Assessing depression recurrence, cognitive burden, and neurobiological homeostasis in late life: Design and rationale of the REMBRANDT study

Abstract

Background

Late-life depression is characterized by disability, cognitive impairment and decline, and a high risk of recurrence following remission. Aside from past psychiatric history, prognostic neurobiological and clinical factors influencing recurrence risk are unclear. Moreover, it is unclear if cognitive impairment predisposes to recurrence, or whether recurrent episodes may accelerate brain aging and cognitive decline. The purpose of the REMBRANDT study (Recurrence markers, cognitive burden, and neurobiological homeostasis in late-life depression) is to better elucidate these relationships and identify phenotypic, cognitive, environmental, and neurobiological factors contributing to and predictive of depression recurrence.

Methods

Across three sites, REMBRANDT will enroll 300 depressed elders who will receive antidepressant treatment. The goal is to enroll 210 remitted depressed participants and 75 participants with no mental health history into a two-year longitudinal phase focusing on depression recurrence. Participants are evaluated every 2 months with deeper assessments occurring every 8 months, including structural and functional neuroimaging, environmental stress assessments, deep symptom phenotyping, and two weeks of ‘burst’ ecological momentary assessments to elucidate variability in symptoms and cognitive performance. A broad neuropsychological test battery is completed at the beginning and end of the longitudinal study.

Significance

REMBRANDT will improve our understanding of how alterations in neural circuits and cognition that persist during remission contribute to depression recurrence vulnerability. It will also elucidate how these processes may contribute to cognitive impairment and decline. This project will obtain deep phenotypic data that will help identify vulnerability and resilience factors that can help stratify individual clinical risk.

Highlights

• •Neurobiological contributors to depression recurrence in later life following remission are poorly understood.
• •Factors contributing to depression recurrence may also contribute to cognitive decline in older adults.
• •Combined laboratory-based and ecological-based monitoring may elucidate causes of depression recurrence.

1. Introduction

Late-life depression (LLD) is associated with negative outcomes including dementia, disability, and increased mortality [1.], [2], [3]. These negative longer-term outcomes are partly related to the recurrent nature of the disorder and repeated depressive episodes. Even after successful treatment, emergence of new depressive episodes is common [4], including both short-term relapse and longer-term recurrence (for simplicity, we refer to both as recurrence). Even with ongoing treatment, high recurrence rates in LLD are observed in both longitudinal cohort studies and maintenance clinical trials, ranging from 35% to 43% over two years, and as high as 57% over four years [2], [5], [6].

We have only limited data on neurobiological predictors of depression recurrence and most available data focus on clinical predictors [2]. Recurrence is most strongly associated with greater numbers of prior depressive episodes and greater severity of residual depressive symptoms [7], [8], [9]. Recurrence can be triggered by stressful life events [10], [11], [12] while maintenance antidepressant treatment reduces recurrence risk [6], [13]. Clinical features such as greater medical morbidity, disability, sleep disturbances and comorbid anxiety may further increase risk [5], [6], [12], [14]. In contrast, greater social support may reduce recurrence risk [3], [5], [15].

LLD is further associated with cognitive impairment and risk for further decline. Depressed elders exhibit impairments in executive functions and in other cognitive domains including episodic memory, visuospatial skills, and information processing speed [16], [17], [18], [19]. However, not all individuals with LLD exhibit impaired cognitive performance [20]. While performance on some cognitive measures improves with successful treatment, impairments typically persist [21], [22], [23], and older depressed adults have an increased dementia risk [24]. Poorer cognitive performance during remission does not clearly influence recurrence risk [5], [25], [26].

2. Scientific model and study objectives

It is unclear what neurobiological factors are prospectively associated with recurrence risk, when these factors trigger recurrence, and how they may be related to the high rates of cognitive impairment observed in LLD. The first study goal tests a model of neural network homeostasis [2]. We posit that depressive episodes develop from disrupted homeostasis in key neural networks involved in affect regulation and cognitive function. Various neurobiological, medical, and environmental factors, including exposure to stressful events, may predispose networks towards disequilibrium [3]. In context of stress or other vulnerability factors, networks first become unstable, a change we hypothesize that is associated with the emergence of subsyndromal clinical symptoms. This can progress to system disequilibrium and emergence of a depressive episode [2], [3]. Successful treatment changes connectivity across networks [27], hypothetically restoring network equilibrium, and results in clinical improvement. We hypothesize that individuals exhibiting greater stress reactivity and residual functional network alterations in remission remain at high risk of recurrence. Specifically, we hypothesize that recurrence will be associated with greater stress reactivity, assessed experimentally using a fMRI task, and environmentally, assessed using ecological momentary assessment (EMA). Similarly, we hypothesize that recurrence risk is increased by residual alterations in the Default Mode Network (DMN; increased intra-network connectivity and reduced disengagement during tasks), Executive Control Network (ECN; decreased inter- and intra-network connectivity), and Salience Network (SN; increased intra-network connectivity) [2], [3].

The second study goal examines how cognitive performance may both predict recurrence risk and also be influenced by neurobiological processes contributing to depression recurrence. Impairment or variability in cognitive performance may serve as markers of functional network vulnerability and signal increased recurrence risk. In parallel, stressful events and resultant physiological stress responses including hypercortisolemia and inflammation are associated with changes in cognitive performance [28], [29], [30], [31], [32]. Markers of high allostatic load are associated with neurodegeneration, tau protein phosphorylation, and other neuropathological markers [33]. These processes may accelerate aging processes in LLD, including brain aging [34], [35] and cognitive decline. We specifically hypothesize that greater stress reactivity, assessed by fMRI and EMA, along with decreasing intra-network ECN functional connectivity, will be associated with two-year cognitive decline. Additionally, we hypothesize that greater variability in ecological smartphone-based assessment of cognitive performance will be predictive of both depression recurrence risk and greater two-year cognitive decline.

3. Study design and methods

REMBRANDT (Recurrence markers, cognitive burden and neurobiological homeostasis in depression) is a three site, two-year observational longitudinal study of remitted LLD. Assessments include repeat clinical phenotyping, neuroimaging, neuropsychological testing, and ecological assessments.

3.1. Participants

3.1.1. Enrollment goals and eligibility

We plan to recruit 300 individuals with LLD to yield 210 remitted LLD participants to enter the two-year longitudinal phase. LLD participants enter through an “Initial Treatment Phase” if currently depressed or are recruited directly into the longitudinal phase if they recently remitted to clinical treatment. We additionally will enroll 75 comparison participants without any lifetime mental health history. Estimating an approximate 20% attrition rate, this will result in a targeted final sample of 170 depressed participants and 60 comparison participants.

Common inclusion criteria across study cohorts include: 1) Age 60 years or older; 2) a Montreal Cognitive Assessment (MoCA) score of 24 or greater, or a MoCA-BLIND score of 18 or higher during telemedicine evaluations, scores sensitive to dementia that ensure participants can complete protocol tasks; 3) Fluent in English. Common exclusion criteria include: 1) History of a substance use disorder (other than nicotine) in the last year; 2) Other DSM5 mental disorders, except for co-occurring anxiety symptoms or anxiety disorders present during the depressive episode (specifically generalized anxiety disorder or panic disorder symptoms); 3) History of a developmental disorder or history of IQ < 70; 4) Acute suicidal ideation within 3 months of study entry; 5) Acute grief (less than 1 month); 6) Current or past psychosis, including history of Major Depressive Disorder with psychotic features; 7) Primary neurological disorders, including major neurocognitive disorders related to dementia, Parkinson’s disease, stroke, epilepsy, etc. Peripheral neurological illnesses such as essential tremor or regular headaches are allowable. However, as cognitive difficulties are common in LLD [19], [36], a diagnosis of Mild Cognitive Impairment (MCI) is not exclusionary; 8) Presence of unstable medical illness requiring urgent treatment; 9) Magnetic resonance imaging (MRI) contraindication; 10) Electroconvulsive therapy in last 6 months; 11) Current use of transcranial magnetic stimulation, ketamine, or esketamine.

Additional criteria are specific to participant groups. Currently depressed participants require a DSM5 diagnosis of major depressive disorder, recurrent, with a Montgomery-Asberg Depression Rating Scale (MADRS) [37] score of 15 or greater. Remitted depressed participants require a DSM5 diagnosis of major depressive disorder, recurrent, in partial or full remission, and MADRS score of 10 or less. Remission from a documented depressive episode had to occur no more than 4 months prior. Comparison participants must exhibit a MADRS score of 8 or less, cannot endorse any lifetime psychiatric history, and have no history of psychotropic medication use. Past brief therapy for specific challenges or losses (such as marital therapy or grief therapy) is allowable.

3.1.2. Recruitment, informed consent, and enrollment

The study is conducted at: The University of Illinois-Chicago (UIC; Chicago, Illinois); The University of Pittsburgh Medical Center (UPMC: Pittsburgh, Pennsylvania); and Vanderbilt University Medical Center (VUMC; Nashville, Tennessee). Participants are recruited using research registries, community outreach, clinical referrals, and advertisements, including newspapers, mass transit services, public service announcements, and internet-based advertising. On initial contact, study staff use a phone script to provide elements of consent and study details. Potentially eligible participants are scheduled for a screening visit conducted in-person or by Zoom videoconference.

After providing written informed consent, participants complete psychiatric diagnostic testing using the MINI International Neuropsychiatric Interview (MINI) [38] for DSM5 and cognitive screening using the MoCA, or the MoCA – BLIND for telemedicine visits. Vital signs (height, weight, blood pressure, pulse) are obtained for in-person participants. A study clinician (a psychiatrist or mental health nurse practitioner supervised by a study psychiatrist) conducts a clinical interview including review of medications and past medical and psychiatric history. They assess symptom severity using the MADRS and Hamilton Anxiety Rating Scale (HARS) [39].

Participants were compensated for their time and study procedures. The compensation schedule is detailed in Supplemental Methods: Participant Reimbursement (Supplemental Tables 1–3).

3.1.3. Assessment and management of suicidal ideation

Acute suicidal ideation in the previous three months is exclusionary. This is assessed at the initial evaluation through the MINI’s suicidality module [38] that assesses suicidal thoughts and behaviors occurring both over the last month and over the lifetime, MADRS scoring of item 10, and clinician interview including medical chart review, assessment of current symptoms, and review of past psychiatric history. Operationally, acute suicidal ideation is defined as an intent or plan to harm themselves, regular or frequent thoughts of suicide, recent suicidal behavior, or help seeking to avoid suicidal behavior. It does not include passive thoughts of death or rare thoughts of suicide with no intent or plan.

If a participant is thought to be suicidal, a physician investigator will immediately be contacted to assess severity of the suicidal ideation, imminent risks, and to develop an immediate plan of care. Participants who meet this threshold at initial evaluation are withdrawn from the study. Depending on the individual’s severity and current treatment, the follow-up plan may include increased frequency of outpatient clinic visits, further evaluation at the psychiatric emergency room, or immediate psychiatric hospitalization. If suicidal symptoms improve with treatment, participants could later be enrolled in the study after the three-month eligibility window.

Participants may develop suicidal ideation over the course of the study. Trained raters will probe for passive death wishes, and suicidal ideation, intent or plan when they administer the MADRS. If a participant endorses suicidal ideation, intent, or plan, the rater will immediately contact a study clinician to assist with evaluation and management. The site principal investigators or another study physician will always be available by cellular phone to discuss and assume management of suicidal patients. In case of extreme emergency, raters are instructed to call their hospital security team or 911 for immediate help. The study clinician will work with patients expressing suicidal thoughts to assure safety and treat their symptoms, including medication changes, initiation of psychotherapy, neuromodulation treatment, or hospitalization. Development of suicidality does not immediately lead to study withdrawal unless the study clinician concludes that a return to clinical care is in the participant’s best interest.

3.2. Initial treatment phase (ITP) for currently depressed enrolled participants

The goal of the ITP is not scientific inquiry, but to provide a structured pharmacological treatment approach to facilitate remission in depressed participants. Participants can be treated for up to 20 weeks, although this can be extended for an additional 4–8 weeks for participants who improve but do not remit by 20 weeks. Individuals who do not remit are referred for clinical care. Remission is defined as MADRS score of 10 or less at 2 time points: the end of the ITP, and at the longitudinal study baseline visit, which must occur within 4 months but no sooner than 1 month following remission.

The ITP provides algorithm-guided treatment informed by STAR*D [40] and the Duke Neurocognitive Outcomes of Depression in the Elderly (NCODE) studies [41], with medication selection informed by participants’ history and preferences. STAR*D reported a cumulative remission rate of 56% after Level 2 [40], while NCODE reported a 65% remission rate [41]. With aripiprazole augmentation, an additional 41% of LLD who did not respond to a SNRI will remit [42]. Based on these data, we estimated a 70% remission rate.

During the ITP, participants are assessed at least every 4 weeks by clinical interview, MADRS, Clinical Global Impression (CGI) scale and assessment for adverse events. The ITP algorithm (Fig. 1) begins with a selective serotonin reuptake inhibitor (SSRI). Treatment decisions are informed by the change in MADRS assessed every 4 weeks. The algorithm specifies: A) a < 25% MADRS change will lead to a switch to a serotonin-norepinephrine reuptake inhibitor (SNRI) or mirtazapine; B) a 25–50% change will lead to first-level augmentation with bupropion or mirtazapine; or C) a > 50% improvement will lead to continued treatment until plateau or remission. If participants do not improve with either a swap to a SNRI or first-level augmentation, aripiprazole can be used as a final augmentation option.

Fig. 1.

Initial Treatment Phase Algorithm. The initial treatment phase provides a structured algorithm including decision points occurring every 4 weeks. This provides for continuation treatment for robust improvement (MADRS change of 50% or greater), augmentation (for MADRS change between 25% and 50%) or switch strategies (for MADRS change less than 25%). The algorithm is meant to serve as general guidance and can be modified as needed according to participant history or preference. This phase lasts up to 20 weeks, although can be extended for individuals showing progression at that time frame.

As the goal of the ITP is to get participants well, study clinicians can modify the medication algorithm for a given participant as needed to accommodate past treatment history. If a participant has not tolerated or responded to available SSRIs, they can start a SNRI, bupropion or mirtazapine at Step 1. Dose increases can be substituted for medication change or augmentation decisions if the medication is tolerated and not currently at the maximum dose. Participant preference similarly informs treatment decisions. Psychotherapy is allowable but is not integrated into the treatment algorithm. Due to challenges with access, psychotherapy is not routinely offered to participants.

3.3. REMBRANDT longitudinal phase design

Remitted depressed participants and comparison participants enter the two-year longitudinal phase. This phase involves scheduled contacts every 2 months (Table 1), with additional visits occurring as clinically indicated. Assessment “Biomarker” visits include the baseline visit and occur every 8 months (Months 0, 8, 16, 24), involving clinical and questionnaire assessments, MRI, initiation of the ‘burst’ ecological assessments, and neuropsychological testing. Clinic or telemedicine visits occur between Assessment visits at 4-month intervals (Months 4, 12, 20). Telephone visits occur at the 2-month interval between other visits (Months 2, 6, 10, 14, 18, 22) and can be converted to in-person visits as needed.

Table 1.

Rembrandt longitudinal phase clinical assessments and procedures.

Assessment or Procedure	Screening Visit	Assessment Visits	Bimonthly Clinic Visits
Initial Eligibility •Consent process •MINI for DSM5 [38] •Montreal Cognitive Assessment (MoCA) •MRI Screening	X
Psychiatric and Depression History	X
Medical History •Cumulative Illness Rating Scale-Geriatric (CIRS-G) [86] •Pain severity (VAS, 0–10) •World Health Organization Disability Assessment Scale 2.0 (WHODAS) [87]	X	X
Clinical Monitoring •Health and Medication Review •Montgomery Asberg Depression Rating Scale (MADRS) [37] •Hamilton Anxiety Rating Scale (HARS) [39] •Frequency, Intensity, Burden of Side Effects Rating Scale (FIBSER) [88] •Medication Reconciliation Tracking Form (MRTF) [89] •Antidepressant Treatment History Form (ATHF; current treatment intensity) [90], [91] •Vital signs (in-person visits)	X	X	X
Personality: NEO Five Factor Inventory [45]	Baseline visit only
Symptom Phenotyping •Apathy Evaluation Scale (AES) [92] •Fatigue Severity Scale (FSS) [93] •Ruminative Response Scale (RRS) [69]		X	X(Every 4 months)
•Connor Davidson Resilience Scale-10 [94] •Penn State Worry Questionnaire (PSWQ) [95]		X
Environmental (Social / Stress) Assessments •Major Experiences of Discrimination Scale [44]	Baseline visit only
•Duke Social Support Index [43] •Perceived Social Support Conflict (MIDUS II) [96] •PROMIS Social isolation scale (4-item) •Holmes-Rahe Life Stress Inventory [38] •Perceived Stress Scale (PSS) [97]		X
Ecological “Burst” Assessments •Actigraphy (2 weeks) •EMA of depressive symptoms (1 week) •EMA of cognitive performance (1 week)		X
Magnetic Resonance Imaging (MRI)		X
Neuropsychological Testing	Baseline and Month 24 visits only

Once participants enter the longitudinal phase, visits occur every 2 months, with more involved Assessment “Biomarker” Visits occur at baseline, and every 8 months afterwards. Ad hoc visits can occur as clinically indicated.

Abbreviations: EMA: Ecological Momentary Assessment; MIDUS: Midlife in the United States; MINI: Mini-International Neuropsychiatric Inventory; VAS: Visual-Analogue Scale.

During the longitudinal phase, depressed participants generally remain on the regimen that led to remission. Treatment during this phase can be provided by either study clinicians or a clinical provider outside the study. Medication changes are neither prohibited nor limited and occur as needed for tolerability, worsening symptoms, or other indications. There are no restrictions to the types of antidepressant medications that remitted depressed participants can receive. Engagement in psychotherapy is allowed as is initiation of neuromodulation treatment (ECT or TMS) for depression recurrence. Participants can continue in the study if they elect to stop taking their antidepressant medications.

Participants continue in the study even after recurrence, defined as a MADRS of 15 or greater plus meeting DSM-5 criteria for a new major depressive episode, including the requirement for a 2-week duration of symptoms. Remission is defined as subsequent MADRS of 10 or less and return to normal function. Individuals who experience a recurrent episode and are identified within the first four weeks from start of symptoms are requested to attend an additional ad hoc Assessment Biomarker Visit, involving neuroimaging and self-report symptom phenotyping questionnaires.

3.3.1. Clinical and questionnaire assessments

Participants are broadly characterized throughout the study (Table 1). Each bimonthly clinic visit assesses depression severity with the MADRS, anxiety severity with the HARS, and frequency, intensity and burden of adverse events using the Frequency, Intensity, Burden of Side Effects Rating (FIBSER) scale. Adverse event tracking similarly includes tracking falls, other significant injury, acute medical events, hospitalizations, or other serious adverse events.

Treatment changes are tracked using the Medication Reconciliation Tracking Form (MRTF) and a modified Antidepressant Treatment History Form (ATHF). Medications used for anxiety (such as low-dose benzodiazepines), sleep (such as zolpidem or sedating antidepressants such as trazodone), or wakefulness-promoting agents (such as modafinil, used for obstructive sleep apnea) are similarly tracked on the MRTF. These can be used in exploratory analyses.

The Assessment Visits occurring every 8 months quantify medical comorbidity with the Cumulative Illness Rating Scale – Geriatric (CIRS-G), pain severity using a visual-analog scale, current substance use via clinical interview, and self-reported disability with the World Health Organization Disability Assessment Schedule (WHODAS) 2.0. Environmental assessments include the Duke Social Support Index [43], the Perceived Social Support Conflict Scale, and the social isolation scale. Stress exposure is measured with the Holmes-Rahe Life Stress Inventory, focused on the last 8 months, and subjective stress with the Perceived Stress Scale (PSS). History of discrimination is assessed at baseline using the Major Experiences of Discrimination Scale [44].

Deeper phenotyping includes baseline personality assessment using the Five Factor Inventory [45]. Questionnaires assessing apathy, fatigue, and rumination are administered every 4 months, while questionnaires assessing resilience and worry are completed every 8 months (Table 1).

3.4. Magnetic resonance imaging

Participants complete 3Tesla MRI that acquire data for automated tissue identification, assessment of white matter hyperintensities, diffusion weighted imaging, and both resting and task-based BOLD fMRI. Scanning is conducted on a Siemens Prisma scanner (UPMC), a Philips Elition (VUMC), and GE Discovery MR750 System (UIC), using 32-channel head coils at all sites. Given variability in scanner types across study sites, to assure MRI data harmonization we relied on the Adolescent Brain Cognitive Development (ABCD) Study MRI Protocol [46]. Imaging protocols were thus based on published ABCD Protocol parameters for each scanner type (Supplemental Tables 4 and 5), modifying sequence acquisition times as needed to achieve study goals. We obtain two shorter resting state fMRI acquisitions in contrast to a single longer acquisition as concatenating multiple shorter scans leads to greater overall reliability [47]. Quality assurance processes include assessment of inter-system variability via monthly acquisition of two MR phantoms across all sites, including the ADNI (Alzheimer’s Disease Neuroimaging Initiative) structural phantom and the fBIRN functional phantom.

Standard procedures for MRI pre-screening and participant monitoring are conducted per each institution’s standard practices. Participants are oriented to the button box or glove to be used in scanner and positioned so their head and neck were relaxed, but without rotation. Once positioned, head, back, and leg support is used as needed for motion reduction, comfort, and stabilization. The participant is centered so the head coil’s centering crosshairs align on the participant’s nasion. During the two 5-minute resting state fMRI scans, participants are asked to keep their eyes open and fixate on a crosshair.

3.4.1. Functional MRI task

Participants complete a performance-titrated version of the Multi-Source Interference Task (MSIT) [48]. The MSIT evokes a stress response through cognitive conflict, response inhibition, unpredictable stimuli, time pressure, and negative feedback [48]. During the MSIT, participants identify the number that differs between two other numbers by pressing 1 of 3 buttons corresponding to specific numbers in the display (Fig. 2). For congruent (control) trials, the target number in the display appears in a location compatible with its position on the button box or glove. For incongruent trials, the target number appears in a position that is incompatible with its position.

Fig. 2.

Multi-source interference task design. Figure displays the stimulus of the multi-source interference task (MSIT) used during functional MRI. Represented are the stimulus (top) and response (bottom) for control and stress conditions. In both conditions, the participant is instructed to push the button corresponding to the numerical value of the digit that is different from the other two, and suppress the urge to indicate the digit’s location.

During the incongruent condition, each participant’s accuracy at target number identification is titrated to approximately 50% by varying the inter-trial intervals (ITI). Thus, more accurate performance within an incongruent condition block prompts shorter ITIs, shorter response time windows, and more trials per block. Failure to respond within the designated time provides participants with negative feedback. This “stressor” incongruent condition elicits subjective distress, significant cardiovascular stress reactivity [48], [49], [50], and evokes neural responses in stress-related brain regions [51]. This titrated version does not produce a ceiling effect and enables the examination of individual differences in neural stress responses [48], [51]. To capture individual in-scanner experience, participants complete the Amsterdam Resting State Questionnaire (ARSQ) [52] after the functional MRI sequences.

3.5. Neuropsychological assessments

Neuropsychological testing is conducted by site-based research assistants who complete virtual training from the UPMC neuropsychology team (supervised by MAB). Training includes watching video demonstrations, direct instruction from trainers and repeat administration of the battery to a senior experienced psychometrist via videoconference, and administrations to study participants monitored on-site by a certified team member.

Participant cognitive performance is assessed at baseline and 24 months using a comprehensive test battery. Although initially planned to be conducted in-person, with the advent of the COVID-19 pandemic, cognitive testing was shifted to Zoom video conference administration. This required slight modification of some instructions given to study participants to accommodate the virtual testing environment. To maintain uniformity, and based on positive early experiences, we elected to continue this remote neuropsychological testing throughout the study. Testing can be conducted if the participant is at home, assuming the home environment is conducive to testing, or at the research facility, with the participant and examiner being in different rooms. This involves using two cameras for the study participant, one focused on their face and one focused on their hands resting on the test table to monitor their hand movements during tasks that involve writing or drawing.

The battery includes the Wide Range Achievement Test 5 (WRAT5) Word Reading subtest [53], Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) [54], Wechsler Adult Intelligence Scale, 4th edition (WAIS-4) Letter-Number Sequencing [55], Trail Making Test (TMT), Controlled Oral Word Association Test (FAS and Animals) [56], and the Stroop Color and Word Test (SCWT) [57]. It also includes the Everyday Cognition Scale (E-Cog; short form) [58], which is both a self- and informant-report measure of perceived cognitive changes at present compared to 10 years ago, and the Performance Assessment of Self-Care Skills (PASS short form) [59] to assess current functional competence on instrumental activities of daily living. The RBANS has six alternate forms; the two with the highest concordance (Forms A and D) are administered in a counterbalanced manner over time to minimize practice effects. The content of the remaining tests should not be especially amenable to practice effect. Ultimately, for outcome analyses, these tests are used to quantify cognitive status using a composite measure based on the Preclinical Alzheimer Cognitive Composite (PACC) [60]. Modeled on the PACC, we examine a composite score including the Delayed Memory Index from the RBANS, to represent executive functions the mean standardized score of the SCWT and the TMT Part B, the RBANS Total Index Score (minus Delayed Memory Index) and the PASS summary score. Beyond the composite measure, individual test scores can additionally be examined in exploratory analyses to identify latent variables or individual factors across tests that can be used as predictors.

As the study allowed participants with MCI, we plan to use the broad-based battery to conduct cognitive diagnostic adjudication. All relevant baseline and follow-up data (including clinical data and neuropsychological test results, E-Cog and functional status) will be reviewed at Diagnostic Adjudication Conferences by a neuropsychologist, a site geriatric psychiatrist, and the cognitive assessor familiar with the participant. We will apply University of Pittsburgh ADRC procedures and use the NIA-AA criteria to diagnose dementia [61]. Participants with likely dementia will be referred to the site’s ADRC or memory clinic for further evaluation. We will use NIA-AA criteria [62] and the National Alzheimer’s Coordinating Center / Revised Petersen Comprehensive Criteria [63] to diagnose MCI. Assessment of change in cognitive performance over time includes consideration of interval medical events and changes in overall health.

3.6. Ecological assessments

After completing an exploratory smartphone familiarity survey (Supplemental Table 6) [64], smartphone-administered ecological momentary assessments (EMA) occur over two weeks after each 8-month assessment biomarker visit (Table 1). These non-concurrent EMAs include one week assessing depressive symptoms and one week assessing cognitive performance. EMA notifications occur four times daily at quasi-random intervals during waking hours with no less than two hours between each session. Actigraphy data are collected concurrently over the entire two-week period. Ecological data management varies across the modalities, however ultimately all data are uploaded into a secure server at UPMC. From there, study-specific programming scripts extract key data elements from the raw data files for integration into the study REDCap database.

While older adults with depression and mild cognitive difficulties can reliably complete smartphone-administered EMA batteries [65], [66], we anticipated a need for participant support to avoid technical issues, frustration, and to facilitate data capture. We simplify procedures by providing a standardized study Android-based smartphone supporting all EMA and connecting to the study Fitbit. At participant request, to reduce the number of devices they carry, we can link the LifeData app and Fitbit to their personal smartphone. All systems are setup and participants logged into the study account while they attend their in-clinic visit. They are reminded that while it is important for them to respond to prompts as quickly and often as possible, it is acceptable to miss a session and they should not respond to a prompt in an unsafe situation, such as driving. Participants are provided an illustrated guide detailing common technical problems and solutions, such as how to connect the study smartphone to their home Wi-Fi system. Study staff are additionally available for further technical assistance. Ecological assessments are monitored remotely by study staff. Participants who are not engaging in assessments and data are not being generated are contacted to address motivational or technical hurdles.

3.6.1. Ecological momentary cognitive assessments

Unsupervised EMA of cognitive performance uses the Ambulatory Research in Cognition (ARC) App [66] on a standard study Android smartphone. The ARC app includes three validated, reliable cognitive tasks [66]: Symbols, Prices, and Grids. Over a 7-day administration period, tests have a reliability over 0.85. All tests are scored so higher scores indicate worse performance. Data from the ARC App is uploaded via smartphone WiFi directly to the secure sever at UPMC.

Symbols is a 12-trial processing speed task. Participants are shown abstract shapes and asked to rapidly indicate which of two pairs match one of three target pairs. The outcome measure is the median response time of correct trials.

Prices is an associative memory task. During learning, participants see ten common shopping items paired with a price for 3 s. During recognition, participants are given two prices to choose from for each item, with each price separated by at least $3.00. To protect against interference effects, items are never repeated within the same day, and item-price pairs are never repeated over the study week. The outcome measure is the total error score.

Grids is a spatial working memory task. Participants are shown three common objects on a 5 × 5 grid. After a distraction task, a blank 5 × 5 grid is presented, and participants tap the squares where the objects were located. The outcome variable is a Euclidian distance estimate agnostic to the item.

3.6.2. Ecological momentary symptom assessment

Symptom self-report is obtained using the RealLife app via LifeData (www.lifedatacorp.com). Data are collected through the native RealLife iOS or Android apps are initially transferred to the LifeData servers, then subsequently pulled to UPMC study servers. Participants have 30 min to complete each assessment trial, with each trial lasting for 2–3 min.

EMA question content (Table 2) queries the participant’s social setting and subjective stress level. Mood is assessed with the 4-item Patient Reported Outcome Measurement Information System (PROMIS) Short Form 1.0 Emotional Distress-Depression 4a [67], and fatigue using the first two items of the PROMIS Short Form 1.0 Fatigue 4a [68], with questions phrased as how participants feel “right now.” We measure rumination using the 4-item brooding scale from Ruminative Response Scale [69]. Additional questions examine broader feelings of negative affect and, to examine potential resilience factors, positive affect [70], [71], [72], [73].

Table 2.

Ecological momentary assessment symptom questions.

Instrument	Question
Context	Are you alone or with others?
PROMIS – Emotional Distress - Depression 4a	I feel worthless
	I feel helpless
	I feel depressed
	I feel hopeless
PROMIS – Fatigue 4a	I feel fatigued.
	I have trouble starting things because I am tired.
Stress	Rate the severity of your stress (0–10)
	What type of stress are you experiencing?
Rumination (Brooding)	What am I doing to deserve this?
	Why do I always react this way?
	Are you thinking about a recent situation, wishing it had gone better?
	Why do I have problems other people don’t have?
	Why can’t I handle things better?
Positive Affect / Negative Affect	Right now, are you feeling…. Down?
	Guilty?
	Lonely?
	Anxious?
	Happy?
	Cheerful?
	Satisfied?

All questions were modified from the original instruments to focus on how participants were feeling at that moment rather than over some past period. PROMIS instruments used short-form 1.0 versions. Other than the context stress items, all questions were scored with a Likert scale, ranging from 1 (not at all) to 5 (very much). Context and stress questions were created specifically for this study. Context was simply a dichotomous question (alone/not alone). Stress severity was rated 0–10. Type of stress was operationalized to include work, family, financial, health, social (non-family), and other, with participants being able to choose all that apply.

3.6.3. Actigraphy

Actigraphy data are obtained using Fitbit Charge3 devices over two weeks concurrently with EMA. Data collected by the wearable device are saved by the native Fitbit smartphone apps and transferred to the Fitbit servers, then subsequently pulled to UPMC study servers. We selected the commercial-grade Fitbit over research actigraphs as it obtains the required data at a more effective cost and could be translated more directly into clinical practice. Fitbit measures of activity are valid and reliable in older adults [74] and over one week consistent with research actigraphs [75]. Actigraphy data can usefully identify activity patterns that may serve as distinct phenotypes [76]. The device’s ‘coaching’ functions are disabled, although participants can reference device data to track their activity. Participants are asked to wear them for 24 h daily over the two-week period, aside from showering and device charging. They are provided guidance on how to size the Fitbit band, instruction on its charging schedule, and reminders to open the Fitbit app regularly to facilitate syncing with the device.

3.7. Analytic strategy and power analysis

3.7.1. General approach

Study data are collected and managed using REDCap electronic data capture tools hosted at VUMC [77], [78]. REDCap (Research Electronic Data Capture) is a secure, web-based software platform that supports data capture for research studies, providing 1) an intuitive interface for validated data capture; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for data integration and interoperability with external sources.

Prior to hypothesis testing, we will evaluate distributional assumptions and check for outliers. For clinical, imaging, and ecological data, we will apply advances in analyses for multivariate functional data that use the concept of data depth to allow for the analysis of high-dimensional data [79]. Primary hypotheses will be tested using ANCOVA and regression models, with effects selected depending on the hypothesis tested, including random subject effects and auto-correlated errors when outcomes are collected at each time point. All tests will be two-sided at level 0.05, controlling the false discovery rate via a resampling method.

3.7.2. Missing data

The proposed mixed effects models are appropriate and unbiased if missing data exist and are missing at random. Sensitivity analyses will be conducted by fitting two models: one when missing data are randomly imputed from the lower quantile of observed values and one when missing data are randomly imputed with values from the upper quantile of observed values. The distribution of values from the multiple imputations will be explored and compared to estimates from the fitted model assuming data are missing at random.

3.7.3. Aim 1 strategy: neurobiology of recurrence

Aim 1 / hypothesis 1 tests the association between neural reactivity during the fMRI stress tasks in remitted LLD relative to comparison participants. We will fit mixed-effects repeated measures ANCOVA models that include a fixed effect for participant diagnosis, a fixed effect for time, interactions between participant type and time, fixed effects controlling for age and sex, random subject effects, and autocorrelated errors for observations within subject. Our primary hypothesis of mean differences between remitted depressed and comparison participants at any point in time will be examined through the simultaneous test of the effect of participant type and the effect for the interaction between participant type and time. If the null hypothesis of no effect is rejected, we will investigate the interaction term. If the interaction term is found to not be significant, we will report the estimated effect of participant type. If the interaction term is found to be significant, we will compute appropriate contrasts to quantify estimated effect of participant type at each point in time, and changes in effects over time. Similar models and tests will examine LLD participants to test the association between neural reactivity and risk of recurrence. We will also examine EMA symptom instability. These involve analogous ANCOVA models fit to measures of affective instability to test associations between affective instability and both LLD and risk of recurrence. Affect instability in each weekly EMA session will be computed through the mean-adjusted absolute successive difference (MAASD) for each variable (mood, rumination, and negative affect). Possible effects of informative missing data due to technological difficulties will be evaluated through a sensitivity analysis.

Aim 1 / Hypothesis 2 examines resting state fMRI data. We will fit ANCOVA models to: a) the global measure of between-network connectivity obtained through the correlation of each network’s associated principal time series; b) global measure of intra-network connectivity defined as the mean connectivity across all voxels that belong to the same network, and c) global measure of network instability defined as the tSNR within canonical resting state networks. These will be fit separately using a main effect for patient-type to all participants and using a main effect for recurrence for only LLD participants to test associations between intra- and inter-network connectivity as well as network instability with LLD and risk of recurrence, respectively.

3.7.4. Aim 2 strategy: cognitive decline and LLD recurrence

Aim 2 / Hypothesis 1 evaluates associations between 2-year cognitive decline and stress reactivity. Primary measures are the composite cognitive score (PACC) and measure of stress reactivity (task-related ROI activity and MAASD for mood, rumination and negative affect). We will fit a linear regression model regressing PACC decline simultaneously on stress reactivity at each of the observation times. The omnibus F-test will be used to test associations with stress reactivity at any time point, and tests of individual coefficients and partial R2 will be used to test and quantify effects of stress reactivity at each point in time.

Aim 2 / Hypothesis 2 examines EMA cognitive performance relationships with recurrence and cognitive decline. It will consider both the sample variance of each of the three EMA cognitive tests within each burst and the sample variance across each burst. To evaluate the effect of EMA cognitive performance variability on LLD recurrence, a logistic regression model will be used to regress recurrence on EMA cognitive variability. To evaluate the effect of EMA cognitive variability on cognitive decline over two years, a linear regression model will be used to regress two-year neuropsychological testing decline on EMA cognitive variability. In exploratory analyses, we will consider the volatility (i.e. the standard deviation of log-differences) of assessments within each burst.

3.7.5. Power analyses

Power analyses were conducted using preliminary data estimating a 20% attrition rate and 43% recurrence rate over two years [5], resulting in 60 comparison participants and 170 LLD participants, of whom 73 experience recurrence and 97 do not. Given these sample sizes, for Aim 1 the power to detect differences between LLD and comparison participants under all possible values of observed effect sizes and autocorrelation values was greater than 97%. Given these sample sizes, the power to detect differences between recurrent and remitted participants under all possible values of observed effect sizes and autocorrelation values was greater than 94%. For both sets of analyses, to investigate the possible effects of higher rates of dropout, for our highest within-subject autocorrelation of ρ = 0.75 and our smallest observed effect size of η² = 0.10, there is greater than 82% power even with a 60% attrition rate.

Power is comparable for Aim 2. With the proposed sample size, the power to detect associations between cognitive decline and stress reactivity and network stability under all possible values of observed effect sizes was greater than 93%. Similarly, the power to detect the predictive association of EMA cognitive performance variability on recurrence, under all possible values of effect sizes, is greater than 88%. In both sets of analyses, to investigate the possible effects of higher rates of dropout on power, for the smallest observed effect size of η² = 0.10, there is greater than 80% power even with a 33% attrition rate.

3.8. Governance and safety monitoring

The study is governed by an Executive Committee chaired by the VUMC Principal Investigator (WDT) and including the two other PIs (OA and CA). This committee meets twice monthly. One monthly Executive Committee meeting focuses on clinical operations, including recruitment, retention, protocol implementation, and unexpected adverse events. This meeting is supplemented by weekly or biweekly operational meetings between the site study coordinators. The second monthly Executive Committee meeting includes other study co-investigators, focuses on the scientific progress of the study, and serves as the Publication Committee for the study.

Given the long duration of the study, the study also has an independent Data Safety and Monitoring Board (DSMB). The DSMB meets every six months to monitor study conduct and participant safety. It advises the team on issues related to safety, data quality, and overall study conduct.

4. Discussion

There is a significant gap in our understanding of the underlying mechanisms and longitudinal interrelationships between depression recurrence, brain network function, brain aging, and cognitive decline [2], [3]. The current study seeks to better elucidate these relationships in context of stress, as individuals with remitted depression exhibit different physiological responses to stress [80], [81], [82] and stressful life events and greater perceived stress increase recurrence risk [5], [12]. Stressors may alter homeostasis of key intrinsic functional networks, predisposing to brain network dysfunction and clinical symptoms. The resultant increase in allostatic load from each depressive episode may contribute to accelerated brain aging, vulnerability to future recurrent episodes, and risk of cognitive decline [2], [3].

The study design provides close monitoring, with regular assessment of depression symptom change, environmental influences, and cognitive performance. The frequent neuroimaging, occurring four times over the two-year period, provides an opportunity to capture functional changes related to depression recurrence, but also identify trajectories of how network function may change with either repeated depressive episodes or sustained remission. Thus the frequency of data collection across different modalities provides a novel opportunity to identify individual trajectories.

This study has high potential clinical significance. It would be extremely useful to move beyond the ‘past history predicts the future’ paradigm that guides our current prognostics. An approach that individually stratifies risk of recurrence in recently remitted individuals could guide clinical care, identifying individuals who may be able to reduce the frequency of clinical encounters and those who will need continued frequent ongoing care despite a recent remission. A better understanding of factors that protect against recurrence may allow us to target them in prevention efforts. Although maintenance antidepressant treatment reduces recurrence risk [6], [13], there are also likely benefits for improving physical activity, social engagement, or targeted psychotherapy techniques. These questions are relevant to understand how depression recurrence may influence cognitive decline and how we can reduce the dementia risk in LLD.

As with any study, limitations are related to what is not being measured. We obtain comprehensive measures of cognitive performance and corresponding structural MRI measures of vascular injury, gray matter atrophy, and loss of white matter integrity. However, the study does not systematically obtain peripheral or central markers of Alzheimer’s disease pathology, limiting our ability to consider how such neuropathology may influence depression recurrence. Similarly, the study does not assess all potential factors contributing to depression vulnerability [3], such as earlier life head trauma. While presence of substance use disorders are exclusionary, assessment of ongoing substance use occurs informally during clinical interview and is not based on using structured assessments. Likewise, despite conducting burst actigraphy assessments quantifying activity, we do not otherwise assess physical activity or assess frailty, despite frailty being a challenging LLD phenotype [83], [84]. Additionally, while our actigraphy approach can assess activity patterns informative of sleep habits [76], it does not accurately capture sleep quality as would polysomnography. Finally, while our scientific model focuses on stress exposure and reactivity, other models may also be relevant to understanding depression recurrence, including examination of reward function [85]. Relatedly, despite focusing on stress, we do not examine physiological processes that arise from stress exposure, including inflammatory or hypothalamic-pituitary-adrenal axis changes.

4.1. Conclusions and future directions

Scientifically, this study will improve our understanding of how alterations in neural circuits and cognitive processes that persist during remission contribute to vulnerability for new depressive episodes. It will also elucidate how these processes may or may not contribute to cognitive impairment and decline. Clinically, this project will obtain deep phenotypic data that will identify vulnerability and resilience factors that can stratify individual risk.

This work will be broadly applicable to prevention research and clinical practice. It will inform the identification of “early warning signs” to recognize those patients at highest risk of recurrence, informing clinical planning and management. It will also help define neurobiological targets for recurrence prevention studies. Even if our hypotheses about stress reactivity involvement in recurrence risk are not supported, we will still gain a better understanding of the interrelationship between brain aging, recurrence, and cognitive decline.

Declaration of Competing Interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Warren D Taylor, MD reports financial support was provided by National Institute of Mental Health. Olusola Ajilore reports financial support was provided by National Institute of Mental Health. Carmen Andreescu reports financial support was provided by National Institute of Mental Health.

Appendix A. Supplementary material

Supplementary material

.