Artificial intelligence is fast becoming the digital front door of modern medicine. In the face of overwhelming patient demand, healthcare systems are rapidly deploying AI-enabled triage tools to sort symptoms, assess risk, and route patients before they ever step into a clinic.
Key Takeaways |
|
Prioritize Digital Maturity over Shiny Toys: Policymakers and healthcare leaders often neglect backend standards in favor of flashy, proprietary software. True digital transformation requires achieving operational maturity through frameworks like the Digital Health Profile and Maturity Assessment Toolkit (DHPMAT), with standardization serving as the critical keystone for safe system interoperability. |
| Acknowledge the Complexity of Clinical Routing: Triage is not a simple mathematical equation; it is a nuanced biopsychosocial process. AI models built in a vacuum fail to capture the complex probability diagnoses and time-critical threats that human clinicians identify, necessitating the use of high-quality, localized data that reflects specific demographic and clinical realities. |
| Protect Public Health Equity: The aggressive monetization of medical vocabularies creates "equity blind spots" that marginalize resource-limited populations. To prevent automated systems from reinforcing existing healthcare disparities, stakeholders should favor open-source digital public goods and adopt the "Quintuple Aim" framework, which demands the lifecycle co-creation of tools by vendors, clinicians, and consumers. |
However, introducing automated clinical sorting into a fractured digital environment poses an immediate threat to patient safety and global health equity. In a powerful commentary published in the Journal of Medical Internet Research, Emeritus Professor Siaw-Teng Liaw from UNSW Sydney responds directly to a foundational study by Alamoudi et al regarding the real-world safety of automated clinical routing. Liaw argues that safe automation requires healthcare networks to look past flashy, proprietary software and invest heavily in the fundamental digital maturity of the entire medical ecosystem.
AI triage applications collect patient-reported data to determine the urgency of care, recommending paths ranging from simple self-care to immediate emergency transfers.
The danger lies in treating triage as a simple mathematical equation. Alamoudi et al note that while these systems show high technical accuracy in controlled hospital tests, 0% of published data evaluates AI triage using real-world patient data in routine general practice. In family medicine, clinical routing depends on a deeply nuanced biopsychosocial diagnostic process. Human clinicians do not just read checklists; they build probability diagnoses, rule out time-critical threats, navigate overlapping chronic conditions, and uncover hidden patient agendas. To match this complex reality safely, AI models cannot be built in a vacuum. They require high-quality, localized data that reflects the specific cultural values, clinical baselines, and demographic priorities of the actual communities they serve.
Deploying an advanced machine learning algorithm into a weak digital environment often backfires. To map out an organization's actual capacity to manage automated systems, Liaw advocates for the use of the Digital Health Profile and Maturity Assessment Toolkit (DHPMAT).
The toolkit evaluates readiness by measuring five core digital foundations, including infrastructure, platform data sharing, and quality improvement frameworks, across a strict five-stage evolutionary scale:
Stage 1: Adapting & Assessing – Initial exploration, baseline profiling, and early pilot adjustments.
Stage 2: Controlling – Implementing basic risk management, security constraints, and data oversight.
Stage 3: Standardizing – Harmonizing data, systems, and terminologies across the enterprise architecture.
Stage 4: Optimizing – Refining workflows, maximizing tool performance, and expanding data integration.
Stage 5: Innovating – Deploying advanced, predictive systems like automated AI triage safely at scale.
Liaw highlights Standardization as the absolute keystone of this progression. Without well-implemented data standards, true interoperability is impossible.
Unfortunately, Liaw argues, policymakers and healthcare executives frequently try to skip vital developmental steps. Driven by political incentives that favor highly visible innovations over unglamorous backend standards, decision-makers consistently prioritize purchasing proprietary, shiny new toys. This bias toward commercial novelty, combined with a highly competitive market of closed systems, causes leadership to neglect the critical upstream standardization work required to build a shared enterprise architecture and make health systems talk to one another.
| In this video, Emeritus Professor Siaw-Teng Liaw from UNSW Sydney addresses a critical tension in modern digital health transformation: the rapid deployment of artificial intelligence (AI)–enabled primary care triage systems despite a low global evidence base for real-world safety, bias control, and quality. |
The Private Good Versus Public Good Conflict
A major bottleneck to global health equity is the aggressive monetization of medical vocabularies and infrastructure. Alamoudi et al warn of an equity blind spot where aggregated data hides severe performance gaps for vulnerable populations. When essential diagnostic terminologies are commodified and sold as proprietary servers, they introduce artificial financial and access barriers for resource-limited areas.
By contrast, open-source digital public goods, such as the World Health Organization’s (WHO) ICD-11 system, offer accessible alternatives that promote global semantic consistency without restrictive corporate licensing.
This tension is especially visible in the underlying business ethics of big tech conglomerates. Corporations have scraped massive troves of personal and public internet data without formal consent or compensation to build massive, proprietary models.
This corporate concentration can widen global health divides during public health crises. As witnessed during the COVID-19 pandemic, the manipulation of digital assets by profit-driven entities directly contributed to vaccine, antiviral, and therapeutic distribution disparities. This systemic inequality disproportionately impacts marginalized populations living on the wrong side of socioeconomic, age, and racial divides.
Moving Toward Sustainable AI Governance
To prevent automated triage from reinforcing existing healthcare disparities, Liaw says that automated tools must be evaluated through the lens of the Quintuple Aim. This framework insists that digital health transformations optimize five interconnected factors symmetrically:
Achieving this standard requires a complete shift toward the lifecycle cocreation of digital tools. Technology vendors, clinicians, and consumers must work as equal partners to develop, implement, monitor, and continuously improve AI assets. Moving forward, health systems must anchor automated clinical routing in transparent, fair regulatory frameworks that protect local social capital and guarantee access to high-quality, interoperable data.
Why JMIR?
The author chose the Journal of Medical Internet Research to share these insights due to the journal's leadership in clinical informatics, e-health policy, and digital divide research. As healthcare systems globally undergo digital transformations, this analysis provides a clear governance framework to ensure early-stage AI evaluations yield accurate, actionable data for technology integration.
Curious to see how digital maturity and equity tracking are reshaping the future of healthcare integration? Watch the video featuring Siaw-Teng Liaw and read the full commentary article to explore the ecosystem-informed framework and the strategic roadmap for responsible AI adoption in primary care.