Alzheimer disease (AD) is the most common cause of dementia, with a worldwide prevalence of about 25 million in 2010, expected to be doubled by 2030 because of increased life expectancy.¹ The earliest recognizable pathological event in AD is cerebral amyloid-β aggregation.² This pathology may be present up to 20 years before the onset of dementia.^3,4 Novel research criteria for AD in individuals without dementia emphasize the presence of amyloid pathology to define the first stage of the disease.^5,6

Importance One characteristic histopathological event in Alzheimer disease (AD) is cerebral amyloid aggregation, which can be detected by biomarkers in cerebrospinal fluid (CSF) and on positron emission tomography (PET) scans. Prevalence estimates of amyloid pathology are important for health care planning and clinical trial design. Objective To estimate the prevalence of amyloid abnormality in persons with normal cognition, subjective cognitive decline, mild cognitive impairment, or clinical AD dementia and to examine the potential implications of cutoff methods, biomarker modality (CSF or PET), age, sex,APOEgenotype, educational level, geographical region, and dementia severity for these estimates. Design, Setting, and Participants This cross-sectional, individual-participant pooled study included participants from 85 Amyloid Biomarker Study cohorts. Data collection was performed from January 1, 2013, to December 31, 2020. Participants had normal cognition, subjective cognitive decline, mild cognitive impairment, or clinical AD dementia. Normal cognition and subjective cognitive decline were defined by normal scores on cognitive tests, with the presence of cognitive complaints defining subjective cognitive decline. Mild cognitive impairment and clinical AD dementia were diagnosed according to published criteria. Exposures Alzheimer disease biomarkers detected on PET or in CSF. Main Outcomes and Measures Amyloid measurements were dichotomized as normal or abnormal using cohort-provided cutoffs for CSF or PET or by visual reading for PET. Adjusted data-driven cutoffs for abnormal amyloid were calculated using gaussian mixture modeling. Prevalence of amyloid abnormality was estimated according to age, sex, cognitive status, biomarker modality,APOEcarrier status, educational level, geographical location, and dementia severity using generalized estimating equations. Results Among the 19 097 participants (mean [SD] age, 69.1 [9.8] years; 10 148 women [53.1%]) included, 10 139 (53.1%) underwent an amyloid PET scan and 8958 (46.9%) had an amyloid CSF measurement. Using cohort-provided cutoffs, amyloid abnormality prevalences were similar to 2015 estimates for individuals without dementia and were similar across PET- and CSF-based estimates (24%; 95% CI, 21%-28%) in participants with normal cognition, 27% (95% CI, 21%-33%) in participants with subjective cognitive decline, and 51% (95% CI, 46%-56%) in participants with mild cognitive impairment, whereas for clinical AD dementia the estimates were higher for PET than CSF (87% vs 79%; mean difference, 8%; 95% CI, 0%-16%;P = .04). Gaussian mixture modeling–based cutoffs for amyloid measures on PET scans were similar to cohort-provided cutoffs and were not adjusted. Adjusted CSF cutoffs resulted in a 10% higher amyloid abnormality prevalence than PET-based estimates in persons with normal cognition (mean difference, 9%; 95% CI, 3%-15%;P = .004), subjective cognitive decline (9%; 95% CI, 3%-15%;P = .005), and mild cognitive impairment (10%; 95% CI, 3%-17%;P = .004), whereas the estimates were comparable in persons with clinical AD dementia (mean difference, 4%; 95% CI, −2% to 9%;P = .18). Conclusions and Relevance This study found that CSF-based estimates using adjusted data-driven cutoffs were up to 10% higher than PET-based estimates in people without dementia, whereas the results were similar among people with dementia. This finding suggests that preclinical and prodromal AD may be more prevalent than previously estimated, which has important implications for clinical trial recruitment strategies and health care planning policies.

Cerebral amyloid-β aggregation is an early pathologic event in Alzheimer disease (AD), starting 2 to 3 decades before dementia onset.^1,2 Approximately 25% of cognitively normal elderly individuals and 50% of patients with mild cognitive impairment (MCI) have biomarker evidence of amyloid pathology.^1,3 These persons are at increased risk for developing AD-type dementia,¹ but the extent to which amyloid-β aggregation is associated with cognition in persons without dementia is unclear. Understanding the association between amyloid pathology and cognitive functioning is important for a better understanding of the course of AD and for the design of AD-prevention trials.

Establishing collaborations between cohort studies has been fundamental for progress in health research. However, such collaborations are hampered by heterogeneous data representations across cohorts and legal constraints to data sharing. The first arises from a lack of consensus in standards of data collection and representation across cohort studies and is usually tackled by applying data harmonization processes. The second is increasingly important due to raised awareness for privacy protection and stricter regulations, such as the GDPR. Federated learning has emerged as a privacy-preserving alternative to transferring data between institutions through analyzing data in a decentralized manner. In this study, we set up a federated learning infrastructure for a consortium of nine Dutch cohorts with appropriate data available to the etiology of dementia, including an extract, transform, and load (ETL) pipeline for data harmonization. Additionally, we assessed the challenges of transforming and standardizing cohort data using the Observational Medical Outcomes Partnership (OMOP) common data model (CDM) and evaluated our tool in one of the cohorts employing federated algorithms. We successfully applied our ETL tool and observed a complete coverage of the cohorts' data by the OMOP CDM. The OMOP CDM facilitated the data representation and standardization, but we identified limitations for cohort-specific data fields and in the scope of the vocabularies available. Specific challenges arise in a multi-cohort federated collaboration due to technical constraints in local environments, data heterogeneity, and lack of direct access to the data. In this article, we describe the solutions to these challenges and limitations encountered in our study. Our study shows the potential of federated learning as a privacy-preserving solution for multi-cohort studies that enhance reproducibility and reuse of both data and analyses.