Using machine learning to assess the extent of busy ambulances and its impact on ambulance response times: A retrospective observational study

Lars Eide Næss; Andreas Jørstad Krüger; Oddvar Uleberg; Helge Haugland; Jostein Dale; Jon-Ola Wattø; Sara Marie Nilsen; Andreas Asheim

doi:10.1371/journal.pone.0296308

Using machine learning to assess the extent of busy ambulances and its impact on ambulance response times: A retrospective observational study

PLoS One. 2024 Jan 5;19(1):e0296308. doi: 10.1371/journal.pone.0296308. eCollection 2024.

Authors

Lars Eide Næss^{1

2

3}, Andreas Jørstad Krüger^{1

2

3}, Oddvar Uleberg^{2

4}, Helge Haugland^{1

2

3}, Jostein Dale², Jon-Ola Wattø², Sara Marie Nilsen^{5

6}, Andreas Asheim^{5

7}

Affiliations

¹ Department of Research and Development, The Norwegian Air Ambulance Foundation, Oslo, Norway.
² Department of Emergency Medicine and Pre-Hospital Services, St. Olav's University Hospital, Trondheim, Norway.
³ Department of Circulation and Medical Imaging, Norwegian University of Science and Technology, Trondheim, Norway.
⁴ Division of Emergencies and Critical Care, Department of Research and Development, Oslo University Hospital, Oslo, Norway.
⁵ Center for Health Care Improvement, St. Olav's University Hospital, Trondheim, Norway.
⁶ Department of Clinical and Molecular Medicine, Norwegian University of Science and Technology, Trondheim, Norway.
⁷ Department of Mathematical Sciences, Norwegian University of Science and Technology, Trondheim, Norway.

Abstract

Background: Ambulance response times are considered important. Busy ambulances are common, but little is known about their effect on response times.

Objective: To assess the extent of busy ambulances in Central Norway and their impact on ambulance response times.

Design: This was a retrospective observational study. We used machine learning on data from nearby incidents to assess the probability of up to five different ambulances being candidates to respond to a medical emergency incident. For each incident, the probability of a busy ambulance was estimated by summing the probabilities of candidate ambulances being busy at the time of the incident. The difference in response time that may be attributable to busy ambulances was estimated by comparing groups of nearby incidents with different estimated busy probabilities.

Setting: Medical emergency incidents with ambulance response in Central Norway from 2013 to 2022.

Main outcome measures: Prevalence of busy ambulances and differences in response times associated with busy ambulances.

Results: The estimated probability of busy ambulances for all 216,787 acute incidents with ambulance response was 26.7% (95% confidence interval (CI) 26.6 to 26.9). Comparing nearby incidents, each 10-percentage point increase in the probability of a busy ambulance was associated with a delay of 0.60 minutes (95% CI 0.58 to 0.62). For incidents in rural and urban areas, the probability of a busy ambulance was 21.6% (95% CI 21.5 to 21.8) and 35.0% (95% CI 34.8 to 35.2), respectively. The delay associated with a 10-percentage point increase in busy probability was 0.81 minutes (95% CI 0.78 to 0.84) and 0.30 minutes (95% CI 0.28 to 0.32), respectively.

Conclusion: Ambulances were often busy, which was associated with delayed ambulance response times. In rural areas, the probability of busy ambulances was lower, although the potentially longer delays when ambulances were busy made these areas more vulnerable.

Copyright: © 2024 Næss et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Publication types

Observational Study

MeSH terms

Ambulances*
Machine Learning*
Norway
Probability
Reaction Time
Retrospective Studies

Grants and funding

LEN was supported by The Norwegian Air Ambulance Foundation (SNLA, https://norskluftambulanse.no/) as part of a PhD project (grant number not applicable). The funder had no role in the design of the study, data collection, analysis, decision to publish, or preparation of the manuscript. AA and SMN were supported by The Norwegian Research Council (forskningsradet.no) grant number 295989. The funder had no role in the design of the study, data collection, analysis, decision to publish, or preparation of the manuscript.