Participants consisted of 122 cognitively unimpaired older adults, between the ages of 60 and 80 years of age, recruited from the Butler Alzheimer’s Prevention Registry, a local database of older adults interested in AD research at the Butler Hospital Memory and Aging Program (MAP) [43]. All study procedures were carried out remotely, and detailed recruitment and procedural information has been described previously [38,41,44]. Study procedures are presented in Figure 1.

We used a targeted recruitment to enroll individuals with previous amyloid positron emission tomography (PET) data (elevated [Aß+] or nonelevated [Aß-] determined by clinical read by a radiologist) (final n=73) [41] as well as those without PET data. PET results were available if recorded in our Registry as part of previous research participation at the MAP. Individuals (n=256) were invited to participate in the study through an email or phone call. Individuals who consented (n=146) completed an online survey before the initiation of the remote testing protocol, which included various screening measures, including the Cognitive Functioning Index (CFI) [45] (subjective cognitive concern screen), Geriatric Depression Scale 5-item version (GDS-5) [46] (depressive symptoms screen; score range 0‐5, with higher scores indicating worse depressive symptoms), and the University of California, Los Angeles (UCLA) 3-Item Loneliness Scale [47] (loneliness screen; score range 0‐6, with higher scores indicating worse loneliness). The modified Telephone Interview for Cognitive Status (TICSm) [48] (objective cognition screen; score range 0‐50, with lower scores indicating worse cognitive functioning) was also completed before the remote testing protocol by phone, administered by a research assistant. During screening, 23 participants were excluded, and 1 participant was lost to follow up. After the completion of the 8 day remote study protocol, participants (final n=122) were invited to provide feedback (including study satisfaction and motivation) on their experience through an online survey, and a US $20 gift card was provided for compensation.

Inclusion criteria involved unimpaired cognition, daily use of a smartphone, and familiarity with smartphone features. Unimpaired cognition was defined as a TICSm cutoff score of ≥34 [48]. Exclusion criteria included self-reported history of cognitive impairment or dementia, history of neurologic disease or severe mental illness, and physical inability to complete smartphone-based testing.

The CFI is a 14-item self-report measure which probes changes in cognitive and functional domains over the past year. The CFI was selected for inclusion in this study as it is a widely used, briefly administered measure of SCC that has been well-validated for used in cognitively healthy and preclinical dementia populations [42,45,49,50]. Items reflect memory concerns, increased use of compensatory strategies, changes in driving, or difficulty managing instrumental activities of daily living. Participants respond with yes (1 point), no (0 points), or maybe (0.5 points), with total scores ranging from 0 to 14 (continuous variable). Higher scores reflect greater SCC. As a sensitivity analysis, we also created a SCC variable that included CFI items that solely assessed cognition (CFI-Cog), by removing the functioning related items to create a refined examination of SCC. CFI-Cog included CFI items 1, 2, 3, 4, 5, 6, 8, 11, and 13. This approach is consistent with a recent study which identified that cognitively focused CFI items, rather than functionally related CFI items, were endorsed at a higher rate and more likely to be associated with Aß+ in cognitively unimpaired older adults [42].

Adherence was quantified as the number of testing sessions completed across the 8-day ambulatory protocol (out of 24 sessions).

Participants’ feedback covered various aspects, including their satisfaction with the protocols and their level of motivation to engage with the study tasks. Feedback related to satisfaction was measured with 2 items: “Completing the brain games was fun” and “I became bored with the brain games.” Participants rated their responses on a 6-point Likert scale (1=strongly disagree to 6=strongly agree). For the purposes of this study, responses were dichotomized by those who agreed (Likert scale responses 4, 5, and 6) or those who disagreed (Likert scale responses 1, 2, and 3) on these items due to a largely positive response bias (most individuals found the tasks to be fun and not boring).

Feedback regarding motivation was measured on the item: “It was hard to motivate myself to complete the games each time.” Participants rated their response on a 5-point Likert scale (1=Agree to 5=Disagree). As above, responses were dichotomized into those who agreed or were neutral (Likert scale responses 1, 2, and 3) and those who disagreed (Likert scale responses 4 and 5) for the purpose of this analysis due to a largely positive response bias (most individuals found themselves to be easily motivated to complete the tasks).

Descriptive statistics were conducted to characterize the sample in terms of demographics (self-reported biological sex, race, ethnicity, and years of education), self-reported symptoms of depression (by the Geriatric Depression Scale 5-item version [46]), loneliness (by the UCLA Loneliness Scale [47]), and SCC (CFI total and CFI-Cog), relevant app-based metrics (ie, adherence, satisfaction and motivation), and objective cognitive performance (ie, visual working memory, processing speed, and episodic memory). Pearson correlation coefficient (for continuous variables) and point biserial correlation (for binary variables) were implemented as appropriate to assess the relationship between SCC and demographic variables, depression, loneliness, and outcomes of interest (eg, adherence, satisfaction, motivation, and performance on app-based cognitive assessment). To address aims 1‐3 and secondary analyses, SCC (continuous variable) was the main predictor of interest and we adjusted for covariates, including age, sex, depression, and loneliness, in all models. Covariates were selected based on significant association with independent and dependent variables of interest. A priori significance threshold (α) was set at P≤.05. For our moderation analyses, we examined α levels up to P≤.10.

To test aim 1, linear regression models were used to evaluate the association between SCC and adherence (continuous variable) over the 8-day protocol. For the purposes of this analysis, the CFI total score was the main SCC predictor of interest. We present results for CFI-Cog only when they differ from the CFI total score. To test aim 2, logistic regression models were conducted to evaluate the association between SCC and satisfaction (dichotomized self-report items related to level of fun and boredom), and SCC and motivation (dichotomized self-report items related to motivation). Individual regression models were implemented for each motivation and satisfaction outcome of interest. To test aim 3, separate linear regression models were constructed to evaluate the association between SCC and performance on each M2C2 based cognitive assessment (Prices, Color Shapes, and Symbol Search, all continuous variables). Linear-mixed effects models were used to evaluate whether SCC predicted M2C2 subtest performance over time while controlling for age, sex, and loneliness. Finally, as a secondary analysis in a subsample of participants with Aβ PET data, we fit linear regression models with SCC and the interaction of SCC and Aβ PET positivity as predictors, to test the moderation effect of Aβ PET positivity on the association between SCC and study adherence and performance on M2C2 tasks, adjusting for covariates. All descriptive and regression analyses were conducted using SPSS version 28.0.1.1 (IBM Corp). Linear mixed-effects models were conducted using the nlme R package.

The project received approval from the Butler Hospital Institutional Review Board (#1882523), and all the participants provided informed consent. All data was de-identified. Participants were compensated for their time with a $20 gift card.

Overall sample demographics and SCC, adherence, satisfaction, motivation, and performance variables are presented in Table 1. The sample (N=122) was 68.85 (SD 4.93) years old (range=60‐81 y) with 16.52 (SD 2.39) years of formal education. The majority of the sample identified as female (n=82, 66.7%), White (n=106, 86.2%), and non-Hispanic (n=109, 88.6%). In the subset of individuals with Aβ PET scans, there were 25 Aβ positive participants and 48 Aβ negative participants. Subjective cognitive concerns, as measured on the CFI, were fairly minimal in this sample. On average, the sample scored 1.8 (SD 1.7) out of 14 possible points on the overall CFI ; however, the most common score on the CFI was 0 (mode=0), with 18% of the sample expressing no concerns about their cognition on the CFI.

Bivariate associations are presented in Table 2. The CFI did show a small to moderate correlation with worse cognitive status (TICSm, r=−0.202, P=.03), greater loneliness (UCLA Loneliness Scale, r=0.248, P=.006), and worse adherence (r=−0.207, P=.02). In terms of overall M2C2 performances, the CFI showed a small correlation with episodic memory (ie, Prices) performance (r=−0.196, P=.03).

Overall, remote assessment protocol adherence was high in the sample, with participants completing on average 93.5% of the 24 testing sessions over the 8-day period. CFI-Cog endorsement significantly predicted overall remote testing adherence over 8 days. This suggests that higher levels of SCC were associated with worse adherence (a diminished number of test sessions completed across the protocol) (ß=−.197, P=.04, 95% CI −.647 to −.021). However, this association was not observed when using the CFI total score as the predictor (Table 3).

In terms of overall protocol satisfaction, participants endorsed that completing the assessments was fun (n=102, 85%) and not boring (n=92, 75.2%). Participants also reported that they did not encounter difficulties in motivating themselves to give their best performance (n=93, 77.5%). CFI total score was not associated with metrics of study satisfaction in the adjusted models, including how fun or boring participants found the protocol to be. In terms of motivation, CFI total score was not predictive of self-reported effort across the protocol period (more details in Table 3).

SCCs and their association with overall objective cognitive performances on digital cognitive tests was also investigated (more details in Table 3). High SCC endorsement on the CFI total score was significantly predictive of poorer overall performance on the Prices task (ß=−.200, P=.02, 95% CI −.020, to −.002). There were no significant associations between the CFI and performance metrics for Color Shapes (P=.86) or Symbol Search (P=.94).

We also investigated the association of SCC and longitudinal trends of cognitive performance over the 8-day testing period (M2C2 app performance descriptives for each study day have been previously reported) [38]. There was a main effect of the CFI total score suggesting that higher levels of SCC were associated with worse performance on Color Shapes (working memory) at day 1 (unstandardized estimate [b]=−1.047, SE=0.47, P=.03). Furthermore, there was a significant interaction between SCCs and Color Shapes performance (b=0.048, SE=0.02, P=.03), such that higher CFI total scores were associated with improved performance on this subtest over time. In other words, although high SCC was associated with worse working memory performance initially, individuals seemed to benefit from repeated practice and ultimately performed better (more details in Figure 3). There were no significant associations between SCC and intraindividual variability in Prices or Symbol Match performances over the protocol period.

The moderation of Aβ positivity on the relationship between CFI and app-based metrics, as well as CFI and cognitive performance (ie, Prices, Color Shapes, and Symbol Search) was investigated in a subset of individuals with Aβ PET scans (n=73). The presence of Aβ did not moderate the relationship between CFI total score and adherence. There was evidence for moderation at the P<.10 level of Aβ positivity on the relationship between CFI total score and Prices was observed, but did not fall within the α=.05 threshold (ß=.179, P=.09, 95% CI −.003 to .036). Aβ status did not moderate the association between SCC and Color Shapes or Symbol Search.

This study sought to characterize the relationship between SCC and mobile app–based protocol engagement (adherence, satisfaction, and motivation) and performance on cognitive tasks in a cohort of cognitively healthy older adults. While protocol satisfaction and motivation were not impacted by SCCs, overall adherence was associated, such that higher levels of SCCs predicted lower rates of adherence. Overall, we show that SCCs were associated with worse objective performance on mobile app–based assessments. Our findings suggest that mobile app–based assessments hold promise as sensitive tools to detect subtle neurodegenerative changes in individuals at risk for dementia, with potential application for early detection, diagnosis, and treatment.

We observed an association between SCCs, as measured on the Cognitive Function Index (CFI), and overall worse performance on a test of episodic memory (Prices). A significant association was not observed between SCC and overall performance on tasks of processing speed or working memory. Findings in the broader SCD literature are quite mixed as to whether SCCs are or are not associated with concurrent cognitive functioning abilities [52]; as such, our results may indicate that the use of sensitive digital cognitive tools could enhance the association between SCC and objective cognitive metrics. Our result reflecting a unique association with episodic memory is perhaps clinically meaningful, as memory concerns specifically are frequently considered to be the hallmark presenting symptom of AD [2,53], particularly in individuals over the age of 65 [53]. This observation is underscored by our finding of a moderating effect of cerebral Aβ positivity (at the P<.10 level) on the relationship between SCC and episodic memory (Prices) in a subsample of participants with biomarkers available for analysis (secondary aim), such that the association of SCC with episodic memory was stronger in the presence of Aβ. This result adds to the growing evidence base for an association between SCC and higher Aβ [52]. Replication of this analysis in a larger sample may clarify the interplay between SCC, AD biomarkers, and subtle memory deficits observed on mobile app–based assessment in otherwise cognitively and clinically normal older adults. That being said, it is worth considering that the unique association between SCC and episodic memory in our sample may be at least somewhat secondary to the way the CFI is constructed, as this instrument mostly focuses on episodic memory-related cognitive concerns [45]. Previous studies have demonstrated the importance of querying beyond just memory concerns to most sensitively detect dementia risk [54,55]. Indeed, SCC can be sampled by a multitude of techniques and approaches (eg, memory only vs multiple cognitive domains; traditional paper and pencil vs digital assessments; self vs informant report; capturing current ability vs change in ability, etc) [24,25], all of which may impact the sensitivity of the SCC measures to detect current and risk for future cognitive impairment. Future research should carefully weigh SCC assessment approaches, and consider incorporating a more comprehensive SCC screener which queries across a broader range of cognitive domains, as this may be valuable for detecting more meaningful associations between SCC and mobile app–based cognitive performances.

We also evaluated patterns of cognitive performance across the 8-day assessment period. Analyses revealed that more SCC was associated with initially worse working memory (Color Shapes) performance on day one. However, individuals with higher SCC showed progressively improving working memory performance over the assessment period, and by day 8 were performing in a similar range with those with lower SCC. This pattern was not observed on the episodic memory or processing speed subtests. This finding is consistent with recent results from Aschenbrenner et al [56], which showed variability over high frequency app-based assessments in attention and working memory-based skills, but not in episodic memory, in older adults who were at increased genetic risk for AD [56]. Our finding of an initial relative weakness in working memory skills could reflect the presence of mild cognitive inefficiencies in older adults with SCC, which may be compensated for with repeated exposure to the task over time. This may be due to early dysfunction in working memory processes, which have been characterized in preclinical AD [57,58]. Subjectively, this may be experienced by these individuals as thinking which requires increased concentration, that is punctuated by occasional lapses, and is associated perhaps with a sense of being cognitively overwhelmed, all of which may be driving the report SCC in daily life, but may be subtle enough that it may not be detected on traditional neuropsychological tests. Hence, this type of unique marker of cognitive inefficiency would only be afforded through high frequency, mobile app–based cognitive assessment in this population. The plausible role of performance anxiety may also explain variability in cognitive performance overtime. Although beyond the scope of the present analysis, research which comprehensively measures the role of in-the-moment psychological and physiological states by ecological momentary assessment techniques in relation to in the moment SCC and objective cognition will be critical next steps.

Our findings related to SCC and metrics of protocol engagement and satisfaction with unsupervised, high frequency digital cognitive assessments have potentially broad implications for aging research, especially in light of increasing interest in using these techniques to remotely assess cognitive functioning in this population [31,32]. Indeed, whether or not a study is directly examining SCC or includes SCD as a clinical group of interest, aging studies must contend with the fact that SCCs are widespread and may be clinically meaningful feature among the older adults in their sample [2]. We show that SCC was not associated with protocol satisfaction (including how fun or how boring the experience was) nor did SCC effect how motivated a participant was to engage and put forth their best effort. However, higher levels of SCC did impact protocol adherence in our sample, with mildly worse compliance those with higher SCC observed. For context, protocol adherence was overall quite high in our population (93.5%) and similarly high compliance rates have been described in several other recent smartphone-based studies [32], yet no previous studies have directly explored the impact of SCCs on protocol adherence. It may be that high frequency, mobile assessments are able to detect mild patterns of forgetfulness (manifested in slightly worse adherence rates) in older adults with SCC, so future research should be attuned to this possibility. Replication of this association between SCC and adherence to remote protocols in larger, more demographically diverse samples is warranted.

Some limitations of this study, as well as potential future directions of this research, warrant acknowledgment. Our sample was largely White, disproportionally female, and highly educated. This limits the generalizability of the study findings to broader demographic groups. Our sample was from a relatively small single cohort and only a subset had AD biomarkers available for analysis, restricting study power and may have impeded our ability to obtain significant findings. A measure of anxiety symptomology was not available in the present sample, limiting our ability to understand the plausible role of anxiety, SCC, and performance on digital assessments [32,59]. The main predictor of interest in this study, the CFI, shows adequate evidence for validity and reliability [42], but is somewhat limited in the domains of cognitive concerns it measures (eg, memory and daily functioning). Research has shown that comprehensive assessment of concerns across a broad range of cognitive domains may be most sensitive to risk in aging populations [60], so future smartphone-based research may consider including a broader range of cognitive concern items to better understand this relationship. We were unable to assess informant-report of SCC in this sample, which has been demonstrated to be optimized to predict risk above and beyond self-report [25], and this should be addressed in future smartphone studies. Although such metrics were not available in this study, there is also a small but growing literature which has examined the use of smartphones to assess SCC in a digital remote format in tandem with cognitive tests and relevant clinical outcomes in older adult samples [61-63]. Future research on SCC and smartphone-based digital cognitive assessments should investigate the relative value of traditional in person reported SCC (eg, the CFI) versus smartphone based digital SCC assessments. Finally, this cross-sectional study was not able to investigate the relationship between SCC and longitudinal decline on smartphone-based cognitive performance in individuals with positive AD biomarkers. It will be essential for future studies to follow participants over time to understand if SCC at baseline may predict incident decline on these novel cognitive metrics. Due to the limited nature of this pilot study, we cannot provide specific recommendations about optimal SCC assessment approach for use in future digital cognitive studies.

SCCs are frequently one of the earliest clinical symptoms of dementia [2]. However, clinical interpretations of SCCs are complicated by the ubiquitous endorsement of these concerns [2], their potential for non-neurodegenerative etiology (eg, psychiatric, medical, or sociodemographic factors) [2,3], and frequent lack of association with traditional objective cognitive testing [1]. Smartphone-based, digital cognitive assessments, which are increasingly used in aging research, may offer potential for improved ecological validity and sensitivity to subtle markers of cognitive decline [30-32,35]. Hence, such smartphone-based digital tools may be able to capture quite mild deficits in individuals with SCC who perform within normal limits on traditional neuropsychological tests. Results from the current study showed that SCCs are associated with worse overall memory performance on mobile app assessment, and patterns of cognitive inefficiency (variable working memory) and mild forgetfulness (diminished adherence) across an 8-day assessment period in cognitively intact older adults. Findings indicate that mobile app assessments may be uniquely sensitive to very subtle neurodegenerative changes in at risk older adults, with critical implications for early detection and timely intervention.