Author response: Multimodal brain age estimates relate to Alzheimer disease biomarkers and cognition in early stages: a cross-sectional observational study

Abstract

Article Figures and data Abstract Editor's evaluation eLife digest Introduction Methods Results Discussion Data availability References Decision letter Author response Article and author information Metrics Abstract Background: Estimates of ‘brain-predicted age’ quantify apparent brain age compared to normative trajectories of neuroimaging features. The brain age gap (BAG) between predicted and chronological age is elevated in symptomatic Alzheimer disease (AD) but has not been well explored in presymptomatic AD. Prior studies have typically modeled BAG with structural MRI, but more recently other modalities, including functional connectivity (FC) and multimodal MRI, have been explored. Methods: We trained three models to predict age from FC, structural (S), or multimodal MRI (S+FC) in 390 amyloid-negative cognitively normal (CN/A−) participants (18–89 years old). In independent samples of 144 CN/A−, 154 CN/A+, and 154 cognitively impaired (CI; CDR > 0) participants, we tested relationships between BAG and AD biomarkers of amyloid and tau, as well as a global cognitive composite. Results: All models predicted age in the control training set, with the multimodal model outperforming the unimodal models. All three BAG estimates were significantly elevated in CI compared to controls. FC-BAG was significantly reduced in CN/A+ participants compared to CN/A−. In CI participants only, elevated S-BAG and S+FC BAG were associated with more advanced AD pathology and lower cognitive performance. Conclusions: Both FC-BAG and S-BAG are elevated in CI participants. However, FC and structural MRI also capture complementary signals. Specifically, FC-BAG may capture a unique biphasic response to presymptomatic AD pathology, while S-BAG may capture pathological progression and cognitive decline in the symptomatic stage. A multimodal age-prediction model improves sensitivity to healthy age differences. Funding: This work was supported by the National Institutes of Health (P01-AG026276, P01- AG03991, P30-AG066444, 5-R01-AG052550, 5-R01-AG057680, 1-R01-AG067505, 1S10RR022984-01A1, and U19-AG032438), the BrightFocus Foundation (A2022014F), and the Alzheimer’s Association (SG-20-690363-DIAN). Editor's evaluation This is a useful study exploring multi-modality brain age (structural plus resting state MRI) in people in the early stages or at risk of Alzheimer's disease. They found solid evidence that people with cognitive impairment had older-appearing brains and that older-appearing brains were related to Alzheimer's risk factors such as amyloid and tau deposition. Their data suggest that the multi-modality brain age model is more accurate than a unimodal structural MRI model. https://doi.org/10.7554/eLife.81869.sa0 Decision letter Reviews on Sciety eLife's review process eLife digest The brains of people with advanced Alzheimer’s disease often look older than expected based on the patients’ actual age. This ‘brain age gap’ (how old a brain appears compared to the person’s chronological age) can be calculated thanks to machine learning algorithms which analyse images of the organ to detect changes related to aging. Traditionally, these models have relied on images of the brain structure, such as the size and thickness of various brain areas; more recent models have started to use activity data, such as how different brain regions work together to form functional networks. While the brain age gap is a useful measure for researchers who investigate aging and disease, it is not yet helpful for clinicians. For example, it is unclear whether the machine learning algorithm could detect changes in the brains of individuals in the initial stages of Alzheimer’s disease, before they start to manifest cognitive symptoms. Millar et al. explored this question by testing whether models which incorporate structural and activity data could be more sensitive to these early changes. Three machine learning algorithms (relying on either structural data, activity data, or combination of both) were used to predict the brain ages of participants with no sign of disease; with biological markers of Alzheimer’s disease but preserved cognitive functions; and with marked cognitive symptoms of the condition. Overall, the combined model was slightly better at predicting the brain age of healthy volunteers, and all three models indicated that patients with dementia had a brain which looked older than normal. For this group, the model based on structural data was also able to make predictions which reflected the severity of cognitive decline. Crucially, the algorithm which used activity data predicted that, in individuals with biological markers of Alzheimer’s disease but no cognitive impairment, the brain looked in fact younger than chronological age. Exactly why this is the case remains unclear, but this signal may be driven by neural processes which unfold in the early stages of the disease. While more research is needed, the work by Millar et al. helps to explore how various types of machine learning models could one day be used to assess and predict brain health. Introduction Alzheimer disease (AD) is marked by structural and functional disruptions in the brain, some of which can be observed through multimodal magnetic resonance imaging (MRI) in preclinical and symptomatic stages of the disease (Frisoni et al., 2010; Brier et al., 2014a). More recently, the ‘brain-predicted age’ framework has emerged as a promising tool for neuroimaging analyses, leveraging recent developments and accessibility of machine-learning techniques, as well as large-scale, publicly available neuroimaging datasets (Cole and Franke, 2017b; Franke and Gaser, 2019). These models are trained to quantify how ‘old’ a brain appears, as compared to a normative sample of training data - typically consisting of cognitively normal participants across the adult lifespan (e.g., Cole et al., 2015). Thus, the framework allows for a residual-based interpretation of the brain age gap (BAG), defined as the difference between model-predicted age and chronological age, as an index of vulnerability and/or resistance to underlying disease pathology. Indeed, several studies have demonstrated that BAG is elevated (i.e. the brain ‘appears older’ than expected) in a host of neurological and psychiatric disorders, including symptomatic AD (Franke et al., 2010; Franke and Gaser, 2012; Gaser et al., 2013), as well as schizophrenia (e.g., Koutsouleris et al., 2014), HIV (e.g., Cole et al., 2017c), and type-2 diabetes (e.g., Franke et al., 2013), and moreover, predicts mortality (Cole et al., 2018). Conversely, lower BAG is associated with lower risk of disease progression (Gaser et al., 2013; Wang et al., 2019; Bocancea et al., 2021). Critically, at least one comparison suggests that BAG exceeds other established MRI (hippocampal volume) and CSF (pTau and Aβ42) biomarkers in sensitivity to AD progression (Gaser et al., 2013). Thus, by summarizing complex, non-linear, highly multivariate patterns of neuroimaging features into a simple, interpretable summary metric, BAG may reflect a comprehensive biomarker of brain health. Several studies have established that symptomatic AD and mild cognitive impairment (MCI) are associated with elevated BAG (Cole and Franke, 2017b; Franke and Gaser, 2019). However, the sensitivity of these model estimates to AD in the presymptmatic stage (i.e. present amyloid pathology in the absence of cognitive decline [Sperling et al., 2011]) is less clear. The development of sensitive, reliable, non-invasive biomarkers of preclinical AD pathology is critical for the assessment of individual AD risk, as well as the evaluation of AD clinical prevention trials. Recent studies have demonstrated that greater BAG is associated with greater amyloid PET burden in a Down syndrome cohort (Cole et al., 2017a) and with greater tau PET burden in sporadic MCI and symptomatic AD (Lee et al., 2022). One approach to maximize sensitivity of BAG to presymptomatic AD pathology may be to train brain age models exclusively on amyloid-negative participants. As undetected AD pathology might influence MRI measures, and thus confound effects otherwise attributed to ‘healthy aging’ (Brier et al., 2014b), including the patterns learned by a traditional brain age model, an alternative model trained on amyloid-negative participants only might be more sensitive to detect presymptomatic AD pathology as deviations in BAG. Indeed, one recent study demonstrated that an amyloid-negative trained brain age model (Ly et al., 2020) is more sensitive to progressive stages of AD than a typical amyloid-insensitive model (Cole et al., 2015). However, this comparison included amyloid-negative and amyloid-positive test samples from two separate cohorts and thus may be driven by cohort, scanner, and/or site differences. To validate the applicability of the brain-predicted age approach to presymptomatic AD, it is important to test a model’s sensitivity to amyloid status, as well as continuous relationships with AD biomarkers, within a single cohort. Another recent comparison demonstrated that both traditional and amyloid-negative trained brain age models were similarly related to molecular AD biomarkers, but that further attempts to ‘disentangle’ AD from brain age by including more advanced AD continuum participants in the training sample significantly reduced relationships between brain age and AD markers (Hwang et al., 2022). Thus, in this study, we will apply the amyloid-negative training approach to a multimodal MRI dataset in order to maximize sensitivity to AD pathology in the presymptomatic stage. Most of the brain-predicted age reports described above focused primarily on structural MRI. However, other studies have successfully modeled brain age using a variety of other modalities, including metabolic PET (Goyal et al., 2019; Lee et al., 2022), diffusion MRI (Cherubini et al., 2016; Petersen et al., 2022), and functional connectivity (FC) (Dosenbach et al., 2010; Liem et al., 2017; Eavani et al., 2018; Nielsen et al., 2019). Integration of multiple neuroimaging modalities may maximize sensitivity of BAG estimates to preclinical AD. Indeed, recent multimodal comparisons suggest that structural MRI and FC capture complementary age-related signals (Eavani et al., 2018; Dunås et al., 2021) and that age prediction may be improved by incorporating multiple modalities (Liem et al., 2017; Engemann et al., 2020). One recent study has shown that BAG estimates from an FC graph theory-based model are significantly elevated in autosomal dominant AD mutation carriers and are positively associated with amyloid PET (Gonneaud et al., 2021). Furthermore, we have recently demonstrated that FC correlation-based BAG estimates are surprisingly reduced in cognitively normal participants with evidence of amyloid pathology and elevated pTau, as well as in cognitively normal APOE ε4 carriers at genetic risk of AD (Millar et al., 2022). Thus, incorporating FC into BAG models may improve sensitivity to early AD. This project aimed to develop multimodal models of brain-predicted age, incorporating both FC and structural MRI. Participants with presymptomatic AD pathology were excluded from the training set to maximize sensitivity. We hypothesized that BAG estimates would be sensitive to the presence of AD biomarkers and early cognitive impairment. We further considered whether estimates were continuously associated with AD biomarkers of amyloid and tau, as well as cognition. We hypothesized that FC and structural MRI would capture complementary signals related to age and AD. Thus, we systematically compared models trained on unimodal FC, structural MRI, and combined modalities to test the added utility of multimodal integration in accurately predicting age and whether each modality captures unique relationships with AD biomarkers and cognition. Methods Participants We formed a training sample of healthy controls spanning the adult lifespan by combining structural and FC-MRI data from three sources, as described previously (Millar et al., 2022): the Charles F. and Joanne Knight AD Research Center (ADRC) at Washington University in St. Louis (WUSTL), healthy controls from studies in the Ances lab at WUSTL (Thomas et al., 2013; Petersen et al., 2021), and mutation-negative controls from the Dominantly Inherited Alzheimer Network (DIAN) study of autosomal dominant AD at multiple international sites including WUSTL (McKay et al., 2022). To minimize the likelihood of undetected AD pathology in our training set, participants over the age of 50 were only included in the training set if they were cognitively normal, as assessed by the Clinical Dementia Rating (CDR 0; Morris, 1993), and had at least one biomarker indicating the absence of amyloid pathology (CN/A−, see below). We excluded 59 participants who did not have available CDR or biomarker measures (see Figure 1—figure supplement 1). As CDR and amyloid biomarkers were not available in the Ances lab controls, we included only participants at or below age 50 from this cohort in the training set. These healthy control participants were randomly divided into a training set (~80%; N=390) and a held-out test set (~20%; N=97), which did not significantly differ in age, sex, education, or race, see Table 1. Table 1 Demographic information of the combined samples. MeasureTraining sets (total N=390)Test sets (total N=97) §Analysis sets (total N=452)Ances Controls(CN/<50)DIAN Controls(CN/A−)Knight ADRC Controls(CN/A−)Ances Controls(CN/<50)DIAN Controls(CN/A−)Knight ADRC Controls(CN/A−)CN/A−CN/A+CIN136120134382633144154154Age (mean, SD)29.92 (9.92)40.02 (10.26)64.97 (10.57)26.68 (7.11)41.46 (12.34)64.73 (10.57)66.93 (8.53)72.56 (7.15)‡75.67 (6.86) ‡CDR (N 0 / N 0.5 / N 1.0 / N 2.0)NA120 / 0 / 0 / 0134 / 0 / 0 / 0NA26 / 0 / 0 / 033 / 0 / 0 / 0144 / 0 / 0 / 0154 / 0 / 0 / 00 / 119 / 35 / 2Amyloid status (N + / N -)NA120 / 0134 / 0NA26 / 033 / 0144 / 00 / 1540 / 57Biomarkers available (N PET / CSF / both)NA30 / 6 / 7911 / 22 / 91NA3 / 1 / 215 / 0 / 2824 / 0 / 12017 / 0 / 13714 / 0 / 43APOE ε4 carrier status (N + / N -)NA76 / 4499 / 34NA19 / 728 / 5115 / 2971 / 83 ‡55 / 98 ‡MMSE (mean, SD)NANA29.26 (1.05)NANA29.45 (0.94)29.13 (1.17)28.97 (1.33)25.37 (3.55) ‡Sex (N female / N male)70 / 6485 / 3584 / 5019 / 1816 / 1022 / 1189 / 5591 / 6368 / 86†Years of education (mean, SD)13.68 (2.16)14.78 (3.04)16.16 (2.43)13.95 (1.99)14.92 (2.83)16.48 (2.43)15.71 (2.65)15.90 (2.64)15.05 (2.97)*Race (N American Indian or Alaska Native)100100000Race (N Asian)112000010Race (N Black)670201707171620Race (N Native Hawaiian or Other Pacifc Islander)200200000Race (N White)57118112172626127137134SiteWUSTLMultiple sitesWUSTLWUSTLMultiple sitesWUSTLWUSTLWUSTLWUSTLScannerSiemens TrioSiemens Trio / VerioSiemens Trio / BiographSiemens TrioSiemens Trio / VerioSiemens Trio / BiographSiemens Trio / BiographSiemens Trio / BiographSiemens Trio / BiographField strength3T3T3T3T3T3T3T3T3T CN = Cognitively Normal, <50 = less than age 50, A− = amyloid negative, A+ = amyloid positive, CI = cognitively Impaired, DIAN = Dominantly Inherited Alzheimer Network, ADRC = Alzheimer Disease Research Center, AD = Alzheimer disease, CDR = Clinical Dementia Rating, MMSE = Mini Mental State Examination, WUSTL = Washington University in St. Louis, T = Tesla. Group differences from the CN/A− analysis set were tested with t tests for continuous variables and χ2 tests for categorical variables. * p < 0.05, ^ p < 0.10. † p < 0.01. ‡ p < 0.001. § Test sets include randomly-selected, non-overlapping subsets of participants drawn from the same studies as the training sets. Finally, independent samples for hypothesis testing included three groups from the Knight ADRC: a randomly selected sample of 144 CN/A− controls who did not overlap with the training or testing sets, 154 CN/A+ participants, and 154 cognitively impaired (CI) participants (CDR > 0 with a biomarker measure consistent with amyloid pathology [see below] and/or a primary diagnosis of AD or uncertain dementia [McKhann et al., 2011]). See Table 1 for demographic details of each sample. All participants provided written informed consent in accordance with the Declaration of Helsinki and their local institutional review board. All procedures were approved by the Human Research Protection Office at WUSTL (IRB ID # 201204041). PET and CSF biomarkers Amyloid burden was imaged with PET using (11 C)-Pittsburgh Compound B (PIB; Klunk et al., 2004) or (18 F)-Florbetapir (AV45; Wong et al., 2010). Regional standard uptake ratios (SUVRs) were modeled from 30 to 60 min after injection for PIB and from 50 to 70 min for AV45, using cerebellar gray as the reference region (Su et al., 2013). Regions of interest were segmented automatically using FreeSurfer 5.3 (Fischl, 2012). Global amyloid burden was defined as the mean of partial-volume-corrected (PVC) SUVRs from bilateral precuneus, superior and rostral middle frontal, lateral and medial orbitofrontal, and superior and middle temporal regions (Su et al., 2013). Amyloid summary SUVRs were harmonized across tracers using a centiloid conversion (Su et al., 2018). Tau deposition was imaged with PET using (18 F)-Flortaucipir (AV-1451; Chien et al., 2013). Regional SUVRs were modeled from 80 to 100 min after injection, using cerebellar gray as the reference region. A tau summary measure was defined in the mean PVC SUVRs from bilateral amygdala, entorhinal, inferior temporal, and lateral occipital regions (Mishra et al., 2017). CSF was collected via lumbar puncture using methods described previously (Fagan et al., 2006). After overnight fasting, 20–30 mL samples of CSF were collected, centrifuged, then aliquoted (500 µL) in polypropylene tubes, and stored at –80°C. CSF amyloid β peptide 42 (Aβ42), Aβ40, and phosphorylated tau-181 (pTau) were measured with automated Lumipulse immunoassays (Fujirebio, Malvern, PA, USA) using a single lot of assays for each analyte. Aβ42 and pTau estimates were each normalized for individual differences in CSF production rates by forming a ratio with Aβ40 as the denominator (Hansson et al., 2019; Guo et al., 2020). As pTau/Aβ40 was highly skewed, we applied a log transformation to these estimates before statistical analysis. Amyloid positivity was defined using previously published cutoffs for PIB (SUVR > 1.42; Vlassenko et al., 2016) or AV45 (SUVR > 1.19; Su et al., 2019). Additionally, the CSF Aβ42/Aβ40 ratio has been shown to be highly concordant with amyloid PET (positivity cutoff < 0.0673; Schindler et al., 2018; Volluz et al., 2021). Thus, participants were defined as amyloid-positive (for CN/A+ and CI groups) if they had either a PIB, AV45, or CSF Aβ42/Aβ40 ratio measure in the positive range. Participants with discordant positivity between PET and CSF estimates were defined as amyloid-positive. Cognitive battery Knight ADRC participants completed a 2 hr battery of cognitive tests. We examined global cognition by forming a composite of tasks across cognitive domains, including processing speed (Trail Making A; Schindler et al., 2018), executive function (Trail Making B; Schindler et al., 2018), semantic fluency (Animal Naming; Armitage, 1946), and episodic memory (Free and Cued Selective Reminding Test free recall score; Goodglass and Kaplan, 1983; Grober et al., 1988). This composite has recently been used to study individual differences in cognition in relation the preclinical AD biomarkers and structural MRI (Aschenbrenner et al., 2018), as well as functional MRI measures (Millar et al., 2021). MRI acquisition All MRI data were obtained using a Siemens 3T scanner, although there was a variety of specific models within and across studies. As described previously (Millar et al., 2022), participants in the Knight ADRC and Ances lab studies completed one of two comparable structural MRI protocols, varying by scanner (sagittal T1-weighted magnetization-prepared rapid gradient echo sequence [MPRAGE] with repetition time [TR] = 2400 or 2300 ms, echo time [TE] = 3.16 or 2.95 ms, flip angle = 8 or 9°, frames = 176, field of view = sagittal 256×256 or 240×256 mm, 1 mm isotropic or 1×1×1.2 mm voxels; oblique T2-weighted fast spin echo sequence [FSE] with TR = 3200 ms, TE = 455 ms, 256×256 acquisition matrix, 1 mm isotropic voxels) and an identical resting-state fMRI protocol (interleaved whole-brain echo planar imaging sequence [EPI] with TR = 2200 ms, TE = 27 ms, flip angle = 90°, field of view = 256 mm, 4 mm isotropic voxels for two 6 min runs [164 volumes each] of eyes open fixation). DIAN participants completed a similar MPRAGE protocol (TR = 2300ms, TE = 2.95ms, flip angle = 9°, field of view = 270 mm, 1.1×1.1×1.2 mm voxels; McKay et al., 2022). Resting-state EPI sequence parameters for the DIAN participants differed across sites and scanners with the most notable difference being shorter resting-state runs (one 5 min run of 120 volumes; see Supplementary file 1 for summary of structural and functional MRI parameters; McKay et al., 2022). FC preprocessing and features All MRI data were processed using common pipelines. Initial fMRI preprocessing followed conventional methods, as described previously (Shulman et al., 2010; Millar et al., 2022), including frame alignment, debanding, rigid body transformation, bias field correction, and normalization of within-run intensity values to a whole-brain mode of 1000 (Power et al., 2012). Transformation to an age-appropriate in-house atlas template (based on independent samples of either younger adults or CN older adults) was performed using a composition of affine transforms connecting the functional volumes with the T2-weighted and MPRAGE images. Frame alignment was included in a single resampling that generated a volumetric time series of the concatenated runs in isotropic 3 mm atlas space. As described previously (Fox et al., 2009; Millar et al., 2022), additional processing was performed to allow for nuisance variable regression. Data underwent framewise censoring based on motion estimates (framewise displacement [FD] > 0.3 mm and/or derivative of variance [DVARS] > 2.5 above participant’s mean). To further minimize the confounding influence of head motion on FC estimates (Power et al., 2012) in all samples, we only included scans with low head motion (mean FD < 0.30 mm and > 50% frames retained after motion censoring). BOLD data underwent a temporal band-pass filter (0.005 Hz < f < 0.1 Hz) and nuisance variable regression, including motion parameters, timeseries from FreeSurfer 5.3-defined (Fischl, 2012) whole brain (global signal), CSF, ventricle, and white matter masks, as well as the derivatives of these signals. Finally, BOLD data were spatially blurred (6 mm full width at half maximum). Final BOLD time series data were averaged across voxels within a set of 300 spherical regions of interest (ROIs) in cortical, subcortical, and cerebellar areas (Seitzman et al., 2020). For each scan, we calculated the 300×300 Fisher-transformed Pearson correlation matrix of the final averaged BOLD time series between all ROIs. We then used the vectorized upper triangle of each correlation matrix (excluding auto-correlations; 44,850 total correlations) as input features for predicting age. Since site and/or scanner differences between samples might confound neuroimaging estimates, we harmonized FC matrices using an empirical Bayes modeling approach (ComBat; Johnson et al., 2007; Fortin et al., 2017), which has previously been applied to FC data (Yu et al., 2018). Structural MRI processing and features All T1-weighted images underwent cortical reconstruction and structural segmentation through a common pipeline with FreeSurfer 5.3 (Fischl et al., 2002; Fischl, 2012). Structural processing included segmentation of subcortical white matter and deep gray matter, intensity normalization, registration to a spherical atlas, and parcellation of the cerebral cortex based on the Desikan atlas (Desikan et al., 2006). Inclusion and exclusion errors of parcellation and segmentation were identified and edited by a centralized team of trained research technicians according to standardized criteria (Su et al., 2013). We then used the FreeSurfer-defined thickness estimates from 68 cortical regions (Desikan et al., 2006), along with volume estimates from 33 subcortical regions (Fischl et al., 2002) as input features for predicting age. We harmonized structural features across sites and scanners using the same ComBat approach (Johnson et al., 2007; Fortin et al., 2017), which has also been applied to structural MRI data (Fortin et al., 2018). Gaussian process regression As described previously (Millar et al., 2022), machine-learning analyses were conducted using the Regression Learner application in Matlab (MathWorks, 2021). We trained two Gaussian process regression (GPR; Rasmussen et al., 2004) models, each with a rational quadratic kernel function to predict chronological age using fully-processed, harmonized MRI features (FC or structural) in the training set. The σ hyperparameter was tuned within each model by searching a range of values from 10–4 to 10*SDage using Bayesian optimization across 100 training evaluations. The optimal value of σ for each model was found (see Figure 1—figure supplement 2) and was applied for all subsequent applications of that model. All other hyperparameters were set to default values (basis function = constant and standardize = true). Model performance in the training set was assessed using 10-fold cross validation via the Pearson correlation coefficient (r), the proportion of variance explained (R2), the mean absolute error (MAE), and root-mean-square error (RMSE) between true chronological age and the cross-validated age predictions merged across the 10 folds. We then evaluated generalizability of the models to predict age in unseen data by applying the trained models to the held-out test set of healthy controls. Finally, we applied the same fully-trained GPR models to separate analysis sets of 154 CI, 154 CN/A+, and 144 CN/A− controls to test our hypotheses regarding AD-related group effects and individual difference relationships. Unimodal models were each constructed with a single GPR model. The multimodal model was constructed by taking the ‘stacked’ predictions from each first-level unimodal model as features for training a second-level GPR model (Liem et al., 2017; Engemann et al., 2020; Dunås et al., 2021). For each participant, we calculated model-specific BAG estimates as the difference between chronological age and age predictions from the unimodal FC model (FC-BAG), structural model (S-BAG), and multimodal model (S+FC BAG). To correct for regression dilution commonly observed in similar models (Le et al., 2018; Smith et al., 2019; Liang et al., 2019), we included chronological age as a covariate in all statistical tests of BAG (Cole et al., 2017a; Le et al., 2018). However, to avoid inflating estimates of prediction accuracy (Butler et al., 2021), only uncorrected age prediction values were used for evaluating model performance in the training and test sets. Statistical analysis All statistical analyses were conducted in R 4.0.2 (R Development Core Team, 2020). Demographic differences in the AD samples were tested with independent-samples t tests for continuous variables and χ2 tests for categorical variables, using CN/A− controls as a reference group. Differences in brain age model performance were tested using Williams’s test of difference between dependent correlations sharing one variable, i.e., Pearson’s r between age and each model prediction of age. To correct for age-related bias in BAG (Le et al., 2018; as previously mentioned), we controlled for age as a covariate during all statistical tests. Group differences in each BAG estimate were tested using an omnibus ANOVA test with follow-up pairwise t tests on age-residualized BAG estimates, using a false discovery rate (FDR) correction for multiple comparisons. Assumptions of normality were tested by visual inspection of quantile-quantile plots. Assumptions of equality of variance were tested with Levene’s test. Linear regression models tested the effects of cognitive impairment (CDR > 0 vs. CDR 0) and amyloid positivity (A− vs. A+) on BAG estimates from each model, controlling for true age (as noted above), sex, and years of education. Given the potential confounding influence of head motion on FC-derived measures (Power et al., 2012; Van Dijk et al., 2012; Satterthwaite et al., 2012), we also included mean FD as an additional covariate of non-interest in the FC and S+FC models. We tested continuous relationships with AD biomarkers and cognitive estimates using linear regression models, including the same demographic and motion covariates. Since the range of amyloid biomarkers was drastically reduced in the CN/A− sample, we excluded these participants from models testing continuous amyloid relationships. Effect sizes were computed as partial η2 (ηp2). Results Sample description and demographics Demographic characteristics of the training sets, test sets, and analysis sets are reported in Table 1. CN/A+ participants were older (t = 6.15, p < 0.001) and more likely to be APOE ε4 carriers (χ2 = 34.73, p < 0.001) than amyloid-negative controls. Furthermore, CI participants were older (t = 9.71, p < 0.001), more likely male (χ2 = 8.60, p = 0.003), more likely to be APOE ε4 carriers (χ2 = 56.67, p < 0.001), and had fewer years of education (t = 2.03, p < 0.043), and lower MMSE scores (t = 12.46, p < 0.001) than amyloid-negative controls. Comparison of model performance All models accurately predicted chronological age in the training sets, as ass