Study II — Breathing emergencies: conditional risk and response time
This study uses interpretable, exploratory modelling to investigate non-linear associations and heterogeneity in breathing-problem missions,
focusing on how response time interacts with age and sex in shaping conditional risk.
Design
- Type: observational, registry-based study.
- Approach: interpretable exploratory machine learning with partial dependence (PD) and individual conditional expectation (ICE) plots.
- Interpretation: explanatory/associational patterns; not causal effects and not deployable automation.
Why PD/ICE (and why this is exploratory)
The study prioritises PD and ICE plots to reveal how predicted risk changes with one variable (e.g., response time) while holding others constant,
and to detect heterogeneous effects that average trends can mask.
Key results (high-level)
- Among evaluated models, gradient boosting had the highest discrimination (AUC reported in the paper) but metrics are explicitly not presented as readiness for clinical triage decisions.
- The analysis emphasises how the model captures the influence of response time, age, and sex on predicted risk.
Citation
Hill P, Jonsson D, Lederman J, et al. Uncovering nonlinear patterns in time-sensitive prehospital breathing emergencies: an exploratory machine learning study.
BMC Medical Informatics and Decision Making. (Details on the Outputs page.)
Adaptive EMS Allocation 