The Evidence Gap: Bia Health's AI Ambitions and the Risk of the Black Box
AI-generated image · Bay Street Wire
A Québec startup claims its AI assistant can slash unnecessary medical interventions and save millions, but the path from venture funding to clinical validation remains opaque.
In the high-stakes arena of health tech, the promise of reducing 'low-value care' is a powerful siren song. For a healthcare system under immense pressure, the idea of an AI assistant that can steer clinicians away from unnecessary imaging and redundant interventions is not just a fiscal win—it is a systemic necessity. However, as a tech columnist who has seen countless 'revolutionary' tools vanish into the ether of vaporware, I find the current pitch from Bia Health to be missing the most critical component of medical innovation: peer-reviewed clinical data.
As BetaKit first reported, the Québec City-based startup Bia Health recently announced it has raised $2.15 million CAD to scale its 'Bia Assistant.' The funding consists of $1.4 million in debt provided by Montréal-based Finalta Capital, with equity led by Anges Québec and participation from Desjardins via its Startup en résidence program. The goal, as stated by Bia Health CEO and co-founder Anne-Sophie Paquet in an email to BetaKit, is to help clinicians make more appropriate decisions for patients, thereby reducing unnecessary and low-value care—which Paquet estimates represents roughly 30 percent of healthcare in Canada.
On paper, the logic is sound. Paquet, who is a physiotherapist, told BetaKit that the company was founded in 2016 by healthcare professionals to make clinical knowledge more accessible through medical content libraries and e-learning. Over time, the founders realized the core challenge was not the availability of evidence, but the application of that evidence at the point of care. The Bia Assistant is designed to address this by providing 'evidence-based recommendations' to help clinicians determine diagnoses based on a library of scientific and medical knowledge.
Bia Health claims that its technology can identify tests and interventions that are not in line with evidence-based guidelines or are unlikely to improve patient outcomes. The stakes are high: the company suggests that wide adoption of its tool across primary care could save the province of Québec up to $100 million annually. This focus on waste is not unique to Canada; BetaKit notes a study from the Journal of the American College of Radiology indicating that unnecessary medical imaging costs the U.S. medical system approximately $12 billion USD ($17 billion CAD) every year.
Yet, here is where the evidence-first approach meets a wall of ambiguity. While Bia Health asserts that its tool provides 'evidence-based recommendations,' there is a profound difference between a tool that references a library of medical knowledge and a tool that has been clinically validated to improve patient outcomes. To date, there is no mention of peer-reviewed studies or published clinical trial data proving that the Bia Assistant actually reduces unnecessary interventions without compromising patient safety.
Without this data, the Bia Assistant remains a 'black box.' We are asked to trust that the AI's interpretation of medical guidelines is accurate and that its recommendations are superior to the existing clinical judgment of the providers using it. While Paquet told BetaKit that the tool is not intended to replace clinical judgment, the very purpose of the software is to influence that judgment. If the AI steers a doctor away from a diagnostic image that would have caught a rare but critical pathology, the 'cost saving' becomes a medical failure.
Bia Health is currently attempting to move from theory to practice. The company has entered into an agreement with Santé Québec to conduct a six-month test in the Capitale-Nationale – Universitaire region, involving 100 clinicians. This trial specifically focuses on musculoskeletal care—a category including bone fractures and osteoarthritis. Paquet told BetaKit this focus is intended to create a 'strong foothold' before the company expands into other diagnostic areas.
From a business perspective, Bia Health is not a pre-revenue gamble. Paquet informed BetaKit that the enterprise is already established, bringing in $500,000 in annual recurring revenue. However, revenue is not a proxy for clinical efficacy. The company's ambition to target the larger U.S. market will require more than just a successful six-month pilot in one Québec region; it will require rigorous, transparent data that proves the AI does not trade patient safety for budgetary efficiency.
***
**Opinion:** In my view, the danger of AI in the clinic is the illusion of certainty. When a software interface presents a recommendation as 'evidence-based,' it can create a cognitive bias that discourages clinicians from questioning the output. Until Bia Health submits its results to the scrutiny of the scientific community through peer review, the Bia Assistant is less of a medical breakthrough and more of a sophisticated search engine for clinical guidelines. Noble intentions to reduce waste are commendable, but in medicine, the only currency that matters is validated data. Until then, we are simply trusting a black box to play doctor.