AI for diagnostics of dyspnoea in the emergency department
Author Information
Author(s): Heyman Ellen T., Ashfaq Awais, Ekelund Ulf, Ohlsson Mattias, Björk Jonas, Khoshnood Ardavan M., Lingman Markus
Primary Institution: Department of Emergency Medicine, Halland Hospital, Region Halland, Sweden
Hypothesis
Can a deep learning model accurately classify dyspnoeic emergency department visits into specific diagnoses using comprehensive healthcare data?
Conclusion
The CareNet model achieved a median AUROC of 87.0% in diagnosing dyspnoea-related conditions in emergency department patients.
Supporting Evidence
- The most prevalent diagnoses among the 10,315 dyspnoeic ED visits were AHF (15.5%), eCOPD (14.0%), and pneumonia (13.3%).
- CareNet median sensitivity for AHF, eCOPD, and pneumonia was 74.5%, 92.6%, and 54.1%, respectively.
- CareNet's performance was significantly better than the baseline CatBoost model, which had a median performance of 81.4%.
Takeaway
Doctors can use a new AI tool to help figure out why patients are having trouble breathing when they come to the emergency room.
Methodology
This cross-sectional study analyzed all dyspnoeic ED visits of patients aged 18 and older over a two-and-a-half-year period, using a deep learning model to classify diagnoses based on comprehensive healthcare data.
Potential Biases
Potential bias in care consumption and variability in clinician behavior were not compensated for in the model.
Limitations
The study only included structured data and did not account for free text, images, or ECGs, which may limit the model's effectiveness.
Participant Demographics
The median age of participants was 75 years, with slightly more than half being female.
Statistical Information
Confidence Interval
84.8–88.3%
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website