AI model helps kids get the right medication sooner

Choosing the right medication and dose for each child isn’t always simple. Sometimes it requires trial and error, and side effects we’d rather avoid. In a lot of cases, genetic make-up matters — a child’s genes can influence how they respond to a medication or what side effects are more likely.

Now, a machine learning model developed in-house at The Hospital for Sick Children (SickKids) is helping kids get the right medication at the right dose sooner. 

Pharmacogenomic (PGx) testing is a tool that looks at a patient’s genes to uncover why a medication may or may not be working. When used in the right context, it helps clinicians choose the right drug, at the right dose, for each child — an example of Precision Child Health in action.

The challenge is knowing who could benefit and when. 

“Typically, clinicians would recognize that a patient might benefit from PGx testing and engage the PGx team only when they’re about to prescribe a medication,” explains Dr. Adam Yan, Oncologist and Associate Chief Medical Information Officer (Academic Pediatrics). “But by then, the patient needs the medication right away, and it could take weeks before testing results return, and adjustments are made.” 

The model helps predict who could benefit from PGx testing up to three months before certain medications are prescribed, ensuring patients, families and clinicians can make informed decisions about treatment plans right from the start.  

Where pharmacogenomic and machine learning expertise meet

Both Dr. Yan and Dr. Lillian Sung, Oncologist and Chief Clinical Data Scientist, have worked with the pharmacogenetic team in their own clinical roles, helping cancer patients access and benefit from PGx testing where appropriate.

They’ve also spent the last few years establishing Pediatric Real-world Evaluative Data sciences for Clinical Transformation (PREDICT): an initiative at SickKids to develop, deploy, evaluate, and maintain machine learning models aimed at improving outcomes for patients and families. PREDICT models use data from a cleaned and validated hospital-wide data source called the SickKids Enterprise-wide Data in Azure Repository (SEDAR), purpose-built to support clinical, research and machine learning applications.

Together they saw an opportunity: maybe machine learning can help us better utilize PGx testing by alerting us to consider it earlier.

They brought the idea to Iris Cohn, Director of SickKids’ Pharmacogenomics Program and Dr. Ruud Verstegen, Staff Physician in the Division of Clinical Pharmacology & Toxicology, to collaborate on a model. Since the program was first established in 2017, around 1,500 children have received PGx testing with the support of the SickKids Foundation. Historically, most of those children were tested after multiple treatments had failed or unwanted side effects appeared, when opportunities to prevent those outcomes had already been missed.

Cohn and Verstegen saw the potential in the model to change this, helping identify nine specific medications where earlier testing could help tailor dosing more precisely and reduce the risk of treatment failure and/or severe side effects.

Using retrospective (past) data, they developed a machine learning model that could accurately predict who would receive these medications within three months of admission. Next, they validated the accuracy of the model through a silent trial, meaning they ran the model in the clinical environment without it influencing patient care. Proving its success, they were ready to implement it into daily practice. 

"It’s important with these medications, because of potential side effects, that you dose more precisely and within a specific window of treatment,” says Cohn. “With this algorithm, we can get the information we want at the right time.”

According to Cohn, with pharmacogenomics being a relatively new field used in clinical practice, the model serves as a nudge, reminding clinicians that PGx testing could be an option for their patients.

“It’s all about visibility,” she says. “The alert is right there in front of you. It's already changing the culture around awareness and access to PGx testing.”

The model is now actively deployed in Oncology, Nephrology, Gastroenterology, Bone Marrow Transplant and Neurology to help inform when to offer testing.

It’s part of a growing ecosystem at SickKids focused on developing, deploying and scaling responsible AI solutions. That ecosystem includes programs like SickKids AI (SKAI), which provides the enterprise-wide expertise and oversight needed to ensure AI tools are developed responsibly and in ways families and clinicians can trust.

Since the model began clinical use in July 2025, it’s helped identify almost 80 patients who could benefit from PGx testing. Without, it’s expected that only one per cent of those patients would have received testing. The team has gradually expanded its reach to other units as capacity and access to PGx testing grows, with the goal of scaling the approach across SickKids to enable more individualized care.

“This is how you deliver Precision Child Health,” says Sung. “You do it at scale, for all who can benefit.”