Dynamic Influence of Mood on Subjective Cognitive Complaints in Mild Cognitive Impairment: A Time Series Network Analysis Approach

Abstract

Objectives:

Subjective cognitive complaints (SCC) are common and clinically relevant in mild cognitive impairment (MCI) but are intertwined with mood states. Using Ecological Momentary Assessment (EMA) of SCC and network analyses we sought to uncover the links between mood and SCC and how these links may vary by the presence or absence of MCI.

Design:

We used EMA to collect intensive longitudinal data. In addition to analyzing the data at an aggregate level to estimate between-person associations of affect and SCC variables, we used time series analyses to estimate contemporaneous and time-lagged relations between the variables.

Setting:

EMA survey and mobile cognitive testing in subjects’ natural environments.

Participants:

The sample included 100 participants, 48 with Normal Cognition (NC), and 52 with MCI.

Measurements:

Participants completed 30-day EMA protocols in which surveys sampling SCC and moods were delivered 3 times per day.

Results:

The association between SCC (as measured by EMA) and standard in-lab measures of SCC was significant in MCI, but not in NC. Despite no average level differences in severity of SCC, there was a strong association between negative affect (as measured by EMA) with SCC in NCs compared to MCI.

Conclusions:

EMA maybe a useful and valid approach to measuring SCCs in MCI. Network analyses indicated that negative affect was linked with later SCCs. This finding was strong in persons with NC than in persons with MCI. The basis of the attenuated association between negative affect and SCCs in MCI deserves further study.

1. INTRODUCTION

Subjective cognitive complaints (SCC) are common among older adults with and without cognitive impairments(Bassett and Folstein, 1993), and are seen as clinically relevant in understanding cognitive decline(Hong and Lee, 2023; Jacob et al., 2019; Mark and Sitskoorn, 2013; Molinuevo et al., 2017; Ávila-Villanueva et al., 2016). Several current criteria (e.g., the National Institute on Aging-Alzheimer’s Association recommendations) for mild cognitive impairment (MCI) include subjective cognitive complaints (SCC) by the patient or an informant(Albert et al., 2011; Truong et al., 2022). However, the inclusion of SCC in these criteria is controversial(Jeste, 2022; Kliegel et al., 2005). For example, research has repeatedly shown that self-reported SCC in older adults is often associated with depressive symptoms(Pfund et al., 2022), rather than objective cognition(Bell et al., 2023; Hill et al., 2016; Markova et al., 2017; Mogle et al., 2020), with some research showing that inclusion of SCC in MCI criteria leads to an overdiagnosis of MCI(Edmonds et al., 2014). The association between SCC and mood states is complex, particularly given that neuropsychiatric symptoms, such as depression, are common in MCI(Martin and Velayudhan, 2020). Therefore, it is difficult to discern if negative affect and SCC are associated because they both are present in the early stages of cognitive decline or if increased negative affect leads to greater SCC.

To date, most research on the link between emotions and SCC has relied on global assessments of emotion or affective symptoms predominantly assessed in a lab or clinic-based setting rather than momentary assessments which can capture the temporal dynamics of these experiences(De Vries et al., 2021). EMA has been shown to be a feasible methodology to momentarily assess a wide-array of variables in studies of older adults with and without cognitive impairment or affective symptoms(Kim et al., 2020; Ramsey et al., 2016), and stroke survivors (Lau et al., 2022; Lau et al., 2023). To our knowledge, there have only been a few studies utilizing frequent sampling to examine the dynamic associations between SCC and emotional experiences in older adults. In an 8-day daily diary study of normal cognitive aging involving 333 older adults who were US Veterans, greater stress had both a concurrent and lagged (i.e., the next day) association with greater self-report of memory failure(Neupert et al., 2006). In addition, another smaller (n=45) but longer (3 week) EMA study in Multiple Sclerosis patients showed significant positive correlations between SCC and fatigue, anxiety, depression, and pain(Chen et al., 2023). To our knowledge, there have been no studies evaluating SCC and mood states in mixed samples of older adults with MCI.

EMA could be useful not only to understand the timing of the co-occurrence of negative moods and SCC, but also the time-lagged relationships between mood and SCC using time series network analyses(Heyse et al., 2021). In network analysis, contemporaneous and time-lagged relationships can be represented as a single network of nodes and connections among nodes according to contemporaneous and lagged effects. Such networks combine all relationships between variables, potentially uncovering feedback loops that are indicative of homeostasis (maintenance) or buildup processes that regulate the relationships between emotional experience and SCC. Our past studies that combined EMA with causal network models in other populations provided insights into complex dynamics involving a variety of variables: mood states, cognition, social context, self-assessments of competence, and behaviors(Badal et al., 2022a; Badal et al., 2022b; Badal et al., 2021).

Therefore, this study aimed to evaluate relationships between moods and SCC using a sample of 52 people with MCI and 48 normal cognition (NC) who completed 30-day EMA protocols in which surveys sampling SCC and moods were delivered 3 times per day. To examine these effects comprehensively, we 1) evaluated the association of averages of EMA-based measures of SCC in MCI and NC with global measures of self-rated cognition at baseline, 2) examined averaged correlations of positive and negative emotional experience with SCC in NC and MCI, and 3) performed network analyses of SCC and moods in NC and MCI. We hypothesized that EMA measures of SCC would be significantly associated with global self-report measures of the same construct and that mood states, across measurement strategies, would be significantly associated with SCC in both MCI and NC. We explored differences in these associations between MCI and NC and the incremental value of network analyses for detecting concurrent and lagged links between moods and SCC in both samples for the benefit of hypothesis generation.

2. METHODS

2a. Sample

Analysis included 52 participants who met criteria for MCI (MCI group) and 48 cognitively unimpaired participants (normal cognition; NC group). All participants were aged 50 and over, English-proficient, and had a familiar other who served as an informant. Exclusion criteria included (1) history of a neurological disorder that may affect brain function (e.g., stroke, epilepsy, movement disorder),(2) presence of dementia (determined via existing diagnosis and/or established cutoff scores on the MoCA considering education(Rossetti et al., 2011), (3) head injury with loss of consciousness ≥15 minutes, (4) sensory impairments (vision or hearing) that would interfere with a participants ability to complete the study protocol, (5) presence of intellectual disability (as defined as a WRAT-4 Reading subtest standard score < 70), (6) current diagnosis of a substance use disorder, or (7) history of a psychotic disorder or bipolar disorder.

Data were gathered from December 2020 to December 2021 at three different academic sites: The University of Texas at Dallas (UTD), University of California San Diego (UCSD), and the University of Miami Miller School of Medicine (UM). UTD recruited participants via community advertisements and from the participant pool at UTD’s Center for Vital Longevity. UCSD participants were recruited from advertisements in the Stein Institute for Successful Aging newsletter and word of mouth. UM participants were recruited from advertisements in the local community, the participant pool at the Florida Alzheimer’s Disease Research Center, and clinical programs at the Miller School of Medicine Memory Disorders Center.

2b. Study Protocol

All procedures were approved by each university’s respective Institutional Review Board prior to protocol implementation; all participants provided written, informed consent. Participants completed a brief phone screen with study staff to confirm initial eligibility. Participants then completed a baseline assessment to collect sociodemographic information, the Montreal Cognitive Assessment (MoCA)-BLIND version 7(Pendlebury et al., 2013), neuropsychological testing for diagnostic classification, and subjective cognitive functioning measures, including the Cognitive Assessment Interview (CAI). Participants were compensated $50 for completing the baseline assessment. Due to restrictions during the COVID-19 pandemic, some participated through remote visits (68%) while other individuals completed the baseline assessment in person (32%). Additional details on COVID-19 modifications can be found in (Moore et al., 2022). There were no significant differences in neuropsychological test performances based on remote or in person administration.

Following the baseline assessment, eligible participants then participated in an EMA survey and mobile cognitive testing period for 30 consecutive days. This study only includes data from the EMA surveys; details about the mobile cognitive tests can be found in (Moore et al., 2022). Participants were able to use their own smartphone device or a study-provided Android smartphone. If participants chose to use a study-owned device, they were trained on how to operate the device and provided with a user manual. All participants were trained by study staff on the EMA protocol and completed a mock EMA protocol with study staff to ask questions and troubleshoot any difficulties. During the survey period, participants were sent a text message notification three times per day to complete EMA surveys via the NeuroUX platform. To encourage adherence and to provide assistance if there were any technical difficulties, study staff contacted participants if three consecutive surveys were missed. Participants were compensated $0.88 for each completed survey.

2c. Baseline Measures

Cognitive Assessment Interview (CAI).

The CAI is a 10-item interview-based measure designed to assess the severity and change in cognitive deficits and how deficits affect everyday functioning(Ventura et al., 2010). Trained raters interview participants and their informant separately about 6 domains including working memory (2 items), attention vigilance (2 items), verbal learning and memory (2 items), reasoning and problem solving (2 items), speed of processing (1 item), and social cognition (1 item). Based on responses to semi-structured questions and prompts, the clinical rater rates each item on a 7-point scale ranging from 1=normal, not at all impaired to 7=cognitive deficits are so severe as to present danger to self/others. The “self-report” scores are based on the participant’s report, “informant” scores are based on informant reports, and the “research associate” scores are generated on the basis of a combination of participant self-report, informant report, and observations during the evaluation. The CAI Global Score represents the average score across the 10 items, with greater scores indicating worse impairment. Additionally, the instrument also includes an additional Global Assessment of Functioning (GAF) rating which is rated on a 100-point scale. For example, scores from 100 to 88 indicate unimpaired, 87 to 74 indicates minimal cognitive deficits but functioning is generally effective, and 73–59 indicates mild cognitive deficits with some consistent effect on functioning.

Determination of MCI status.

Participants completed comprehensive neuropsychological testing to assess MCI status. Additional information on the battery and procedures can be found in (Moore et al., 2022). Participants were classified as MCI if they met criteria for any subtype of MCI using the Jak/Bondi criteria(Jak et al., 2009) indexed by one standard deviation below normative expectations on two different assessments within a single cognitive domain (i.e., memory, attention, language, executive functioning).

2d. EMA protocol

During each EMA survey, participants were asked a series of questions to assess a wide array of variables including subjective cognitive functioning and moods, as well as several other variables not relevant to the current study. To assess momentary subjective cognitive functioning, participants were asked to respond to three separate questions querying difficulty concentrating, forgetfulness, or slow thinking (i.e., “Since the last alarm, how much have you experienced [difficulty concentrating, forgetfulness, or thinking slow]?”) using a scale from 1=not at all to 7=extremely. To assess positive and negative moods, participants were asked to rate, on a scale from 1=not at all to 7=extremely, their current level of NA (sadness and nervousness) and PA (happiness and excited).

2e. Analysis

Pearson’s correlations were calculated between aggregated EMA variables and overlayed on heatmap generated using Python Seaborn(Waskom, 2021) library to assess between-person level associations between variables. We evaluated the associations between aggregated affect, SCCs, and in-lab measures of SCC. We applied R-to-Z transformation to evaluate differences in these aggregated correlations between MCI and NCs.

To assess within- vs. between-person variability, we calculated Mean Squared Successive Differences (MSSD) (Von Neumann, 1941) and its ratio to variance (σ²). MSSD is a dimensionless measure of volatility, a ratio <2 of MSSD to σ² suggests sequential order and autocorrelation.

Network analyses were performed with Tigramite(Runge, 2019), a Python implementation of causal discovery using partial correlations and Momentary Conditional Independence (PCMCI(Runge et al., 2019)). Only the edges with P value <0.01 (using the alpha parameter in Tigramite) were represented in network graphs. To control for Type 1 errors in correlations, the Benjamini-Hochberg Method(Benjamini and Hochberg, 1995) is utilized by the PCMCI algorithm. This method effectively controls the false discovery rate (Type 1 errors) in the constructed networks. The method is popular and effective even in small sample studies(Storey and Tibshirani, 2003), eliminating spurious dependencies while maintaining high sensitivity. A window of size 1(T_max = 1, or, 8-hours) was used for the dataset as focus was on near term and longer lags were deemed potentially uninformative.

The timeseries for a subset of variables for nomothetic analysis for a group (MCI or NC) was constructed by sequentially arranging the individual timeseries. In such series, we marked the missing values, and we invoked the run_pcmci() method, with the constructed timeseries as parameter, and set ParCorr() (partial correlation) as the function to evaluate the associations between two variables. Additionally, we set as the window for lags at one, and set the alpha level of significance to 0.01. The PCMCI method returned us the 3-dimensional array containing the computed associations, p-values and lags. We then performed Benjamini-Hochberg Method(Benjamini and Hochberg, 1995) correction to minimize false positives. Finally, we obtained the link matrix representing the network graphs.

We did not impute missing values. To handle missing EMA data (in our study the affect and SCC variables) without introducing a bias, missing values were flagged and appropriately called by Tigramite. The algorithm is designed to handle some degree of missing values. Although the algorithm flags an error if there are too many missing values for the analysis and terminates the analysis, we did not encounter such a situation.

The networks constructed were compared across the groups based on measures of edge density and goodness of fit. Network density quantifies the interconnectedness of the nodes within the graph (Eq.1).

$$\text{D}=\frac{{\sum }_{i,j}^{\text{N}}{w}_{i\to j}}{\text{N}(\text{N}-1)/2}$$ (1)

Where $w_{i\to j}$ is an edge connecting nodes $i$ and $j$ in the network of N nodes.

Goodness of fit (R²) measures the similarity between two network graphs with an identical set of nodes (Eq. 2). In our case, we used the NC networks as reference. A value of 1 implies identical structure, while values closer to 0 reflects reduced similarity.

$${\text{R}}^2=1-{\sum }_{i,j}^{\text{N}}{\left(w_{i\to j}^{\text{DX}}-w_{i\to j}^{\text{NC}}\right)}^2$$ (2)

Where N is the total number of nodes in the network of participants with diagnosis (DX), $i$ and $j$ are nodes, and ${\text{w}}_{\text{i}\to \text{j}}$ is a weighted edge between them.

3. RESULTS

3a. Group comparisons on Sample Characteristics and Lab Ratings

The sample (Table 1) comprised N=48 NC and N=52 MCI, with no significant differences in age, years of education, sex, race or ethnicity. Adherence was indexed by the proportion of EMA surveys answered, which was 79.5% overall (SD = 21.7). Adherence was significantly higher in the NC group (M = 85.7%, SD = 20.9) compared to the MCI group (M = 76.8%, SD = 21.9), t(df=98) = 2.08, P = .04). Similarly, informant reported as well as research associate reported CAI GAF scores, reflecting better cognitive performance, were significantly higher for NC. EMA measures of mood variables (sad, relaxed, excited, happy, and nervous) or SCCs (difficulty concentrating, forgetfulness, and thinking slow), when aggregated by individuals, did not significantly differ between NCs and people with MCI (Table S1). The correlations among informant reported as well as self-reported CAI GAF scores were stronger and significant in people with MCI (Figure S1).

Table 1.

Group differences in Sociodemographic and Lab measures

	NC (N=48)	MCI (N=52)	Effect size	Statistics
	Mean(SD); min-max	Mean(SD); min-max	Cohen’s d	Mann Whitney U / χ2	P value
Age	70.5(6.5); 60–87	71.3(7.6); 54–85	−0.11	1145.50	0.481
Years of Education	16.2(2.0); 12–23	15.9(2.6); 9–21	0.13	1319.50	0.615
Sex N(%)				0.93	0.334
Male	34(70.8%)	31(59.6%)
Female	14(29.2%)	21(40.4%)
Race N (%)				0.00	1.000
White	43(89.6%)	47(90.4%)
Non White	5(10.42%)	5(9.62%)
Ethnicity N(%)				0.50	0.482
Non-Hispanic	43(89.6%)	43(82.7%)
Hispanic	5(10.4%)	9(17.3%)
Measures
Montreal Cognitive Assessment (MoCA)	20.1(1.4); 16–22	19.1(2.0); 13–22	0.56	1607.50	0.012*
Cognitive Assessment Interview (CAI), Self-Reported, Global	1.3(0.5); 1–2	1.6(0.7); 1–3	−0.55	882.00	0.015*
Severity
Cognitive Assessment Interview (CAI), Self-Reported, Global	87.8(5.6); 75–100	87.4(6.9); 70–100	0.06	1183.00	0.961
Assessment of Functioning (GAF)
Cognitive Assessment Interview (CAI), informant, Global	1.1(0.3); 1–2	1.5(0.8); 1–4	−0.71	667.00	0.002*
Cognitive Assessment Interview (CAI), informant, Global	90.6(14.8); 1–100	88.6(9.5); 60–100	0.16	1109.50	0.104
Assessment of Functioning (GAF)
Cognitive Assessment Interview (CAI), Research Associate, GSCI	1.2(0.4); 1–2	1.9(0.8); 1–4	−1.07	548.00	<0.001*
Cognitive Assessment Interview (CAI), Research Associate, Global Assessment of Functioning (GAF)	88.1(6.0); 70–100	79.8(9.7); 61–95	1.03	1635.50	<0.001*

^*Significant differences

3b. Correlations among aggregated EMA variables

The heatmaps constructed from cross-correlations between aggregated EMA variables for the two groups are shown in Figure 1. The correlation between the two EMA negative mood state variables (e.g., sad, nervous) and the three EMA subjective cognitive complaint variables (Difficulty in concentrating, Forgetfulness, slow Thinking) were uniformly higher for the NC sample. For example, aggregated reports of being sad and thinking slow were highly correlated (r=.89, P <0.01) in NC, but the correlation was lower (r=.48, P <0.01) in MCI.

Figure 1:

Heatmap of cross-correlations among EMA variables aggregated by sample across time. All correlations were significant.

The differences in Z scores, an estimator of effect size for correlations or Cohen’s q (Cohen, 1988), was computed for the association between mood variables and SCC (Table S2). For negative mood, they were interpreted as large as they ranged between 0.776 to 1.011 across the two groups. In contrast, correlations between positive affect and SCC were not materially different and showed mostly small effect sizes between the two samples.

3c. Correlations among averaged EMA cognitive-complaint variables for individuals and their baseline assessments of cognition

In the MCI group, higher EMA SCC ratings were strongly, positively, and significantly correlated with informant, research associate and self-report based Cognitive Assessment Interview Global Severity Scores, reflecting more cognitive impairment (Figure 2). Better CAI GAF measures across sources were negatively correlated with EMA measures of SCC in MCI. The self-report, informant, and research associate CAI GAF measures negatively correlated with EMA measures of SCC in MCI. For NC, the self-report CAI GAF measure negatively correlated with EMA measures of SCCs, but not other informat measures. For NC, self-reported better CAI GAF scores were negatively correlated with EMA measures of SCC.

Figure 2:

Correlation between baseline CAI measures and EMA measures aggregated by sample. * Indicates P<.05, and ** indicates P<.01.

3d. Within- and between person variability

High autocorrelation (MSSD/σ² < 2) suggests between person differences were stronger than within person variations. Also, MCI differed from NC and showed lower variances in their momentary reports of slow thinking. Variability in ratings of difficulty concentrating and forgetfulness did not differ (Table S3).

3e. Influence of Negative Affect on Subjective Cognitive Complaints

Networks constructed from the timeseries of EMA variables for negative affect are displayed in Figure 3a. Overall, negative affect was more strongly associated on a contemporaneous basis with SCCs among NC than people with MCI, displayed by deeper green straight lines in NC networks (Figure 3). In NCs, anxious and sad moods were associated with later complaints of thinking slowly which in turn were correlated with forgetfulness and difficulty in concentrating.

Figure 3a.

Negative Affect Network Models. Straight lines represent contemporaneous associations while curved lines represent lagged associations with numerals in hollow font (namely 1’s) in the figures denoting the lags between the variables. The lag of 1 sampling interval is approximately 8 hours. Shades of green represent positive associations while shades of red represent negative.

The MCI sample displayed weaker coupling between the negative affect variables and SCC, displayed by paler green contemporaneous (straight) edges across them, while the strengths of edges within the SCC variables are greater. The lagged effects between negative affect and SCC were also weaker or not present in MCI contrasted with NCs.

Measures of network density also support stronger relationships between negative affect and SCC among NCs when compared to people with MCI (2.1 vs. 1.7, Table S4). The Density ratio (calculated using NC as reference group) is higher for NC than MCI (1.00 vs. 0.81). The goodness of fit, R² measure of the MCI network was 0.85 showing similarity to NC network.

3f. Influence of Positive Affect on Subjective Cognitive Complaints

Networks constructed from the timeseries of EMA variables for positive affect are displayed in Figure 3b. Overall, better positive moods were more negatively associated on a contemporaneous basis with more SCCs among people with MCI than NC, displayed by larger number of linkages between mood variables and SCC variables in MCI than NC networks (Figure 3b). Some links however were stronger and more negative in NCs, most notably between momentary reports of being happy and having difficulty in concentrating such that happiness seems to promote a perception of better concentration among NC.

Figure 3b.

Positive Affect Network Models. Straight lines represent contemporaneous associations while curved lines represent lagged associations with numerals in hollow font (namely 1’s) in the figures denoting the lags between the variables. The lag of 1 sampling interval is approximately 8 hours. Shades of green represent positive associations while shades of red represent negative.

Measures of network density also support stronger relationships between positive affect and SCC among people with MCI than NC (1.9 vs. 1.6, Table S4). Density ratio (calculated using NC as reference group) is higher for MCI than than HC (1.19 vs. 1.00). The R² measure of the MCI network (0.92) suggest close similarity to the NC network.

4. DISCUSSION

This study is among the first, to our knowledge, to employ EMA in understanding the time course and correlates of subjective cognitive complaints in MCI and NC, focusing on the convergence with momentary negative and positive moods. There were several potential important findings. For one, this study demonstrated that EMA may be a feasible and valid method to probe SCC in MCI. Our study demonstrated high adherence rates to EMA in this older adult group with and without cognitive impairment, albeit with slightly lower adherence in MCI (76% vs. 85%). The feasibility of EMA in older adults has been previously reported(Kim et al., 2020; Ramsey et al., 2016), but this paper is one of the first to demonstrate that assessing SCC via EMA is feasible in MCI population. Secondly, the pattern of coupling of emotion and SCC seemed to vary across MCI and NC samples, with greater independence from negative affect in MCI compared to NC. Requiring replication in a prospective study, differences in the convergence between affective experience and SCC may be an avenue for future research.

Our study also found preliminary support for the convergent validity of SCC assessed via EMA in MCI with lab-based measures in this study. Specifically, in the MCI group, EMA-based SCC were moderately associated with the self-, informant-, and RA-rated CAI, indicating that persons with MCI in this sample had some awareness into their cognitive difficulties. These findings may not reflect all persons with MCI and dementia (e.g., anosognosia in Alzheimer’s disease)(Cacciamani et al., 2021; Edmonds et al., 2018). Despite the convergence of lab-based and momentary assessments of SCC, the SCC reported via EMA did not differ across objectively-defined diagnostic groups (i.e., MCI and NC), who were defined by substantial differences in performance on a neuropsychological assessment battery. This is in-line with previous research demonstrating that lab- or clinic-based assessment of SCC does not help to Identify objective deficits in older adults(Edmonds et al., 2014) and that global reports of subjective complaints can fail to differ across MCI and NC participants(Harvey et al., 2023).

However, the correlation between EMA-based SCC and the CAI was not present in the NC group, potentially related to lower objective cognitive impairment, leading to less agreement between reports across information sources on this instrument within the NC group. Therefore, it appears that reports of subjective cognitive abilities in older adults without objective deficits may reflect a different phenomenon. Further examination of the dynamics (e.g., intra-individual variability) of EMA-assessed SCC in older adults with an dwithout cognitive impairment is needed. In sum, this study does provide preliminary evidence that EMA-assessed SCC may be a valid way to repeatedly assess SCC in MCI.

There are differences in the patterns of intercorrelation between mood state reports in the NC and MCI participants. MCI participants manifested notably smaller intercorrelations between positive and negative affect and smaller negative correlations across positive and negative emotional domains than NC (Figure 1). This finding may suggest some differences in bearings of mood states on subjective complaints across groups. We recently reported on a similar phenomenon in SMI, wherein participants with schizophrenia, assessed with EMA, manifested a reduced negative within-subjects correlation between positive and negative affect among participants with bipolar disorder(Parrish et al., 2023). Those same participants with bipolar disorder manifested a much larger correlation than the participants with schizophrenia between the momentary experience of NA and reports of pooer cognitive performance assessed immediately after test performance(Dalkner et al., 2023). The normative congruence between positive and negative affect may be reduced in MCI, possibly impacting on the convergence of mood states and subjective performance.

Correlational analyses and network models indicated contemporaneous and prospective links between negative affect and subsequent SCC in the NC group that were weaker than in MCI group. The cross-sectional association between negative affect and SCC in older adults has been repeatedly documented in the literature(Bell et al., 2023; Hill et al., 2016; Mogle et al., 2020). However, this analysis adds to this body of work by identifying lagged associations between affect and subsequent SCCs. Network analyses indicated that negative affective experiences precede SCCs, particularly in NCs. It is unclear why people with MCI had attenuated associations between negative affect and SCCs. However, it may be that SCCs that are not linked to affective experiences and that decoupling may be of clinical significance pending replication. The observation that MCI differed from NC and showed lower intra-individual variation in their momentary reports of slow thinking, suggesting greater stability in their responses, may be one reason for diminished correlations with negative affect. Measuring the dynamic between SCC and affect would be difficult without EMA; therefore, further research would be needed to test the feasibility of this approach in samples with greater variability in SCCs and objective cognitive ability.

It is important to consider these findings in the context of limitations. The sample sizes were relatively small and thus the findings would need to be replicated. That said, the SCC and mood ratings were based on greater than 2500 EMA surveys. Importantly, our exploratory comparative findings correspond to a general pattern observed in MCI and NCs and were corroborated by R to Z transformation comparing strength of associations between negative affect and SCC across groups. This, however, needs to be replicated by other studies. In addition, the samples were comprised of primarily white participants, although 17% were Hispanic, and high average education. Future studies would need to accommodate a more representative sample in older adults, particularly in respect to race and ethnicity. We also lacked information on MCI subtypes or biomarker data to confirm Alzheimer’s disease risk, which would be highly valuable correlates to examine in future studies. We also note that the focus on emotion in this sample should not be assumed to reflect the role of depression. This sample was in general experiencing a low level of distress and so the findings reported here may not generalize to the convergence of clinically significant depressive or anxious symptoms on SCC. Because of the small sample size, further potential confounds within MCI and NC could not be examined. There could be several possible confounders in this study, including but not limited to the differences in ownership of, or the device-types used to deliver EMA prompts, or if the study procedures were conducted remotely or in-person, and at which site. Lastly, this analysis focused on SCC and affect and did not examine the dynamic associations with objective cognition (e.g., with ecological momentary cognitive testing); further research examining EMA-based objective cognition in addition to EMA SCC and affect may lend further insights into these relationships. These temporal dynamics (between objective and subjective cognition) are expected to be nuanced as suggested by a study (Snitz et al., 2015), where subjective memory ratings and objective cognitive performance were found to mutually influence each other and were also cognitive-domain dependent.

Overall, this study adds to the small body of literature indicating feasibility and validity of assessing SCCs and related experiences in people with MCI. While SCCs are are clinically relevant given their association with dimnished quality of life(Hill et al., 2017), this study suggests that emotion independence of SCCs in MCI maybe a useful construct to help delineate prognostically meaningful SCCs. As such, there may be utility of EMA and time-series models such as network models in probing emotional and other contextual dynamics of SCC in MCI such as identifying patterns of SCC that are most predictive of objective cognitive decline.

Data availability

This clinical data is not publicly available due to privacy concerns, including HIPAA regulations. For access, qualified researchers may contact the corresponding author.