Advancing a machine learning-based decision support tool for pre-hospital assessment of dyspnoea by emergency medical service clinicians: a retrospective observational study
2025

Machine Learning Tool for Assessing Dyspnoea in Emergency Medical Services

Sample size: 6354 publication 10 minutes Evidence: high

Author Information

Author(s): Wivica Kauppi, Henrik Imberg, Johan Herlitz, Oskar Molin, Christer Axelsson, Carl Magnusson

Primary Institution: University of Borås

Hypothesis

Can a machine learning model improve the assessment of dyspnoea in pre-hospital settings compared to traditional triage tools?

Conclusion

The machine learning model using gradient boosting significantly outperformed traditional triage methods in predicting serious adverse events among patients with dyspnoea.

Supporting Evidence

  • All machine learning models showed better performance than traditional triage tools.
  • The gradient boosting algorithm had the best performance with a ROC AUC of 0.81.
  • 80% of the data was used for model development and 20% for validation.
  • 1,118 serious adverse events were identified among the study cohort.
  • 78% of patients had a history of dyspnoea.

Takeaway

Researchers created a smart computer program to help ambulance workers figure out how serious patients with breathing problems are, and it worked better than older methods.

Methodology

This was a retrospective observational study analyzing data from 6,354 patients who called for an ambulance with dyspnoea as the main symptom.

Potential Biases

There is a risk of incomplete documentation due to the retrospective nature of the study.

Limitations

The study's data is from 2017, which may limit its relevance today, and it was conducted in a specific region of Sweden, affecting generalizability.

Participant Demographics

78% of participants were 65 years or older, and 44% were male.

Statistical Information

P-Value

p<0.05

Confidence Interval

95% CI 0.70–0.76 for RETTS-A, 95% CI 0.78–0.84 for gradient boosting

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1186/s12873-024-01166-9

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication