Causal AI to predict first-in-human clinical outcomes

Causal AI to predict first-in-human clinical outcomes from preclinical and real-world clinico-genomic data

Using AI and Canadian healthcare data to improve the success and safety of new cancer treatments

Many cancer drugs that show early promise often fail when tested in clinical trials, costing time, healthcare resources, and importantly, patients’ opportunities for better care. Our project aims to improve how new cancer therapies are developed by predicting how well they are likely to work in patients before clinical trial programs are launched.

We are developing an artificial intelligence (AI) system that learns from real-world cancer data collected in Canadian hospitals in British Columbia and Quebec. These data include information about patients’ cancers, treatments received, genetic profiles of tumors, and outcomes, such as mortality. All data remain securely stored at the original institutions and are anonymized to protect patient privacy.

To ensure data security, we use federated learning, a method where the AI is trained locally at each data site. This means patient data never leave the hospital or research center, and only encrypted model updates are shared. The AI system combines machine learning with causal modeling, which helps the AI understand not just patterns in the data, but why certain treatments work better for certain patients. We also incorporate information about the chemical and biological properties of drugs to help predict how new therapies may behave in humans.

This project focuses on cancer and precision medicine. By better predicting clinical trial outcomes, we can design safer, more efficient trials, reduce trial failures, and identify which patients are most likely to benefit from specific treatments. Ultimately, this could speed up access to new effective therapies, reduce unnecessary side effects, and improve outcomes for cancer patients.

By the end of the project, we aim to deliver a validated AI platform that supports smarter clinical trial design and the ability to offer more personalized cancer treatment decisions. The technology developed in this project will form the foundation of Synograph, a clinical trial prediction software platform being developed by Onco-Innovations, to support real-world adoption of these methods in research and healthcare.

"Our project impacts the Canadian digital health landscape by breaking down so-called silos through privacy-preserving collaboration. By advancing the SynoGraph Causal AI framework, we are building a prototype to predict clinical outcomes from standardized clinico-genomic data. This initiative utilizes federated learning to keep raw data secure behind institutional firewalls while accelerating breakthroughs in precision oncology. Adhering to FAIR and OMOP standards, we are moving Canadian-led AI from the laboratory to real-world clinical application." - Alind Gupta (Inka Health, Onco-Innovations)