Project CANAIRI

Testing an algorithm in ‘silent mode’ is a period of algorithmic evaluation where the artificial intelligence (AI) tool is running on live or near-live data, making inferences about patients in real-time. These predictions are evaluated for accuracy and provide a test of the model’s performance in its intended clinical setting but without yet affecting patient care or institutional operations. Many believe it is a critical step to provide key assurances for the responsible and efficient integration of machine learning (ML) tools into clinical and administrative settings.

CANAIRI, or the Collaboration for trANslational AI tRIals (CANAIRI) project, will develop consensus-based standards of best practices and key capabilities for conducting clinical trials. The group advocates for a widening of these ‘silent’ trials toward one that is sociotechnical in nature and wants to draw attention to the importance of this critical testing stage for AI in healthcare.

As a first step, the group uses the term ‘translational trial’ to widen the scope and practices of silent trials to include human factors, implementation science, operational/systems integration, social license, legal and ethical, economics, environmental, and regulatory considerations. It calls these ‘translational trials’ to emphasise that when technology is translated from bench to bedside, it needs to be tested to ensure that it is as closely mapped as possible to what the bedside looks like and what the needs are.

Project CANAIRI will undertake an international consensus-generating methodology to identify current practices for translational/silent evaluations, develop guidance for health settings, and generate knowledge products for a variety of stakeholders. This project is the first of its kind and looks to fill a critical gap in current AI translation frameworks.

History and collaborators

CANAIRI is spearheaded by AIML Deputy Director, Dr Melissa McCradden who helped develop the first ethical framework for AI translation. The framework included a silent testing phase and focused on patient outcomes with other CANAIRI members in 2020.

“The genesis of this idea really came about when I first started my postdoc [research] with Anna Goldenberg,” said Dr McCradden. Dr Goldenberg is the Varma Family Chair in Biomedical Informatics and Artificial Intelligence at The Hospital for Sick Children (SickKids).

 “She was the first to publicise the importance of silent trials [1] and I was really fascinated by this idea, because there really is no equivalent to it in traditional research,” said Dr McCradden.

“My experience at SickKids has shown me that the silent trial is the 'algorithm graveyard. If instead we have an interim testing phase to vet the algorithms that won't work, we can focus translation efforts on those that will work in a basic sense in the real clinical setting,” she continued.

“So, building out the requirements for this phase is critical to managing expectations and controlling hype in AI so that we don't end up with a collapse of AI optimism down the line.”

Among the group’s members and co-authors are AIML Senior Research Fellow Dr Lauren Oakden-Rayner, University of Adelaide Professor Carolyn Semmler, AIML Professor Lyle Palmer, and PhD student Lana Tikhomirov. CANAIRI is also partnering with the Aboriginal Health Unit at Women’s and Children’s Health Network [in Adelaide] to ensure that their voices continue to guide CANAIRI’s work towards an outcome that is ethically defensible and resonates with the communities who are most impacted by AI.

Relevant AIML news articles

AI’s canary in the coal mine: Translational trials for accountable AI integration.

Other related pieces

CANAIRI: the Collaboration for Translational Artificial Intelligence Trials in healthcare. Nature Medicine.

AI's Canary in the Coal Mine: The Ethical Significance of Translational Trials in Healthcare ML. YouTube.