Impact of Challenging Engagement on Cognition in Older Adults

All older adults experience some degree of cognitive compromise as they age and approximately 32 percent of adults aged 85 and older suffer from Alzheimer's disease (AD). The Alzheimer's Association estimates that delaying the onset of AD symptoms by only five years would reduce the rate of incidence by 50 percent! The present clinical trial builds on a wealth of observational work and more recent experimental research conducted in the PI's lab, which suggests that an important element of maintaining cognitive vitality for life is sustained engagement in mentally-challenging activities. In a U.S. sample of cognitively normal adults, investigators recently demonstrated that older adults who were randomly assigned to learn digital photography, quilting, or both, in fast-paced, demanding classes for 15 hours per week for three months, showed enhanced episodic memory function-both at the end of the engagement period and, importantly, one year later (Park et al., 2014). The observed memory improvements were in comparison to two active control conditions that were low in new learning: a social engagement group that had fun but did not engage in active learning, and a placebo condition where participants worked on low-effort cognitive tasks that relied on use of previous knowledge. The investigators also found similar facilitation effects when older adults were trained to use many different applications on an iPad. The lab most recently reported that older participants who participated in high-effort engagement conditions showed an increase in neural efficiency, exhibiting a change in neural activity from a pre-intervention pattern characteristic of older adults to a post-intervention pattern typical of young adults. Based on these findings, which included relatively small numbers of subjects, the investigators will conduct a larger clinical trial to determine whether mentally challenging activities facilitate memory in cognitively normal adults via changes of neural structure and function. The investigators propose to conduct a clinical trial study that will (a) evaluate the efficacy of different types of engagement in improving cognitive function in older adults, (b) examine the likelihood that mental effort invested is the underlying mechanism accounting for engagement effects, (c) show whether engaging in high-demand activities results in reliable brain changes. The investigators expect to demonstrate that when older adults engage for a sustained period of time in high-effort tasks (learning photography), both their memory and the modulation capacity of their brain will increase.

Randomization Procedure and Statistical Analyses:

Potential participants will complete an initial eligibility form. The investigators will contact those who are deemed eligible, will follow up with a TICS cognitive phone interview screening, and provide the link to an on-line demographic enrollment questionnaire.

The investigators will invite subjects who pass these screens to attend an informational session. At these sessions, project RAs will consent potential participants and have them complete the MRI screening form. The investigators will schedule consenting participants for cognitive testing and MRI scans (if applicable).

During the 3-week period set aside for cognitive and MRI testing, subjects will be assigned among the three treatment arms using a centrally created randomization scheme. Because baseline data for all potential participants will be available at the time of randomization, the rerandomization method of Morgan and Rubin (2012 Annals of Statistics 40:1263) will be utilized, which can achieve improved covariate balance in this setting. The investigators will generate a series of randomizations and evaluate them for balance on age, education, and sex, designating balance in terms of the MANOVA F statistic comparing the distributions of the covariates across the treatment groups. Once a randomization that meets this criterion is identified, it will be applied to the eligible participants.

Preliminary Analyses. In initial analyses, we will summarize categorical variables by proportions and continuous variables by means and quantiles. We will graph continuous variables and assess them for skewness, transforming if necessary (for example by logs or square roots) to render them more nearly normally distributed. We will explore relationships among variables by examining scatter plots and correlation matrices. We will conduct all analyses in R (version 3.3.2 or later) or SAS (version 9.4 or later).

Analysis of Primary Outcome Variables. The primary cognitive outcome endpoint will be a composite, scalar episodic memory score, as described in earlier research from the SYNAPSE project (5). This measure will exhibit substantial between-subject variability, in that subjects who give high scores at baseline are likely to give high scores at follow-up as well. To account for this, in primary analyses we will adjust for baseline levels by analysis of covariance - i.e., including baseline values together with treatment arm in a regression model for the post-treatment outcome. Alternatively (and equivalently), we can analyze the outcome variable in a mixed model, evaluating a treatment effect by estimating a time-by-treatment interaction. We will moreover conduct mixed-model analyses including the intermediate (6-week, mid-treatment) and long-term (1-year) values of the cognitive outcomes together with the end-of-treatment (12-week) outcome. As a secondary analysis to further elucidate the magnitude and timing of treatment effects, we will seek to create parsimonious models of this outcome as a function of time, treatment arm, stratification factors (center, age, sex, education) and potentially other factors measured at baseline. The primary brain outcome will be a vector measure of fMRI activation in four brain regions of interest, as described above and in previous work from SYNAPSE (7). This measure is also likely to exhibit substantial between-subject variability. We will again analyze the outcome variable in a mixed model, evaluating the treatment effect by estimating a time-by-treatment interaction, and conduct a secondary analysis where we model activation in the four regions as functions of time, treatment arm, stratification factors (center, age, sex, education) and potentially other factors measured at baseline.

Analysis of Secondary Outcomes. Dose-response. We anticipate that there will be a dose-response relationship, with the control arms having the lowest values, high-engagement arms, the highest values, and moderate-engagement arms having values in between. We will construct mixed models to estimate the sizes of these effects, and to determine whether effects are linear or nonlinear in the degree of engagement.

Subgroup Analyses. We will conduct a number of analyses aimed at estimating treatment effects within strata of age (younger or older than age 72), gender (male or female), education (greater or less than 14 years), and center (Dallas or Hamburg). We expect each of these strata to comprise roughly half of the subjects, except that our sample will likely be 65% female, reflecting the sex imbalance in the elderly. We will replicate our main analyses in each of the stratum subgroups, and additionally test for interactions.

Incomplete data. A major concern in any follow-up study is that there will be substantial dropout, eroding trial power. As indicated above, we expect no more than 15% to 20% of subjects to fail to complete the followup evaluation schedule. This is not a large fraction of dropout, and we have provided for its effects on power in our sample-size calculations. A second concern is that dropouts may differ systematically from completers, potentially introducing bias into estimated treatment contrasts. Our primary approach to analysis is to use mixed models, which give correct results as long as the dropout mechanism is missing at random - i.e., the probability of dropout, given the potentially missing observation and all prior observed data, does not depend on the potentially missing observation. Moreover, as long as the dropout is roughly balanced between treatment arms, it is unlikely to have a substantial biasing effect on estimated treatment effects, even if the dropout mechanism is not missing at random. In any event, if dropout is excessive or is unbalanced between arms, or there is concern that it is not missing at random, we are prepared to conduct analyses for sensitivity to nonignorable (i.e., biasing) dropout using general methods that Dr. Heitjan has developed.

Latent Factor Modeling. As a further form of secondary analysis, we will analyze the cognitive outcome variables simultaneously using a latent-variable approach. With this method, one models the several cognitive variables at each measurement time as being statistically independent given an unobserved, subject-specific latent variable. One accommodates serial correlation within subjects by estimating correlation of the latent variable within subjects over time. The approach evaluates treatment and time effects by modeling the mean of the latent variable as a function of the predictors, and each subject's estimated latent trajectory serves as a summary of his outcome status. We will also apply such models to the four-variate fMRI primary outcome.

Neuroimaging Analyses. We will test for a Group x Time interaction on the primary measure of modulation capacity. For the large-scale brain network analyses, we will utilize measures of connectivity between major nodes within the networks as indicators to develop constructs for the executive fronto-parietal network, salience network, and default network at each interval of data collection. To measure increases in hippocampal volume, we will segment the left and right hippocampus into four regions of interest (subiculum, CA1 and CA2/CA3/CA4, dentate gyrus, and entorhinal cortex) for all individual high-resolution MRI images. This division into 4 regions of interest (ROI) will ensure sufficient reliability and reproducibility of the subfield distinctions. These ROIs will be entered into the same Group x Time ANOVA as in the fMRI analyses.