- Email: [email protected]
Simulation of emergency care for patients with ACS in Saint Petersburg for ambulance decision making
Available online at www.sciencedirect.com
ScienceDirect Procedia Computer Science 108C (2017) 2210–2219
International Conference on Computational Science, ICCS 2017, 12-14 June 2017, Zurich, Switzerland
Simulation of emergency care for patients with ACS in Saint Petersburg for ambulance decisionwith making Simulation of emergency care for patients ACS in Saint Petersburg for ambulance decision making 1 1 Ivan Derevitskiy , Evgeniy Krotov , Daniil Voloshin1, 1Alexey Yakovlev1,2 , Ivan Derevitskiy 1, Evgeniy Krotov1, 1 Sergey V. Kovalchuk , Vladislav Karbovskii 1 1,2, , Daniil Voloshin Alexey Yakovlev 1 ITMO University, Saint-Petersburg, Russian Federation. 1 1 2 V.Medical , Vladislav KarbovskiiRussian Kovalchuk Federal AlmazovSergey North-West Research Centre, Saint-Petersburg, Federation
1 ITMO University,[email protected], Saint-Petersburg, Russian Federation. [email protected], [email protected], Federal Almazov North-West Medical Research Centre, Saint-Petersburg, Russian Federation [email protected], [email protected], [email protected] [email protected], [email protected], [email protected], [email protected], [email protected], [email protected] 2
Abstract One of the stages of emergency medical care in case of Acute Coronary Syndrome (ACS) (if there are Abstractconditions for surgical intervention) is directly linked to the time between the first contact medical One of stagesand of the emergency medical in case ofcoronary Acute Coronary Syndrome (if there are Time with thethe patient and inflating the care balloon in the artery (in a medical(ACS) institution). medical conditions surgical facility intervention) is on directly linked to the delivery time between the first of the operation startfora medical depends the time of patient to hospital, as contact well as withthethewaiting patienttime and the andqueue inflating theinstitution. balloon in the artery (ina adevelopment medical institution). Time Thiscoronary paper describes of ambulance on in the in the of the operation start aaggregate medical facility depends on thetwo timeperiods of patient delivery to hospital, as time well as model for obtaining estimation of these of time. The estimation is on the waiting time of in the queue in paper describes a development of ambulance ambulance obtained by means described in the the institution. article the This decision support system (DSS) in the model obtaining aggregate estimation of these time. The estimation time(the is service.for Unlike modern navigation systems DSS takestwo intoperiods accountofambulance vehicle behavior obtained of described in the the decision system (DSS) in the the help ambulance With of the ability to by exitmeans into oncoming traffic) andarticle availability of freesupport operation rooms. service. Unlike modern navigation DSSservice takes into accountout ambulance behavior (the described simulation model of the systems ambulance we carried the timevehicle distribution analysis ability to exit into oncoming traffic) and availability of free operation rooms. With the help of the (between the first contact with the patient and surgical intervention in case of ACS) in St. Petersburg, described simulation modeluses of the carried out the in time analysis Russia. Simulation scenario realambulance data on theservice work ofwe ambulance service thedistribution city. (between the first contact with the patient and surgical intervention in case of ACS) in St. Petersburg, © 2017 The Authors. Published by Elsevier B.V. Russia. Simulation scenario uses real data on the work of ambulance serviceambulance in the city. Keywords: acute coronary syndrome, transport modeling, decision support system, service Peer-review under responsibility of the scientific committee of the International Conference on Computational Science Keywords: acute coronary syndrome, transport modeling, decision support system, ambulance service
1 Introduction Modern ambulance service 1 Introduction
is a dynamic multi-functional system, which has a complicated maintenance mechanism. Relevance of the research decision support systems in the ambulance service Modern service is a dynamic multi-functional system, has of a patient complicated is caused ambulance by the necessity to optimize decision-making, in order to improvewhich the quality care. maintenance mechanism. decision support in the ambulance service Among the key features ofRelevance the work of the research ambulance service, we cansystems highlight increased need for is caused by of theevery necessity to optimize decision-making, in orderlife to improve theThus, quality patient care. "reliability" decision, due to its importance for human and health. theofuse of DSS is Among the key features of the work of the ambulance service, we can highlight the increased need for "reliability" of every decision, due to its importance for human life and health. Thus, the use of DSS is 1877-0509 © 2017 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the scientific committee of the International Conference on Computational Science 10.1016/j.procs.2017.05.178
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
an effective approach that automates processing large amounts of information. Another key ambulance service feature is the high dynamics of the service work. High dynamics results from little time to make a decision that healthcare providers usually have. Consequently, the DSS can be used to reduce the time of decision-making in complicated cases by automating the analysis of the situation. This paper describes the methodology of development of DSS in the ambulance service. One of the DSS modules solves the problem of time estimation between the first contact with the patient and the beginning of the surgical intervention for patients with ACS. This estimate was obtained using a complex model, which includes a transport model, (that assessed the time of the patient delivery to the hospital), flow of calls model, as well as the queuing model that evaluates the waiting time. According to the paper, by (Tereshchenko & Zhirov 2010), in the provision of care for patients with ACS (patients with medical conditions for Percutaneous Coronary Intervention, PCI) the algorithm of actions of a medical brigade depends on the predicted time of start PCI. Tereshchenko and Jirov state that if the time between the first contact with medical personnel and inflating the balloon in the SC (coronary artery) exceeds 90 minutes, it is preferable to use thrombolytic therapy. If the predicted time is less than 90 minutes, then the use of invasive strategy is preferable. Thus, the task of predicting the time between the first contact of medical staff with patients and the beginning of the operation is an urgent one. The quality of the forecast impacts the patient's treatment strategy.
2 Related Works 2.1 Time estimation of the patient transportation to a hospital The period time between the first contact with the patient and the start of the operation can be divided into two parts. The first step is to calculate the patient delivery time. Delivery time is calculated on the basis of the selected route from the patient initial location to the hospital. Thus, when calculating the patient delivery time, it is necessary to find the fastest route. The problem of finding the fastest routes for ambulances may be considered from the perspective of the search of the optimal path in a weighted graph. To find the way we must take into account dynamic changes in the traffic situation and also the specific behavior of the ambulance vehicles. One of solutions to the problem of the ambulance fastest path search is proposed in (Gayathri & Chandrakala 2014). The authors apply Dijkstra algorithm of the shortest path search in a weighted graph, which takes into account the current state of traffic. The paper (Nordin et al. 2011) proposes a more efficient (in terms of performance) algorithm A * of finding the shortest path in the graph. In (Maxwell et al. 2010) the shortest route is searched using ADP dynamic programming techniques. In (Gendreau et al. 2001) ambulance route search problem is solved by a heuristic tabu search algorithm. In (Kergosien et al. 2015) it is proposed to seek the shortest path between two points and estimate the time of its passage, on the basis statistical data of prior ambulance trips, and comparing estimated routs to the current ones. Data of base previous ambulance trips helping in choosing the optimal route is also described in (López et al. 2005). In (Créput et al. 2011) proposed to use the method of neural networks (Kohonen self-organizing maps) to find the shortest path for an ambulance. One of the disadvantages of these approaches is the lack of consideration of the special behavior of ambulances on the roads. For example, the possibility of entering the opposite lane can change route selection. Unlike the above mentioned approaches, in the ambulances movement model, described in this paper, simulates travel into the oncoming lane.
2.2 Queuing models in ambulance service The second step of estimating time from the first contact with patient to the moment of surgery beginning is to calculate waiting time which patient has to spent in hospital after arrival.
2211
2212
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
In cardiology that time interval is called door-to-balloon time. Due to limited resources of PCI facilities in hospitals queues can emerge and in some cases queuing time can exceed the time of transportation. Applications of queuing theory for solving optimization problems in the ambulance is a widely used practice. One of the most well-known example is the hypercube model for solving the problem of the distribution ambulance fleet (Brandeau & Larson 1986). The advantage of that model is the possibility of finding the analytical solutions of the optimization problem. However, with increasing complexity of the model, as an instance introduction of priority levels (Siddharthan et al. 1996) of calls significantly complicates the solution of the problem. Also for simplicity of patient flow simulation arrival rates are modeled as a homogeneous Poisson process (Green et al. 2006), however, in order to take into account time dependence of the arrival rate we need to use non-homogeneous process, but finding parameters of its distribution is a complex task. In our work we use the empirical distribution of arrival rate instead of analytical distribution. There are some other applications of queuing theory, in (Cochran & Roche 2008) queuing model is used to find the optimal number of beds in hospitals. However, the main idea of the model is the same. Complex model presented in this paper which combined a queueing and transportation model allow us not only to solve optimization problems of ambulance fleet and hospitals resources allocation, but also a problem of finding the optimal ambulance dispatching policy.
3 Description of the model 3.1 Simulation environment description In this paper, we present the imitational simulation environment. The described instrument is based on Pulse tool, described in (Karbovskii et al. 2016). The function of the environment is to simulate the operation of the ambulance at a given locality. The ambulance service model consists of two autonomous units: transport model and health facilities queuing model. A transport model for ambulance services has been devised on previous work described in (Ivanov & Knyazkov 2014). It imitates the generation of incoming ambulance calls and dispatching (assignment of crews and vehicles to patients) process. The realization of the model used here has been implemented using C# programming language. A comprehensive description of the model is presented further in section 3.2. The queuing model used in this research has been previously discussed in (Kovalchuk et al. 2016). Based on the data about the flow of planned and emergency patients the model imitates the ques inside health facilities. More detailed description of the model can be found in section 3.3. Interaction between the described units of the imitational simulation environment is performed through POST http requests sent by the transport model to the server component of the health facilities queuing model. Transport model issues requests of the following types: initiation of the queuing model, request for id of the health facility patient shall be transported to (if the medical indicators speak in favor of hospitalization), request-notification of patients’ arrival to the hospital, termination of the queuing model work. At the beginning of the ambulance service operation simulation, the POST request that initiates the queuing model is sent. When the decision to hospitalize a patient is made, a request for a receiving health facility is sent to the transport model. Arguments for the request are the estimates of the time needed to transport patients to the health facilities suitable for a them based on the diagnosis-specific requirements, current model simulation time and patient’s id. The location of the facility that has been retrieved through the id request is considered to be optimal for patient’s transportation. Notificationrequest indicating the arrival of the patient serves to assign the patient to the que of one of the health facilities. Here arguments are patients and health facility id’s and current simulation time. After the
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
request of the last type has been sent and simulation statistics are aggregated, simulation terminates. The outline of the Interaction between simulation environment’s components is presented in Figure 1. Health facility queuing model Simulation initiation
Ambulance service transport model Start request
Optimal hospital request Assignment to health facility
Addition of the patient to the que Simulation statistics aggregation Termination of the simulation
Optimal hospital information Patient adding request
Finish request
Simulation initiation
Travel to patient s address Hospitalization decision-making Transportation of patient to hospital Termination of the simulation
Request from A to B
Figure 1: – Scheme of the interaction between models
3.2 Description of the multiagent transport model The transport model was developed based on multi-agent transport simulation package is described in the (Ivanov & Knyazkov 2014) and (Karbovskii et al. 2014). Imitational modeling of the work of ambulance services allows to assess the effectiveness of different dispatching decisions without interfering with the quality of treatment of actual patients. The imitational model used in this research consists of multiagent transport model reproducing the relocations of patients and a model of registration and procession of incoming calls for ambulance, described in section 1.3. It is based on the time-discreet leader-follower behavior model known as IDM (Intelligent Driver Model Treiber et al. 2000). Strictly speaking, he transport model proposed by this paper is a modification of IDM. In our model, IDM agents are only ambulances. Values of our model parameters are presented in Table 1. Automobile traffic is simulated using average values of speeds on the roads. When an ambulance is driving along the road, its desired speed is assigned the average speed value from the data set. Moving to the oncoming lane is simulated using the data on the road network. An ambulance moves to the oncoming lane if average speed on the current road is lower than the lowest speed boundary, and the speed of the oncoming lane is higher than the upper speed boundary. This method is based on a rough assumption, that high speed indicates a low density of cars. Boundary speed values are also presented in Table 1. On the oncoming lane, the ambulance drives with the desired speed.
2213
2214
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
Model parameter Vehicle Length Maximum Acceleration Desired deceleration Desired velocity Safe Time Interval Minimum distance Acceleration exponent The lowest speed boundary The upper speed boundary Table 1: Simulation parameters.
Parameter value 7 meters 8 meters per second 3 meters per second 60 kilometers per hour 1.8 second 0.5 meter 0.7 25 kilometers per hour 45 kilometers per hour
The algorithm for describing ambulance vehicle travel through road network is based on the regulations for transporting ambulance patients currently utilized in Saint Petersburg (Russia). At the beginning of the simulation, a crew and a vehicle are placed at the ambulance stations. When an incoming call is registered, a vehicle that has shortest estimated arrival time of is assigned to the address. At the next step, an optimal route regarding current traffic situation is generated. Upon the arrival to the address the vehicle is not serving other patients during the treatment time. If hospitalization is required, an optimal traffic-adjusted route to the hospital is generated. Upon transporting the patient to the hospital, an optimal route for a vehicle to reach the ambulance station is generated and it becomes available to new patients.
Figure 2: User interface devised for the imitational model software package As mentioned above, ambulance vehicle, as a special type of transport, has privileges over other types of transport. However, as literature survey suggests, the vast majority of the approaches to modeling ambulance vehicle travel do not account for such benefits. In order to tackle this issue, the transport model used in this research previously received a module for reproducing the authorized counter-flow movement of ambulance vehicles. In order to filter off the streets with a single-lane traffic, transport model uses the information about the availability of the second lane on the given road is derived from the OpenStreetMap data for Saint-Petersburg. In cases, where second lane is present, a decision to switch to it depends on two factors: a) own speed of the vehicle; b) mean traffic speed at the edges of the graph representing left (for the right-hand direction of traffic) or opposite-flow lane (if the density at the opposite lane is low and the speed of vehicles is high, ambulance vehicle can shift
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
lanes). Thus, the lower the ambulance vehicle’s own speed and density of the traffic at the opposing flow, the higher is the probability of shifting lanes. In this study, we have set thresholds for the own speed to 20 km/h and the density of the counter-flow traffic equal to the conditions where average speed exceeds 45 km/h. An example of the user interface from the imitational model software package is presented in Figure 2.
3.3 Mass Servicing System model in ambulance service In order to simulate in-hospital workload we use queueing model introduced in (Kovalchuk et al. 2016). This model describes activities of the hospital with several PCI facilities in case of defined inflow of patients. Simulation performed by the developed model provides detailed information on patients inflow processing. E.g., analysis of time spent by patients in queues can be used for predictive estimation of mortality rates. Next component
Planned flow generator
Planned queue Scheduler
Urgent flow generator
PCI unit 1
PCI unit 2
Urgent queue PCI unit 3
Figure 3: Mass Servicing System model. This model includes two patients flows – that is, planned and urgent patients. In waiting time distribution (including arrival rate and time of PCI) was estimated using data from Almazov Federal Medical Research Center, which serves (among others) as a destination for ambulances transporting the patients with ACS as well as regular patients scheduled for cardiological surgeries. In this paper urgent flow generator is replaced with flow generated according to chosen decision policy. Model structure is presented in fig. 3. It has two FIFO (first in, first out) queues for planned and urgent patients. The urgent queue always has a higher priority than the planned queue, but it has to be noticed if a planned operation started it cannot be interrupted. When one of the PCI units is available (operation is finished) the model scheduler selects a patient from the queues according to their priority status and directs them to the available PCI unit. The number of PCI units and their operational mode is based on real data for every hospital. In addition to patient waiting time and the time of PCI, the model takes into account the time needed to prepare the patient for the surgery and time of switching facilities from one PCI to another. As a result, the model provides rough estimation of door-to-balloon time (Soon et al. 2007) which correlates with mortality and complications rates after PCI. The amount of door-to-balloon time together with transportation time can be used to predict mortality or complication probability during certain periods of time with better accuracy.
4 Description of the experiment 4.1 Simulation scenario In order to study the distribution of time between the first contact with the patient and the beginning of the operation 2 simulation scenarios were launched. Each scenario simulates the work of emergency
2215
2216
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
services in St. Petersburg for 7 days. The scenarios differ in the selection method for the hospital to transport patients with ACS. In the first scenario (when hospitalization is required) the hospital with the shortest path time in selected. In the second scenario, in addition to the path time the hospital, availability of operating rooms is taken into account. In these scenarios calls to the ambulance service are divided into 2 types. The first type includes treatment with the ACS diagnosis. These scenarios are generated on the basis of real data on the intensity and location of calls with the diagnosis of acute coronary syndrome received in 2015 in St. Petersburg. The data were obtained using the ambulance service in St. Petersburg. According to the ambulance service, on the average they daily receive about 43 calls from of patients with ACS. In the simulation scenario, after receiving the first type call patient hospitalization in one of the city's medical institutions with equipment for PCI, is simulated. The second type of calls includes calls with all other diagnoses. These calls are simulated to take into account ambulances workload with other calls and its impact on providing service to patients with ACS. Data on the daily all intensity and the number of working ambulances were obtained from of the Saint-Petersburg Ambulance Service station website (www.03spb.ru). Calls location is simulated on the basis of the density of population of the city. The greater the number of people registered at a particular address, the more likely is a call from the given address. The data on the density of St. Petersburg population was obtained from the Federal Migration Service. In contrast to the calls of the first type, hospitalization of the second type patients (after providing initial medical care) is not simulated. When providing care to patients of the second type the average time of the process care is taken into account. The average time is known from the report by the Committee of Health of St. Petersburg. The methodology of obtaining data on the road network congestion is described in detail in (Derevitskiy et al. 2016). Data on the structure of the road network of St. Petersburg was obtained by OpenStreetMap. The used in the simulation scenarios data on the number of ambulances, ambulance stations, the number and location of health facilities receiving patients for PCI, the intensity of both types calls are illustrated in Table 2. Model parameter Number of ambulances (normalized to the intensity of the calls ACS) The daily calls rate ACS The daily calls to the service rate (normalized to the number of ambulances) The number of hospitals with PCI The number of ambulance stations The average time of providing initial medical care to patients with ACS The average time helping the rescue squad patients of the second type Table 2: Simulation parameters.
Parameter value 43 43 473 15 46 10 min 30 min
4.2 Time distribution analysis At the start of each simulation scenario statistics of the model were collected. For each patient with ACS the following data were recorded: ambulance travel time to the patient, the time of transporting the patient to the hospital, patient waiting time in the hospital. Distribution of the transportation times for both scenarios is illustrated in Figure 4 (in right). The distribution of the waiting time in hospitals is illustrated in Figure 4 (in left). Figure 4 (in left) illustrates the increase in the average transportation time in the second scenario, when compared with the first one. This is due to the fact that in the second scenario, the patient is not
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
always taken to the nearest hospital (in terms of transportation time). Figure 4 (in right) illustrates the reduction in patient waiting time in the second scenario if compared with the first one. The decrease results from the usage of the algorithm that takes into account queue to the operating room, in the second scenario. In fig. 4 we can observe a few number of significant delays (over 1500 minutes). In our experiment it happened because of occurrence of extremely long planned surgery, which is actually a rare event (less than 15 cases in a year), and combined with intense income flow it resulted in a long waiting time. Normally, in such cases patients from queue should be transferred to another hospital.
Figure 4: Distribution of time indicators
Figure 5: Distribution of time between the first contact with the patient and the beginning of PCI Figure 5 illustrates the distribution of time between the first contact with the patient and the beginning of PCI. Time is calculated as the sum of the time of providing the initial medical care on the
2217
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
2218
spot (in our model it is 10 minutes), the time of the patient transportation to the hospital, and the waiting time before operation. Results of the research show a reduction of the total time when the second strategy of choice of medical institution is applied. Thus, it can be concluded that the second strategy is more effective in terms of reducing the time between the first contact with the patient and the start of PCI.
5 Conclusion In this article, we present a methodology of the development of multicomponent simulation environment. The suggested tool is capable of simulating the ambulance service in town. One of the environment modules calculates the predicted time passing between the first contact with patient and the start of surgical intervention for patients with ACS. The described module can be helpful when making a decision on the thrombolytic therapy for a patient. Unlike existing navigation aids, transport model which is part of the described instrument takes into account the possibility of the ambulance moving into the oncoming lane. In future we are going to compare the effectiveness of ambulance travel time prediction made with the help of the suggested model, with the efficiency of the forecast by modern navigation systems. One of the perspectives of our further research is the use of distributed computing in the described multi-agent model. Tools and methods for distributed multi-agent simulation in the social sciences are described in (Gulyás et al. 2011). With the help of the simulation environment, we analyze the effectiveness of the strategy which takes into account queues in hospitals (in case of patients with ACS transported to hospitals). We conclude that this strategy is more effective in comparison with the strategy based on the forecast of the patient transportation time.
6 Acknowledgements This research is financially supported by The Russian Scientific Foundation, Agreement #14-1100823 (15.07.2014).
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
References Brandeau, M.L. & Larson, R.C., 1986. Extending and applying the hypercube queueing model to deploy ambulances in Boston. TIMS Studies in Munagement Science, 22, pp.121–153. Cochran, J.K. & Roche, K., 2008. A queuing-based decision support methodology to estimate hospital inpatient bed demand. Journal of the Operational Research Society, 59(11), pp.1471–1482. Créput, J.-C. et al., 2011. Dynamic vehicle routing problem for medical emergency management. In: Self Organizing Maps. Applications and Novel Algorithm Design, pp.233–250. Derevitskiy, I. et al., 2016. Traffic estimation on full graph of transport network using GPS data of bus movements. Procedia Computer Science, 101, pp.207–216. Gayathri, N. & Chandrakala, K.R.M.V., 2014. A novel technique for optimal vehicle routing. In Electronics and Communication Systems (ICECS), 2014 International Conference on Electronics and Communication System. pp. 1–5. Gendreau, M., Laporte, G. & Semet, F., 2001. A dynamic model and parallel tabu search heuristic for real-time ambulance relocation. Parallel computing, 27(12), pp.1641–1653. Green, L. V et al., 2006. Using queueing theory to increase the effectiveness of emergency department provider staffing. Academic Emergency Medicine, 13(1), pp.61–68. Gulyás, L. et al., 2011. Tools for large scale (distributed) agent-based computational experiments. Proceedings of CSSSA-11, Computational Social Science Society of America Annual Conference, Santa Fe, NM, USA, October 2011. Ivanov, S. V & Knyazkov, K. V, 2014. Evaluation of in-vehicle Decision Support System for Emergency Evacuation. Procedia Computer Science, 29, pp.1656–1666. Karbovskii, V. et al., 2016. Multimodel agent-based simulation environment for mass-gatherings and pedestrian dynamics. Future Generation Computer Systems. Karbovskii, V., Ivanov, S. & Knyazkov, K., 2014. A Multi-agent Simulation Of Human Behavior During Emergency Evacuations With In-vehicle Decision Support System. WIT Transactions on Information and Communication Technologies, 56, pp.635–644. Kergosien, Y. et al., 2015. A generic and flexible simulation-based analysis tool for EMS management. International Journal of Production Research, 53(24), pp.7299–7316. Kovalchuk, S. V et al., 2016. Distributed data-driven platform for urgent decision making in cardiological ambulance control. Future Generation Computer Systems. López, B. et al., 2005. A multi-agent system to support ambulance coordination in time-critical patient treatment. In 7th Simposio Argentino de Intelligencia Artificial-ASAI2005. Maxwell, M.S. et al., 2010. Approximate dynamic programming for ambulance redeployment. INFORMS Journal on Computing, 22(2), pp.266–281. Nordin, N.A.M. et al., 2011. An application of the A* algorithm on the ambulance routing. In Humanities, Science and Engineering (CHUSER), 2011 IEEE Colloquium on. pp. 855–859. Siddharthan, K., Jones, W.J. & Johnson, J.A., 1996. A priority queuing model to reduce waiting times in emergency care. International Journal of Health Care Quality Assurance, 9(5), pp.10–16. Soon, C.Y., Chan, W.X. & Tan, H.C., 2007. The impact of time-to-balloon on outcomes in patients undergoing modern primary angioplasty for acute myocardial infarction. Singapore medical journal, 48(2), p.131. Tereshchenko, S.N. & Zhirov, I.V., 2010. Treatment of the acute coronary syndrome with ST segment elevation at the pre-hospital care. Russian Cardiology Research and Production Complex. (In Russian), 6(3), pp.363–369. Treiber, M., Hennecke, A. & Helbing, D., 2000. Congested traffic states in empirical observations and microscopic simulations. Physical review E, 62(2), pp.1805–1824.
2219
ScienceDirect Procedia Computer Science 108C (2017) 2210–2219
International Conference on Computational Science, ICCS 2017, 12-14 June 2017, Zurich, Switzerland
Simulation of emergency care for patients with ACS in Saint Petersburg for ambulance decisionwith making Simulation of emergency care for patients ACS in Saint Petersburg for ambulance decision making 1 1 Ivan Derevitskiy , Evgeniy Krotov , Daniil Voloshin1, 1Alexey Yakovlev1,2 , Ivan Derevitskiy 1, Evgeniy Krotov1, 1 Sergey V. Kovalchuk , Vladislav Karbovskii 1 1,2, , Daniil Voloshin Alexey Yakovlev 1 ITMO University, Saint-Petersburg, Russian Federation. 1 1 2 V.Medical , Vladislav KarbovskiiRussian Kovalchuk Federal AlmazovSergey North-West Research Centre, Saint-Petersburg, Federation
1 ITMO University,[email protected], Saint-Petersburg, Russian Federation. [email protected], [email protected], Federal Almazov North-West Medical Research Centre, Saint-Petersburg, Russian Federation [email protected], [email protected], [email protected] [email protected], [email protected], [email protected], [email protected], [email protected], [email protected] 2
Abstract One of the stages of emergency medical care in case of Acute Coronary Syndrome (ACS) (if there are Abstractconditions for surgical intervention) is directly linked to the time between the first contact medical One of stagesand of the emergency medical in case ofcoronary Acute Coronary Syndrome (if there are Time with thethe patient and inflating the care balloon in the artery (in a medical(ACS) institution). medical conditions surgical facility intervention) is on directly linked to the delivery time between the first of the operation startfora medical depends the time of patient to hospital, as contact well as withthethewaiting patienttime and the andqueue inflating theinstitution. balloon in the artery (ina adevelopment medical institution). Time Thiscoronary paper describes of ambulance on in the in the of the operation start aaggregate medical facility depends on thetwo timeperiods of patient delivery to hospital, as time well as model for obtaining estimation of these of time. The estimation is on the waiting time of in the queue in paper describes a development of ambulance ambulance obtained by means described in the the institution. article the This decision support system (DSS) in the model obtaining aggregate estimation of these time. The estimation time(the is service.for Unlike modern navigation systems DSS takestwo intoperiods accountofambulance vehicle behavior obtained of described in the the decision system (DSS) in the the help ambulance With of the ability to by exitmeans into oncoming traffic) andarticle availability of freesupport operation rooms. service. Unlike modern navigation DSSservice takes into accountout ambulance behavior (the described simulation model of the systems ambulance we carried the timevehicle distribution analysis ability to exit into oncoming traffic) and availability of free operation rooms. With the help of the (between the first contact with the patient and surgical intervention in case of ACS) in St. Petersburg, described simulation modeluses of the carried out the in time analysis Russia. Simulation scenario realambulance data on theservice work ofwe ambulance service thedistribution city. (between the first contact with the patient and surgical intervention in case of ACS) in St. Petersburg, © 2017 The Authors. Published by Elsevier B.V. Russia. Simulation scenario uses real data on the work of ambulance serviceambulance in the city. Keywords: acute coronary syndrome, transport modeling, decision support system, service Peer-review under responsibility of the scientific committee of the International Conference on Computational Science Keywords: acute coronary syndrome, transport modeling, decision support system, ambulance service
1 Introduction Modern ambulance service 1 Introduction
is a dynamic multi-functional system, which has a complicated maintenance mechanism. Relevance of the research decision support systems in the ambulance service Modern service is a dynamic multi-functional system, has of a patient complicated is caused ambulance by the necessity to optimize decision-making, in order to improvewhich the quality care. maintenance mechanism. decision support in the ambulance service Among the key features ofRelevance the work of the research ambulance service, we cansystems highlight increased need for is caused by of theevery necessity to optimize decision-making, in orderlife to improve theThus, quality patient care. "reliability" decision, due to its importance for human and health. theofuse of DSS is Among the key features of the work of the ambulance service, we can highlight the increased need for "reliability" of every decision, due to its importance for human life and health. Thus, the use of DSS is 1877-0509 © 2017 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the scientific committee of the International Conference on Computational Science 10.1016/j.procs.2017.05.178
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
an effective approach that automates processing large amounts of information. Another key ambulance service feature is the high dynamics of the service work. High dynamics results from little time to make a decision that healthcare providers usually have. Consequently, the DSS can be used to reduce the time of decision-making in complicated cases by automating the analysis of the situation. This paper describes the methodology of development of DSS in the ambulance service. One of the DSS modules solves the problem of time estimation between the first contact with the patient and the beginning of the surgical intervention for patients with ACS. This estimate was obtained using a complex model, which includes a transport model, (that assessed the time of the patient delivery to the hospital), flow of calls model, as well as the queuing model that evaluates the waiting time. According to the paper, by (Tereshchenko & Zhirov 2010), in the provision of care for patients with ACS (patients with medical conditions for Percutaneous Coronary Intervention, PCI) the algorithm of actions of a medical brigade depends on the predicted time of start PCI. Tereshchenko and Jirov state that if the time between the first contact with medical personnel and inflating the balloon in the SC (coronary artery) exceeds 90 minutes, it is preferable to use thrombolytic therapy. If the predicted time is less than 90 minutes, then the use of invasive strategy is preferable. Thus, the task of predicting the time between the first contact of medical staff with patients and the beginning of the operation is an urgent one. The quality of the forecast impacts the patient's treatment strategy.
2 Related Works 2.1 Time estimation of the patient transportation to a hospital The period time between the first contact with the patient and the start of the operation can be divided into two parts. The first step is to calculate the patient delivery time. Delivery time is calculated on the basis of the selected route from the patient initial location to the hospital. Thus, when calculating the patient delivery time, it is necessary to find the fastest route. The problem of finding the fastest routes for ambulances may be considered from the perspective of the search of the optimal path in a weighted graph. To find the way we must take into account dynamic changes in the traffic situation and also the specific behavior of the ambulance vehicles. One of solutions to the problem of the ambulance fastest path search is proposed in (Gayathri & Chandrakala 2014). The authors apply Dijkstra algorithm of the shortest path search in a weighted graph, which takes into account the current state of traffic. The paper (Nordin et al. 2011) proposes a more efficient (in terms of performance) algorithm A * of finding the shortest path in the graph. In (Maxwell et al. 2010) the shortest route is searched using ADP dynamic programming techniques. In (Gendreau et al. 2001) ambulance route search problem is solved by a heuristic tabu search algorithm. In (Kergosien et al. 2015) it is proposed to seek the shortest path between two points and estimate the time of its passage, on the basis statistical data of prior ambulance trips, and comparing estimated routs to the current ones. Data of base previous ambulance trips helping in choosing the optimal route is also described in (López et al. 2005). In (Créput et al. 2011) proposed to use the method of neural networks (Kohonen self-organizing maps) to find the shortest path for an ambulance. One of the disadvantages of these approaches is the lack of consideration of the special behavior of ambulances on the roads. For example, the possibility of entering the opposite lane can change route selection. Unlike the above mentioned approaches, in the ambulances movement model, described in this paper, simulates travel into the oncoming lane.
2.2 Queuing models in ambulance service The second step of estimating time from the first contact with patient to the moment of surgery beginning is to calculate waiting time which patient has to spent in hospital after arrival.
2211
2212
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
In cardiology that time interval is called door-to-balloon time. Due to limited resources of PCI facilities in hospitals queues can emerge and in some cases queuing time can exceed the time of transportation. Applications of queuing theory for solving optimization problems in the ambulance is a widely used practice. One of the most well-known example is the hypercube model for solving the problem of the distribution ambulance fleet (Brandeau & Larson 1986). The advantage of that model is the possibility of finding the analytical solutions of the optimization problem. However, with increasing complexity of the model, as an instance introduction of priority levels (Siddharthan et al. 1996) of calls significantly complicates the solution of the problem. Also for simplicity of patient flow simulation arrival rates are modeled as a homogeneous Poisson process (Green et al. 2006), however, in order to take into account time dependence of the arrival rate we need to use non-homogeneous process, but finding parameters of its distribution is a complex task. In our work we use the empirical distribution of arrival rate instead of analytical distribution. There are some other applications of queuing theory, in (Cochran & Roche 2008) queuing model is used to find the optimal number of beds in hospitals. However, the main idea of the model is the same. Complex model presented in this paper which combined a queueing and transportation model allow us not only to solve optimization problems of ambulance fleet and hospitals resources allocation, but also a problem of finding the optimal ambulance dispatching policy.
3 Description of the model 3.1 Simulation environment description In this paper, we present the imitational simulation environment. The described instrument is based on Pulse tool, described in (Karbovskii et al. 2016). The function of the environment is to simulate the operation of the ambulance at a given locality. The ambulance service model consists of two autonomous units: transport model and health facilities queuing model. A transport model for ambulance services has been devised on previous work described in (Ivanov & Knyazkov 2014). It imitates the generation of incoming ambulance calls and dispatching (assignment of crews and vehicles to patients) process. The realization of the model used here has been implemented using C# programming language. A comprehensive description of the model is presented further in section 3.2. The queuing model used in this research has been previously discussed in (Kovalchuk et al. 2016). Based on the data about the flow of planned and emergency patients the model imitates the ques inside health facilities. More detailed description of the model can be found in section 3.3. Interaction between the described units of the imitational simulation environment is performed through POST http requests sent by the transport model to the server component of the health facilities queuing model. Transport model issues requests of the following types: initiation of the queuing model, request for id of the health facility patient shall be transported to (if the medical indicators speak in favor of hospitalization), request-notification of patients’ arrival to the hospital, termination of the queuing model work. At the beginning of the ambulance service operation simulation, the POST request that initiates the queuing model is sent. When the decision to hospitalize a patient is made, a request for a receiving health facility is sent to the transport model. Arguments for the request are the estimates of the time needed to transport patients to the health facilities suitable for a them based on the diagnosis-specific requirements, current model simulation time and patient’s id. The location of the facility that has been retrieved through the id request is considered to be optimal for patient’s transportation. Notificationrequest indicating the arrival of the patient serves to assign the patient to the que of one of the health facilities. Here arguments are patients and health facility id’s and current simulation time. After the
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
request of the last type has been sent and simulation statistics are aggregated, simulation terminates. The outline of the Interaction between simulation environment’s components is presented in Figure 1. Health facility queuing model Simulation initiation
Ambulance service transport model Start request
Optimal hospital request Assignment to health facility
Addition of the patient to the que Simulation statistics aggregation Termination of the simulation
Optimal hospital information Patient adding request
Finish request
Simulation initiation
Travel to patient s address Hospitalization decision-making Transportation of patient to hospital Termination of the simulation
Request from A to B
Figure 1: – Scheme of the interaction between models
3.2 Description of the multiagent transport model The transport model was developed based on multi-agent transport simulation package is described in the (Ivanov & Knyazkov 2014) and (Karbovskii et al. 2014). Imitational modeling of the work of ambulance services allows to assess the effectiveness of different dispatching decisions without interfering with the quality of treatment of actual patients. The imitational model used in this research consists of multiagent transport model reproducing the relocations of patients and a model of registration and procession of incoming calls for ambulance, described in section 1.3. It is based on the time-discreet leader-follower behavior model known as IDM (Intelligent Driver Model Treiber et al. 2000). Strictly speaking, he transport model proposed by this paper is a modification of IDM. In our model, IDM agents are only ambulances. Values of our model parameters are presented in Table 1. Automobile traffic is simulated using average values of speeds on the roads. When an ambulance is driving along the road, its desired speed is assigned the average speed value from the data set. Moving to the oncoming lane is simulated using the data on the road network. An ambulance moves to the oncoming lane if average speed on the current road is lower than the lowest speed boundary, and the speed of the oncoming lane is higher than the upper speed boundary. This method is based on a rough assumption, that high speed indicates a low density of cars. Boundary speed values are also presented in Table 1. On the oncoming lane, the ambulance drives with the desired speed.
2213
2214
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
Model parameter Vehicle Length Maximum Acceleration Desired deceleration Desired velocity Safe Time Interval Minimum distance Acceleration exponent The lowest speed boundary The upper speed boundary Table 1: Simulation parameters.
Parameter value 7 meters 8 meters per second 3 meters per second 60 kilometers per hour 1.8 second 0.5 meter 0.7 25 kilometers per hour 45 kilometers per hour
The algorithm for describing ambulance vehicle travel through road network is based on the regulations for transporting ambulance patients currently utilized in Saint Petersburg (Russia). At the beginning of the simulation, a crew and a vehicle are placed at the ambulance stations. When an incoming call is registered, a vehicle that has shortest estimated arrival time of is assigned to the address. At the next step, an optimal route regarding current traffic situation is generated. Upon the arrival to the address the vehicle is not serving other patients during the treatment time. If hospitalization is required, an optimal traffic-adjusted route to the hospital is generated. Upon transporting the patient to the hospital, an optimal route for a vehicle to reach the ambulance station is generated and it becomes available to new patients.
Figure 2: User interface devised for the imitational model software package As mentioned above, ambulance vehicle, as a special type of transport, has privileges over other types of transport. However, as literature survey suggests, the vast majority of the approaches to modeling ambulance vehicle travel do not account for such benefits. In order to tackle this issue, the transport model used in this research previously received a module for reproducing the authorized counter-flow movement of ambulance vehicles. In order to filter off the streets with a single-lane traffic, transport model uses the information about the availability of the second lane on the given road is derived from the OpenStreetMap data for Saint-Petersburg. In cases, where second lane is present, a decision to switch to it depends on two factors: a) own speed of the vehicle; b) mean traffic speed at the edges of the graph representing left (for the right-hand direction of traffic) or opposite-flow lane (if the density at the opposite lane is low and the speed of vehicles is high, ambulance vehicle can shift
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
lanes). Thus, the lower the ambulance vehicle’s own speed and density of the traffic at the opposing flow, the higher is the probability of shifting lanes. In this study, we have set thresholds for the own speed to 20 km/h and the density of the counter-flow traffic equal to the conditions where average speed exceeds 45 km/h. An example of the user interface from the imitational model software package is presented in Figure 2.
3.3 Mass Servicing System model in ambulance service In order to simulate in-hospital workload we use queueing model introduced in (Kovalchuk et al. 2016). This model describes activities of the hospital with several PCI facilities in case of defined inflow of patients. Simulation performed by the developed model provides detailed information on patients inflow processing. E.g., analysis of time spent by patients in queues can be used for predictive estimation of mortality rates. Next component
Planned flow generator
Planned queue Scheduler
Urgent flow generator
PCI unit 1
PCI unit 2
Urgent queue PCI unit 3
Figure 3: Mass Servicing System model. This model includes two patients flows – that is, planned and urgent patients. In waiting time distribution (including arrival rate and time of PCI) was estimated using data from Almazov Federal Medical Research Center, which serves (among others) as a destination for ambulances transporting the patients with ACS as well as regular patients scheduled for cardiological surgeries. In this paper urgent flow generator is replaced with flow generated according to chosen decision policy. Model structure is presented in fig. 3. It has two FIFO (first in, first out) queues for planned and urgent patients. The urgent queue always has a higher priority than the planned queue, but it has to be noticed if a planned operation started it cannot be interrupted. When one of the PCI units is available (operation is finished) the model scheduler selects a patient from the queues according to their priority status and directs them to the available PCI unit. The number of PCI units and their operational mode is based on real data for every hospital. In addition to patient waiting time and the time of PCI, the model takes into account the time needed to prepare the patient for the surgery and time of switching facilities from one PCI to another. As a result, the model provides rough estimation of door-to-balloon time (Soon et al. 2007) which correlates with mortality and complications rates after PCI. The amount of door-to-balloon time together with transportation time can be used to predict mortality or complication probability during certain periods of time with better accuracy.
4 Description of the experiment 4.1 Simulation scenario In order to study the distribution of time between the first contact with the patient and the beginning of the operation 2 simulation scenarios were launched. Each scenario simulates the work of emergency
2215
2216
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
services in St. Petersburg for 7 days. The scenarios differ in the selection method for the hospital to transport patients with ACS. In the first scenario (when hospitalization is required) the hospital with the shortest path time in selected. In the second scenario, in addition to the path time the hospital, availability of operating rooms is taken into account. In these scenarios calls to the ambulance service are divided into 2 types. The first type includes treatment with the ACS diagnosis. These scenarios are generated on the basis of real data on the intensity and location of calls with the diagnosis of acute coronary syndrome received in 2015 in St. Petersburg. The data were obtained using the ambulance service in St. Petersburg. According to the ambulance service, on the average they daily receive about 43 calls from of patients with ACS. In the simulation scenario, after receiving the first type call patient hospitalization in one of the city's medical institutions with equipment for PCI, is simulated. The second type of calls includes calls with all other diagnoses. These calls are simulated to take into account ambulances workload with other calls and its impact on providing service to patients with ACS. Data on the daily all intensity and the number of working ambulances were obtained from of the Saint-Petersburg Ambulance Service station website (www.03spb.ru). Calls location is simulated on the basis of the density of population of the city. The greater the number of people registered at a particular address, the more likely is a call from the given address. The data on the density of St. Petersburg population was obtained from the Federal Migration Service. In contrast to the calls of the first type, hospitalization of the second type patients (after providing initial medical care) is not simulated. When providing care to patients of the second type the average time of the process care is taken into account. The average time is known from the report by the Committee of Health of St. Petersburg. The methodology of obtaining data on the road network congestion is described in detail in (Derevitskiy et al. 2016). Data on the structure of the road network of St. Petersburg was obtained by OpenStreetMap. The used in the simulation scenarios data on the number of ambulances, ambulance stations, the number and location of health facilities receiving patients for PCI, the intensity of both types calls are illustrated in Table 2. Model parameter Number of ambulances (normalized to the intensity of the calls ACS) The daily calls rate ACS The daily calls to the service rate (normalized to the number of ambulances) The number of hospitals with PCI The number of ambulance stations The average time of providing initial medical care to patients with ACS The average time helping the rescue squad patients of the second type Table 2: Simulation parameters.
Parameter value 43 43 473 15 46 10 min 30 min
4.2 Time distribution analysis At the start of each simulation scenario statistics of the model were collected. For each patient with ACS the following data were recorded: ambulance travel time to the patient, the time of transporting the patient to the hospital, patient waiting time in the hospital. Distribution of the transportation times for both scenarios is illustrated in Figure 4 (in right). The distribution of the waiting time in hospitals is illustrated in Figure 4 (in left). Figure 4 (in left) illustrates the increase in the average transportation time in the second scenario, when compared with the first one. This is due to the fact that in the second scenario, the patient is not
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
always taken to the nearest hospital (in terms of transportation time). Figure 4 (in right) illustrates the reduction in patient waiting time in the second scenario if compared with the first one. The decrease results from the usage of the algorithm that takes into account queue to the operating room, in the second scenario. In fig. 4 we can observe a few number of significant delays (over 1500 minutes). In our experiment it happened because of occurrence of extremely long planned surgery, which is actually a rare event (less than 15 cases in a year), and combined with intense income flow it resulted in a long waiting time. Normally, in such cases patients from queue should be transferred to another hospital.
Figure 4: Distribution of time indicators
Figure 5: Distribution of time between the first contact with the patient and the beginning of PCI Figure 5 illustrates the distribution of time between the first contact with the patient and the beginning of PCI. Time is calculated as the sum of the time of providing the initial medical care on the
2217
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
2218
spot (in our model it is 10 minutes), the time of the patient transportation to the hospital, and the waiting time before operation. Results of the research show a reduction of the total time when the second strategy of choice of medical institution is applied. Thus, it can be concluded that the second strategy is more effective in terms of reducing the time between the first contact with the patient and the start of PCI.
5 Conclusion In this article, we present a methodology of the development of multicomponent simulation environment. The suggested tool is capable of simulating the ambulance service in town. One of the environment modules calculates the predicted time passing between the first contact with patient and the start of surgical intervention for patients with ACS. The described module can be helpful when making a decision on the thrombolytic therapy for a patient. Unlike existing navigation aids, transport model which is part of the described instrument takes into account the possibility of the ambulance moving into the oncoming lane. In future we are going to compare the effectiveness of ambulance travel time prediction made with the help of the suggested model, with the efficiency of the forecast by modern navigation systems. One of the perspectives of our further research is the use of distributed computing in the described multi-agent model. Tools and methods for distributed multi-agent simulation in the social sciences are described in (Gulyás et al. 2011). With the help of the simulation environment, we analyze the effectiveness of the strategy which takes into account queues in hospitals (in case of patients with ACS transported to hospitals). We conclude that this strategy is more effective in comparison with the strategy based on the forecast of the patient transportation time.
6 Acknowledgements This research is financially supported by The Russian Scientific Foundation, Agreement #14-1100823 (15.07.2014).
Ivan Derevitskiy et al. / Procedia Computer Science 108C (2017) 2210–2219
References Brandeau, M.L. & Larson, R.C., 1986. Extending and applying the hypercube queueing model to deploy ambulances in Boston. TIMS Studies in Munagement Science, 22, pp.121–153. Cochran, J.K. & Roche, K., 2008. A queuing-based decision support methodology to estimate hospital inpatient bed demand. Journal of the Operational Research Society, 59(11), pp.1471–1482. Créput, J.-C. et al., 2011. Dynamic vehicle routing problem for medical emergency management. In: Self Organizing Maps. Applications and Novel Algorithm Design, pp.233–250. Derevitskiy, I. et al., 2016. Traffic estimation on full graph of transport network using GPS data of bus movements. Procedia Computer Science, 101, pp.207–216. Gayathri, N. & Chandrakala, K.R.M.V., 2014. A novel technique for optimal vehicle routing. In Electronics and Communication Systems (ICECS), 2014 International Conference on Electronics and Communication System. pp. 1–5. Gendreau, M., Laporte, G. & Semet, F., 2001. A dynamic model and parallel tabu search heuristic for real-time ambulance relocation. Parallel computing, 27(12), pp.1641–1653. Green, L. V et al., 2006. Using queueing theory to increase the effectiveness of emergency department provider staffing. Academic Emergency Medicine, 13(1), pp.61–68. Gulyás, L. et al., 2011. Tools for large scale (distributed) agent-based computational experiments. Proceedings of CSSSA-11, Computational Social Science Society of America Annual Conference, Santa Fe, NM, USA, October 2011. Ivanov, S. V & Knyazkov, K. V, 2014. Evaluation of in-vehicle Decision Support System for Emergency Evacuation. Procedia Computer Science, 29, pp.1656–1666. Karbovskii, V. et al., 2016. Multimodel agent-based simulation environment for mass-gatherings and pedestrian dynamics. Future Generation Computer Systems. Karbovskii, V., Ivanov, S. & Knyazkov, K., 2014. A Multi-agent Simulation Of Human Behavior During Emergency Evacuations With In-vehicle Decision Support System. WIT Transactions on Information and Communication Technologies, 56, pp.635–644. Kergosien, Y. et al., 2015. A generic and flexible simulation-based analysis tool for EMS management. International Journal of Production Research, 53(24), pp.7299–7316. Kovalchuk, S. V et al., 2016. Distributed data-driven platform for urgent decision making in cardiological ambulance control. Future Generation Computer Systems. López, B. et al., 2005. A multi-agent system to support ambulance coordination in time-critical patient treatment. In 7th Simposio Argentino de Intelligencia Artificial-ASAI2005. Maxwell, M.S. et al., 2010. Approximate dynamic programming for ambulance redeployment. INFORMS Journal on Computing, 22(2), pp.266–281. Nordin, N.A.M. et al., 2011. An application of the A* algorithm on the ambulance routing. In Humanities, Science and Engineering (CHUSER), 2011 IEEE Colloquium on. pp. 855–859. Siddharthan, K., Jones, W.J. & Johnson, J.A., 1996. A priority queuing model to reduce waiting times in emergency care. International Journal of Health Care Quality Assurance, 9(5), pp.10–16. Soon, C.Y., Chan, W.X. & Tan, H.C., 2007. The impact of time-to-balloon on outcomes in patients undergoing modern primary angioplasty for acute myocardial infarction. Singapore medical journal, 48(2), p.131. Tereshchenko, S.N. & Zhirov, I.V., 2010. Treatment of the acute coronary syndrome with ST segment elevation at the pre-hospital care. Russian Cardiology Research and Production Complex. (In Russian), 6(3), pp.363–369. Treiber, M., Hennecke, A. & Helbing, D., 2000. Congested traffic states in empirical observations and microscopic simulations. Physical review E, 62(2), pp.1805–1824.
2219