A neural implementation of cognitive reserve: insights from a longitudinal fMRI study of set-switching in aging

The dataset (Data.mat) comprises variables extracted from demographic information, cognitive task performance, and MRI/fMRI imaging of a longitudinal study of 52 adults aged 60–71, evaluated at baseline and after a 5-year follow-up. The analysis code is structured into two main codes: (A) Main Analysis in MATLAB (Code1_MainAnalysis_Matlab.m), which utilizes the primary dataset (Data.mat) to examine brain variables, model cognitive trajectories, and prepare data for further interaction analysis. (B) Interaction Visualization in R (Code2_Johnson-Neyman_R.R), which produces interaction effect visualizations using the output from MATLAB analysis.

Methods
This dataset does not contain any direct fMRI data. The fMRI data in our study were analyzed using a specific technique called Ordinal Trend Canonical Variates Analysis (OrT CVA). Through this approach, we derived a measure known as the OrT score, which was calculated for both single and dual-task conditions. This OrT score is the only fMRI-related variable included in the dataset that we uploaded.

To provide further details, OrT CVA is a multivariate data-driven technique that identifies patterns of regional functional activation that show a monotonic change across multiple experimental conditions (in the current study single and dual conditions). The extracted functional activation patterns, called ordinal trends (OrT), indicate sustained activity across graduated increase in task demand (Habeck, Krakauer, et al., 2005; Habeck, Rakitin, et al., 2005). The technique utilizes a specialized design matrix to enhance variance contributions from patterns that exhibit within-subject increases in pattern scores from single to dual conditions. The test statistic that is used to assess the significance of the task condition relationship of the derived activation pattern is the number of exceptions i.e. the number of individuals showing decreased pattern expression from single to dual condition and thus violate the majority rule of an increase. A null distribution is generated using a permutation test with 1,000 iterations, where condition assignments are randomized within participants. The p-value is determined by the fraction of times the permutation test yields a number of exceptions as low or lower than the point estimate. To ensure the robustness of voxel loadings in the derived pattern, a simple bootstrap technique is employed. The data are resampled with replacement (without randomizing subject and condition assignments), and the analytic point-estimate process is repeated 500 times.  Z-values for the voxel loadings are computed as the ratio of the point estimate of the loading divided by the bootstrap standard deviation around this point estimate. The OrT score which is the average level of activation of the identified patterns is computed for both single and dual conditions at baseline.