The BHR was established by the University of California, San Francisco and is reviewed by an institutional review board. Enrollment began in March 2014 and is ongoing. Details of the study design and sampling procedures have been previously published (15).
The present analysis includes N = 19476 participants aged 56 to 90 years who completed online, self-report questionnaires on a variety of demographic factors (age, gender, education level, race, and ethnicity), medical history (self-reported mild cognitive impairment (MCI), AD, and dementia, memory concern, family history of AD), and an unsupervised version of the CBB. Participants self-administer the test through the BHR website. Before each subtest, participants complete a practice session to get acquainted. Participants are invited by email to complete the CBB at 6-month intervals.
Table 1 contains participant information. Of the 19476 participants, 2459 self-reported an MCI diagnosis, and 388 self-reported an AD diagnosis. Figure 1 displays the distribution of repeated measurements for the sample stratified by self-reported disease status.
CogState Brief Battery
The CBB is a computerized cognitive assessment that has been previously validated with repeated assessment in healthy participants (18–20) and those with mild cognitive impairment (MCI) or early AD (21–23) in both supervised (16, 24, 25) and unsupervised (8, 26, 27) contexts. The CBB consists of four subtests. The Detection (DET) task tests information-processing speed, attention and motor speed. Participants engage in a simple reaction time (RT) paradigm where they must respond as soon as the presented card changes. The Identification (IDN) test measures visual attention and is a choice RT paradigm in which participants must decide as quickly as possible whether the presented card is red. The One-Card Learning (OCL) test measures visual learning and memory and is based on a pattern separation paradigm in which participants must decide whether a presented card has been seen previously in the task; some of the cards have been shown previously, some have not, and some have not been shown but are similar to those that have. The One-Back (ONB) test measures working memory and is based on the one back paradigm in which participants must decide whether the card they are looking at is the same as the card that was presented on the immediately previous trial. On each trial on each test, participants are instructed to respond Yes or No as quickly and as accurately as possible and measures of the speed and accuracy of responses are obtained for each test.
Family History of Alzheimer’s Disease
Family history of AD was obtained from participants from BHR. From BHR’s inception to 8/28/2019, participants answered the following question: “Have you, your sibling(s), or parent(s) ever been diagnosed with Alzheimer’s Disease?”. Beginning on 8/28/2019 to 9/28/2020, the question was worded as such: “Have your children, your sibling(s), or parent(s) ever been diagnosed with Alzheimer’s Disease? Finally, from 9/28/2020 onwards, the question was worded: “Do you have any biological parents, full siblings, or biological children who have been diagnosed with Alzheimer’s Disease?“
Subjective memory concern
Subjective memory concern was collected by asking the following question: “Are you concerned that you have a memory problem?”.
Self-reported Alzheimer’s disease/mild cognitive impairment
As part of their medical history questionnaire, BHR participants were asked, “Please indicate whether you currently have or have had any of the following conditions in the past.”, where AD or MCI were possible options for participants to select. Participants could self-report having both MCI and AD, leading to overlaps.
A CBB decliner was defined based on the identification of worsening performance on each CBB subtest over the study period. For the DET, IDN, and ONB subtests, the 95th percentile subject-specific slope was identified as an objective threshold for decliner status. For the OCL subtest, this cut-off was set at the 5th percentile. This methodology is consistent with that applied in previous studies of cognitive change using CBB performance in supervised contexts (2, 3).
For this analysis, we included all BHR participants with the following inclusion criteria: (1) age 55 years or older (2) had completed at least one CBB subtest (3) had completed all other self-report information described above. The raw data from each CBB subtest is skewed, so the data (mean RT of correct responses) were transformed using a logarithmic base 10 transformation and accuracy data (proportion correct) using an arcsine transformation (17, 19, 28). These transformations were applied by CogState. CBB data were subject to completion and integrity checks as described in previous research (29). Data that failed these checks were excluded.
Performance on each CBB subtest was modeled separately using a linear mixed-effects model (LMEM) with random intercepts and time effects. For each model, performance was modeled as a function of time (scaled to years), baseline age, gender, education level, self-reported AD diagnosis, self-reported MCI diagnosis, family history of AD, and memory concern, as well as their interactions with time. In each LMEM, the regression coefficient for a variable described the association of within-subject change in the predictor with change in the CBB outcome. An interaction term described the magnitudes of the associations of the baseline variables with time. For each CBB subtest, we fit a full model including all interactions effects and a reduced model with no interactions. We then assessed the statistical significance of the baseline variable by time interaction using a likelihood ratio test (LRT) between the full and reduced models.
In separate analyses, we used a series of multivariable binomial logistic regression models to examine associations between CBB “decliner status” and BHR variables. Statistical analyses were conducted using software R version 4.0.2 (30).