Contributions of the ADNI Biostatistics Core

Abstract

The goal of the Biostatistics Core of the Alzheimer's Disease Neuroimaging Initiative (ADNI) has been to ensure that sound study designs and statistical methods are used to meet the overall goals of ADNI. We have supported the creation of a well‐validated and well‐curated longitudinal database of clinical and biomarker information on ADNI participants and helped to make this accessible and usable for researchers. We have developed a statistical methodology for characterizing the trajectories of clinical and biomarker change for ADNI participants across the spectrum from cognitively normal to dementia, including multivariate patterns and evidence for heterogeneity in cognitive aging. We have applied these methods and adapted them to improve clinical trial design. ADNI‐4 will offer us a chance to help extend these efforts to a more diverse cohort with an even richer panel of biomarker data to support better knowledge of and treatment for Alzheimer's disease and related dementias.

Highlights

• The Alzheimer's Disease Neuroimaging Initiative (ADNI) Biostatistics Core provides study design and analytic support to ADNI investigators.

• Core members develop and apply novel statistical methodology to work with ADNI data and support clinical trial design.

• The Core contributes to the standardization, validation, and harmonization of biomarker data.

• The Core serves as a resource to the wider research community to address questions related to the data and study as a whole.

1. INTRODUCTION

The goal of the Biostatistics Core of the Alzheimer's Disease Neuroimaging Initiative (ADNI) since the initial phase of ADNI started in 2004 has been to ensure that sound study designs and statistical methods are used to meet the overall goals of ADNI. As such, Core members provide analytic support to ADNI investigators; contribute to validation, standardization, and harmonization efforts of the data; conduct and support analyses that span multiple domains of data (neuropsychological testing, neuroimaging including magnetic resonance imaging [MRI] and positron emission tomography [PET], fluid markers from cerebrospinal fluid [CSF] or plasma, and clinical assessments); develop methodology to handle multivariate patterns of change or other complexities in the data; and foster communication regarding the study design and data to researchers across the world. In addition, Core members have provided statistical leadership to studies using ADNI infrastructure or those modeled after ADNI.

2. STUDY DESIGN

At the initial stages of planning for ADNI‐1, Dr. Beckett was named Director of the Core. She has collaborated closely with Dr. Weiner and other ADNI leaders to ensure for each study wave that the planned sample sizes and frequency and timing of observation of individuals who were cognitively normal (CN), mildly cognitively impaired, or had mild dementia would give adequate precision for characterizing clinical and biomarker decline. Table 1 summarizes the baseline diagnoses, demographics, and cognitive characteristics for each ADNI wave, including the first 38 clinically evaluated enrollees to ADNI‐4. Overall, ADNI enrollment has aligned well with the goal of ≈ 40% CN, 40% with mild cognitive impairment (MCI), and 20% with dementia. Participants are overwhelmingly non‐Hispanic or Latino (95%) and White (88%) and highly educated (mean 16 years of education).

TABLE 1.

Demographic summaries by ADNI wave.

	ADNI‐1 (N = 819)	ADNI‐GO (N = 131)	ADNI‐2 (N = 790)	ADNI‐3 (N = 692)	ADNI‐4 (N = 38) a	Total (N = 2470)
Baseline diagnosis
CN	229 (28.0%)	0 (0.0%)	294 (37.2%)	374 (54.8%)	16 (38.1%)	913 (37.0%)
LMCI	402 (49.1%)	131 (100.0%)	345 (43.7%)	237 (34.7%)	16 (38.1%)	1131 (45.9%)
Dementia	188 (23.0%)	0 (0.0%)	151 (19.1%)	72 (10.5%)	10 (23.8%)	421 (17.1%)
Age
Mean (SD)	75.2 (6.8)	71.5 (7.9)	72.7 (7.2)	70.7 (7.4)	72.6 (7.6)	72.9 (7.4)
Range	54.4–90.9	55.5–88.3	55.0–91.4	50.4–90.6	55.5–85.2	50.4–91.4
Sex
Female	342 (41.8%)	60 (45.8%)	379 (48.0%)	379 (54.8%)	29 (69.0%)	1190 (48.1%)
Male	477 (58.2%)	71 (54.2%)	411 (52.0%)	313 (45.2%)	13 (31.0%)	1284 (51.9%)
Education
Mean (SD)	15.5 (3.0)	15.8 (2.7)	16.3 (2.6)	16.4 (2.3)	16.0 (2.5)	16.1 (2.7)
Range	4.0–20.0	10.0–20.0	8.0–20.0	10.0–20.0	12.0–20.0	4.0–20.0
Ethnicity
Hispanic or Latino	19 (2.3%)	8 (6.2%)	31 (3.9%)	58 (8.4%)	3 (7.3%)	119 (4.8%)
Not Hispanic or Latino	794 (97.7%)	122 (93.8%)	755 (96.1%)	633 (91.6%)	38 (92.7%)	2342 (95.2%)
Race
American Indian or Alaskan Native	1 (0.1%)	1 (0.8%)	1 (0.1%)	2 (0.3%)	0 (0.0%)	5 (0.2%)
Asian	14 (1.7%)	1 (0.8%)	14 (1.8%)	29 (4.2%)	6 (15.0%)	64 (2.6%)
Native Hawaiian or other Pacific Islander	0 (0.0%)	0 (0.0%)	2 (0.3%)	0 (0.0%)	0 (0.0%)	2 (0.1%)
Black or African American	39 (4.8%)	4 (3.1%)	34 (4.3%)	104 (15.2%)	9 (22.5%)	190 (7.7%)
White	762 (93.0%)	118 (91.5%)	728 (92.3%)	535 (78.3%)	22 (55.0%)	2165 (88.0%)
More than one Race	3 (0.4%)	5 (3.9%)	10 (1.3%)	13 (1.9%)	3 (7.5%)	34 (1.4%)
CDR‐SB
Mean (SD)	1.8 (1.8)	1.2 (0.7)	1.5 (1.9)	1.0 (1.6)	1.7 (1.9)	1.5 (1.8)
Range	0.0–9.0	0.5–4.0	0.0–10.0	0.0–10.0	0.0–7.0	0.0–10.0
MMSE
Mean (SD)	26.7 (2.7)	28.3 (1.5)	27.4 (2.7)	27.9 (2.5)	26.6 (3.2)	27.4 (2.7)
Range	18.0–30.0	23.0–30.0	19.0–30.0	16.0–30.0	19.0–30.0	16.0–30.0

Abbreviations: ADNI, Alzheimer's Disease Neuroimaging Initiative; CDR‐SB, Clinical Dementia Rating, Sum of Boxes; CN, cognitively normal; LMCI, late mild cognitive impairment; MMSE, Mini‐Mental State Examination; SD, standard deviation.

^a Data for ADNI‐4 as of June 17, 2024. Planned new enrollment for ADNI‐4 is 500 participants: 40% CN, 40% MCI, 20% dementia, with at least 50% drawn from under‐represented populations.

The design for ADNI‐1 also included a 50% random sample to receive 18F‐fluorodeoxyglucose (FDG) PET imaging and a further 25% random sample to receive 3T MRI. All participants received 1.5T MRI at regularly scheduled visits according to their diagnosis at screening. In addition, participants could opt into providing CSF samples. ¹ Sampling in ADNI‐1 was stratified by baseline diagnosis and age to ensure comparability. Drs. Anthony Gamst and Michael Donohue, then at the University of California, San Diego, worked closely with the Clinical Core and Informatics Core to implement and oversee the random sampling. ADNI‐2 switched from 1.5T to 3T MRI and added amyloid imaging to the types of PET imaging, each of which was obtained on everyone. Experimental MRI sequences (resting‐state functional MRI, diffusion tensor imaging, and arterial spin labeling) were performed depending on scanner manufacturer at the site. ADNI‐2 had no randomization, but recruitment was stratified on diagnosis to ensure representation within the study across the diagnostic groups. In ADNI‐3, tau imaging ([¹⁸F]Flortaucipir [AV‐1451] PET) and a second amyloid tracer (newly recruited participants received [¹⁸F] florbetaben PET, while roll‐overs from ADNI‐2 continued to receive [¹⁸F] florbetapir PET) were added to the protocol. Twenty percent of amyloid‐negative and 80% of amyloid‐positive (identified by amyloid‐PET imaging) CN and mildly cognitively impaired participants were randomly selected for two additional AV‐1451 PET scans, overseen by Dr. Donohue and the Clinical Core. Some of these changes are illustrated in Table 2, highlighting the challenges for data analysis of harmonizing across study phases and coping with missing data, both sporadic and systematic.

TABLE 2.

Challenges to ADNI data analysis from major study design changes over time. ^a

Study component	ADNI phase: In‐clinic participants
	ADNI‐1	ADNI‐GO	ADNI‐2	ADNI‐3	ADNI‐4
ECog	–	All	All	All	All b
MoCA	–	All	All	All	All
MRI: 1.5T	All	Some roll‐overs	–	–	–
MRI: 3T	25%	All new participants and some roll‐overs	All	All	All
MRI: fMRI	–	Scanner manufacturer dependent	Scanner manufacturer dependent	All	All
MRI: DTI	–	Scanner manufacturer dependent	Scanner manufacturer dependent	All	All
MRI: ASL	–	Scanner manufacturer dependent	Scanner manufacturer dependent	All	All
PET: FDG	50%	All	All	MCI, Dementia, initial visit only	–
PET: Amyloid c
PiB	Small subset	–	–	–	–
AV45	–	All	All	Roll‐overs	Roll‐overs + New
FBB	–	–	–	New	Roll‐overs + New
NAV‐4694	–	–	–	–	New
PET: Tau c
AV1451	–	–	Small subset	All	Roll‐overs + New
MK‐6240	–	–	–	–	New
PI‐2620	–	–	–	–	New
CSF	Opt‐in	All	All	All	Preferred but not required

Abbreviations: ADNI, Alzheimer's Disease Neuroimaging Initiative; ASL, arterial spin labeling; AV45, florbetapir; CSF, cerebrospinal fluid; DTI, diffusion tensor imaging; ECog, Everyday Cognition; FBB, florbetaben; FDG, fluorodeoxyglucose; fMRI, functional magnetic resonance imaging; MoCA, Montreal Cognitive Assessment; MRI, magnetic resonance imaging; PET, positron emission tomography; PiB, Pittsburgh compound B; SD, standard deviation.

^a Notes: Assessment schedule varied across phases by diagnosis and by study component. Refer to the study protocols/study procedures manuals for the different phases of ADNI for specifics (https://adni.loni.usc.edu/methods/documents/).

^b ECog‐12.

^c In ADNI‐4, the specific amyloid‐ and tau‐tracers are assigned for new participants based on proximity to tracer distribution centers.

RESEARCH IN CONTEXT

1. Systematic review: Alzheimer's Disease Neuroimaging Initiative (ADNI) Biostatistics Core papers were published in 2010 and 2015, reviewing the work of the Core during ADNI‐1 and ADNI‐GO/2. Here, the work over the past 20 years by the Core is reviewed.

2. Interpretation: The Biostatistics Core has contributed to the overall study design across the phases of ADNI, validation of biomarkers and the understanding of progression of biomarkers and association with clinical change, and has developed methods for capturing multivariate patterns of change and improving clinical trial design. Finally, the Core has and continues to provide support to related studies.

3. Future directions: In ADNI‐4, the Core will be instrumental in the adaptive design to refer participants from the remote cohort to the cohort providing blood samples and to the clinical cohort. The Core will continue to provide analytic support to ADNI investigators and serve as a resource for investigators across the world.

As noted earlier, a major limitation of the ADNI study to date is the lack of diversity among participants in race, ethnicity, and education. True equity in cognitive aging research would extend to inclusion across the whole population, not merely statistical adjustment for underrepresentation. A key goal for ADNI‐4 is to recruit a more diverse sample, using a wider outreach and dynamic sampling scheme. ² This plan gave rise to additional statistical challenges. The Biostatistics Core participated in evaluation of the performance of candidates for remote testing and screening, which ultimately led to use of a novel speech‐based test (Novoic Storyteller; see Nosheny et al., in this special issue, for more details ³ ). Previous ADNI data on the Everyday Cognition 12‐item test were used to help determine responses that may indicate cognitive impairment, combined with speech‐based testing. The second stage of the dynamic sampling plan involves remote blood testing, again supported by statistical analysis from the Biostatistics Core, to create a more precise and potentially longitudinal recruitment pathway toward in‐clinic assessment. The Biostatistics Core uses ongoing monitoring of recruitment to see if we are meeting our participation goals. We will coordinate with other Cores to adjust outreach and screening criteria as needed to ensure diversity and inclusiveness in this wave of ADNI, while also preserving scientific validity. Practical limitations on sample size and participant burden for the in‐clinic sample mean that we cannot do everything we might wish, but future supplementary studies may build on this sample by drawing on the remote‐participation group. Inclusion of participants from all strata of the initial remote cohort into in‐clinic evaluation will eventually allow estimation of the broader relevance of our biomarker findings to the remote‐participation community. The large, diverse recruitment pool of ADNI‐4 at each stage is thus a major strength, especially with the flexibility of dynamic sampling and longitudinal follow‐up of the remote cohort.

2.1. Imaging and fluid biomarker standardization and validation

During the 20 years of ADNI data collection, both imaging (MRI and PET) and fluid (CSF and blood) markers have been collected longitudinally on participants. The Biostatistics Core has played a key role in tracking and standardizing the data from these complicated measurement paradigms.

The MRI data are complicated from a statistical standpoint by high dimensionality, especially when considered over time. In addition, the technical details varied both at baseline (different 1.5T scanners, different manufacturers, different software releases) and over time and ADNI phases (shift to 3T scanners and new protocols, as shown in  Table 2). PET data have also reflected changes over time, with multiple tracers for different targets, as well as the need to coordinate with evolving MRI structural data. Fortunately, Dr. Danielle Harvey, who joined the Biostatistics Core at its inception as a junior faculty member and is now co‐director, has collaborated closely with both the MRI and PET Cores.

Dr. Harvey's contributions to imaging validation began with the ADNI Preparatory Phase Study for the MRI Core, assessing the longitudinal stability of repeated MRI for measuring brain change using tensor‐based morphometry. ⁴ The findings of the ADNI Preparatory Phase led to the final MRI protocol for ADNI‐1. ⁵ She then worked with Dr. Paul Thompson's group (an MRI Core lab), to identify the most sensitive processing pipeline within the tensor‐based morphometry (TBM) framework, ⁶ validate TBM measures of change, ⁷ and compare measures obtained from 1.5T and 3T MRI. ⁸ With Dr. Reiman's lab within the PET Core, Dr. Harvey also contributed to the validation of a new FDG PET summary measure, the hypometabolic convergence index. ⁹

The Biomarker Core has carried out a series of validation studies to determine reproducibility of fluid biomarker measurements across replicate samples within and between batches, over time, and across assay platforms. ¹⁰ Dr. Beckett has provided guidance on study design, sample sizes, and assay performance criteria both for this within‐ADNI comparison and for confirmatory studies using standardized samples to detect bias as well as consistency. ¹¹ The Biostatistics Core has also worked with the Biomarker Core to harmonize cut‐off criteria for CSF amyloid beta (Aβ_1–42) positivity across platforms using ADNI data, to support a transition at the beginning of ADNI‐3.

2.2. Description and predictors of clinical and biomarker progression

Using ADNI‐1 data, in collaboration with other ADNI investigators, annual percent change of regional volumes (from MRI) was compared across groups and higher atrophy rates were found among AD and individuals with MCI who progressed to AD within 12 months in key regions associated with AD. ¹² Another series of analyses, also with ADNI investigators, found that white matter hyperintensity (WMH) volumes were associated with cognitive decline in 1 year, ¹³ baseline FDG PET and episodic memory were associated with progression from MCI to AD, while the CSF phosphorylated tau (p‐tau)181/Aβ_1–42 ratio was associated with cognitive decline in those with MCI, ¹⁴ and reduced glucose metabolism at baseline and change was associated with cognitive and functional decline ¹⁵ indicating the potential of these markers to be used in participant selection or as covariates in clinical trials.

Integrating data across ADNI‐1, ADNI‐GO, and ADNI‐2 allowed comparison of multiple biomarker trajectories, by placing each participant on a long‐term continuum of disease. We obtained a picture of how each biomarker or clinical marker considered might progress, relative to the others. We then compared how participants classified at baseline as CN, subjective memory complaint, early or late MCI, and dementia compared, finding the expected ordering. ¹⁶ Further analyses compared progression of biomarkers versus baseline values and their associations with cognitive decline and found that biomarker progression for most biomarkers explained more variability in cognitive decline than just a single assessment of the marker. ¹⁷ An analysis of self‐reported and informant‐reported Everyday Cognition (ECog), a measure added during ADNI‐GO/2, found that informant‐reported ECog scores were more strongly associated with objective measures of disease, such as neuropsychological test scores, hippocampal volume, CSF Aβ_1–42, and a summary measure of amyloid burden obtained from amyloid PET imaging. ¹⁸

The ADNI datasets present special challenges to traditional statistical methods because of the very large number of potential variables, either as predictors or as outcomes. Machine learning is designed to detect patterns and structures in very high‐dimensional data such as ADNI biomarkers, imaging, or genomics data. Another series of papers, in collaboration with the Biostatistics Core, used machine learning methods to develop prediction models relevant to the study of AD. The first compared the performance of six different machine learning methods and four different feature sets in predicting AD versus CN. The best‐performing model was then applied to individuals with MCI to identify them as AD‐like or CN‐like based on baseline characteristics and found that the model had good specificity (> 75%), identifying those who remained stable (i.e., not progressing to AD) at 24, 36, and 48 months. ¹⁹ A second paper built a prediction model for amyloid status, based on less costly measures (demographics, apolipoprotein E [APOE] ε4 status, cognitive test scores) and more expensive or invasive procedures (MRI regional volumes and CSF markers), and found, not surprisingly, that models that included CSF biomarkers performed best. ²⁰ Finally, using an approach similar to the first paper (though different machine learning approaches), prediction models for AD versus CN were built using demographics, APOE ε4 status, and cognitive scores and the best‐performing models were applied to those with MCI to classify them as AD‐ or CN‐like. MCI participants were further categorized as having amyloid pathology (based on amyloid PET); being amyloid negative, but evidence of either neurodegeneration (based on hippocampal volume) or tau pathology (based on CSF p‐tau); or having normal biomarker levels. Those who were identified as AD‐like in general progressed to dementia faster than those identified as CN‐like. Further, within the AD‐like and CN‐like groups, those with evidence of amyloid pathology progressed the fastest. ²¹ A follow‐up paper from this group identified MCI and AD participants as biomarker‐negative (amyloid [A], tau [T], neurodegeneration [N]), amyloid‐negative (A–) but other biomarker‐positive (T+ and/or N+), or amyloid positive (A+) and then assessed associations between WMH volume and cognitive function in those subgroups. Increased WMH volume was associated with executive function only in the amyloid‐negative, other biomarker‐positive subgroup, but was not associated with memory in any of the subgroups. ²² The nature of machine learning approaches poses the risk of over‐fitting, so it is critical that ADNI findings be reproducible in other studies. The availability for validation of international studies designed in parallel with ADNI strengthens the appeal of these very powerful techniques.

More recently, using data from ADNI‐3, the Biostatistics Core supported an analysis of individuals with amyloid pathology, but low tau burden (A+T–) as characterized by tau‐ and amyloid PET imaging ²³ and found that low tau burden is not rare in those with amyloid pathology and among impaired individuals, compared to those with higher tau (A+T+), the A+T– group had more cerebrovascular risk factors.

A challenge across all of these works is the ability to compare performance across biomarkers. The Biostatistics Core has developed a standardized framework for the comparison of biomarkers, when multiple biomarkers are collected on the same individuals. Steps for comparison based on a variety of criteria including sensitivity to change and clinical utility are provided (see Harvey et al. in this special issue for more details ²⁴ ).

2.3. Statistical models for clinical and biomarker progression

Given the complexity of the multivariate data collected in ADNI, there was a need to develop methods to examine patterns across all modalities. The Biostatistics Core worked on methods to estimate the sigmoidal progression curves that Jack et al. ²⁵ hypothesized to represent the AD cascade. Early approaches to this problem used an alternating conditional expectations approach. ²⁶ Later approaches used hierarchical Bayesian methods and Markov chain Monte Carlo. ²⁷ Members of the Core were also involved in the development and validation of composite scores for memory ²⁸ and executive function ²⁹ derived from item‐level and overall test scores from the ADNI neuropsychological battery, using advanced psychometric approaches, which enabled researchers across the world to study specific cognitive domains.

2.4. Understanding clinical and biomarker heterogeneity

The Biomarker Core has developed methodology not only for characterizing trajectories of biomarkers, but also for detecting heterogeneity in baseline characteristics and patterns of progression among ADNI participants. Clustering CN individuals based on imaging and CSF biomarkers identified three groups of participants: two relatively healthy and one that, while CN at baseline, showed more rapid decline in Alzheimer's Disease Assessment Scale‐Cognitive subscale scores. ³⁰ A similar analysis in participants who had MCI at baseline found four clusters; one was similar to CN participants and rarely progressed, while the other three had different pathology profiles but were all likely to progress to dementia. ³¹ These works highlighted how biomarkers could be used to identify subgroups with different progression patterns and that although efforts were made in ADNI to enroll an MCI group that was on the path to AD, other pathologies are also likely represented. Recent work by Filshtein, in this current issue, developed new methods to identify subgroups for whom the sequence of biomarker progression was different, based on distances between trajectories rather than age. ³²

2.5. Support for clinical trial designs

Analyses using data from the first few years of ADNI‐1 already demonstrated that measurable change in both PET and MRI imaging and CSF bioassay measurements occurs well before the formal diagnosis of dementia. We presented sample size calculations for hypothetical 12‐month clinical trials using such biomarkers as outcome measures. We also considered study designs using cognitive test scores as outcome measures but with inclusion criteria based on biomarkers and showed the feasibility of such targeted studies with practical sample sizes. ³³ Other papers, in collaboration with the Biostatistics Core, evaluated improvements in power (i.e., reduced sample sizes) using a statistically defined region of interest based on TBM ³⁴ , ³⁵ , ³⁶ or voxel‐based analysis of FDG PET, ³⁷ indicating the value of having sensitive measures of change. A mathematical and statistical investigation into sample size calculations based on only two assessments illustrated the potential bias in the estimated sample sizes depending on the length of the clinical trial. ³⁸

Work from the Biostatistics Core supported the development of the first trials in preclinical, or presymptomatic, AD. ADNI data were used to demonstrate (1) the relative inefficiency of time‐to‐event analyses compared to analyses of longitudinal continuous outcomes, ³⁹ (2) the feasibility of a Preclinical AD trial using a Preclinical Alzheimer's Cognitive Composite, ⁴⁰ (3) the risk associated with amyloid in the Preclinical AD population ⁴¹ and (4) improvements in power and robustness to coronavirus disease (COVID‐19) interruptions by using natural cubic splines. ⁴² Other researchers have demonstrated that progression models derived from ADNI data can in fact perform well in placebo data from existing clinical trials and potentially improve clinical trial design. ⁴³ Clinical trial applications continue to be a key focus for ADNI and the Biostatistics Core. ⁴⁴

2.6. Enhancing use of ADNI data in related studies and by other users

ADNI has played a leading role as a pioneer in open science by sharing data in real time. The complexity of the data, however, can pose challenges to investigators outside the ADNI team. The Biostatistics Core has made substantial contributions to accessibility through its online documentation and support and through direct collaborations with partner projects. Dr. Donohue's popular ADNIMERGE R package (https://adni.bitbucket.io/) provides users with access to the ADNI datafiles (once approved for Laboratory of Neuro Imaging [LONI] access) as well as a curated longitudinal dataset comprising key demographic and clinical variables and imaging and fluid biomarker data, which is also available as a single downloadable file from LONI. The Core supports a Google Group dedicated to ADNI data (https://groups.google.com/g/adni‐data) that enables members to ask questions about the data and their use in research; there are currently > 550 members and > 450 conversations. Members of the Biostatistics Core are answering questions as they come in, often 5 to 10 per week.

One challenge for incorporating MRI results from ADNI, especially for investigators wanting to compare proposed image‐processing techniques, is the multiple images available. A standard dataset for comparison was proposed by Wyman et al. ⁴⁵ for MRI acquired in ADNI‐1, facilitating comparisons by other researchers. Dr. Harvey, as the Core's imaging expert, took part in this effort in collaboration with the MRI Core and the Informatics Core, to establish lists of participants and corresponding image collections (of good quality images with no major scanner changes in a given sequence of visits for individuals) for different sets of assessment visits (such as individuals with all images during the first year of the study [baseline, month 6, month 12]).

The Biostatistics Core helped to design the Japanese Alzheimer's Disease Neuroimaging Initiative (J‐ADNI) parallel study, working closely with Dr. Iwatsubo and his team. The first comparison of baseline enrollees in J‐ADNI found very similar profiles of decline among amyloid‐positive participants with MCI in the two studies, but milder disease in those with dementia. These results supported the possibility of international trials in the earlier stages. ⁴⁶

Several studies of early‐onset AD have been designed to make use of ADNI data to compare disease progression to late‐onset AD, and to explore the possibility that clinical trials can use similar study designs and biomarkers. The Dominantly Inherited Alzheimer Network (DIAN) faced statistical challenges in comparing the trajectories of autosomal dominant AD to those of late‐onset AD, because age was no longer a viable anchor as the metric for “time.” Dr. Beckett and the DIAN statistical team developed and validated a metric based on time of clinical disease onset that has enabled comparisons of level of pathophysiology at clinical onset, and rates of change prior to and after the clinical onset. ⁴⁷ These methods are being extended to the Longitudinal Early‐Onset Alzheimer's Disease Study (LEADS), where Dr. Beckett is a consultant on comparisons to ADNI and DIAN. ⁴⁸

In addition, the ADNI infrastructure enabled the study of traumatic brain injury (TBI) and post‐traumatic stress disorder (PTSD) and their association with AD pathology in Vietnam veterans, collecting many of the same imaging and CSF measures as ADNI on the participants. ⁴⁹ Dr. Harvey led the statistics and worked along other ADNI Core leaders to both design and implement the study. Results from the study found no significant association between TBI or PTSD and AD. ⁵⁰ , ⁵¹

The public access to ADNI has made it popular for education as well as research. The Biostatistics Core has helped to foster this access for the next generation of researchers. Masters and doctoral students in statistics, biostatistics, epidemiology, neuroscience, and other fields, at University of California, Davis; University of California, San Francisco; and University of Southern California, have made use of ADNI data as part of their dissertation or thesis research, with support from Biostatistics Core leaders as advisors and committee members. Online training sessions during ADNI‐1 and our online documentation since then have helped to make ADNI data available to students and post‐doctoral fellows nationally and internationally, bringing new researchers into the AD field.

3. SUMMARY AND FUTURE GOALS

The Biostatistics Core has helped to shape the design of ADNI studies for 20 years. Core members played a key role in analyses addressing questions about measurement of biomarkers, their changes before and during disease progression, their relationship to each other and to clinical progression, and their potential to improve clinical trial design. ADNI‐4 will add new complexities with its dynamic recruitment plan, which may call for updates in study design as enrollment and follow‐up proceed. Our previous work sets an initial template for analysis of the behavior of new biomarkers. A key role for the Biostatistics Core moving forward will be to ensure that analyses give insight into how well our past work generalizes into this new, more diverse study population, and how clinical trials can best be designed to address the challenges of AD and related dementias in the wider population, especially those who have had little opportunity for participation in AD research in the past. In addition, a key focus of the Biostatistics Core will be contributing to the harmonization of the data across all of the phases of ADNI. We are excited to continue to be part of ADNI in its third decade.

CONFLICT OF INTEREST STATEMENT

Dr. Beckett, Dr. Harvey, Dr. Donohue, and Ms. Saito receive support from research grants from NIH. Dr. Harvey serves as a Statistical Advisor for PLOS ONE. Author disclosures are available in the supporting information.

Supporting information

Supporting Information

Beckett LA, Saito N, Donohue MC, Harvey DJ. Contributions of the ADNI Biostatistics Core. Alzheimer's Dement. 2024;20:7331–7339. 10.1002/alz.14159