Effect of topography and weather on call to automatic electrical debrillator attach time by drone for out-of-hospital cardiac arrest: a virtual ight simulation study

Background Delivery of automatic electrical debrillator (AED) by unmanned aerial vehicle like drones was suggested to improve early debrillation for out-of-hospital cardiac arrest. We developed a drone-AED ight virtual simulator using 3-dimensional topographic and meteorological information. The goal of this study is to assess the effect of topography and weather on call to AED attach time in drone-AED program. Methods We included patients from 2013 to 2016 in Seoul, South Korea, registered in the Korean out-of-hospital cardiac arrest registry. We developed a drone-AED ight simulation using topographic information of Seoul for Euclidean ight pathway and topographic ight pathway including vertical ight to overcome high-rise structures. We used 4 drone ight scenarios according to weather conditions or visibility: ight and control advanced drone, ight advanced drone, control advanced drone and basic drone. Primary outcome was emergency medical service call to AED attach time. Secondary outcome was success rate of call to AED attachment within 5 or 10 minutes, and pre-arrival rate of drone-AED before AED delivery by ground ambulance. Results 16,596 patients were included. Median ight time of drone-AED was 2.6 and 1.0 minute for topographic ight simulation and Euclidean pathway. Call to AED attach time in topographic pathway was 7.0 minutes in ight and control advanced drone and 8.0 minutes in basic drone. The time in Euclidean pathway was 6.5 minutes in ight and control advanced drone and 7.0 minutes in basic drone. Pre-arrival rate of drone-AED in Euclidean pathway was 38.0% and 16.3% for ight and control advanced drone and basic drone. whereas, pre-arrival rate in the topographic pathway was 27.0% and 11.7%, respectively. Conclusions Drone-AED took longer call to AED attach time in basic drone than ight and control advanced drone. Pre-arrival rate of ight and control advanced drone was decreased in topographic ight pathway compared to Euclidean pathway. Trial registration This study used cases retrospectively registered in the Korean out-of-hospital cardiac arrest registry.

topographic ight simulation and Euclidean pathway. Call to AED attach time in topographic pathway was 7.0 minutes in ight and control advanced drone and 8.0 minutes in basic drone. The time in Euclidean pathway was 6.5 minutes in ight and control advanced drone and 7.0 minutes in basic drone. Pre-arrival rate of drone-AED in Euclidean pathway was 38.0% and 16.3% for ight and control advanced drone and basic drone. whereas, pre-arrival rate in the topographic pathway was 27.0% and 11.7%, respectively.
Conclusions Drone-AED took longer call to AED attach time in basic drone than ight and control advanced drone. Pre-arrival rate of ight and control advanced drone was decreased in topographic ight pathway compared to Euclidean pathway. Trial registration This study used cases retrospectively registered in the Korean out-of-hospital cardiac arrest registry.

Background
Out-of-hospital cardiac arrest (OHCA) is one of the leading causes of death worldwide. (1-3) Early de brillation is one of the important links to improve the survival rate of OHCA patients. (4-7) Public access de brillation (PAD) programs have been implemented to reduce time between cardiac arrest to the rst de brillation. (4,8,9) However, incidence of OHCA located nearby PAD is limited. Moreover, it is di cult for the bystander to nd the nearest PAD and apply it to the victims rapidly.
To overcome the limitation of current PAD installation strategy, delivery of automatic electrical de brillator (AED) to the OHCA scene by unmanned aerial vehicle (UAV) such as drone has been suggested. The delivery of AED using drones was proposed to improve AED application rate and reduce de brillation time. (12)(13)(14)(15) However, previous studies on UAV delivering AED(UAV-AED) program showed several limitations on its implementation to real clinical practice. First, control of UAV ight to the scene could be limited by weather conditions like rain, snow, wind speed and temperature. Poor visibility during ight due to nighttime or short sight distance by fog could also affect ight permission of UAV. The UAV could not safely and rapidly deliver AED in extreme weather conditions or with poor visibility. Second, topographic conditions could also increase the UAV ight time to the scene and weaken the bene t of UAV-AED program. Previous simulation studies reported the effect of UAV-AED program based on the ight route of UAV by the Euclidean distance from drone installation site to the location of OHCA victims. (12)(13)(14)(15) However, to y across the high-rise buildings in real life, vertical movement should be added to horizontal movement regarding Euclidean distance.
In this study, we developed a virtual UAV-AED ight simulation using topographic information such as natural terrain and buildings in Seoul, South Korea. We also added the meteorological information into the simulation scenario to permit ight of UAV-AED into the scene safely in metropolitan city. The goal of this study is to assess the effect of topography and weather on call to AED attach time by UAV-AED program for OHCA. The hypothesis of this study is that adding topographic and weather conditions on the UAV-AED simulation program would increase call to AED attach time, unlike prior studies which did not consider these conditions.

Study design
This study is a retrospective observation study using a computerized virtual simulator. We included all OHCA cases registered in Korea OHCA Registry (KOHCAR) from 2013 to 2016, transported by emergency medical services(EMS) across South Korea. The Korea Disease Control and Prevention Agency (KDCA) approved use of all data, and the study was approved by the Institutional Review Board of the study site.

Data source
The KOHCAR is a nationwide database including all cardiac arrest patients transported by EMS ambulances operated by re departments across South Korea since 2006. This database registers prehospital information written by emergency medical technicians of National Fire Agency. Trained medical record reviewers of KCDC visited the hospitals, in which the OHCA patients were transported by National Fire Agency. They collected in-hospital information and outcome of OHCA victims. Then, prehospital and in-hospital data of each case was merged using the Utstein guidelines. (16,17) To develop a UAV-AED ight simulation, we initially extracted data on topography and altitude of Seoul city from Google maps. We constructed a geographic information database including height of buildings by combining the data about altitudes of each terrain with the altitude of all facilities supported by Ministry of Government Administration and Home Affairs. We also added meteorological information including wind speed, precipitation, snowfall, temperature, visibility and weather phenomenon, which were hourly recorded on Korea Meteorological Administration database. These data were added for scenario models of limited UAV-AED ight due to extreme weather condition.

Study setting
Seoul has a population of 9.7 million and covers a total surface area of 605.2 km 2 . Seoul is a metropolitan city with many high-rise buildings and mountains. The EMS system of Seoul is a two-tiered public service model with service level of EMT-intermediate. Seoul is divided into 25 districts. Each district has 3 to 8 re stations with ambulance vehicles and there are 116 re stations in total. (18) Each ambulance vehicle is usually staffed with 3 EMTs. (2,19) One dispatch center covers all EMS calls across Seoul (20). The PAD installation is mandatory on public health o ces, ambulances, airport, train, and apartments with more than 500 households. Approximately 8,000 AEDs are installed in Seoul. (21)

Study population
All OHCA cases in Seoul from January 1, 2013 to December 31, 2016 with age of ≥ 9 years at cardiac arrest recognition time were eligible for this study. Patients with OHCA occurrence in the ambulance during EMS transport was excluded. Moreover, we excluded pediatric OHCA cases aged under 8 years. This is because pediatric cases aged under 8 were recommended for dose attenuator usage to optimize de brillation energy. (22) Cases with missing AED attach time or cardiac arrest recognition time were also excluded.

Variables
We used the Utstein variables from KOHCAR database such as gender, age, witnessed status, location of event (private vs public vs unknown), bystander CPR, initial electrocardiogram (ECG) rhythm, and EMS de brillation. (23) We collected the EMS time pro les including EMS call time, EMS arrival at the scene time, EMS departure from the scene time, EMS hospital arrival time, call to cardiac arrest recognition time and call to AED attach time.
The address recorded on KOHCAR was used as the place of cardiac arrest. Geo-coding for the place of cardiac arrest was performed using Google Maps APIs (Google, California, United States). Regarding the weather-related variables, we collected hourly wind speed, precipitation, snowfall, temperature, visibility and weather phenomenon. Daytime and nighttime of OHCA occurrence was divided by 6AM and 6PM.

UAV-AED station allocation
All of 116 re stations in Seoul were used as the candidates for UAV-AED installation. Among the 116 stations, the optimal location for each number of stations increased by 5 (i.e. 5, 10, 15, etc.) was selected for simulation from 5 to 116 stations. A multicriteria evaluation was conducted for selecting optimal combinations of possible UAV-AED installed stations to reduce call to AED attach time. We generated an OHCA occurrence layer by heat-map analysis of the OHCA occurrence location from 2013 to 2016 in Seoul (Appendix. 1). The EMS call to scene arrival time was analyzed by inverse distance weight interpolation to obtain EMS-response time layer. (Appendix. 2) The OHCA risk map was calculated by adding 1: 1 weighting of OHCA occurrence layer and EMS call to scene arrival time layer. GIS analysis was performed using qGIS 3.4. The OHCA risk map was constructed with a lattice with a resolution of 50 m x 50 m. For optimal number of UAV-AED stations, the estimated coverage of each UAV-AED station was obtained by allocating a 3km circle for each station. The optimal location for each number of stations, increased by 5, was selected according to the location with maximum score of OHCA risk map, which was calculated using the genetic algorithm. (24)  Based on meteorological information of Seoul during study period, we simulated 4 scenarios according to ight performance of the UAV regarding weather and visibility. The rst scenario was basic UAV model. In this model, ight of UAV-AED was restricted if EMS call for OHCA occurred during extreme weather conditions, which were de ned as strong winds of 10km/h or higher, rain, snow, and temperature below 0℃. Also, if the call was made during nighttime or if the sight distance was less than 1km, ight of UAV in the basic model was not permitted. The second model, control advanced UAV could y regardless of time or limited visibility during ight. However, it was prohibited for use during extreme weather conditions. The third model, ight advanced UAV could y in extreme weather conditions, but it could not y in situations of poor visibility. Lastly, the ight and control advanced UAV model could y whenever during the study period regardless of weather conditions or poor visibility. We simulated these 4 types of UAV model scenarios for 2 different ight pathways. The rst ight pathway is the direct ight route through Euclidean distance from UAV-AED station to OHCA site. The second topographic ight pathway was generated by UAV-AED ight simulation developed in this study using topographic information.
Outcome Primary outcome was call to AED attach time. Call to AED attach time pro les by current EMS practice was measured by time pro les in the KOHCAR database; and time pro les by UAV-AED was measured by pro les derived from UAV-AED simulation. Secondary outcome was success rate of call to AED attach within 5 minutes or 10 minutes, and pre-arrival rate of UAV-AED before current EMS based AED delivery.

Statistical analysis
The paired Wilcoxon rank sum test was used to compare the call to AED attach time between current practice and UAV-AED program. Call to AED attach success rate within 5 or 10 minutes before and after UAV-AED program implementation was compared using McNemar test. Call to AED attach time was The weather and visibility related factors during study period are described in Table 2. Total of 6,749(40.7%) OHCA cases occurred at night and 1,780(10.7%) cases had sight distance of less than 1 km.
The number of OHCA calls during extreme weather conditions was 2,658 (16.0%) cases at temperature of < 0℃, 1,585 (9.6%) during rain, 884 (5.3%) during snow, 84 (0.5%) during lightning, and 1 case during wind speed higher than 10m/s.   Fig. 3.). Median ight time of UAV-AED in Euclidean ight pathway was 1.0 minute (Table 3.). The median call to attach time was 6.5 minutes in ight and control advanced UAV and 7.0 minutes in the other 3 UAV models. Success rate of call to AED attach time within 5 minutes was 34.8%, 31.3%, 28.8%, 26.7% in each UAV model. Pre-arrival rate of UAV-AED before current EMS based AED delivery was 38.0% in ight and control advanced UAV model and 16.3% in basic UAV model.   The effect of UAV-AED program on reducing call to AED attach time in this study was lower than the results of previous studies (12)(13)(14)(15). First, operation of UAV was limited by environment with poor visibility during ight. In situations of poor visibility, current UAV operated by remote control could not guarantee safe delivery of AED to the scene without collision with housing buildings or laypersons. Second, ight performance of current UAV was restricted under extreme weather conditions like raining or high wind speed. Third, tra c environment and EMS resources of metropolitan city showed shorter delivery time of AED to the scene by ambulance vehicle compared to rural areas. In Seoul, median call to AED attach time of current EMS was 8 minutes in this study.
UAV-AED was also advantageous in UAV-AED arrival time before current EMS based AED delivery in this study. However, optimized UAV-AED installation in each community was required to maximize the bene t of UAV-AED. Floating commercial drones with limited ight operation at night or in bad weather did not reduce call to AED attach time despite all EMS stations in Seoul were used for UAV-AED installation. Drones with augmented visibility or ight performance could overcome this restriction. Flight of UAV in metropolitan city by remote control was another limitation when nding the best route in metropolitan city with high-rise buildings. Designation of the ight pathway for UAV-AED based on 3-dimensional topographic information should be preceded for effective implementation of the UAV-AED program in metropolitan city.

Limitations
First, this study was a simulation study using a computerized virtual ight simulator. Clinical trial of UAV-AED is required to assess clinical outcome or other operational factors affecting UAV-AED ight. Second, we used ight scenarios with ight speed of 50km per hour and limited UAV ight operation Improvement of UAV ight performance may affect the result. Third, there is a limitation of generalizability. The goal of this study was to assess the effect of UAV-AED in metropolitan city like Seoul, South Korea. Difference of surface area, density or height of high-rise buildings, natural terrain and weather can affect the result.

Conclusions
We developed a virtual UAV-AED ight computer simulation using 3-dismensional topographic information and meteorological information. Call to AED attach time in Euclidean pathway was 6.5 minute in ight and control advanced UAV model and 7.0 minutes in basic UAV model. The call to AED attach time in topographic pathway was 7.0 minutes in ight and control advanced UAV model and 8.0 minutes in basic UAV model. Pre-arrival rate of UAV-AED in Euclidean pathway was 38.0% and 16.3% in ight and control advanced UAV model and basic UAV model, respectively; whereas pre-arrival rate in the topographic pathway was 27.0% and 11.7%, respectively.   Figure 1 The topographic ight pathway used in UAV-AED virtual ight simulation. A: UAV-AED allocated station, B:

Figures
The site where out of hospital cardiac arrest occurred