Recruitment

A sampling frame was developed from New York City Registered Voter Lists (RVL) obtained from the Board of Elections. Based on their age, individuals on the RVL were categorized into 10-year age bins and assigned to sampling blocks consisting of 450 potential participants in each bin. Individuals in a sampling block were sent letters of introduction, explaining the project’s goals and how the recipient was identified. Within two weeks of sending recruitment letters, a follow up telephone call was conducted to establish rapport, identify exclusions, and enroll participants in the study.

Baseline and follow-up surveys

Lab visits

At the initial lab visit for each wave, participants received training in the use of custom-configured smart phones for the measurement burst protocol and complete lab-based cognitive tasks.

Training session

In the training session, participants learned how to operate the study smartphone, practiced the smartphone survey questions and cognitive tests (described in Measures section) with the research assistant, and asked questions about the smartphones and protocol. Participants also received training in the use of the saliva collection kits (described in Measures section). The research assistant explained the schedule of data collection and showed participants the Help Line number on the back of the phone which they could call if they encountered problems with the smartphone or questions about the data collection.

Lab-based cognitive assessments

Lab-based cognitive assessments occurred at each wave and consisted of face-to-face and computer-based tests of working memory, episodic memory, and processing speed (see Measures section and Table 1).

EMA practice phase

In the first wave of the study, participants completed a 2-day EMA practice phase after their initial lab visit and training. This served as a period in which participants became familiar with the study protocol. During the EMA practice phase, data was collected through smartphones and salivary assessments, described below. Participants returned to the lab at the end of the EMA practice phase and their compliance with the study protocol was calculated. Those participants who completed 80 % or more of morning, beeped, and bedtime smartphone surveys were invited into the burst phase of the study. Participants who did not meet compliance criteria were thanked and provided payment commensurate with their participation.

Smartphone surveys

Each morning, participants completed a brief smartphone survey about their previous night’s sleep and their expectations regarding how pleasant and stressful the day ahead will be. The smartphone produced an audible alert (“beep”) 5 times during the day, signaling participants to complete a survey about their recent stressors, current activities and emotions, and recent thoughts. At the completion of each of these surveys, the smartphone launched three brief cognitive tasks for the participant to complete. Beeps were scheduled in order to sample the entire waking day, with quasi-random timing to ensure that participants do not anticipate the beeps. At the end of each day, participants completed a separate bedtime survey in which they reported on their physical symptoms. The smartphone data was stored on the phone and data were sent in an encrypted format to a secure server upon completion of each survey.

Saliva assessments

Saliva was self-collected four times daily during the first 5 days of this period. Upon waking each morning, participants placed a synthetic Salivette (Sarstedt Inc., Newton NC) into their mouth and were instructed to chew on it gently until saturated. They then placed the Salivette into a pre-labeled collection tube and stored it in their refrigerator or freezer until their return visit to the clinic. Saliva samples were collected again in an identical manner 15 min and 30 min post-waking, and at bedtime. The specific times of sample collection were recorded by the participant on a form that was returned to investigators at the end of the burst.

Burst phase

Participants who met compliance criteria in the EMA practice phase were invited to follow the same protocol (i.e., 1 morning, 5 beeped, 1 evening survey each day) for 14 consecutive days. For the first 5 days of the burst phase, participants completed saliva assessments as in the EMA practice phase (i.e., upon waking, 15 min post-waking, 30 min post-waking, bedtime). Research staff made follow-up calls after the first day of smartphone surveys and again at the end of week 1 to motivate compliance and address any questions that arise during the burst assessments.

Wrap-up visit

Participants returned to the lab for a brief wrap-up visit after the 14-day burst phase and to return the study smartphones and saliva samples. During this visit a qualified phlebotomist obtained 12-h fasting blood samples, which were brought to the Institute for Clinical and Translational Research (ICTR) at Albert Einstein College of Medicine. The ICTR distributed samples for immediate analysis to the appropriate laboratories; samples were processed and aliquots of whole blood, plasma, and serum were stored at −70 °C for future analysis. Physical assessments were also obtained at this visit.

Longitudinal assessment schedule

There are 3 waves of follow-up data collection scheduled for 9, 18, and 27 months post baseline. As the study is currently in progress and has rolling recruitment, participants are currently in different waves. To limit participant burden, participants do not complete the EMA practice phase in follow-up waves. The first two lab visits were condensed into a single visit in which participants completed a brief refresher training on smartphone use and saliva collection, and they completed lab-based cognitive assessments. Prior to this visit, they completed a survey packet that included questionnaires to assess physical health, mental health, stress, personality, and psychosocial factors (e.g., social support). Participants next completed the burst phase, and then the wrap-up visit.

Demographic

In addition to conventional demographic variables describing stable characteristics of individuals (e.g., age, race, ethnicity, gender) assessed at baseline, we conducted follow-up assessments of aspects of individuals’ life context that may vary over time (e.g., income, work, marital status) that may vary over time and affect exposure and response to stressors.

Stress

The term “stress” describes a multicomponent process that, as measured in ESCAPE, has four major components that are aligned with different time-scales. Specific measures used to assess each component are shown in Table 1. Chronic stress describes sources of stress (e.g., caregiving, dangerous neighborhood; [35, 36]), many of which do not have clear onset or endings. Major life events are infrequent, dramatic stressors, and include events such as divorce, widowhood. Subjective stress describes general appraisal of the state of feeling distressed or overwhelmed over the last month. Daily events are possibly more frequent and/or recurrent stressors that are less dramatic and include events such as arguments and deadlines. Table 1 describes the specific instruments employed in ESCAPE to measure these components of stress (i.e., chronic stress, major life events, subjective stress, and daily events).

Unconstructive Repetitive Thoughts (URT)

We also assessed several different ways in which URT may play a role in individuals’ responses to major life events, chronic stressors, and daily hassles. Table 1 describes the specific measures used to assess both dispositional and state (i.e., momentary) URT. First, dispositional URT describes tendencies for some individuals to engage in recurrent negative thoughts more than others, both in general [37, 38] as well as in response to specific sources of major life stress [39]. We expect that dispositional measures (i.e., WBSI, RRQ, IES-R) will operate through influencing the frequency and intensity of momentary experiences of URT, both in response to a stressor as well as in the absence of external triggers. State URT was assessed using 4 items developed for this study. These items ask individuals to use a continuous visual analogue scale (a slider) to rate the thoughts they had experienced in the 5 min prior to completing the smartphone survey in terms of their overall valence (unpleasant—pleasant), negative self-focus (“Were you thinking about personal problems or worries?”), and controllability (“Were you experiencing a train of thought that you couldn’t get out of your head?” and “Were you preoccupied with thoughts of something about to happen or that might happen in the future?”).

Cognition

We used two complementary approaches to assessing cognitive function. First, lab-based cognition was assessed during visits to the research clinic by a trained technician in a controlled testing environment. Second, ambulatory cognition was assessed repeatedly via smartphones in naturalistic settings as people go about their daily activities. We expect that ambulatory assessments will provide a more ecologically valid characterization of individuals’ cognitive functioning that will complement and extend traditional lab-based assessments.

Lab-based cognitive assessments included fluid intelligence and crystallized intelligence, assessed by the Raven’s Progressive Matrices [40], and the WRAT/Vocabulary [41], in addition to working memory, episodic memory, and processing speed. Working memory (WM) reflects a person’s ability to manipulate and maintain information in an active form. Three standardized tasks were used to measure WM in the lab: counting span, operation span [42, 43], and the backwards letter-span task [44]. Episodic memory was measured using the Auditory Verbal Learning Test (AVLT) [45] and a word-number paired associates learning task [46]. The “Card-memory Task” (CMT) required participants to learn the location of 6 playing cards that appear on a computer monitor, and recall those locations after completing a visual distractor task. Participants completed two choice reaction time tasks as measures of processing speed. In two separate tasks, participants completed two strings of characters (letters or numbers) as quickly as possible. Strings were 3, 6, or 9 characters in length to provide a variety of difficulty levels for analysis. The Symbol-symbol comparison task required participants to decide as quickly as possible whether a target pair of non-verbalizable symbols matches any of a set of five comparison pairs. Stimuli were randomly generated for each of cognitive tasks to avoid stimulus specific learning effects across repeated measures.

Ambulatory cognition was assessed on the smartphone by three tasks presented after each beeped survey in the following fixed order: processing speed, spatial working memory, and verbal working memory. Screen shots are provided in Fig. 2. The smartphone processing speed tasks required participants to compare three symbol pairs at the top of the screen and with two symbol pairs at the bottom of the screen and decide as quickly as possible which of the two pairs presented at the bottom of the screen is among the pairs at the top of the screen (see Fig. 2.1). The location memory task assessed spatial working memory and required participants to memorize the location of three red dots that appear on a 5 x 5 grid for 3 s (see Fig. 2.2a). After a visual distractor exercise of 8-s (see Fig. 2.2b), participants then recalled the locations of the 3 dots (see Fig. 2.2c). The verbal working memory was adapted from the columnized n-back paradigm [47], and included a 0-back and 2-back condition. Participants saw a series of 3 standard playing cards (see Fig. 2.3a) slide from one box on the right of the screen to the second box on left of the screen (see Fig. 2.3b). They completed a 0-back and 2-back variation of the standard n-back task. In the 0-back, cards face up and participants were asked to determine whether the cards in the two boxes match. In the 2-back variation, the previous two cards face down (see Fig. 2.3c). Each assessment consisted of 16 symbol comparison trials, 2 dot-memory trials, and 12 2-back trials.

Physiological measures of stress and endocrine function

Our primary marker of endocrine function in ESCAPE is cortisol, which is the primary glucocorticoid (GC) in humans and an important stress hormone. The majority of our cortisol variables were derived from salivary assessments, including cortisol awakening response (CAR) and area under the curve (AUC). The CAR is of particular importance to the ESCAPE study because it can be used to identify a blunted CAR, which is a signature of chronic stress [48]. Blood samples were also used to determine plasma cortisol and dehydroepiandrosterone sulfate (DHEAS). DHEAS is an adrenal hormone that both suppresses cortisol production and antagonizes its immunosuppressive (anti-inflammatory) effects [49, 50]. As a low DHEAS/cortisol ratio has been linked with a range of clinical outcomes, including cognitive impairment and senile dementia [51], the ESCAPE study will evaluate the effects of stress on this ratio and its role in explaining the impact of stress on cognition. Further, DHEAS declines with age and reductions in the DHEAS/cortisol ratio may explain why the deleterious consequences of HPA axis activation increase with age [52–56].

Assay details

At the end of each burst, cortisol levels were determined employing a competitive solid phase time-resolved fluorescence immunoassay with fluorometric end point detection (DELFIA). The intra-assay coefficient of variation (CV) is 4.0-6.7 % and the inter-assay CV is 7.1-9.0 %. Plasma, obtained at the end of each burst, is assayed to determine cortisol and DHEAS levels by enzyme-linked immunosorbent assay (ELISA) (IBL International Corp., Toronto ON). The minimum detection limit for cortisol is 2.46 ng/ml, and its inter- and intra-assay CVs are 2-5 %. The minimum detection limit for DHEAS is 4 pg/ml and its inter- and intra-assay CVs are 4-7 %. All of the above assays were performed in duplicate.

Physiological measures of inflammation

The majority of studies linking cognition and inflammation have not examined more than one or two inflammatory markers at a time. This study takes a multivariable approach, and simultaneously assesses ten pro- and anti-inflammatory biomarkers: interleukin (IL)-1β, IL-2, IL-4, IL-5, IL-6, IL-8, IL-10, tumor necrosis factor (TNF)-α, interferon (IFN)-γ and granulocyte macrophage colony-stimulating factor (GM-CSF; assay details described below). We also determined circulating plasma levels of C-reactive protein (CRP), which has widespread use clinically as a well-validated and sensitive marker of systemic inflammation [57]. Stimulated ex vivo inflammatory cytokine responses were also measured. For this purpose, a subset of whole blood (1 mL) was incubated with 1 μg/mL lipopolysaccharide (LPS). LPS is an antigen that stimulates immune cells and can thus be used to quantify the inflammatory response that cells in whole blood generate to an immune challenge. This provides a dynamic measure of inflammatory change that is unique from that of basal inflammation, and which may differentially relate to stress and cognition. Because not all cytokine levels rise significantly in response to ex vivo LPS stimulation, a subset of cytokines were assessed in LPS-stimulated blood: IL-1β, IL-6, IL-8, IL-10, TNF-α, and GM-CSF.

Importantly, the various inflammatory biomarkers determined in the ESCAPE study serve different functions and may therefore exhibit different response patterns to stress. This study thus allows for an inflammatory profile to be determined for each individual at a given time point, which is more reliable, valid, and informative than the rise or fall of a single biomarker. We expect the pattern in which these profiles change across years will be more predictive of cognitive aging than changes in any single measure of inflammation.

Assay details

Basal and LPS-stimulated cytokine levels were determined using multiplex magnetic bead arrays (Life Technologies, Grand Island NY). Multiplex bead arrays have advantages over standard ELISAs, including simultaneous measurement of interrelated cytokines, high sensitivity, and reductions in inter-plate variability, cost of analysis, and the amount of time and sample needed. The minimum detection limit for these assays is less than 0.5 pg/ml for each analyte and inter-assay CVs are 4.4 %-8.6 %. CRP is determined from blood plasma using ELISA (Cayman Chemical, Ann Arbor MI); the minimum detection limit is .047 ng/ml (.000047 mg/L) and intra-assay CV is 2 %-7 %. All assays were performed in duplicate.

Physical and mental health

ESCAPE includes both objective and subjective assessments of health. In order to examine longitudinal changes in indicators of overall physical health, participants completed blood draws and physical assessments (e.g., BP, waist-to-hip ratio). Participants also provided subjective reports of their physical functioning, medical history, and medication use.

Overall mental health was assessed via self-reports of anxiety, depression, and anger using standardized measures from the National Institutes of Health (NIH) Patient Reported Outcomes Measurement Information System (PROMIS). In the momentary smartphone surveys, participants reported on their current emotional states (anger, anxiety, depression), which will allow us to track predictors of dynamic shifts in affective states as well as obtain more ecologically valid assessments of mental health than single-shot questionnaires can provide.

Risk and protective factors

Individual difference characteristics (i.e., personality, psychosocial resources, life orientation) as well in as lifetime experiences (i.e., childhood adversity, positive life events) may moderate the relationships among stress, URT, and cognitive function. ESCAPE includes measures of both risk and protective factors, as described in Table 1.

Short-term tests

With regard to shorter term effects, we will evaluate the hypothesis that URT contributes to short-term (daily) stress effects on cognition via two sets of analyses. First, we will fit an autoregressive multilevel mixed model (MLM; [58]) to burst data to establish a lagged association between daily stress and ambulatory cognitive function, controlling for current stress, fatigue, mood, and other momentary covariates. Multilevel modeling is required because the burst data consists of repeated observations nested within persons. After establishing the lagged effect of daily stressors on working memory, we will then follow multi-level mediation procedures described in Bauer and colleagues [59] to test whether daily URT significantly mediates this relationship. Second, we will use structural MLM [60] to examine whether components of daily stress (e.g., exposure and reactivity) predict higher levels of systemic inflammation and altered inflammatory responses, and whether inflammation explains the association between daily stress and ambulatory memory function.

Long-term tests

To elucidate longer-term effects, we will use structural equation modeling (SEM; [61]). In these models, we will test the hypotheses that URT (at the psychological level), and evidence of HPA-axis dysregulation and inflammation (at the physiological level), mediate long-term effects of chronic stress on cognitive function. We will use baseline data to first determine whether individual differences in the tendency to engage in URT mediates the effects of chronic stress on current levels of ambulatory (assessed by smartphone) and lab-based measures of cognitive function. These analyses will be followed by SEM analyses that examine whether 1) those who have experienced greater chronic stress will have lower (blunted) values for cortisol area under the curve (AUC) and CAR, and elevated basal and inflammatory responses, 2) a blunted AUC and CAR predict greater inflammation, and 3) inflammation account for associations between chronic stress and cognition measures. Next, we will conduct longitudinal analyses using MLM to test the prediction that URT mediates the effects of chronic stress on the rates of cognitive change across measurement bursts. In this analysis, chronic stress will be treated as a time-varying predictor of intraindividual cognitive change across the four year follow-up, and URT will be examined as a time-varying mediator. Similarly, in analyses adding in HPA-axis and inflammatory markers we expect that AUC, CAR, and inflammatory changes will mediate the effects of chronic stress on intraindividual cognitive change. Sliwinski and colleagues provide a detailed description and example of this analytic approach in a study that examined long-term intraindividual changes in daily stress and negative affect [62].

Missing data/attrition

Some participants will not complete all planned measurements. Our use of general linear mixed models will insure unbiased results as long as missing data depend only upon observed variable (i.e., missing at random [MAR]). We will conduct sensitivity analysis using pattern mixture methods (e.g., [63]) to assess possible effects of informative attrition on study conclusions. Item-level missing data will be imputed using standardized guidelines specific to each questionnaire when available (e.g., PROMIS measures). In other cases, if the amount of missing data for a questionnaire is small (e.g. <5 %), we will consider calculating prorated scale scores to minimize casewise deletion.

Sample size calculations

Our power analysis considered sample size requirements for conducting both between-person and within-person statistical tests with an alpha level set to .05. For testing between-person effects we determine the necessary sample for detecting a small effect of stress on cognition (partial r = .16) using a general linear model. We selected this effect size based on our previous work [14]. This yielded a necessary sample size of approximately N = 310. Determining power for within-person tests conducted by multilevel analysis is complex, as values for many different parameters (e.g., intercept, slope, residual variance, covariances) must be specified. For simplicity, we assumed dependent variables were standardized (mean = 0, variance = 0) and the intraclass correlation ranged between .4 and .6 for the smartphone cognitive measures. Using these assumptions we conducted Monte Carlo (using SAS IML and MPLUS) simulations to determine the minimum detectable within-person effect assuming an initial sample of 300 and attrition rate of 20 % at burst 2, and 10 % at each burst thereafter. Results from our simulations yielded power of at least .80 for detecting small within-person effects on cognition that could account for at least 3 % of the variance.