Real-World AI in Chest Diagnostics: An NHS Evaluation
The RSET evaluation assessed AI deployed across multiple NHS trusts to support chest X-ray interpretation. Using mixed methods and health economic analysis, the study measured actual clinical impact rather than controlled performance alone. The result is a practical view of what happens when diagnostic AI moves into routine care.
AI’s Promise: Prioritisation and Positive Outlook
Streamlining diagnoses
AI systems successfully flagged high-suspicion chest X-rays, helping services prioritise urgent cases. Around 90 percent of suspected cancers were prioritised for review within 24 hours in sites using the tools. This faster triage supported earlier clinical review and contributed to favorable health economic estimates for rapid pathways.
Staff and patient reception
Clinicians and patients generally viewed AI as a helpful second pair of eyes that provided reassurance and supported workflow. Where radiography teams were prepared and supported, staff reported clearer lists and a perceived reduction in diagnostic delay.
Unveiling Implementation Realities
Unexpected workload and variation
Implementation costs and operational effort were greater than many trusts anticipated. The scale of local configuration, staff training, and process redesign varied widely. Some sites struggled with resourcing and found that adopting AI without solving underlying bottlenecks limited benefits.
Monitoring and infrastructure gaps
Many trusts lacked robust data pipelines and governance to monitor AI performance and outcomes at scale. Incomplete monitoring capacity made it hard to track false positives, workflow impact, and long-term clinical value. Supplier stability and contract management also emerged as risks.
Key takeaways for wider adoption
Lessons learned point to practical steps: invest in data and monitoring infrastructure, fund implementation and staffing, adopt problem-first deployment, define clear metrics for clinical and economic impact, run staged rollouts, and build clinician familiarity with tools. With targeted work on these areas, AI can support faster diagnosis in routine NHS pathways while avoiding common implementation pitfalls.
