Machine learning model identifies cost-effective diagnostic sequences for moderate-to-severe breathlessness, indicating initial lung-focused testing reduces diagnostic costs.
Key Points
To develop an artificial intelligence reinforcement learning framework that determines optimal, low-cost diagnostic pathways for individuals presenting with chronic breathlessness.
Trained a reinforcement learning model using Swedish population data from individuals with moderate to severe breathlessness.
Modeled 16 clinically relevant conditions alongside their associated diagnostic tests and standard healthcare costs across subgroups stratified by sex and smoking history.
The AI model produced consistent, low-cost diagnostic pathways with high diagnostic yield across all evaluated subgroups.
The optimal diagnostic sequence prioritized evaluations of body mass index, anxiety, depression, physical activity, and spirometry, followed by diffusing capacity, chest CT, and hemoglobin testing, placing pulmonary investigations ahead of cardiac testing.