Uncovering nonlinear patterns in time-sensitive prehospital breathing emergencies: an exploratory machine learning...

Area of Science:

• Prehospital emergency medicine
• Health informatics
• Machine learning applications in healthcare

Background:

• Timely prehospital care is critical for high-risk time-sensitive (HRTS) conditions.
• Understanding the impact of response time and demographics on breathing emergencies is limited.
• This study explores machine learning to analyze these factors.

Purpose of the Study:

• To investigate the combined influence of emergency medical response time, patient age, and sex on the likelihood of encountering HRTS conditions.
• To identify nonlinear patterns and interactions between these variables in prehospital settings.
• To leverage machine learning for a deeper understanding of HRTS risk factors.

Main Methods:

• Retrospective analysis of 132,395 prehospital missions (Stockholm, 2017-2022).
• Trained multiple machine learning models (random forest, gradient boosting, neural networks, logistic regression).
• Utilized partial dependence and individual conditional expectation plots to visualize relationships between response time, age, sex, and HRTS likelihood.

Main Results:

• Older age (>60 years) consistently correlated with a higher probability of HRTS.
• Patients over 60 showed increased risk with response times exceeding two hours.
• Gradient boosting models provided modest classification metrics (AUC 0.66, F1-score 0.55), with a focus on relationship characterization.

Conclusions:

• Machine learning effectively identified complex interactions between response time, age, and sex in time-sensitive breathing emergencies.
• Findings suggest refining dispatch protocols and developing targeted screening questions.
• Opportunities exist to re-evaluate lower-priority calls after extended delays and for future predictive modeling.