AI in Healthcare: The Evidence Is More Complicated Than the Headlines Suggest

The pace of AI adoption in healthcare is accelerating faster than the evidence base supporting it. A 2025 American Medical Association survey found that 66% of physicians are already using AI health tools, up from 38% in 2023, and 68% believe AI positively contributes to patient care in some way. At the same time, substantial clinical caution remains: many physicians worry about AI influencing diagnosis and treatment decisions, fearing errors, bias, and misuse.

Both the enthusiasm and the caution are justified by the evidence. What is not justified is the tendency in health system planning to treat AI as either a transformative solution to every system challenge or a threat to professional clinical judgment that must be resisted. The actual picture is considerably more nuanced, and understanding that nuance matters for the significant investments BC's health system is being asked to make in digital transformation.

Where the Evidence Is Strong

AI is delivering measurable, replicable value in specific, bounded applications. AI-powered imaging tools trained on large radiological databases have demonstrated exceptional accuracy in identifying abnormalities in cancer, stroke, and lung disease detection, in some cases surpassing human performance and enabling earlier interventions. In medication adherence, a pharmacist-led, AI-supported program demonstrated cost savings of 31% for hypertension, 25% for hyperlipidemia, and 32% for diabetes through improved adherence, a significant return on a targeted, workflow-integrated intervention.

Administrative AI applications show consistent efficiency gains. AI tools that reduce documentation burden, streamline prior authorization, and support scheduling optimization free clinician time for patient care without requiring the same evidence threshold as clinical decision-making applications. In a health system where administrative burden is one of the primary drivers of physician burnout and inefficiency, these applications represent high-value, lower-risk investments.

Where the Evidence Is More Complicated

Clinical decision support AI is where the evidence becomes significantly less straightforward, and where the gap between product claims and clinical reality is most consequential for health system leaders.

A September 2025 scoping review found that AI-based clinical decision support does not reliably improve clinicians' decision-making performance, even when the AI performs well on the underlying diagnostic task. One study cited in the review found that providing physicians with a large language model assistant did not improve diagnostic reasoning performance, even though the AI model outperformed physicians on the same task in isolation. Another found that AI decision support for chest X-ray diagnosis did not improve on clinicians' diagnostic performance in practice.

This counterintuitive finding has an explanation. The effectiveness of AI clinical decision support is strongly shaped by human-computer interaction factors, how the recommendation is presented, how much clinical context is provided, whether the clinician's workload allows for meaningful engagement with the AI output, and whether the tool is designed around actual clinical workflow or imposed on top of it. An accurate AI recommendation that appears at the wrong moment in a workflow, is poorly explained, or conflicts with a clinician's clinical intuition without adequate justification, does not improve care. It creates friction.

What BC's Digital Health Strategy Needs to Reckon With

Budget 2026 removed digital initiatives from the BC health care budget and moved this line item to the Ministry of Citizens Services, a structural signal that digital health is being repositioned as an IT function rather than a clinical transformation priority. The implications depend entirely on what governance and clinical integration frameworks accompany that shift.

The OECD's 2024 rethinking of health system performance assessment explicitly incorporates digital health infrastructure and information systems as core dimensions of sustainable system capacity. Health systems that treat digital transformation as an IT procurement exercise rather than a clinical redesign challenge consistently underperform those that embed digital tools into clinical workflow redesign from the beginning, with frontline clinician co-design and rigorous outcome evaluation.

A value-based healthcare framework, which focuses on maximizing health outcomes relative to the cost of care, rather than on volume of services, provides the most useful evaluative lens for AI investments. The question for every AI investment is not "does this technology work in a controlled study?" It is "does this technology improve health outcomes per dollar spent, in the actual workflows of BC clinicians, for the populations BC is trying to serve?" That question requires evidence, co-design, evaluation infrastructure, and the organizational humility to stop using tools that are not delivering the outcomes they promised.

The AI opportunity in BC's health system is real. So is the risk of investing in tools that impress in demonstrations and underperform in practice. The difference is in how rigorously that question is asked, and whether the answers change the investment decisions.

IHSTS supports evidence-based digital health integration within BC's health system, with a focus on clinical co-design, equity, and outcome measurement. Learn more at ihsts.org.