AI for primary care proficient in concept but sluggish in practice: Lit review

There’s no shortage of technically impressive AI applications for primary care. Many could ably assist providers in clinical diagnostics, chronic disease management and operational support. However, these tools tend to lag well behind AI models aimed at clinical specialties when it comes to integration into routine practice.

That’s among the important findings to shake out of a literature review conducted by researchers with the University of Debrecen and the Medical Futurist Institute, both in Hungary.  

Gellert Katonai, MD, Nora Arvai, MSc, and Bertalan Mesko, MD, PhD, concentrated on studies published in English and had their work published in August by the Journal of Medical Internet Research. 

In a section discussing the technical potential of primary care AI vs. its real-world constraints, the researchers make three key points:  

1. Despite promising technical performance during early pilot testing, most AI tools for primary care remain at the proof-of-concept stage.

These tools have shown limited integration into clinical workflows and unclear real-world impact, the authors state. Bridging this gap, Katonai and co-authors add, “requires tools that demonstrate clinical value while fitting into existing workflows, which is essential to address ongoing implementation challenges, including usability, workflow integration and cost-related concerns.” More:  

‘This gap between technical feasibility and clinical usability underscores the need for AI solutions tailored to primary care’s specific workflow demands, resource constraints and the effort required to transform routine practice.’

2. Primary care deals with broad, often undifferentiated presentations, requiring AI systems to handle multimodal data and variable clinical reasoning.

This contrasts with specialized care, which tends to favor task-specific AI tools, the researchers point out. The challenge to primary care is evident in triage tools and symptom checkers, “which perform inconsistently depending on use case and clinical context,” the authors remark. More: 

‘These variabilities highlight the difficulty of designing AI systems that can replicate the nuanced and situation dependent–sensitive reasoning of primary care, which often relies on patient history, symptom presentation and social context.’ 

3. Primary care providers worldwide face high levels of administrative burden and burnout, often driven by staffing shortages, complex EHR systems and increasing time pressures. 

“The COVID-19 pandemic further intensified these issues by accelerating the shift toward asynchronous, electronic and nonvisit care models,” the authors note, “while also fostering novel diagnostic pathways and forms of doctor-patient interaction.” 

‘In other sectors of healthcare, such as hospital administration, AI has already begun to ease such burdens through tools such as ambient digital scribes, suggesting that successful models for reducing workload exist but have yet to be fully adapted for primary care settings.’

“Persistent challenges”—from usability concerns to training gaps to organizational barriers—“continue to limit broader adoption,” Katonai and colleagues conclude. “These findings emphasize that the future of AI in primary care depends not only on technological capability but also on thoughtful integration into the relational and practical realities of primary care.”

Read the whole thing. 

• In other research news:
   
  • Johns Hopkins: AI devices with no clinical validation tied to more recalls, study finds
     
  • San Diego State: Public health expert awarded a Fulbright to investigate AI in healthcare
     
  • Icahn School of Medicine: New AI model accurately identifies which atrial fibrillation patients need blood thinners to prevent stroke