How Holland Bloorview is understanding medication use with federated learning

Holland Bloorview logo
Doctor doing exam on child


Neurodevelopmental conditions, including autism and attention-deficit/hyperactivity disorder (ADHD), can be associated with significant distress and disability in children. Neurodevelopmental conditions are the largest identifiable subpopulation of children with disabilities and account for up to 14% of all children. The diagnostic labels of autism and ADHD in particular are broad and often lack precision to guide care. This creates a unique challenge when prescribing medications, which can significantly improve outcomes, as there is a lack of markers to guide individualized medication choice. This often results in clinicians prescribing neurodivergent children medications through a trial and error approach. The Bloorview Research Institute is housed in Holland Bloorview Kids Rehabilitation Hospital, Canada’s largest pediatric rehabilitation hospital recognized globally for its unique client population and leadership in childhood care.


The Bloorview Research Institute wanted to explore innovative ways to pilot the analysis of demographic, phenotypic, and medication data from the POND Network. They wanted a cloud-based solution that can analyse datasets from collaborating hospitals and universities with the goal of federating insights on biases in medication prescription patterns based on demographic identities, as well applying machine learning techniques to predict medication use in children with neurodevelopmental conditions.


The Bloorview Research Institute leveraged Omics AI to connect disparate datasets from POND partners and analyse it through Workbench. The solution uses Publisher to connect data and Workbench to process it. Machine learning services were also used to support the creation of novel machine learning algorithms.


The Bloorview Research Institute partnered with DNAstack and to help create a modular software solution to analyse demographic, phenotypic, and medication data collected through POND. The resulting insights confirmed findings surfaced in previous scientific publications, as well as generated new hypotheses related to potential biases in medication use based on a child’s ethnicity and family income status. This represents the first machine learning model to be trained across a GA4GH compliant federated network, and has generated significant interest from collaborators to grow the effort to help answer questions that can improve precision healthcare in children.


Samples analysed


Sites networked


Insights generated
The ability to explore multiple datasets from different jurisdictions will give us both the volume and diversity of data to have the power to make transformative discoveries for the lives of kids, youth, and adults with developmental differences and disabilities. Crucially, doing so without having to move the data from local environments will enable us to respect the requests of various communities to protect their privacy and confidentiality.
Dr. Evdokia Anagnostou
Vice President of Research

Ready to Get Started?

Accelerate your science with Omics AI.

Have questions? Contact us