Castellani and Perry, 2014) emphasize the importance of identifying the earliest clinical manifestations of disease. Subjective cognitive decline (SCD), defined as the perception of cognitive decline despite normal performance on traditional neuropsychological testing, is likely to be one such early manifestation of illness with studies increasingly pointing to the potential relevance of SCD as an inexpensive and easily obtainable “pre-clinical” marker of AD (Geerlings et al., 1999; Reisberg et al., 2008; Sperling et al., 2011; Rabin et al., 2017; Jessen et al., 2020).

Research in AD as well as in aging generally supports an association between SCD and objective memory both cross-sectionally and longitudinally, and there is emerging evidence of the association between SCD and AD biomarkers (Gilewski et al., 1990; Hertzog et al., 1990; Pearman and Storandt, 2004; Beaudoin and Desrichard, 2011; Amariglio et al., 2012; Perrotin et al., 2012; Hülür et al., 2014; Snitz et al., 2015; Chen et al., 2019, 2021). However, the utility of SCD as a marker of cognitive functioning and biomarker status appears to vary as a function of multiple factors including task factors (e.g., measurement and operationalization issues) and person factors (e.g., individual characteristics) which together obscure its association with objective markers of disease (Schmidt et al., 2001; Jessen et al., 2010; Tandetnik et al., 2015; Ossenkoppele and Jagust, 2017). For example, the perceptions that memory is worse than others of the same age (i.e., age-anchored SCD) maps on more closely to AD biomarkers than perceptions of memory being bad in general, or worse than before, for example (Perrotin et al., 2012; Tandetnik et al., 2015; Chapman et al., 2021). With regard to person factors, there is recognition that personality and mood are likely important in the conceptualization of SCD; however, other factors remained to be explored (Pearman and Storandt, 2004; Slavin et al., 2010; Merema et al., 2013; Steinberg et al., 2013).

From a self-awareness perspective, SCD may be considered a hyperaware state (hypernosognosia) indicative of early dysfunction not yet detectable, or which does not reach a formal threshold for impairment, on clinical neuropsychological measures. As disease progresses, disordered awareness in the form or lack of awareness of deficits (anosognosia) likely follows SCD in a subset of individuals with mild cognitive impairment; this disordered awareness can be a prognostic indicator of disease progression as well as important clinical outcomes (Starkstein, 2014; Vannini et al., 2017; Munro et al., 2018). Knowledge of one’s own cognitive abilities (e.g., metacognition) has been examined extensively in healthy young and older adults (Nelson, 1990; Price et al., 2010; Hertzog and Dunlosky, 2011; Souchay and Isingrini, 2012; Cauvin et al., 2019; Siegel and Castel, 2019; Gagliardi et al., 2020) and has proven useful in understanding the clinical phenomenon of anosognosia, particularly disordered awareness of memory loss (Cosentino et al., 2007; Galeone et al., 2011; Rosen et al., 2014; DeLozier and Davalos, 2016).

Indeed, several groups have used metamemory testing to measure memory awareness in AD, and this type of assessment may offer a unique vantage point into the accuracy of SCD. As a direct measure of one’s memory awareness, metamemory is a critical person factor that should be considered in the interpretation of SCD. Specifically, individuals who demonstrate good metamemory (i.e., who have good awareness of their actual memory function), may be expected to have a more accurate subjective report of cognitive decline than those who have poor metamemory. Despite its clear relevance for understanding the prognostic relevance of SCD, metamemory has rarely been examined in relation to SCD (Buckley et al., 2016; Vannini et al., 2019; Chi et al., 2020; Gagliardi et al., 2020), perhaps because metamemory as a construct evolved primarily in the field of cognitive psychology and is not a formal component of clinical neuropsychological evaluations (Sunderaraman and Cosentino, 2017; Chapman et al., 2020).

The aim of this paper is to examine the extent to which metamemory moderates the relation between SCD and objective memory. As performance on traditional neuropsychological assessments of memory is by definition “normal” in individuals with SCD, we must utilize more challenging and sensitive neuropsychological tests to more rigorously examine the accuracy of SCD. The current study includes two memory measures shown to be sensitive to SCD as well as to AD biomarkers among clinically normal older adults. As stated above, our hypothesis postulates that those with better metamemory will have more accurate SCD; defined as a stronger association between SCD and objective memory testing on sensitive tasks.

This study examined the extent to which metamemory moderated the association between SCD and memory abilities in older adults. Consistent with previous work showing an association between SCD and rigorous measures of subtle cognitive dysfunction (Chapman et al., 2021), bivariate associations revealed that individuals with higher SCD had weaker performance on select list learning measures including greater susceptibility to both proactive interference and retroactive interference. With regard to the moderating role of metamemory, results from this study support the idea that in general, SCD is more strongly linked to memory abilities among individuals with better metamemory. Indeed, metamemory moderated the association between SCD and susceptibility to proactive interference Metamemory did not, however, moderate the association between SCD and retroactive interference or short-term memory binding. Below we offer potential interpretations for these findings and discuss current issues in the measurement and conceptualization of SCD more broadly, beginning with the variable associations between SCD and the memory outcomes selected for the current study.

The selective associations between SCD and only two of four memory outcomes, all previously shown to be sensitive to preclinical AD (Parra et al., 2010; Loewenstein et al., 2016; Crocco et al., 2018), was somewhat unexpected. For example, both proactive and retroactive interference on the LASSI-L have been linked to total cortical loading of amyloid and the precuneus specifically, among cognitively normal older adults (Loewenstein et al., 2016). In fact, the ability to recover from proactive interference has repeatedly been shown to be more sensitive to pre-clinical AD than other LASSI markers (Loewenstein et al., 2016, 2017). It is thus not immediately clear why SCD relates differently to each of these metrics. Susceptibility to proactive interference, associated with SCD in the current study, is assessed by measuring recall of List B after two study trials of List A. Recovery from proactive interference, not currently associated with SCD, is defined as recall of List B after its second presentation. It may be that in the current cognitively normal sample, there is little variability in performance after studying this list twice, limiting the degree to which it maps onto SCD. Indeed, average scores were higher (11.9) and the minimum score higher (6) than on the susceptibility metric (8.3 and 1, respectively). Nevertheless, the selective associations between SCD and increased susceptibility to proactive and retroactive interference may reflect specific early dysfunctions in cognitive control mechanisms. Previous research has shown that individuals with reduced working memory capacity (Rosen and Engle, 1998; Brewin and Smart, 2005) or inhibitory control (Anderson et al., 2000; Anderson, 2003; Anderson and Levy, 2007) tend to be more susceptible to interference effects and intrusive thoughts. Subtle changes in these cognitive control mechanisms could impact the use of specific and more effective retrieval mechanisms (Anderson and Levy, 2007; Unsworth, 2016, 2019).

Unexpectedly, SCD was also unrelated to short-term memory binding, the latter measure having previously been associated with SCD in a subset of this same cohort (Chapman et al., 2021). It is important to keep in mind, however, that while both the LASSI-L and STMB are sensitive to preclinical AD, their neural underpinnings are not synonymous. As highlighted earlier, LASSI-L measures have been associated with amyloid load in key AD regions such as cingulate, precuneus, frontal lobe as well as volumetric and cortical integrity of medial temporal lobe regions including the hippocampus. In contrast, the STMB has been associated primarily with posterior parietal-occipital regions implicated in the ventral visual stream, regions hypothesized to be affected during the sub-hippocampal stages of AD (Parra et al., 2014). As such, depending on the regional distribution of potential brain changes among individuals in a given sample, the extent to which SCD maps onto one or another cognitive measure will likely differ.

The inconsistency of metamemory as a moderator was also unexpected. While the size and direction of the moderation effect were generally comparable across different outcome measures, the moderating effect was only significant for SCD and measures of proactive interference (susceptibility to and recovery from), but not retroactive interference or short-term memory binding. There are several factors that could have led to this discrepancy. First, the link between SCD and memory itself is variable as discussed above. It may not be feasible to detect a significant moderation effect in situations where SCD is not even weakly associated with a specific memory outcome, as was the case for STMB in the current study. A second potential issue is that metamemory itself is heterogeneous, consisting of two broad categories: monitoring (i.e., what you know about your memory) and control (i.e., how you manage your memory). Monitoring, the focus of the current study, is itself multi-dimensional and can be operationalized in a number of ways that capture individuals’ confidence level (i.e., calibration) as well as their ability to adjust their expectations for performance as it varies over the course of a test (i.e., resolution). Furthermore, metamemory can be measured at different levels including an item-by-item basis (e.g., will you know the answer to this question?), or a summary level (e.g., how many answers will you know overall?) as well as at different points in time, including prior to or following memory performance (Nelson, 1984, 1990). Different studies have revealed nuances in the correlates of individual metamemory measures depending on a variety of factors including the score that is used (calibration versus resolution), the level at which it is measured (item versus summary), and the population in which it is measured (cognitively normal older adults versus AD) (Kikyo et al., 2002; Maril et al., 2003; Kikyo and Miyashita, 2004; Chua et al., 2006, 2009; Cosentino et al., 2007; Bertrand et al., 2018). From a cognitive perspective, aging studies have shown that confidence in retrieval judgments may be susceptible to variations in memory functioning (Hertzog et al., 2010, 2021). In line with this, reduced memory abilities in older adults may limit their access to diagnostic cues necessary to make accurate metacognitive judgments (Dunlosky and Metcalfe, 2008). Alternatively, older adults might have access to adequate cues but be unable to make valid inferences to reach accurate metacognitive judgments, possibly due to age-related changes in pre-frontal networks (Perrotin et al., 2008; Thomas et al., 2011; Fleming and Dolan, 2012). Given the seeming susceptibility in the current cohort to interference effects, and the moderating effects in this domain, we could also speculate that early vulnerability in frontal medial regions results in compromise to inferential judgments and resultingly to less accurate metacognitive judgments. Additional work is needed to tease apart the underlying cognitive as well as neuroanatomical substrates of both the susceptibility to interference and the moderating effects of metamemory ability.

In conclusion, results partially support our hypothesis that metamemory would moderate the association between SCD and memory performance, and provide rationale for consideration of metamemory when evaluating the accuracy of SCD. However, this study was not without limitations. First, the current sample included only participants with all available measures which reduced the sample significantly. However, in order to address this limitation, a multiple imputation approach was conducted in sensitivity analyses which revealed no significant differences between the initial model and the imputed model with the exception of the interactive effect of gamma and SCD on B2, also indicated in the False Discovery Rate adjusted p-values applied to complete-case analyses. A second limitation was that in 24 participants, gamma was not computed due to ties (i.e., no variability in either their FOK judgments or performance accuracy, with the majority of these cases always indicating “yes” for the FOK judgment with accuracy scores = 1). These cases could be considered as having perfect metamemory, highlighting a possible limitation of our task which for some participants may have a ceiling effect. A greater number of items within each learning trial would increase the likelihood of calculable gamma scores and provide a more comprehensive measure of metamemory in older adults. Another possible limitation was the relatively low level of SCD reported within this sample, along with possible ceiling effects on some cognitive measures which also may have reduced the strength of associations between SCD and cognition, as well as the moderating role of metamemory. Finally, the cohort included in this sample primarily included individuals drawn from other ongoing research studies rather than individuals presenting to a memory disorders clinic, which could skew not only the distribution of SCD but the level of concern regarding SCD, a factor known to increase SCD’s utility as a maker of preclinical AD (Jessen et al., 2010). Ideally, this study would have included sensitivity analyses to explore the effects of community/research recruited versus clinically recruited. This analysis, however, was not possible given that only 13/157 individuals were clinic recruited. There are numerous ways in which we are currently tailoring our ongoing study of SCD, including increasing SCD screenings and referrals from the community and local clinical practices to enroll individuals with higher levels of SCD. Moreover, we are tracking participants longitudinally to examine the extent to which SCD predicts decline over time, as well as the extent to which change in SCD is more predictive than a single SCD assessment. The current literature is mixed; For example, while Drouin et al. (2021) found that subjective memory change predicted longitudinal memory change, Hertzog et al. (2018) found that subjective memory change was more related to current memory complaint rather than an indicator of actual memory change.

This study also had a number of considerable strengths including the prospective, rigorous assessment of SCD using an age-anchored framework shown to relate more closely than other measurement frameworks (e.g., comparing one’s memory to 5 years ago) to objective measures of cognition (Perrotin et al., 2012; Tandetnik et al., 2015; Chapman et al., 2021). Another notable strength was the inclusion of objective metamemory testing, as well as two novel memory tests sensitive to pre-clinical AD, all of which have rarely if ever been combined in a single cohort. Finally, all participants completed comprehensive neuropsychological testing to ensure that they did not meet criteria for Mild Cognitive Impairment. Ongoing work, in addition to enriching our sample with individuals who present to the clinic with complaints, is examining not only the relative contribution of metamemory as a moderator, but of other person factors such as mood, personality, and attitudes about aging (Chapman et al., in preperation). Together, these analyses will continue to inform the way in which SCD can be optimized as a marker of pre-clinical AD.

The datasets presented in this article are not readily available because due to IRB restrictions we cannot share the data. Requests to access the datasets should be directed to StC, sc2460@cumc.columbia.edu.

The studies involving human participants were reviewed and approved by IRB COLUMBIA UNIVERSITY. The patients/participants provided their written informed consent to participate in this study.

All authors listed have made a substantial, direct, and intellectual contribution to the work, and approved it for publication.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.