Mild cognitive impairment (MCI) represents an intermediate stage of cognitive impairment, intended to identify the transitional phase between normal aging and dementia (Petersen et al., 1999; Winblad et al., 2004; Petersen et al., 2014; Scharre, 2019). When the concept of MCI was first introduced, documentation of memory decline was largely drawn from cognitive concerns expressed during the clinical interview, with informant reports of decline and objective neuropsychological data used to corroborate abnormal memory performance (Petersen et al., 1999). Currently, subjective cognitive concern (SCC) is a core criterion for the most widely-used definition of MCI in clinical and research settings (Winblad et al., 2004; Kelley and Petersen, 2007; Albert et al., 2011; Jack et al., 2018; Kasper et al., 2020; Mayo Clinic, 2020; UCSF Weill Institute for Neurosciences, 2022). For other prodromal dementia conditions (e.g., subjective cognitive decline, motoric cognitive risk syndrome), research has examined the assessment of SCC in a more systematic, psychometrically validated manner with some studies using formal SCC screens including cognitive domains beyond memory (i.e., executive functioning, language, attention/concentration, visuospatial navigation, etc.) (Rabin et al., 2020; Diaz-Galvan et al., 2021; Nester et al., 2021; Wasef et al., 2021). However, the measurement of SCC, especially as a criterion for MCI classification, lacks standardization (Stephan et al., 2013; Rabin et al., 2015; Molinuevo et al., 2017).

There are a multitude of studies on mild cognitive impairment (more than 18,000 articles) with highly variable approaches for (1) how to measure SCC (i.e., clinical interview vs. questionnaire; single- vs. multiple-item questionnaire; “homegrown” questionnaires vs. standardized/validated questionnaires; assessing multiple domains vs. only memory; capturing current ability versus change); (2) optimal sources of the SCC report (i.e., self-, informant-, clinician-report); (3) whether corroboration from informant is required; or (4) how to capture and quantify SCC (i.e., any concern, use of a cutoff score or percentage, median or average score, age-appropriate normative scores) (Stephan et al., 2013; Rabin et al., 2015). Further systematic review of how SCC has been measured for the diagnosis of MCI is warranted but beyond the scope of this paper. Supplementary Table 1 provides a sampling to demonstrate the extensive variability in the field but is not intended to be exhaustive. A large proportion of studies list SCC as a criterion for MCI diagnosis, citing Petersen/Winblad’s original definition, but do not provide specific information on how SCC was defined. Some studies do not include SCC as a criterion. Others describe the measure(s) used but do not provide information on the nature of the measurement, source of information, or precise operational definitions of SCC. In studies using SCC questionnaires, items also vary in their referents (relative to previous state personal cognitive status or age-matched mates), time periods (now, over the past year, over the past 5 years), and response option formats (e.g., dichotomous yes/no vs. Likert-like scales for severity, frequency, degree of change).

The original rationale for the inclusion of SCC as a criterion for MCI diagnosis was to identify individuals undergoing cognitive change. Objective cognitive tests measure status, but unless baseline measures are available, they do not capture change. At baseline, cognitive tests may flag individuals with long-standing cognitive difficulties (e.g., learning disabilities, neurodevelopmental disorders, etc.) as impaired (Petersen, 2004). For individuals with above average intellectual abilities, cross-sectional cognitive tests may not capture early stages of cognitive decline if scores fall within normal limits. Despite the compelling rationale, the predictive validity of SCC for incident dementia remains uncertain. For the original Petersen criteria, SCC needed to be self-reported (Petersen et al., 1999), requiring an individual to possess a level of awareness and a willingness to acknowledge memory changes. Though self-reported SCC are associated with objective cognitive performance, they are also associated with depression and anxiety (Buckley et al., 2013; Edmonds et al., 2014; Ryu et al., 2016; Yates et al., 2017; Topiwala et al., 2021; Scholz and Donders, 2022), resulting in overreporting of SCC from cognitively unimpaired individuals (Yates et al., 2017; Edmonds et al., 2018). Many studies show that individuals with MCI demonstrate anosognosia or poor awareness of their own cognitive and functional deficits (Vogel et al., 2004; Edmonds et al., 2014; Fragkiadaki et al., 2016; Gerretsen et al., 2017; Bastin et al., 2021; Ilardi et al., 2021), resulting in underreporting of cognitive changes (Edmonds et al., 2018; Ryu et al., 2019). Further, anosognosia independently predicts conversion from MCI to dementia and is associated with biomarkers, such as reduced brain metabolism (Gerretsen et al., 2017), suggesting that the absence of anosognosia (i.e., insight and SCC) may actually be clinically useful to identify individuals who are unlikely to convert from MCI to dementia (Gerretsen et al., 2017; Bastin et al., 2021). This appears to be contradictory to SCC being a core criterion for defining MCI, and as such, some research efforts support the removal of SCC (Lenehan et al., 2012; Edmonds et al., 2014, 2018; Hackett et al., 2020) on the basis that they do not meaningfully contribute to diagnosis or may even contribute to misdiagnosis of MCI (Vogel et al., 2004; Edmonds et al., 2014; Fragkiadaki et al., 2016; Ilardi et al., 2021). Consistent with this approach and with the various problematic aspects of requiring SCC as a criterion in MCI, Jak, Bondi, and colleagues introduced an approach to MCI classification that uses comprehensive neuropsychological criteria without SCC (Jak et al., 2009; Bondi et al., 2014).

Recent reports suggest that SCC can be assessed by an individual or knowledgeable informant, including clinicians, friends, or family members (Albert et al., 2011; Tangalos and Petersen, 2018). Some research supports that informant-reported SCC or mutual report from both the individual and informant better correlate with an individual’s objective cognitive performance and may better predict progression to dementia than only self-reported SCC (Tierney et al., 1996; Rabin et al., 2012; Gifford et al., 2015; Edmonds et al., 2018; Numbers et al., 2023; Peng et al., 2023). Some investigators address the differences in self- and informant-report with a discrepancy score that may correlate well with objective cognitive impairment (Edmonds et al., 2018) and is a validated approach to measure anosognosia (Starkstein et al., 2006)—which may be clinically useful to improve identification of progression from MCI to dementia. However, informant-reported SCC still has its limitations and can be affected by factors, including type and quality of relationship, frequency of contact, expectations, and affective states of both individual and/or informant (Jessen et al., 2014; Hackett et al., 2020). For example, there may be limited opportunity to notice changes in cognition for older adults who are socially isolated or do not see their families or physicians except for rare occasions.

Whether to include SCC as a criterion for MCI depends critically on its predictive validity for incident dementia. Much of the research on this topic has been carried out in samples of predominantly highly educated, non-Hispanic White older adults, limiting the generalizability of this work to more diverse populations. The current study investigates three operational definitions of self-reported concerns and two definitions of informant-reported concerns about cognition to determine whether their inclusion in two widely-used MCI definitions (i.e., Petersen and Jak/Bondi) add to the predictive validity for incident dementia. Using a more representative cohort of older adults than is typical for aging studies, we additionally investigated discrepancy patterns between self- and informant-reported SCC and their association with incident dementia. Based on prior research (Buckley et al., 2013; Edmonds et al., 2014; Ryu et al., 2016; Yates et al., 2017; Topiwala et al., 2021), we predict that the inclusion of any SCC endorsement (compared to no SCC or a high average cutoff) as a criterion for MCI diagnosis will be the least predictive of incident dementia. Beyond this, there is no prior literature that can specifically guide hypotheses for other operational definitions of SCC as it pertains to the diagnosis of MCI. As such, this work will be exploratory and is an important step forward toward standardizing SCC measurement in MCI.

Whether or not to include SCC as a criterion in the classification of MCI is an uncertainty in the field—but one with important implications. A research-driven approach to the inclusion, exclusion, and/or specific operationalization would be ideal to address this question; however, relatively little attention has been given to standardizing and optimizing the SCC criterion. Moreover, research examining MCI criteria in diverse and/or health disparate populations is limited, which not only restricts the generalizability of the construct, but also upholds health inequities in aging populations (Turner et al., 2022). The present study investigated three operational definitions of self-reported SCC and two operational definitions of informant-reported SCC, within the classification of MCI, in a demographically diverse cohort of older adults. Overall, results indicated that not including self-reported SCC in either Jak/Bondi or Petersen MCI classifications provided the best balance between sensitivity and specificity for incident dementia, across 2-, 3-, 5-, and 7-years of follow-up. In a subset of the sample that had informants available, not including self-reported SCC was comparable to definitions that included any self-reported SCC.

Results are consistent with a growing literature that has questioned the utility of SCC as a criterion in MCI classification, primarily on the basis that SCC may contribute to the misdiagnosis of MCI (Mitchell, 2008; Lenehan et al., 2012; Edmonds et al., 2014; Yates et al., 2017; Edmonds et al., 2018). Self-perceived SCC are present across the lifespan and typically increase with age, with approximately 11% of adults aged 45–64 years endorsing SCC and growing to 50–80% of those over the age of 70 (Mitchell, 2008; Taylor et al., 2018; CDC, 2019; Jessen et al., 2020). Extensive variability in methods of SCC assessment for aging populations (see Stephan et al., 2013; Rabin et al., 2015; Molinuevo et al., 2017) likely contributes to the difference in the prevalence of SCC between studies and highlights the need for further research toward standardizing SCC measurement. In addition to inconsistency in operationalization, there can be challenges in disentangling the numerous demographic, psychosocial, and health related factors that complicate the relationship between SCC, objective cognition, and neurodegeneration. Emerging research shows inconsistent associations between SCC and objective memory in racial/ethnic minoritized individuals (Sims et al., 2011; Jackson et al., 2017; Hughes et al., 2020; Rodriguez et al., 2021; Chapman et al., 2023; Katz et al., 2023). SCC reporting may differ (1) among racial/ethnic groups; (2) among those with different levels of formal educational attainment; or (3) due to sociocultural factors that impact the perception subjective experience, and meaning of cognitive changes (Sayegh and Knight, 2013; CDC, 2019; Hughes et al., 2020; Dilworth-Anderson and Gibson, 2022; Katz et al., 2023). Overall, additional research investigating how cultural factors such as level of education, language status, or cultural norms and values impact an individual’s understanding of their own cognitive status and how they endorse cognitive concerns. These cultural differences provide an added layer of complexity to measuring SCC in diverse individuals and using it as a reliable criterion for MCI.

Further contributing to the instability of SCC as a diagnostic criterion, cognitive concerns may evolve as a result of biopsychosocial variables, such as mood, somatic conditions, chronic pain, medication use, sleep disorders, psychosocial stress, personality factors, and/or normal aging (Bassett and Folstein, 1993; Comijs et al., 2002; Mitchell, 2008; Elfgren et al., 2010; Yates et al., 2017; Jessen et al., 2020; Jenkins et al., 2021; Scholz and Donders, 2022) and resolution of these factors may impact the course of SCC over time (Glodzik-Sobanska et al., 2007; Vestberg et al., 2010). On the other hand, subsets of older adults who endorse persistent SCC but do not show cognitive impairment on sensitive neuropsychological measures (i.e., do not meet criteria for MCI) are more likely to demonstrate objective cognitive decline in the future and are at higher risk for ultimately developing dementia (Treves et al., 2005; Jessen et al., 2014; Luck et al., 2015; Molinuevo et al., 2017; Rabin et al., 2017). It is notable, however, that many individuals who meet criteria for subjective cognitive decline (SCD), do not ultimately progress to dementia (Jessen et al., 2020) and the association of SCD with incident MCI is also not clearly understood (Warren et al., 2022). In the clinical context, a careful evaluation of SCC is essential to identify and treat any plausibly modifiable contributors and to identify older adults in need of additional work-up. Thus, as a clinical tool, SCC may be valuable during preclinical phases (i.e., when an individual’s insight remains intact and prior to objective cognitive impairment) or when the clinical question is to rule out dementia (i.e., when increased specificity is needed). However, in research settings or when individuals progress to more overt cognitive impairment, the complex, multifactorial contributors to SCC may cloud the predictive validity for dementia when SCC is included in MCI definitions.

Although the association between SCC and dementia risk is complex, our results showed both self- and informant-reported SCC was related to risk of dementia. Additionally, the risk of dementia was significantly higher among those with high levels of informant-reported SCC, regardless of the level of self-reported SCC, but the risk of dementia was not as high in those with high levels of self-reported SCC but low informant-reported SCC. This corroborates previous research showing that informant-reported SCC or mutual report better correlate with an individual’s objective cognitive performance and better predicts progression to dementia than only self-reported SCC (Tierney et al., 1996; Rabin et al., 2012; Gifford et al., 2015; Edmonds et al., 2018; Peng et al., 2023). Self-awareness of cognitive deficits and the discrepancy between self- and informant-report have a U-shaped distributions. Self-reported cognitive concerns tend to be elevated relative to informant-reported concerns in SCD (i.e., high discrepancy, Rami et al., 2014; Ryu et al., 2019), become more aligned in early MCI (i.e., low discrepancy, Rabin et al., 2017; Edmonds et al., 2018; Ryu et al., 2019), and become more discrepant as anosognosia develops in late MCI and dementia (Edmonds et al., 2014; Rueda et al., 2014; Rabin et al., 2017). Given the possible fluctuation in self-reported SCC and self-awareness of cognitive and adaptive functioning, this may explain why utilizing SCC was not additive to predicting incident dementia in our sample.

Importantly, although collateral information about an older adult’s daily functioning can be of great benefit in the clinical setting, requiring the presence of an informant may, in fact, hamper ADRD research efforts. For example, excluding older adults without informants effectively translates to excluding socially isolated older adults, who may be at greater risk for dementia (Huang et al., 2023). In addition to having more social support, older adults who have available informants may be demographically different in other ways (e.g., fewer medical comorbidities, better quality of life, etc., Cacioppo and Cacioppo, 2013; Portacolone et al., 2018). Of note, the EAS recruits from the Bronx, NY, a county with one of the most diverse U.S.-born older adult populations in New York and some of the poorest health metrics. Only 62% of our sample had informants available, and this subset had a higher proportion of non-Hispanic Whites, suggesting an unintended selection bias for studies that exclude those without study partners. To promote health equality, it is necessary to examine diagnostic criteria and direct research efforts to capture the intended at-risk populations.

Several study limitations warrant mention. Only a small subset of participants was joined by their informants, and there was a limited number of incident dementia cases. Additional studies in larger samples are needed. Although non-Hispanic Blacks were well represented, the population of Hispanic older adults or those of other racial/ethnic groups were insufficient to include in the analyses. Additionally, the self- and informant-report SCC questionnaires were limited in their coverage of cognitive domains and not all measures were available for both participants and their informants. As EAS has evolved, we have given much more consideration to the measurement of SCC and have newly implemented an expanded version of the widely-used Cognitive Change Index (CCI-40, Nester et al., 2022). Future research could expand on this current study utilizing standardized measures of SCC and exploring their utility in other health disparate groups.

To our knowledge, this was the first longitudinal study directly comparing five specific operational definitions of self- or informant-reported SCC in a diverse community-based sample. Our findings suggest that not including SCC, across different MCI definitions, provides the best balance of sensitivity and specificity in terms of predicting incident dementia in our sample of diverse individuals. Although the original rationale was to avoid prescribing a diagnosis of MCI to individuals with long-standing cognitive difficulties (e.g., learning disabilities) (Petersen, 2004), the extremely high prevalence of self-reported SCC in our sample warrants consideration as to whether this criterion is diagnostically additive and/or should be removed from MCI classifications. Consistent with prior research, informant-reported SCC was more strongly related to risk of incident dementia than self-perceived cognitive concerns. However, requiring informant reports in diagnostic criteria for MCI or as inclusion criteria in research settings may unintentionally exclude health disparate groups and/or those at elevated risk for dementia. Additional consideration is needed to determine how best to utilize informant-report in MCI screening and diagnosis. Taken together, these results provide valuable information for researchers and clinicians in their understanding of self- and informant-reported SCC and their relation to MCI classifications and incident dementia in diverse community-based populations. Importantly, this research can serve as a springboard for future efforts (e.g., pooling data) and to stimulate increased discussion among experts to arrive at an optimal MCI definition that promotes diagnostic accuracy while also being mindful of factors that may uphold health inequalities in aging populations.

The terms of consent for research participants stipulate that an individual’s data can only be shared outside of the EAS investigators group if the group has reviewed and approved the proposed secondary use of the data. This consent applies regardless of whether data has been de-identified. Access is mediated via a standardized request process via Mindy Katz, mindy.katz@einsteinmed.edu.

All protocols were approved by the Einstein Institutional Review Board (IRB). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent from the participants or the participants’ legal guardians/kin was obtained at the initial clinic visit.

KC and LR are the corresponding authors and guarantors of this manuscript and contributed to the conception and the design of the study. MK and CW organized the database. CW performed the statistical analysis and assisted with interpretation of results. KC, in collaboration with the co-authors, wrote the first draft of the manuscript, incorporated all levels of feedback from co-authors, and prepared the manuscript for submission. RL, DB, and CN provided expert guidance and reviewed this manuscript and revisions. The project was funded by successful project grants secured by MK and RL. All authors contributed to manuscript revision, read, and approved the submitted version.