Optimizing Structural Heart Disease Screening via Longitudinal EHR Pattern Recognition

Article Impact Level: HIGH
Data Quality: STRONG
Summary of  NEJM AI https://doi.org/10.1056/AIoa2500588 
Dr. Evangelos K. Oikonomou  et al.

Points

Researchers developed TARGET-AI as a foundational clinical decision support tool to help health systems deploy artificial intelligence for heart disease screening more effectively and precisely.
The algorithm uses longitudinal data from electronic health records to identify patterns in a patients clinical trajectory that suggest an increased risk for specific cardiovascular diagnoses.
By providing a data driven layer of guidance the tool helps clinicians avoid overusing AI models which often leads to high false positive rates and unnecessary patient anxiety.
Validation across the Yale New Haven Health system and external datasets from the United States and United Kingdom confirmed the models performance across different demographics and demographics.
This innovative framework establishes necessary guardrails for the next phase of AI research by focusing on the streamlined partnership between human clinicians and automated diagnostic detection tools.

Summary

This research evaluated the development and validation of TARGET-AI, a foundational clinical decision support tool designed to optimize the deployment of artificial intelligence in cardiology. While AI-enhanced electrocardiography (ECG) shows significant promise for screening structural heart disease (SHD), its integration into routine practice is often hampered by high false-positive rates when applied to unselected populations. TARGET-AI addresses this limitation by acting as an “AI-informed layer” that identifies patients whose clinical trajectories suggest a higher pre-test probability of disease, thereby providing necessary guardrails for the use of secondary diagnostic models.

The framework utilizes a methodology analogous to large language models, but instead of predicting word sequences, it analyzes deidentified longitudinal events within the electronic health record (EHR). By recognizing complex patterns in a patient’s clinical history and diagnostic markers, the algorithm determines the optimal timing for AI-enabled screening. The model was developed and validated across the Yale New Haven Health system and further tested on external datasets from both the United States and the United Kingdom. This external validation confirms the model’s generalizability across diverse patient demographics and distinct healthcare system characteristics, mitigating the risk of site-specific overfitting.

The implementation of TARGET-AI aims to maximize the precision of AI detection tools while minimizing unnecessary downstream testing, costs, and patient anxiety. By focusing AI resources on patients most likely to benefit, the system streamlines clinical workflows and assists health systems in navigating the surge of new digital health tools. The findings suggest that targeted deployment strategies are essential for the next era of AI research, shifting the focus from model development to the practical partnership between human clinicians and automated diagnostic assistants in real-world clinical settings.

Link to the article: https://ai.nejm.org/doi/10.1056/AIoa2500588

References

Oikonomou, E. K., Batinica, B., Dhingra, L. S., Aminorroaya, A., Coppi, A., & Khera, R. (2026). Target-ai: A foundational approach for the targeted deployment of artificial intelligence electrocardiography in the electronic health record. NEJM AI, 3(2). https://doi.org/10.1056/AIoa2500588