Study II · Peer-reviewed publication · BMC Medical Informatics and Decision Making

Breathing emergencies and nonlinear risk

An exploratory machine learning study of response time, age, sex, and high-risk time-sensitive triage among patients with breathing problems.

Open BMC article Back to thesis All publications

Study profile

Conditional risk in breathing-problem missions

The study examines how response time, age, and sex jointly shape the probability of encountering a high-risk time-sensitive condition among patients initially reported with breathing problems.

Article: Hill P, Jonsson D, Lederman J, Bolin P, Vicente V. Uncovering nonlinear patterns in time-sensitive prehospital breathing emergencies: an exploratory machine learning study. BMC Medical Informatics and Decision Making. 2025;25:205.
DOI: 10.1186/s12911-025-03046-z.
Design: Retrospective observational study using exploratory machine learning.
Participants/data: 132,395 prehospital missions in Stockholm, 2017–2022, initially reported as breathing problems.
Outcome: High-risk time-sensitive condition, operationalised as RETTS Red/Orange at first EMS contact.
Methods: Gradient boosting, random forest, neural networks and logistic regression were evaluated; partial dependence and individual conditional expectation plots were central to interpretation.
Key contribution: The study uses interpretable machine learning to reveal nonlinear and heterogeneous patterns in time-sensitive breathing emergencies, without presenting the model as deployable triage automation.

Core findings

How age and waiting interact with time-sensitive risk

Older age increased conditional risk

Patients over 60 years showed consistently higher predicted probability of high-risk time-sensitive triage among breathing-problem missions.

Risk patterns were nonlinear

The analysis identified complex response-time patterns, including rising risk among older patients during prolonged waits exceeding two hours.

Interpretability before prediction

Model performance was modest and secondary. The scientific value lies in visualising how age, sex, and response time shape heterogeneous risk patterns.

Scientific contribution

Breathing problems as a high-risk dispatch phenotype

The study demonstrates how routine EMS data can reveal clinically meaningful heterogeneity within a broad dispatch category.

Study II supports a more nuanced view of response time: the safety meaning of waiting depends on patient vulnerability and presenting problem. This is especially relevant for older patients with breathing problems, where delayed reassessment may be important.

Breathing problems Age interaction Response time HRTS PD/ICE interpretation

Interpretation note

This study is exploratory and observational. Machine learning is used to characterise nonlinear relationships and heterogeneity, not to provide a ready-to-use dispatch, triage, or clinical decision-support system. Prospective evaluation, human-factors design, and governance would be required before operational translation.