Introduction

Alzheimer's disease (AD) has a long preclinical phase in which multiple pathophysiological changes are observable before the onset of overt cognitive impairment1,2,3. However, it has been widely acknowledged that cognition does decline subtly during this preclinical stage, as evidenced by diverging cognitive trajectories on annually administered standardized neuropsychological tests in cognitively unimpaired (CU) older adults who have abnormal amyloid-β and tau biomarkers or are genetically at risk for AD4,5,6,7. Importantly, these longitudinal studies also revealed that the initial divergence of the AD-related trajectory often does not present as actual decline in test scores, but rather as a lack of practice effects, or a lack of expected improvement, on repeated cognitive testing in those with preclinical AD8. This is especially true for memory tests that are presented and retested using the exact same content, suggesting that AD pathology initially manifests as decreased ability to learn over repeated exposures due to decrements in the process of memory consolidation5,8,9,10. This observation has inspired preclinical AD researchers to shift from the classical annual single timepoint cognitive testing paradigm to novel memory paradigms that leverage remote, digital assessments to assess learning over repeated exposures at much more frequent intervals11,12,13,14.

The Boston Remote Assessment for NeuroCognitive Health (BRANCH) was developed to improve detection of the earliest AD-related cognitive changes, by assessing learning over days using daily, brief, remote digital testing on an individual’s own device13. Previous work has demonstrated the feasibility of the multi-day BRANCH paradigm in CU older adults, as well as good test-retest reliability and validity of personalized multi-day learning curves (MDLCs) captured remotely over one week15. Another study showed that BRANCH MDLCs obtained over seven days revealed very early amyloid-related deficits in memory consolidation that were undetectable using conventional single time-point assessments, suggesting that MDLCs are fundamentally different from a single timepoint measure and may offer a more sensitive way to capture the earliest cognitive signs of preclinical AD16. In the current study, we aimed to further elucidate the impact of concomitant tau pathology and neurodegeneration on these MDLCs among CU older adults.

CU individuals who harbor both amyloid-beta (Aβ) and tau pathology are considered to have a more advanced stage of preclinical AD than individuals with evidence of only elevated Aβ17,18. Mounting evidence suggests that tau deposition in neocortical areas in particular is associated with elevated Aβ19,20 and greater risk for future cognitive decline and progression to dementia21. From a disease process perspective, it would thus be expected that CU Aβ+ individuals with concomitant neocortical tau pathology would show more pronounced decrements in memory consolidation18. This could be further substantiated from a neuroanatomical perspective, as the temporal neocortex plays a critical role in memory consolidation22. Interactions between the hippocampus and neocortex transform temporary memories into more prolonged representations with experience and time23. In turn, disruptions of this pathway likely interfere with this transformation process, leading to less efficient learning over repeated exposure and thus more diminished MDLCs.

One pathway through which tau is thought to affect cognitive performance is through neurodegeneration, which can manifest as structural brain changes, including, but not limited to, gray matter loss. Notably, the emergence of Aβ-related tau pathology in CU older adults appears to affect gray matter changes and cognitive performance differently compared to age-related tau. That is, a recent study showed that long-term memory changes in Aβ-negative older adults are primarily linked to age-related medial-temporal lobe (MTL) tau deposition and hippocampal atrophy, whereas the more accelerated memory decline observed in Aβ-positive older adults is associated with neocortical tau-related cortical thinning specifically in AD-signature areas24. As MDLCs are considered to capture memory consolidation decrements that are uniquely associated with elevated Aβ beyond age-related memory changes, it would thus also be expected that MDLCs would be associated with cortical thinning in these AD-signature areas.

As such, the primary aim of this study was to investigate whether BRANCH MDLCs can detect subtle differences in memory consolidation across CU older adults with different Aβ and tau biomarker profiles based on Aβ and neocortical tau positron emission tomography (PET) uptake. We hypothesized that MDLCs would be most diminished in individuals with abnormal Aβ and abnormal tau biomarkers, as this is considered a more advanced pathophysiological stage of AD18. Our secondary aim was to examine the association between MDLCs and AD-signature cortical thickness25, and we hypothesized that lower MDLCs would be associated with greater AD signature cortical thinning.

Results

Sample characteristics

We included N = 167 participants (mean age 74.3 years, 63% female), of whom n = 106 (63%) were classified as A-T-, n = 46 (28%) were classified as A + T-, and n = 15 (9%) were classified as A + T + . Table 1 shows the characteristics for the total sample as well as each biomarker profile. The A-T- group was younger than the A + T- and A + T+ groups, but the latter two groups did not differ in terms of age. In addition, there were no other demographic differences across groups. Global CDR and MMSE scores did not reveal any clinical differences across groups. As expected, amyloid burden and neocortical tau tracer uptake were lowest in the A-T group and highest in the A + T+ group. Finally, cortical thickness was highest for the A-T- group, but lowest for the A + T- group.

Table 1 Baseline demographic and clinical characteristics for the total sample and by A/T biomarker profile
Full size table

BRANCH MDLCs across A/T biomarker profiles

Figure 1 shows the multi-day learning trajectories on the BRANCH composite including MDLC scores, separately for each biomarker profile. Supplementary Fig. 1 shows the raw learning curve data for each individual participant on the BRANCH composite, showing no evidence of ceiling effects in BRANCH composite scores over seven days of testing. However, some A-T- participants showed evidence of ceiling effects on the Face Name or Groceries Prices tests (see Supplementary Fig. 2). That is, n = 37 (22%) of the individuals reached the maximum score on Face Name on two or more days, and n = 11 (7%) reached the maximum score on Groceries Prices on two or more days.

Fig. 1: Multi-day learning trajectories and corresponding multi-day learning curve (MDLC) scores on the BRANCH composite for each A/T biomarker profile group, corrected for age, sex, and years of education.
figure 1

Fill areas around lines present 95% confidence intervals. BRANCH Boston Remote Assessment for NeuroCognitive Health, MDLC multi-day learning curve, A amyloid, T tau.

Full size image

Linear regression analyses with the BRANCH composite score showed that no group differences were detected using day 1 accuracy scores (Fig. 1, Table 2). However, BRANCH Composite MDLC scores diminished across ascending biomarker groups, with the A + T+ group performing significantly worse (β = –0.58, 95%CI[–1.06 to –0.10], p = 0.018, Hedges’ g = 0.61) compared to the A-T- group. The A + T- group performed numerically worse than the A-T- group but only at trend-level (β = –0.24, 95%CI[–0.55 to 0.07], p = 0.128, Hedges’ g = 0.43) (Table 2).

Table 2 Outputs from linear regression models examining BRANCH Day 1 and MDLC scores across A/T groups, with the A-T- group as reference groupdd
Full size table

Linear regression models using the individual BRANCH tests revealed that amyloid and tau positivity contributed differently to MDLCs on each individual task. That is, being A+ regardless of T-status was associated with diminished MDLCs on Digit Signs (A + T- group: β = –0.31, 95%CI[–0.62 to –0.01], p = 0.047, Hedges’ g = 0.55), whereas only A + T+ was associated with worse MDLCs on the Face Name (A + T + group: β = –0.60, 95%CI[–1.10 to –0.09], p = 0.022, Hedges’ g = 0.58) and Groceries Prices tests (A + T+ group: β = –0.61, 95%CI[–1.13 to –0.08], p = 0.024, Hedges’ g = 0.63) (Fig. 2, Table 2).

Fig. 2: Group differences in multi-day learning curve (MDLC) scores on the BRANCH Composite and each individual BRANCH task.
figure 2

Effects obtained from linear regression models adjusting for age, sex, and years of education. Error-bars present 95% confidence interval. BRANCH Boston Remote Assessment for NeuroCognitive Health, MDLC multi-day learning curve. A = amyloid; T = tau.

Full size image

Exploratory analyses comparing the A + T- group to the A + T+ group showed that the A + T+ group performed numerically worse on the BRANCH Composite and individual tests, with larger estimates observed on the Face Name and Groceries Prices tests compared to the Digit Signs test, but none of these effects reached statistical significance after correcting for covariates (Supplementary Table 1).

BRANCH MDLC scores and AD signature cortical thickness

There was no association between BRANCH composite day 1 scores and cortical thickness in AD signature areas (Fig. 3), and this was similar for day 1 scores for each individual test (all p > 0.5). However, worse BRANCH composite MDLC scores were associated with greater AD-signature cortical thinning (β = 0.18, 95%CI[0.04–0.33], p = 0.013) (Fig. 3), which seemed largely driven by the Digit Signs and Face Name tasks tests (Table 3). Supplementary Table 2 shows the correlations between MDLC scores and AD-signature cortical thickness within each A/T group, suggesting that associations between BRANCH Composite MDLCs and cortical thinning were stronger in the A + T- group. However, it should be noted that these analyses were only exploratory, and that the relatively small sample-size of the A + T+ group may have hampered our ability to detect a statistically significant effect within this group.

Fig. 3: Associations between BRANCH Composite Day 1(left) and MDLC scores (right) and AD-signature cortical thickness (all standardized values).
figure 3

Note: Regression line is corrected for age, sex, and years of education. Fill areas around lines present 95% confidence intervals. Individual data points are uncorrected for covariates. BRANCH Boston Remote Assessment for NeuroCognitive Health, MDLC multi-day learning curve, A amyloid, T tau, AD Alzheimer’s disease.

Full size image
Table 3 Output from linear regression models investigating the association between BRANCH MDLC scores and AD-signature cortical thickness
Full size table

Discussion

We show that assessing learning over multiple days through remote, daily, digital memory testing can reveal differences in memory consolidation among CU older adults with ascending abnormal Aβ and tau biomarkers, implying that these learning curves track with advanced pathophysiological stages of AD. On the BRANCH composite, we observed that MDLCs diminished in individuals who harbor both Aβ and tau pathology, which was detected with a medium-to-large effect-size. However, we also found that Aβ and tau status contributed differently to MDLCs on each individual BRANCH test. That is, elevated Aβ (regardless of tau status) was associated with diminished MDLCs on the Digit Signs test, while it was high tau (in the presence of elevated Aβ) that was associated with worse MDLCs on the Face Name and Groceries Prices tests. Finally, we observed that diminished MDLCs were associated with greater cortical thinning in AD-signature regions. Importantly, none of the associations with A/T biomarker profiles or cortical thickness were detected using single time point BRANCH accuracy scores obtained at day 1. Taken together, these findings highlight how a multi-day learning paradigm such as BRANCH can reveal novel and unique information about cognitive changes during preclinical AD.

Traditionally, neuropsychological models have understood the clinical manifestation of AD in terms of cognitive change over years or even decades, which has been based on annual in-clinic cognitive testing26. However, there is now an emerging field that shows how capturing cognitive datapoints over much more frequent time-intervals, made possible through remote digital testing, may improve our brain-behavior models of AD, particularly in the preclinical stage of the disease11,12,27. The current study contributes to this evolving field, by demonstrating how a multi-day learning paradigm provides novel insights into the nature of cognitive decline in preclinical AD. More specifically, daily digital testing allows for greatly accelerating the ability to detect the phenomenon of (lack of) PE that was previously only characterized over years8. Complementing our previous work showing that MDLCs provide reliable and valid information about cognition and reveal Aβ-related memory deficits among otherwise CU older adults15,16, our current findings indicate that these MDLCs track with AD pathophysiological stages following the AT(N) biological staging framework17,18. Thus, by shifting from an annual test paradigm focusing on detecting decline to a daily paradigm focused on the ability to learn and recall daily repeated information, we can now appreciate that the very earliest cognitive signs of emerging AD pathology may manifest as subtle memory consolidation failures that gradually progress along the preclinical AD spectrum.

We found that A/T biomarker-related group differences in memory consolidation were captured differently by the individual BRANCH tests. That is, elevated Aβ seemed to be the main driver for diminished MDLCs on the Digit Signs test, whereas for both the Face Name and Groceries Prices MDLCs were not diminished until elevated tau was demonstrable in the neocortex. It could be argued that this dissociation is due to different measurement properties between tests. For instance, the Digit Signs task may lend itself more for capturing MDLCs as it has no fixed maximum score and thus, room for improvement over days is technically infinite. In contrast, both the Face Name and Groceries Prices MDLCs are based on accuracy scores which (eventually) may lead to a plateau in the MDLC. As such, it could be the case that the A-T- group reached a plateau on the latter two tasks without having further room to show improvement that could distinguish them from the A + T- group. However, whereas the raw MDLC values indicate that some A-T- individuals reached a plateau on the Face Name (i.e., the maximum MDLC score of 1), this does not seem to be the case on the Groceries Prices. Therefore, it is less likely that the tests' different sensitivity to Aβ and tau pathology is caused by measurement properties alone. Another explanation could be the different neural underpinnings of MDLCs on the Digit Signs, which is a processing speed measure with an associative memory component, versus MDLCs on the Face Name and Groceries Prices, which both primarily rely on memory retrieval without a speed component. Thus, Digit Signs MDLCs may be more sensitive to the more ‘global’ effect of elevated cerebral Aβ on slowing of synaptic transmission knowing to affect processing speed, whereas the Face Name and Groceries Prices MDLCs capture more immediate memory processes that rely on the interaction between MTL and related neocortical areas once tau start to accumulate in those regions. This is in line with previous studies showing that early Aβ-mediated cognitive slowing in CU adults was detected on annually repeated processing speed measures over multiple years, while annual declines on memory retrieval tasks were only detected in those with both emerging Aβ and tau pathology28,29.

Besides implications for theory forming about the nature of cognitive decline in preclinical AD, our study also contributes to other recent efforts showing how digital cognitive assessment may facilitate the assessment of cognition in AD research and clinical trials. Computers, tablets and especially smartphones are ubiquitous and becoming increasingly used and accepted by older adults30,31, which has opened the door to the development and implementation of numerous digital cognitive tools. These include, for example, traditional cognitive screening tools presented on a mobile device32, novel tablet-based cognitive test batteries for in-clinic administration33,34, as well as unsupervised versions of existing cognitive test paradigms35,36. Overall, these initiatives have highlighted the potential of digital assessments to improve the efficiency, precision, and standardization of cognitive testing, thereby providing more scalable and cost-effective methods to identify individuals with or at risk for cognitive decline37. Also, remote assessments on personal devices have the most potential to make cognitive testing more accessible and may thereby provide the much-needed boost for participant diversity in research and intervention studies38,39. Furthermore, remote assessments enable the design of novel cognitive paradigms that may be superior to traditional cognitive paradigms, by leveraging the ability to deploy high frequency testing that is not allowed using in-clinic testing. This has led to promising new paradigms including ecological momentary assessments, or “burst testing”, to obtain more reliable and ecologically valid estimates of cognition that have the potential to reveal subtle difficulties that only become evident on day-to-day tasks in the participant’s own environment30,40, as well as memory paradigms providing more sensitive metrics of short-term learning11,41. Among these, the multi-day BRANCH paradigm is unique in that it assesses the ability to learn and retrieve information that is presented over multiple timepoints, thereby mimicking the formation of memories for learned information in everyday life.

By providing insights about early AD-related failures of learning, that were previously, using extant methods, only detected over multiple years, we believe that remote, digital cognitive paradigms such as BRANCH have the potential to transform the field. For example, implementing multi-day BRANCH may facilitate screening large numbers of individuals without overt symptoms for secondary prevention trials, by aiding in the identification of older adults who may be biomarker-positive of AD and thus at greatest risk of cognitive decline. Moreover, combining multi-day BRANCH with plasma biomarkers may provide a scalable alternative to identify those who are eligible for and most likely to benefit from certain pharmacological interventions. Finally, repeating multi-day BRANCH over time may provide a more rapid method to track therapeutic benefit compared to traditional cognitive single time-point paradigms, as evaluating whether an individual’s multi-day learning curve has diminished or improved over time may provide a more sensitive outcome measure of cognitive change. We are currently collecting multi-day BRANCH longitudinally, which will provide insights into how MDLCs change over time in CU older adults with and without abnormal AD biomarkers.

There are several limitations that should be considered when interpreting our findings. First, our study sample predominantly consists of highly educated, non-Hispanic White participants, and thereby it is unknown how generalizable our findings are to more diverse populations. In addition, it is important to address that a certain level of digital skills, as well as compliance with daily testing, are required to successfully implement the multi-day BRANCH paradigm. The feasibility of remote, mobile test paradigms has been demonstrated in HABS and other cohorts15,30,42, but this should also be determined in individuals with less digital literacy. We are currently testing a Spanish version of BRANCH and rolling out BRANCH in community-cohort studies, which will help us address these important questions in the future. Further, a general challenge with unsupervised cognitive testing is the fact that there is less control over the location, timing of testing, and likelihood of participant distraction, which are all factors likely to interact with task performance and thus may impact the validity of test scores42. However, a previous study indicated that these factors mainly affect speed of performance rather than accuracy scores43, and hence we do not expect that the uncontrolled environment has biased our findings to a large extent. Additionally, there are other factors that cannot be experimentally controlled such as the presence of relatives helping the participant or unauthorized use of memory aides. To mitigate this, we asked participants to attest to completing the task unaided each day for the sake of advancing research.

Finally, the relatively small sample-size of particularly the A + T+ group may have hampered our ability to detect statistically significant differences between the A + T- and A + T+ groups, although the group-wise differences were all in the expected direction. Unfortunately, this small sample-size also did not allow us to further divide A + T+ group into N-/N + , so we were not able to formally test whether MDLCs further diminished with ascending N+ status. It should also be acknowledged that the use of cutoffs in general can be regarded a limitation, as they are imperfect estimates and there is no unique, optimal cut-point for whether someone has true pathology44. Despite this, previous studies have indicated the clinical utility and relevance of dichotomizing Aβ as well as tau status, when cut points are carefully selected45,46,47. We aimed to achieve this by using previously applied methods to derive specific A/T cut-offs separately for the different scanners used in our studies. Given the fact that this resulted in a proportion of A + T- and A + T+ individuals that is in line with previous reports of neuropathology in asymptomatic older adults48, we are confident that our selected cut-offs provided meaningful AD biomarker profile groups. However, the ultimate validation of the BRANCH paradigm will be to examine how MDLCs change over time and whether MDLC scores are predictive of future cognitive decline and progression to MCI or dementia. This will be particularly relevant in A + T- individuals, as having only elevated Aβ in the context of normal tau, especially at older age, may not be sufficient to develop cognitive decline. Longitudinal BRANCH data collection in the current study cohort is ongoing, which will allow us to investigate the predictive validity of the BRANCH MDLCs in each of the different biomarker profile groups in the future. We are also planning to implement BRANCH in other populations that pose substantial diagnostic challenges for memory-clinics, such as individuals with early-onset presymptomatic AD or those genetically at-risk for AD who are typically younger than the current study sample. In the future, it would also be interesting to examine the utility of BRANCH multi-day learning curves in other neurological or psychiatric conditions that are associated with subtle memory changes.

To conclude, we showed how unsupervised, remote, digital cognitive testing on an individual’s own device provides the opportunity to deploy previously understudied but promising cognitive paradigms that can improve our understanding of brain-behavior relationships in the earliest stages of AD pathophysiology. Our results indicate that subtle failures in memory consolidation track with the degree of AD biomarker positivity among CU older adults, thereby providing novel insights into the earliest phases of memory decline during preclinical AD. Ultimately, paradigms such as multi-day BRANCH have the potential to detect and track AD-biomarker related memory decrements more sensitively and rapidly than extant approaches, which is particularly relevant at the preclinical disease stage where intervention and prevention are most likely to be successful.

Methods

Participants

Participants were community-dwelling cognitively unimpaired (CU) older adults (age 60–92 years) from three Mass General Brigham observational cohort-studies: the Harvard Aging Brain Study (HABS, 2P01AG036694-11; R.A.S., K.A.J) and affiliated Instrumental Activities of Daily Living study (R01AG053184 and R01AG067021; G.A.M.) and Subjective Cognitive Decline study (1R01AG058825-01A; R.E.A). HABS is an ongoing longitudinal observational cohort-study of community-dwelling older adults who were clinically normal at the time of enrollment. Inclusion and exclusion criteria for HABS have been described in detail elsewhere49. Since BRANCH was introduced to participants at various stages of HABS, the present study included participants were deemed to be CU as defined by a Clinical Dementia Rating (CDR) global score = 0 at the time of their HABS in-clinic visit that was within a year of their BRANCH assessment. HABS participants who progressed to a global CDR of 0.5 were discussed in a consensus meeting and were included in our study if they were still deemed CU, as determined by clinician consensus based on standardized cognitive and functional test results and medical history50. For the SCD and IADL studies, participants were included at the start of the study, and were thus deemed CU based on study entry criteria, including a Clinical Dementia Rating (CDR) global score = 0, Mini-Mental Status Examination (MMSE) > 25, and Logical Memory Delayed Recall scores above education-adjusted cutoffs ( ≥ 9 for 16+ years of education, ≥5 for 8–15 years of education). This manuscript reports on a subset of the study sample described in our previous validation studies15,16. For the current study, we only selected those participants who had both amyloid and tau PET imaging available within three years of BRANCH testing (see PET Imaging Acquisition and Processing section).

BRANCH data were collected between December 2021 and August 2023. All study procedures contributing to this work were conducted in accordance with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. The study protocol was approved by the Mass General Brigham Institutional Review Board (#2019P000476). All participants provided informed consent prior to study participation.

Multi-day BRANCH paradigm

Multi-day BRANCH is an unsupervised web-based assessment for administration on an individual’s own device, including two associative memory tests (Face Name and Groceries Prices) and a processing speed test with an associative memory component (Digit Signs) that are repeated over seven days using identical stimuli each day to assess learning curves. The development and validation of multi-day BRANCH paradigm is described in detail elsewhere, including comprehensive information on its feasibility, reliability, and validity in CU older adults13,15,16. Briefly, these studies showed that multi-day BRANCH was feasible and well-accepted by older adults, with over 90% participants completing all daily study assessments with no clinical or demographic differences between completers versus non-completers. Further, we found that BRANCH multi-day learning curves exhibit good psychometric properties including high test-retest reliability as well as good concurrent validity15. Finally, our previous study comparing the BRANCH multi-day learning paradigm across Aβ-negative and Aβ-positive older adults showed that there were no Aβ-related differences in feasibility aspects such as days to completion and daily completion time16.

The Face Name test is a modified version of the Face-Name Associative Memory Exam (FNAME)51, for which participants are asked to remember a series of face-name pairs. Following a delay of 5 min, the participant is shown each face and asked to select the first letter of the name paired with that face (first letter name recall). Next, they are asked to identify the correct name via multiple choice (face-name matching). The outcome is the average number of correct responses for first letter name recall and face-name matching combined. The Groceries Prices test is another associative memory task for which participants have to remember a price paired with a pictured grocery item52. Following a delay of 3–5 min, participants identify the correct price among counterbalanced incorrectly paired and partially novel price/grocery distractor pairs. The outcome is number of correct responses. The Digit Signs Test is modeled on the Digit Symbol Substitution Test and reflects a measure of processing speed with an associative memory component53. Participants are shown a key of 6 street signs paired with digits and they must indicate “yes” or “no” whether a series of digit-sign pairs are correct. The outcome is number of correct pairs completed within 120 s.

BRANCH study procedures

Participants completed all three BRANCH tests for seven consecutive days, which took them about 10–15 minutes per day15. Prior to daily testing, participants attested to completing tasks independently and unaided without recording stimuli/responses with the goal of advancing research. Participants specified the time of day they preferred to take the test and were notified at that time each day. They were prompted to take the test at the same time each day but could choose to take it at another time that day if necessary. The same order of tests was used each day: The assessment always started with the Face Name learning trials, followed by the Groceries Prices learning trials, and then the Digit Sign test. After the Digit-Sign test, participants’ recall of the Face-Name pairs was tested using the first letter name recall and face-name matching subtests (delay period between learning trials and recall assessment of ~5 min). Lastly, participants’ recall of the Groceries Prices pairs was tested using the grocery-prices matching subtest (delay period between learning and recall assessment of ~3–5 min).

BRANCH MDLC scores

For each BRANCH task, an MDLC score is computed using an area under the curve (AUC) method summarizing day 1 performance followed by a non-linear learning trajectory over the subsequent six days15,16. To account for an individual’s starting point, a scaled AUC is computed as AUC/AUCmax, where AUCmax is the maximum value of the AUC obtained if the participant scored at the maximum value from the second through the final test administration. More specifically, this scaled AUC aims to remove the effect of an individual’s day 1 performance, to ensure that multi-day learning scores actually reflect memory consolidation over the seven days rather than encoding effects at the first day of testing. Scaled AUC values (hereafter referred to as “MDLC scores”) were computed for an equally weighted composite across the three tasks as well as for each individual task. MDLC scores range from 0 (no learning) to 1 (optimal learning).

PET imaging acquisition and processing

All 167 participants had amyloid PET with [11 C]Pittsburgh compound-B (PiB) and tau PET with [18 F]flortaucipir (FTP) within 1 ± 1.1 years of their BRANCH assessment. Amyloid and tau deposition were quantified in accordance with established protocols for acquisition and analysis54,55. Briefly, PiB images were acquired using a 60-minute dynamic acquisition and FTP images were acquired from 75–105 min post-injection on a Siemens ECAT HR+ or GE PET scanner. For both images, a mean PET image was created and coregistered with the corresponding T1 MR image using the SPM12 package (Wellcome Center for Human Neuroimaing) and the resulting coregistration transformation matrices were saved. FreeSurfer (version 6.0) regions of interest (ROIs) defined by segmenting the MR images were transformed into the PET native space using the inverse transformation matrices.

Amyloid and Tau PET positivity

PiB was expressed as the distribution volume ratio (DVR, estimated with reference Logan graphical method) with cerebellum gray as reference. A global cortical aggregate was calculated for each participant based on the average DVR in frontal, lateral temporoparietal, and retrosplenial (FLR) regions, to group participants into low amyloid (A-) versus high amyloid (A + ). The threshold for amyloid positivity (A + ) was determined for each camera separately. For the HR+ camera, a DVR FLR cutoff of 1.14 was used, based on previous work on defining the lowest threshold for amyloid PET to predict future Aβ accumulation and cognitive decline in clinically normal individuals56. Since the GE camera has higher sensitivity to amyloid uptake, the DVR FLR cut-off for A+ for that camera was adjusted to 1.11.

For FTP, standardized uptake value ratios (SUVRs) were computed for each region of interest (ROI) with cerebellar gray as reference. We grouped participants into low tau (T-) versus high tau (T + ) based on an early neocortical tau composite averaging the SUVRs in inferior-temporal, middle temporal and fusiform regions, which was based on a well-established stereotypical progression of tau pathology in Αβ-positive CU adults47,54. The threshold for tau positivity (T + ) was determined for each camera separately and based on the mean + 2*standard deviation across a reference sample of Αβ-negative clinically normal individuals from HABS-affiliated cohorts that was available for each camera. For the HR+ camera, this led to a SUVR cut-off of 1.315, whereas for the GE camera this led to a slightly lower SUVR cut-off of 1.29.

MR imaging acquisition and processing

Magnetic resonance (MR) imaging was available for n = 162 (of 167) participants and was acquired within 1.1 ± 1.0 years of their BRANCH assessment. MR imaging was performed on a 3 T scanner (TIM Trio; Siemens) with a 12-channel phased-array head coil. A T1-weighted volumetric magnetization–prepared rapid-acquisition gradient-echo (MPRAGE) image was acquired with following parameters: TR = 2300 ms, TE = 2.95 ms, TI = 900 ms, flip angle = 9°, resolution = 1.1 ×1.1 ×1.2 mm. T1-weighted images were processed and quality-assessed using the automated reconstruction protocol in FreeSurfer (version 6.0). Our MRI measure of interest was a FreeSurfer-derived AD-signature composed of the surface-area weighted average of the mean cortical thickness in the following individual ROIs: bilateral entorhinal, parahippocampal, inferior-temporal, middle temporal, and fusiform25,47.

Statistical analyses

Statistical analyses were completed using R (v4.0.3). Statistical significance was set at p < 0.05. We combined participant’s A (+/-) and T (+/-) status to generate three biomarker profile groups: A-T-, A + T-, A + T + . Only 2 individuals were classified as A-T + , hence those were excluded from our analyses.

Demographic and clinical differences between the biomarker profile groups were investigated using Chi-square tests for discrete variables and one-way analyses of variance (ANOVA) followed by Tukeys Honest Significant Difference (HSD) post hoc tests for continuous variables.

First, multi-day trajectories for the BRANCH composite were visualized by plotting the mean accuracy scores over seven days for each A/T biomarker profile group after regressing out the effects of age, sex, and years of education. Next, linear regression analyses were used to examine differences in MDLC scores (outcome) across A/T biomarker groups (predictor of interest) with the A-T- as reference group, and adjusting for age, sex, and years of education. Separate models were run for the BRANCH composite MDLC score as well as MDLC scores for each individual test (Digit Signs, Face Name, Groceries Prices). Similar models were run to examine whether any group differences were detected using BRANCH day 1 accuracy scores as outcome, for the composite as well as the individual tests. Finally, linear regression analyses adjusting for age, sex, and education were used to examine associations between MDLC scores (outcome) AD-signature cortical thickness (predictor of interest), for both the BRANCH composite and the individual tests. These analyses were also repeated using BRANCH day 1 accuracy scores as outcome. All outputs and figures from linear regression models present standardized regression coefficients, to facilitate the comparison of estimates across BRANCH measures. For the A/T group models, Hedge’s g effect-sizes are reported as well, where 0.2 reflects a small effect, 0.5 a medium effect and 0.8 a large effect57.