In the Canadian population, delaying the onset of dementia by just 1 year in populations at risk for Alzheimer's disease (AD) is projected to result in almost 500,000 fewer cases of dementia by 2050 (Alzheimer Society of Canada, 2022). To guide clinical interventions to delay symptom onset, identification and characterization of the earliest signs of cognitive decline, or increased risk for developing said decline, has become an increasing focus in both research and clinical settings. Subjective cognitive decline (SCD) is seen by many as the earliest behavioral harbinger of heightened risk for future cognitive impairment (Rabin et al., 2017).

SCD is a clinical state characterized by reports of “self-experienced persistent decline in cognitive capacity” despite scores within normal ranges on cognitive measures (Jessen et al., 2020, 2014). Although our current clinical tools do not provide objective evidence of cognitive decline, individuals with cognitive complaints are twice as likely as cognitively normal (CN) older adults without complaints to decline to mild cognitive impairment (MCI; Gifford et al., 2014; Mitchell et al., 2014), a clinical state with objective cognitive decline but little impact on an individual's activities of daily living, or to AD-related dementia, where individuals demonstrate both objective cognitive decline and diminished autonomy in daily activities. Studies of AD neuropathology have shown the SCD state to be associated with increased tauopathy within the entorhinal cortex (Buckley et al., 2017) and decreased hippocampal metabolism (Vannini et al., 2017). Furthermore, the presence of certain facets of SCD, sometimes termed SCD-plus (Jessen et al., 2020), carries further indications of brain pathology. For example, memory-specific complaints are related to amyloid-β levels in the brain (Amariglio et al., 2012) and cerebrospinal fluid (Miebach et al., 2019), as is worry over one's self-reported cognitive decline (Miebach et al., 2019; Verfaillie et al., 2019). Given these relationships between elements of SCD and AD-type pathology, SCD has been suggested as a preclinical stage of AD (Jessen et al., 2020), specifically indicative of stage 2 with the National Institute on Aging – Alzheimer's Association (NIA-AA) research framework for AD (Jack et al., 2018).

While more clinical factors within SCD-plus have been shown to increase risk of decline [e.g., seeking medical help (Snitz et al., 2018); corroboration of decline by an informant (Caselli et al., 2014; Valech et al., 2015)], we do not yet understand which cognitive factors modulate SCD risk for progressing onto dementia or clearly distinguish SCD from cognitively normal aging. Here we are presented with a conundrum: how does one find potentially meaningful differences in cognitive performance between cognitively normal older adults and those with SCD, given that subtle cognitive changes reported by individuals with SCD are, by definition, within normal ranges of test performance. Different methods of analyzing cognitive data in SCD versus CN groups has begun to yield some results. For example, Caillaud et al. (2020) observed correlational differences, finding that SCD scores in episodic memory correlated with hippocampal volumes while CN scores did not, even though the groups did not differ in their mean test scores. Koppara et al. (2015) argued that the rate of change in objective measures of memory and verbal fluency prior to the onset of SCD is informative, demonstrating that decline in these areas of cognitive performance was steeper for those that would go on to develop SCD. Still, novel applications of statistical analyses are needed to further uncover and characterize these subclinical cognitive changes within SCD (Mortamais et al., 2017).

Recently, network theory has been applied widely in the field of psychology to capture complex local and global relationships within multivariate data (Borsboom et al., 2021). This approach to analysis allows a researcher to explore the gestalt of a collection of psychological measures, bringing to light not only the relationships between two variables but also the larger web of relationships and patterns within the data. Within a network, individual variables or entities are represented by “nodes,” and these nodes are in turn connected with “edges” representing the association between those two variables. Network science has been implemented in studies of social networks (with nodes representing individuals and edges used to express the connections between them), neural networks (with nodes as neural clusters and edges as connections between those brain areas), and psychometric networks (with nodes representing psychological variables and edges representing conditional associations between variables). In all these applications, a network approach allows researchers to gain insight beyond what is afforded in univariate analyses.

Using networks as a method of understanding interconnected systems, researchers have also begun to explore cognitive functioning with a network perspective. In many ways, cognition is already conceived of as a network: a web of intertwined cognitive domains and abilities that give rise to a whole greater than the sum of its parts (Siew et al., 2019). Cognitive functioning is thus a potential target for network theory and analysis (Garcia-Ramos et al., 2016), although there are inherent challenges to using network models to study cognition (Ferguson, 2022). Previous literature in cognitive science has applied networks to understand connections among elements of cognitive development and intelligence (van der Maas et al., 2017) as well as skills of reading, spelling and mathematics (Zoccolotti et al., 2021) to normative samples. Within neuropsychological studies, networks have been used to understand cognitive abilities in the context of stroke (Massa et al., 2015), Parkinson's disease (Ferguson & Foley, 2023), acquired brain injury (Jonker et al., 2019), epilepsy (Garcia-Ramos et al., 2021; Kellermann et al., 2016), autism spectrum disorder (Ibrahim et al., 2016), attention-deficit/hyperactivity disorder (Eadeh et al., 2021) and, relevant to the current study, cognitive decline (De Marco et al., 2024; Ferguson, 2021; Goodwin et al., 2023; Nevado et al., 2021; Tosi et al., 2020; Wright et al., 2021). Findings from these latter studies have revealed distinct patterns of connectivity in cognitive networks over the aging process and suggest a process of network reorganization in both amnestic and non-amnestic MCI, AD, and vascular encephalopathy networks compared to cognitive networks of health older adults. Using Alzheimer's Disease Neuroimaging Initiative (ADNI) neuropsychological data, Ferguson (2021) found episodic memory nodes in MCI and semantic fluency and processing speed nodes in early AD to be highly influential, as well as an overall shift from intra-domain associations in the cognitively normal network to interdomain clustering in clinical group networks. Findings from Tosi et al. (2020) examining healthy control, AD, and vascular encephalopathy cognitive networks suggested that age and education were influential in healthy control cognitive networks but had little impact in clinical groups while semantic fluency was highly influential in both AD and vascular networks. Notably, none of these studies examining cognitive networks along the AD continuum have included SCD. Because network analysis offers an alternative perspective on cognition by exploring the interrelationships between cognitive test variables, there is potential for this methodological approach to provide evidence from patterns of associations among objective cognitive measures to corroborate subjective reports of impairment in SCD.

The current study aimed to explore the structures of cognitive networks across AD risk state continuum, with the novel inclusion of individuals with SCD. Specifically, the goals of this study were to (1) determine if there was a pattern of cognitive performance that differentiated SCD from CN participants, and (2) to generate cognitive networks across the AD risk continuum with the exploratory aim of identifying within- and across-group patterns of association among cognitive test variables. Networks models of cognitive performance in CN, SCD, MCI and AD groups were constructed and compared, with a specific goal of describing network characteristics in SCD that might identify it as a preclinical stage of AD and as objectively distinct from cognitively normal aging.