Identifying optimal locations for automated external defibrillators (AED) in Freiburg: development and validation of a machine learning model based on demographic and infrastructural data

Identifying optimal locations for automated external defibrillators (AED) in Freiburg: development and validation of a machine learning model based on demographic and infrastructural data

First responder, Smartphone alerting systems, Out-of-hospital cardiac arrest, Automated external
defibrillator, Public access defibrillation, Dispatch centre

Introduction Out-of-hospital cardiac arrest (OHCA) is a critical medical emergency where rapid access to automated

external defibrillators (AED) can significantly improve survival rates. However, there is currently a lack of well-
established frameworks and guidelines concerning the optimal placement of AED. Additionally, historical data on the

locations of OHCA incidents is often unavailable or incomplete. This study seeks to address these gaps by analyzing
the most effective AED placement strategies and evaluating the impact of additional AED locations on suspected
OHCA cases. To achieve this, a machine learning (ML) model is developed that relies exclusively on demographic and
infrastructural factors, without the need for historical OHCA location data.
Methods In this data-driven predictive modelling study, 5,076 alerts of suspected OHCA and 95 AED locations in
Freiburg were analysed (October 7, 2018, to May 28, 2024). Demographic and infrastructural data were integrated into
a three-step approach to identify and prioritize optimal AED placements. A Decision Tree was trained to predict OHCA
risk at possible locations, followed by the application of a greedy algorithm to determine AED locations. The models
were validated using several performance metrics and historical OHCA data to ensure accuracy. Additionally, different
scenarios were evaluated to maximize AED coverage of OHCA incidents.
Results Optimizing AED placement using predicted data increased coverage from 21.6% to 42.4%, without adding
more devices. The ML model’s coverage was only 6.7% lower than that achieved using historical alert data. Adding 19
AEDs (a 20% increase) to the existing network raised coverage to 30.5%.
Conclusion The findings demonstrate the feasibility of using ML models for AED placement in regions lacking
comprehensive historical data. Integrating advanced ML techniques can further refine strategies for AED deployment
in urban areas, ultimately improving emergency response effectiveness.

Julian Ganter1,3*, Hannah Bakker2

, Stefan Nickel2

, Elisa-Sophie Reichling2

, Alicia Wittmer2

, Niklas Werner2
,

Thomas Brucklacher3

, Robert Wunderlich3,4, Georg Trummer3,5, Hans-Jörg Busch3,5 and Michael Patrick Müller3,6

https://doi.org/10.1186/s12873-025-01441-3

2026

PERI IRAWAN

PDF

ENGLISH

TEXT