DC-MARVEL was a two-year multidomain randomized clinical trial (ClinicalTrials.gov Identifier: NCT04559789) conducted as a collaborative effort between Neurotrack Technologies, Inc. and the University of Arkansas. Funding for the project was provided by the National Institutes on Aging (NIA) in the form of an SBIR grant (Grant Number: R44AG063672). DC-MARVEL was evaluated and approved as an ethical research study by the Institutional Review Board at the University of Arkansas (Protocol Number: 2009280813A009), and the NIA’s Data and Safety Monitoring Board provided ongoing safety and regulatory oversight. The study protocol21 and preliminary analyses from four months of intervention10 were published previously.
Sample
A total of 592 individuals were screened for enrollment in DC-MARVEL, and after an evaluation of inclusion/exclusion criteria, 216 participants were ultimately included at baseline. Upon completing the 2-year study duration, 178 participants (128 females, 50 males) remained in the study after accounting for attrition and missing data. A full summary of enrollment, sample size, and attrition is presented graphically in Fig. 1. To meet the inclusion criteria for this study, a participant had to be between the ages of 45 and 75 years, be fluent in English, own a smartphone, be willing to communicate via text message, and have at least two of the following risk factors for ADRD based on the Australian National University–Alzheimer’s disease risk index (ANU-ADRI)22: high school education or less; a body mass index (BMI) ≥ 25 kg/m2 but less than 40 kg/m2; or history of diabetes, hypertension, high cholesterol, smoking, or traumatic brain injury. Exclusion criteria were visual problems impacting the ability to view a screen at a normal distance; history of a learning disability; recent cardiovascular event; current participation in a cognitive training intervention or lifestyle change program; current diagnosis of any mental health condition, neurologic condition, dementia, mild cognitive impairment, or any other serious health condition; or more than one of the following ADRD protective factors based on the ANU-ADRI: high physical activity level, eating non-fried fish or seafood more than 5 times per week, or a high level of cognitive engagement10,21.
The minimum sample size was determined a priori by utilizing both a mathematical and practical power analysis approach. Minimum total sample size requirement was calculated mathematically using G*Power 3.1.9.723 based on the primary statistical test planned for this study (2 × 4 mixed factorial ANOVA). Calculations were based on an alpha level of 0.05, a statistical power of 0.8, and a small effect size (ηp2 = 0.02 or d = 0.2). The largest required sample size returned from mathematical analyses was a total sample of 70 participants. Practical analyses for required sample size were conducted by examining related HC literature and the participant number recruited to demonstrate efficacy. Studies examining the efficacy of HC on improving psychological and cognitive variables known to be modifiable ADRD risk factors were carried out with sample sizes ranging from 40 to 45 total participants and demonstrated significant improvement in outcome measures19,20,24. Taken together, a total sample size of at least 70 participants was set as a minimum for this study.
DC-MARVEL used a convenience sampling recruitment approach. Potential participants were sought through advertising on National Public Radio, advertising on a university daily newswire service, social media, and word of mouth. Individuals expressing interest in the study were emailed a link to an inclusion/exclusion survey instrument, which determined whether they were eligible for study enrollment.
Study Design
DC-MARVEL utilized a parallel arm trial design. Participants were randomized into one of two groups (HC or HE) with an equal allocation ratio. Random assignment was achieved by pre-assigning all study IDs to an arm using a binary random number generator with an equal allocation randomization rule for the full recruited initial sample size at baseline. As participants entered the study, they were sequentially assigned an ID number and assigned to the corresponding study arm. Given the nature of the design and associated interventions, blinding was not possible and therefore, not utilized in this trial. The randomization sequence was generated by the principal investigator, and study enrollment and trial arm assignment carried out by the study coordinator.
Measures
Three outcomes were assessed in this study. First, the participant’s self-reported global cognition was collected as a measure of their subjective assessment of their current cognitive state. Self-reported global current cognition compared to perceived past cognitive state, although subjective, is a valid and reliable predictor of current cognitive state25 and is sensitive to detection of mild cognitive impairment26. Second, neuropsychological cognition was measured as an objective assessment of a participant’s overall cognitive state and subdomains of cognition. Multidimensional neuropsychological cognition assessment batteries are the accepted gold standard measures for objective cognitive assessment, both in total cognitive ability and domain-specific cognition, and these instruments are used in both the research and clinical setting to effectively discriminate between individuals with and without cognitive impairment27,28. Third, ADRD risk was assessed as both a risk measure (positive risk), protection measure (negative risk), as well as a composite risk score aggregating both risk and protection. In long-term follow-up studies, Alzheimer’s risk data obtained via survey instrument has demonstrated predictive validity for determining future ADRD diagnoses within 3–6 years following testing22.
Self-Reported Global Cognitive Status
The Everyday Cognition (ECog-12) survey was used to assess self-reported global cognition. The ECog-12 is a 12-item self-reported survey instrument asking participants to compare their current state of cognition to their cognitive state 10 years in the past across multiple common everyday cognitive tasks25. Each item uses a 4-point Likert-type scale, with higher values indicating greater cognitive impairment compared to the perceived historical cognitive state from one decade prior to testing25. The ECog-12 is reliable and valid, and effectively discriminates participants with clinically relevant cognitive impairment (mild cognitive impairment or greater) from individuals with normal cognitive function25.
Neuropsychological Cognitive Ability
The Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) was used to assess neuropsychological cognitive ability. The RBANS assessment is presented to participants verbally and on a tablet, delivered by a trained test administrator. Participants are asked to complete several cognitive/memory tests including repeating words and stories, drawing geometric shapes, identifying pictures, and matching symbols with numbers from a given key29,30. The battery evaluates five neuropsychological construct dimensions: immediate memory, visuospatial/constructional cognition, cognitive language ability, attention, and delayed memory. Scores from the five dimensions were adjusted based on age and level of education to generate a domain index score, and combined to yield a single, continuous index score indicating overall neuropsychological cognitive ability30. For all RBANS index and total composite scores, higher scores are better. RBANS is a valid and reliable instrument for measuring neuropsychological cognitive domains and overall ability31, and has good sensitivity and specificity for discriminating cases of cognitive impairment from cognitively healthy controls27.
To ensure DC-MARVEL included not only virtual interventions, but also digital, remote assessment mechanisms, the Neurotrack Cognitive Battery (NCB) was also included as a measure of neuropsychological status. The NCB is an entirely self-administered cognitive battery which was presented to participants on a laptop computer. The battery incorporates tests evaluating and returning continuous scores for six cognitive domains: processing speed, attention, task-inhibition, executive function, associative learning, and associative memory32,33. Full descriptions of the six tests in the battery are provided elsewhere32,33. Importantly, the NCB platform attempts to address shortcomings present in other digital cognitive batteries through real-time webcam integration to capture additional information outside of test response choice (eye tracking), objective scoring via algorithm, and self-administration independent of a trained healthcare professional32. For all measured domains aside from executive function and task-inhibition, higher scores indicate better performance. The NCB is a valid and reliable assessment of neuropsychological status32–37 with excellent ability to discriminate individuals with likely cognitive impairment from healthy controls33,37.
Alzheimer’s Disease Risk
The ANU-ADRI is a self-report survey instrument assessing ADRD risk by quantifying several positive and negative risk factors and applying statistically-derived weightings13. Protective (negative risk) factors included are social engagement, cognitive activity, physical activity level, non-fried fish and seafood consumption, and alcohol consumption (if less than 2 drinks per day). Risk (positive risk) factors included in the ANU-ADRI are diabetes/dysregulated blood glucose status, depression status, obesity, history of traumatic brain injury, history of smoking, high cholesterol, high alcohol consumption (3 or more drinks per day), exposure to pesticides, as well as known demographic risk factors such as sex, age, and level of education13. In the ANU-ADRI scoring model higher scores indicate more risk/less protection and lower scores indicate less risk/more protection. The ANU-ADRI is a valid22 and reliable38 measure of ADRD risk.
Another ADRD risk scoring model produced from the Cardiovascular Risk Factors, Aging, and Dementia (CAIDE) study was also calculated and used as a measure of ADRD risk. CAIDE provides a numeric risk value that represents 20-year ADRD risk39. The CAIDE scoring algorithm includes age, years of education, biological sex, BMI, blood pressure, total cholesterol, history of smoking, and apolipoprotein E4 (ApoE4) gene carrier status as risk factors which are weighted according to longitudinally observed 20-year risk impact to generate a final risk score. Unlike the ANU-ADRI, CAIDE does not consider protective factors in its scoring model. Higher CAIDE values are indicative of a higher risk of developing ADRD in the next 20 years39. Elevated (> 7) CAIDE risk score is a good predictor of new ADRD diagnosis in a 20-year timeframe and discriminates individuals who will receive a new diagnoses from individuals who will not with high sensitivity and specificity40.
Testing Protocol
Upon completion of the initial inclusion/exclusion survey, qualified participants were asked to read and sign a digital copy of an informed consent document. Following completion of the informed consent document and formal study enrollment, participants scheduled a baseline testing visit to the laboratory where all study measures were completed. Participants completed three more testing visits approximately 4, 12, and 24 months from the baseline visit. The testing window for DC-MARVEL began on January 18th, 2021, and concluded on July 31st, 2023. All testing visits were conducted in the Exercise Science Research Center at the University of Arkansas. The protocol for this study and all recruiting procedures were approved by the Institutional Review Board at the University of Arkansas.
Prior to each testing visit, participants were asked to complete one additional digital survey to remotely collect demographic and health status data. This demographic and health survey contained the ECog-12 survey and questions designed to collect demographic information such as age, biological sex, and level of education relevant to this analysis. Planned ancillary studies required the collection of additional data via remote survey, not germane to this analysis, and are described in the published protocol21.
For the next scheduled testing task, participants were asked to complete the ANU-ADRI survey instrument on a provided laptop. A member of the research team was present to answer any questions the participant had regarding the ANU-ADRI, though the inventory is otherwise self-guided. After completing ANU-ADRI, basic cardiovascular and anthropometric data were collected from participants. Blood pressure was collected manually by a trained research team member using a standard inflatable sphygmomanometer cuff and stethoscope, and pulse rate was collected with a standard pulse oximeter on the finger. Body mass was measured using a physician’s beam-balance scale, and height was measured using a stadiometer.
At baseline, the RBANS testing battery was administered approximately 20 min after the blood pressure, pulse rate, and anthropometric assessments to ensure a cognitive break following the earlier survey instruments. The RBANS assessment is produced in multiple versions (A, B, C, and D) utilizing the same set of tests with slight changes to the specific utilized words, figures, and numbers to mitigate learning effects between testing sessions29,30. For this study, the RBANS Form A was administered in the first testing visit, Form B in the second, Form C in the third, and Form D in the fourth. The RBANS tests, which require manual scoring, were graded by a trained and experienced rater in accordance with procedures from the RBANS manual30. Roughly 45 minutes after the administration of the RBANS tests, participants were instructed to complete the NCB on a provided laptop. As the NCB platform is designed as a self-guided assessment, only minimal instructions were given. At each subsequent timepoint, the order of the RBANS and the NCB assessments was switched.
Pursuant to planned ancillary studies administered concurrently with DC-MARVEL, additional in-person data collection took place during this study, but those elements are not relevant to the core methodology presented here. The published protocol for this study describes those ancillary protocol components in detail21.
Intervention
Health Coaching
The virtual HC intervention was designed and administered by Neurotrack Technologies, Inc. A trained health coach was assigned to participants randomized into the HC intervention. Unlike other health and lifestyle interventions usually examined in research, HC does not follow a standardized approach. Rather, HC operates within a broad set of principles and practices to provide a personalized intervention to participants making an exact description of the protocol difficult or impossible. For this study, however, the HC intervention included several aspects which were consistent across participants. After the first testing visit to the laboratory, participants were scheduled to have an initial video conference or phone call with their health coach during which they discussed the HC process, were educated about the lifestyle domains known to impact cognitive health, identified which domains they wanted to change, had their motivation and willingness to change assessed by their coach, and established achievable goals to realize their desired future health vision. The following lifestyle domains were specifically suggested by the health coach as areas in which participants could improve cognitive health: nutrition, physical activity, sleep, stress, social engagement, and cognitive activity. The specific intervention for each participant which followed these initial discussions was tailored to each individual’s goals but was formulated within the previously described framework. The decision of which modifiable risk factors to focus on was made based on the participant’s preferences and the coach’s recommendation.
For the entire 2-year study duration, the participant and health coach communicated monthly via a scheduled video conference or phone call, and the coach reached out to participants 1–2 times per week via text messaging app and email. In scheduled monthly meetings, the health coach checked progress, assessed readiness for progression, identified obstacles, and strategized with the participant regarding how the remaining intervention would be implemented going forward. The coach’s more frequent weekly messages and emails to participants provided personalized educational materials based on the specific participant’s current goals or were provided as motivational messages, at the coach’s discretion. The number of HC scheduled remote meetings was tracked as a measure of HC adherence. The health coach recorded adherence data for each de-identified participant in an online database available to the research team.
Health Education
Participants in the HE intervention group received a biweekly email from the research team including educational material outlining how they could potentially improve their cognitive health through lifestyle and behavioral change. All educational materials were based on scientific articles, curated by the Neurotrack team, and reviewed by a subject matter expert to ensure maximum accuracy, readability, and engagement. The emails were designed to be eye-catching and engaging to ensure participants read each when they received it as they were instructed. The same lifestyle domains addressed by the health coach were utilized as topics in HE to allow for better comparison between interventions. Outside of scheduling and basic communication regarding logistics, participants in the HE group only had access to the research team during their scheduled testing visits.
Data Analysis
Analyses were conducted using the entire sample with an additional subgroup analysis on the older adult participants (65 years of age and older) to address the fourth research question and assess age-related effects. Most data analyses were performed using SPSS 27 (IBM Corp, Armonk, NY), and SAS 9.4 (SAS Institute, Cary, NC) was utilized to perform Durbin–Watson and Brown–Forsythe tests. Means, standard deviations, and 95% confidence interval of the mean were calculated for all continuous dependent, demographic, and anthropometric variables. Before completing any inferential statistical tests, all relevant assumptions were checked, and if they were met, statistical analysis was allowed to proceed. Prior to hypothesis testing, independent samples t-tests were utilized to determine if there were differences in mean baseline cognitive scores (ECog-12 and RBANS) or ADRD risk scores (ANU-ADRI and CAIDE) between males and females. Independent samples t-tests also determined if mean cognitive or risk scores were different for the HC and HE groups at baseline. A 2 × 4 (intervention × time) mixed factorial ANOVA was employed to determine if outcomes had changed from baseline to 4 months with each intervention group (HC or HE), and if an observed change in any measured outcome was dependent on intervention. That is, to determine if a significant main effect for time was found for any measured dependent variable or if there was a significant intervention × time interaction effect. If baseline sex differences were found in cognitive scores, sex was included as a blocking factor in inferential analyses to better isolate the effect of intervention and time. An a priori alpha level of 0.05 was used for all analyses. Post hoc pairwise tests were carried out via Fischer’s LSD and all parametric inferential statistical tests were two-tailed.