Federated learning is a technique that enables the use of distributed datasets for machine learning purposes without requiring data to be pooled, thereby better preserving privacy and ownership of the data. While supervised FL research has grown substantially over the last years, unsupervised FL methods remain scarce. This work introduces an algorithm which implements K-means clustering in a federated manner, addressing the challenges of varying number of clusters between centers, as well as convergence on less separable datasets.

Background
Aggregation of cohort data increases precision for studying neurodegenerative disease pathways, but efforts to combine data and expertise are often hampered by infrastructural, ethical and legal considerations. We aimed to unite various cohort studies in the Netherlands to enhance research infrastructure and facilitate research on dementia etiology and its public health implications.

Methods
The Netherlands Consortium of Dementia Cohorts (NCDC) includes participants with initially no established cognitive impairment from 9 Dutch cohorts: the Amsterdam Dementia Cohort (ADC), Doetinchem Cohort Study (DCS), European Medical Information Framework for Alzheimer’s Disease (EMIF-AD), Longitudinal Aging Study Amsterdam (LASA), the Leiden Longevity Study (LLS), The Maastricht Study, the Memolife substudy of the Lifelines cohort, Rotterdam Study and Second Manifestations of ARTerial disease-Magnetic Resonance (SMART-MR) study. The objectives of NCDC are to improve data infrastructure and access to cohorts related to aging and dementia, investigate the role of Alzheimer’s disease and vascular pathology in the development of dementia and estimate the public health impact of established dementia risk factors by assessing their relative contribution to the population burden of dementia.

Results
We increased the findability, accessibility, interoperability and reusability (FAIR) status of the cohorts through harmonization of data across cohorts, implementation of medical imaging repositories for scan management, implementation of the Personal Health Train infrastructure and provision of meta-data in existing cohort catalogues. We established the ethical and legal frameworks required for federated and pooled analyses and performed the first remote federated data analyses using the Personal Health Train infrastructure. To determine biomarkers of Alzheimer’s disease, endothelial dysfunction and inflammation, 2554 plasma samples were analyzed centrally. Federated, pooled, and coordinated meta-analyses have led to multiple publications in the context of NCDC.

Conclusion
The combination of population-based and clinical cohorts, the coordinated assessment of plasma markers in previously collected samples and implementation and use of the Personal Health Train infrastructure for federated analysis are both feasible and promising for future collaborative efforts.

Federated learning is an upcoming machine learning paradigm which allows data from multiple sources to be used for training of classifiers without the data leaving the source it originally resides. This can be highly valuable for use cases such as medical research, where gathering data at a central location can be quite complicated due to privacy and legal concerns of the data. In such cases, federated learning has the potential to vastly speed up the research cycle. Although federated and central learning have been compared from a theoretical perspective, an extensive experimental comparison of performances and learning behavior still lacks. We have performed a comprehensive experimental comparison between federated and centralized learning. We evaluated various classifiers on various datasets exploring influences of different sample distributions as well as different class distributions across the clients. The results show similar performances under a wide variety of settings between the federated and central learning strategies. Federated learning is able to deal with various imbalances in the data distributions. It is sensitive to batch effects between different datasets when they coincide with location, similar to central learning, but this setting might go unobserved more easily. Federated learning seems to be robust to various challenges such as skewed data distributions, high data dimensionality, multiclass problems, and complex models. Taken together, the insights from our comparison gives much promise for applying federated learning as an alternative to sharing data. Code for reproducing the results in this work can be found at: https://github.com/swiergarst/FLComparison.