ArticlePDF Available

Abstract and Figures

In this paper we study the coordination of Emergency Medical Service (EMS) for patients with acute myocardial infarction with ST-segment elevation (STEMI). This is a health problem with high associated mortality. A ``golden standard'' treatment for STEMI is angioplasty, which requires a catheterization lab and a highly qualified cardiology team. It should be performed as soon as possible since the delay to treatment worsens the patient's prognosis. The decrease of the delay is achieved by coordination of EMS, which is especially important in the case of multiple simultaneous patients. Nowadays, this process is based on the First-Come-First-Served (FCFS) principle and it heavily depends on human control and phone communication with high proneness to human error and delays. The objective is, therefore, to automate the EMS coordination while minimizing the time from symptom onset to reperfusion and thus to lower the mortality and morbidity resulting from this disease. In this paper, we present a multi-agent decision-support system for the distributed coordination of EMS focusing on urgent out-of-hospital STEMI patients awaiting angioplasty. The system is also applicable to emergency patients of any pathology needing pre-hospital acute medical care and urgent hospital treatment. The assignment of patients to ambulances and angioplasty-enabled hospitals with cardiology teams is performed via a three-level optimization model. At each level, we find a globally efficient solution by a modification of the distributed relaxation method for the assignment problem called the auction algorithm. The efficiency of the proposed model is demonstrated by simulation experiments.
Content may be subject to copyright.
manuscript No.
(will be inserted by the editor)
Distributed coordination of emergency medical service for
angioplasty patients
Marin Lujak, Holger Billhardt, Sascha Ossowski
Received: date / Accepted: date
Abstract In this paper we study the coordination of Emergency Medical Service (EMS)
for patients with acute myocardial infarction with ST-segment elevation (STEMI). This is
a health problem with high associated mortality. A “golden standard” treatment for STEMI
is angioplasty, which requires a catheterization lab and a highly qualified cardiology team.
It should be performed as soon as possible since the delay to treatment worsens the pa-
tient’s prognosis. The decrease of the delay is achieved by coordination of EMS, which is
especially important in the case of multiple simultaneous patients. Nowadays, this process
is based on the First-Come-First-Served (FCFS) principle and it heavily depends on hu-
man control and phone communication with high proneness to human error and delays. The
objective is, therefore, to automate the EMS coordination while minimizing the time from
symptom onset to reperfusion and thus to lower the mortality and morbidity resulting from
this disease. In this paper, we present a multi-agent decision-support system for the dis-
tributed coordination of EMS focusing on urgent out-of-hospital STEMI patients awaiting
angioplasty. The system is also applicable to emergency patients of any pathology needing
pre-hospital acute medical care and urgent hospital treatment. The assignment of patients
to ambulances and angioplasty-enabled hospitals with cardiology teams is performed via a
three-level optimization model. At each level, we find a globally efficient solution by a mod-
ification of the distributed relaxation method for the assignment problem called the auction
algorithm. The efficiency of the proposed model is demonstrated by simulation experiments.
Keywords EMS Coordination ·Ambulance Coordination ·Angioplasty ·PCI ·Distributed
Optimization ·Auction Algorithm ·Emergency Medical Assistance
1 Introduction
Based on the World Health Organization data, ischemic heart disease (IHD) is the single
most frequent cause of death killing 7.4 million people in 2012, which is 13.2%of all the
deaths worldwide [43]. It is a disease characterized by ischaemia (reduced blood supply) of
the heart muscle, usually due to coronary artery disease. At any stage of coronary artery dis-
ease, the acute rupture of an atheromatous plaque may lead to an acute myocardial infarction
(AMI), also called a heart attack [45].
University Rey Juan Carlos, Madrid, Spain, E-mail: firstname.lastname@urjc.es
Determined by electrocardiographic findings, AMI can be classified into acute myocar-
dial infarction with ST-segment elevations (STEMI) and without ST elevation (NSTEMI).
Effective and rapid coronary reperfusion is the most important goal in the treatment of pa-
tients with STEMI. One of the reperfusion methods is angioplasty or primary percutaneous
coronary intervention (PCI), which is a non-surgical procedure used to treat the stenotic
coronary arteries of the heart. It is the preferred treatment when feasible and when per-
formed within 90 minutes after the first medical contact [48].
Angioplasty should be performed by a highly specialized cardiology team in a catheteri-
zation lab. A catheterization lab is an examination room in a hospital or clinic with diagnos-
tic imaging equipment used to visualize the arteries and the chambers of the heart and treat
any stenosis or abnormality found. Since the equipment of the lab is very costly, it is only
sporadically present in hospitals. Even more, due to high costs of a cardiology team nec-
essary for the realization of angioplasty (usually consisting of a specialized interventional
cardiologist or radiologist, a cardiac physiologist, a nurse and a radiographer), angioplasty-
enabled hospitals are generally sparse.
The mortality of STEMI is influenced by many factors, among them: patient delay time,
age, treatment, history of prior myocardial infarction, and the number of diseased coronary
arteries. The patient delay time, defined as the period from the onset of STEMI symptoms
to the provision of reperfusion therapy, is an important determinant of the effectiveness of
angioplasty and the clinical outcome of STEMI patients [11, 40]. Due to insufficient EMS
coordination and organizational issues, elevated patient delay time remains a major reason
why angioplasty has not become the definitive treatment in many hospitals.
In this paper, we develop a decision-support system for the coordination of EMS focus-
ing on urgent out-of-hospital STEMI patients awaiting angioplasty (angioplasty patients)
but also applicable to emergency patients of any pathology needing pre-hospital acute med-
ical care and urgent hospital treatment. We propose a change of the centralized hierarchy-
oriented organizational structure to a patient-oriented distributed organizational structure of
EMS that increases the flexibility, scalability, and the responsiveness of the EMS system.
The proposed decision-support system is based on the integration and coordination of all
the phases EMS participants go through in the process of emergency medical assistance
(EMA). The model takes into consideration ambulances’, patients’, hospitals’, and cardiol-
ogy teams’ real-time positions for real-time assignment of patients to the EMS resources.
The objective of the proposed system is the reduction of patient delay times by distributed
real-time optimization of decision-making processes.
In more detail, we mathematically model patient delay time and present a three-level
problem decomposition for the minimization of combined arrival times of multiple EMS ac-
tors necessary for angioplasty. For the three decomposition levels, we propose a distributed
EMS coordination approach and modify the auction algorithm proposed by Bertsekas in
[5] for the specific case. The latter is a distributed relaxation method that finds an optimal
solution to the assignment problem.
On the first level, agents representing ambulances find in a distributed way the patient
assignment that minimizes arrival times of available ambulances to patients. After the treat-
ment in situ, on the second optimization level, ambulances carrying patients are assigned
to available hospitals. On the third level, arrival times of cardiology teams to hospitals are
coordinated with the arrival times of patients.
The proposed approach is based on a global view, not concentrating only on minimiz-
ing single patient delay time, but obtaining the EMS system’s best solution with respect to
the (temporal and spatial) distribution of patients in a region of interest. Simulated emer-
gency scenarios demonstrate the efficiency of the coordination procedure and a significant
reduction in the average patient delay time.
This paper is organized as follows. Section 2 treats the State-of-the-Art in EMA coor-
dination for emergency patients in general and angioplasty patients in particular. In Section
3, we introduce the delays related with the EMS coordination for angioplasty patients. The
importance of reducing delays in angioplasty procedure is demonstrated on the case of the
EMS provider of the Community of Madrid in Spain, SUMMA 112, in Section 4. In Section
5, we formulate the EMS coordination problem for simultaneous EMA of multiple angio-
plasty patients. Section 6 describes the proposed multi-agent architecture with the modified
auction algorithm for the coordination of participants in EMA for angioplasty. Section 7
contains simulation results comparing the proposed coordination approach and the bench-
mark First-Come-First-Served (FCFS) principle used presently in most of the Western world
countries. We draw conclusions and outline the directions for future work in Section 8.
2 State-of-the-Art in emergency medical assistance coordination
In the conventional EMA coordination approach, a medical emergency coordination center
(ECC) is the only EMA coordinator. The region of interest is divided into geographically
separated areas for which a team of human operators coordinates the ambulance fleet and
hospitals for emergency patients’ assistance. Such a fragmented EMS coordination approach
often presents a suboptimal coordination solution due to the lack of the system’s global view
and high proneness to human error and delays. Furthermore, its centralized organizational
structure is a bottleneck of the EMS, which, in case of contingencies, can cause the halt of
the whole system.
Usually, the ECC has real-time information of the states of ambulances and hospitals. It
dispatches the nearest available (idle) ambulance with Advanced Life Support (ALS) to the
patient, thus applying the FCFS strategy. Even though this strategy is optimal in the case of
a single patient awaiting EMA, in the case of multiple simultaneously appearing patients, it
provides suboptimal results as it discriminates against patients appearing later [8].
When an ambulance arrives to the scene, it diagnoses AMI by an electrocardiogram,
confirms the diagnosis to the ECC, provides in-situ assistance, and takes the patient to the
hospital assigned by the ECC. The ECC applies the FCFS strategy for hospital assignment
by assigning the patient to the nearest available hospital with a catheterization lab while the
hospital alerts its closest cardiology team to the case. Furthermore, the ECC transfers patient
identity and clinical data, treatment and expected arrival times to all the EMS participants
involved and creates a record of the patient movement.
The right choice of the ambulance to be assigned to a pending out-of-hospital patient re-
duces total patient delay, which can significantly improve patients’ chances and reduce mor-
bidity and mortality caused by any emergency pathology including STEMI. Consequently,
EMSs seek to minimize ambulance arrival times to patients and, if hospitalization is neces-
sary, minimize the time of the patient’s transport to an adequate nearby hospital.
Not all hospitals can treat angioplasty patients due to the lack of technical means or
cardiology teams or both. Usually, each hospital with a catheterization lab has assigned to it
its own cardiology team(s) located at alert outside hospital and obliged to reach the hospital
in the case of emergency. The reason for their outside hospital location is the cardiology
teams’ cost that constitutes a large portion of the costs in surgical services [22]. Furthermore,
significant cost savings can be obtained if cardiology teams work for multiple hospitals [12].
When a patient arrives to the hospital, the angioplasty procedure itself generally takes
about 30 minutes but in specific cases it can last up to 2 and a half hours. The duration is
dependent upon the technical difficulty of the case and the number of balloon catheters that
have to be employed. The procedure’s duration can be forecasted with prior angiography.
The most important factor in maximizing emergency patient’s survivability is a timely
and effective treatment. The EMS coordination process usually implemented in the ECCs is
manual and the management is based on a case-by-case principle with high human workload
necessary for telephone arrangements to find a solution. In the case of the presence of an
increased number of angioplasty patients and different medical teams located in multiple
sites, support for optimized EMS coordination based on information updated in real time is
necessary for efficient angioplasty planning and scheduling.
There is a significant volume of research on EMS efficiency improvements related to
ambulance coordination models for emergency patients e.g., [8, 20] and operating room
planning and scheduling, e.g., [12]. The basis of the proposed methods is mostly queuing
theory, simulations and mathematical programming, e.g., [12, 23, 39].
Domnori et al. in [18] discuss the suitability of agent-based applications to managing
healthcare emergencies and large scale disasters and their application to problems where
the main challenge is coordination and collaboration between participants. Furthermore,
L´
opez et al. in [32] propose a multiagent system using an auction mechanism based on
trust to coordinate ambulances for emergency medical services and in [31] present a multi-
agent system with auctions, MASICTUS, with the aim of supporting the diagnosis of acute
stroke diseases while coordinating ambulance services and expert neurologists for patient
attendance. The auction mechanism here is based on three patient priority cases where the
winning ambulance has the best expected arrival time and a good trust degree.
In [1], Bandara et al. study optimal dispatch of paramedic units to emergency calls to
maximise patients’ survivability. They show that the FCFS policy is not always optimal and
that dispatching ambulances considering the priority of the call leads to an increase in the
average survival probability of patients. However, they do not research possible efficiency
improvements within the patients of the same priority.
In [8], we propose mechanisms that dynamically reduce patient delay through efficient
assignment of ambulances to patients as well as the redeployment of available ambulances
in the region of interest. We test these mechanisms in different experiments using historical
data from SUMMA 112. The results empirically confirm that our proposal reduces signif-
icantly the EMA average response times. Moreover, in [33] we propose an organization-
based multi-agent application for EMA based on the auction algorithm [5] where EMS par-
ticipants are considered with different trust levels. The simulation results confirm further
improvements in shortening EMA delays. Additionally, in [6], we present an abstract event-
based architecture for fleet management systems that supports tailoring dynamic control
regimes for coordinating fleet vehicles and illustrate it for the case of EMS management.
Furthermore, in [7], we present an approach that addresses the problem of timely au-
tomatic transmission of complete, real-time information about the current state of the am-
bulance fleet that is usually transmitted by the ambulance crew members. Due to the often
stressful work of those professionals, the information is frequently not sent in a timely man-
ner. We use a Complex Event Processing architecture (e.g., [15]) to automatically identify
and transmit incidents and changes in the operational states of ambulances. As a result, the
availability of information in the ECC and, thus, the effectiveness of the service is improved.
Availability of real-time information about EMS participants is the base for their effi-
cient assignment to patients. The efficiency of the assignment can be achieved through vari-
ous primal, dual, and primal-dual methods that have been proposed for obtaining an optimal
solution of the assignment problem. The Hungarian method was the first such approach that
solves assignment problem within time bounded by a polynomial expression of the number
of agents [27]. Since it is centralized, it cannot be implemented in the systems that have in-
trinsically distributed information and computation. Therefore, a distributed version of the
Hungarian method was proposed in [21]. Here, agents autonomously perform different sub-
steps of the Hungarian algorithm based on their own information and the one received from
other agents in the system. The algorithm computes a globally optimal solution in O(n3)
cumulative time (O(n2) for each agent), with O(n3) messages exchanged among nagents.
The auction algorithm proposed in [5] is another distributed relaxation method for the
assignment problem that, based on an appropriate choice of bids, determines prices for ob-
jects and renders them more or less attractive for the agents to bid for. It is an iterative
procedure that operates like an auction in which unassigned agents bid simultaneously for
objects by raising their prices. The bidding is typically executed concurrently, where each
agent calculates its bids simultaneously and independently of other agents considering only
its own local information. Once all bids are collected, multiple bids are iteratively compared
to determine the best offer for the system and the objects are awarded to the highest bidder.
Even though it implicitly tries to solve a dual problem, the algorithm actually attains a dual
approximate optimal solution complying with -complementary slackness [4]. Furthermore,
it can be interpreted as a Jacobi-like iterative relaxation method for solving a dual problem.
The Jacobi method is an algorithm for determining the solutions of a diagonally dominant
system of linear equations. The worst case complexity of the auction algorithm with
scaling is O(NAlog(N C)), where Nis the number of agents, A, the number of mutually
assignable pairs of agents and objects, and Cis the maximum absolute object auction value.
The algorithm is competitive with existing methods, and when executed in a distributed sys-
tem or on a parallel machine, the algorithm exhibits substantial speedup [5]. For all these
reasons and due to its economical market-related flavor, we use a modification of the auction
algorithm for the assignment of EMS resources to angioplasty patients.
In spite of an exhaustive quantity of work on the optimization of EMA, to the best of
our knowledge, there is little work on optimization models for the coordination of EMS for
STEMI patients. This case is specific since it includes the coordination of the assignment
of idle ambulances to patients, assignment of catheterization laboratories in available hos-
pitals to diagnosed STEMI patients, and the assignment of available cardiology teams to
hospitals for the angioplasty procedure. All of the three assignments need to be combined to
guarantee good arrival times for all the simultaneous patients awaiting the treatment. This
EMS coordination problem is somewhat similar to the case described in [34] where we
shortly describe a distributed multi-agent coordination model for after-hours urgent surgery
patients that cannot be covered by the in-house surgery teams. Therefore, in this paper, we
extend the model presented in [34] and adapt it to the coordination of EMS for angioplasty.
Furthermore, we propose a three-level distributed optimization approach for this case.
3 Angioplasty and patient delay
STEMI is diagnosed based on the patient’s history of severe chest pain lasting for 20 minutes
or more, not responding to nitroglycerine and having an evident ST-segment elevation in the
electrocardiogram [48]. It is treated as an emergency where an immediate goal is to open
blocked arteries and reperfuse the heart muscles as soon as possible either through urgent
coronary angiography and primary angioplasty or with thrombolysis, whichever is available.
When the formerly mentioned therapies are unsuccessful, it is treated with bypass surgery.
Delays in treating STEMI increase the likelihood and amount of cardiac muscle damage
due to localised hypoxia, which is why the recommended delay between a patient’s arrival
at the hospital and the time he/she receives angioplasty (also called door-to-balloon time)
is no more than 90 minutes [44]. Also, the delay between the first medical contact and
reperfusion therapy is referred to as system delay. It is a good service quality indicator and
patient outcome predictor [47]. Regarding the connection between the system delay and
mortality in the short, medium and long run, Terkelsen et al. in [47] performed a historical
follow-up study of Danish medical registries of STEMI patients, who were transported by
the EMS and treated with angioplasty from January 1, 2002 to December 31, 2008, at 3
high-volume angioplasty centers in Western Denmark. A total of 6209 patients underwent
primary angioplasty within 12 hours of system delay. The mortality rates were as follows:
a system delay of 0 through 60 minutes (n= 347) corresponded to a long-term mortality
rate of 15.4%(n= 43); a delay of 61 through 120 minutes (n=2643) to a rate of 23.3%
(n= 380); a delay of 121 through 180 minutes (n=2092) to a rate of 28.1% (n= 378);
and a delay of 181 through 360 minutes (n=1127) to a rate of 30.8% (n= 275) (with
the probability of type II error Pβ.001). Moreover, patient delay time should be reduced as
much as possible by active hospital-EMS coordination since each 30 minutes of the delay
increases the relative risk of 1-year mortality by 7.5% [16]. By reducing the patient delay
time, we minimize the extent of heart muscle damage and preserve the pumping function of
the heart. The result is lower mortality and the invalidity resulting from STEMI [45].
Different countries have legally established upper bounds on the maximal allowed pa-
tient and/or system delay time. To take a concrete example, the American College of Car-
diology (ACC) and American Heart Association (AHA) recommend that the system delay
for patients with STEMI should be less than 90 minutes [25, 46]. The European Society of
Cardiology in [45] for angioplasty indicates a goal patient delay of 90 min while in high-
risk cases with large anterior infarcts it is 60 min. Anyhow, the prognosis of STEMI patients
varies greatly depending on a person’s health, the extent of the heart damage influenced
by the patient delay time, the treatment given, but also based on the meteorological fac-
tors and the time of appearance. It was noticed that the incidence of non-fatal STEMI is
approximately 40%higher from 06:00 12:00 AM as compared to other times of the day
[29]. Season of the year, daily atmospheric temperature, pressure, and relative humidity are
meteorological factors that also affect the number of STEMI deaths per month. Many ge-
ographically dispersed studies confirm this claim, e.g., Greece [17], Belgium [36], US and
Canada [10], Italy [35], Hungary [26], and Korea [28]. Temperature influences sympathetic
tone, blood pressure, and blood platelet functioning [10, 24, 29, 30, 36].
Morbidity and mortality from STEMI have decreased over the years due to better treat-
ment including greater use of angioplasty. However, more than 25 %of patients with STEMI
die before receiving medical care, most often from ventricular fibrillation. To limit infarct
size and to prevent infarct extension and expansion in patients with STEMI, the main issue
is to coordinate efficiently EMS and initiate reperfusion therapy as soon as possible.
4 Case-study: SUMMA 112, EMS provider in the Community of Madrid, Spain
We demonstrate the criticality of the problem in the case of SUMMA 112, the emergency
medical service provider in the Community of Madrid in Spain covering a population of
more than 6.4 million.
The strategy that SUMMA 112 uses for the coordination of emergency AMI patient
assistance is the following. When SUMMA makes the first contact with a patient, in the
case of fibrinolisis and/or shock, the patient is given the highest priority and is directed
towards primary angioplasty coordinating in the least time possible the EMS participants
necessary for EMA. Otherwise, i.e., if there was no contraindication to fibrinolisis and/ or
shock, then
if the time from symptom onset was from 12 to 24 hours and if the patient still shows
symptoms or instability, or if the time from symptom onset is from two to four hours,
he/she is given the highest priority (priority zero) and is directed to the primary angio-
plasty;
if the time from symptom onset is less than two hours, or it is from two to four hours,
but the expected patient delay time is more than two hours, then the patient is directed
towards fibrinolysis. After that, he/she is assigned priority 1 and is transferred to imme-
diate rescue angioplasty. If reperfusion injury has occurred, then the patient is transferred
to immediate angioplasty after thrombolysis.
The average time from symptom onset to contacting the emergency services in SUMMA
112 with procedures and strategies described previously on average is 70 min; SUMMA 112
attends on average in 10 min and stabilizes the patient, diagnoses ST-Elevation Acute Coro-
nary Syndrome, and administers fibrinolysis on average in 25 min; following fibrinolysis, it
takes on average 38.5 min to reach the hospital [2].
We analyzed SUMMA 112’s patient data for the year 2009. Out of 33 hospitals belong-
ing to the public health service of Madrid, 9 of them had catheterization laboratories and
human cardiology teams capable of performing the angioplasty procedure. SUMMA 112
disposes of 36 ambulances with Advanced Life Support (ALS) whose (fixed) base stations
are located at the 33 hospitals of the Community. Angioplasty patients share EMS resources
with the rest of emergency patients. Moreover, cardiology teams are present in the hospitals
during work hours and available from home after-hours; when at home, they assist patients
only if called by the hospital [41].
The day with the highest number of emergency patients in Madrid in 2009 was January
21. There were 221 most sever emergency patients, 40 of whom were simultaneously wait-
ing for ambulance assignment in the system with at least one more emergency patient [8].
Even though we do not dispose of the data on the pathology of these patients, in the Commu-
nity of Madrid in 2009, cardiovascular diseases caused 11,453 deaths, representing 27.7%
of all deaths (41,268) of the Community in that year [38]. Furthermore, 2146 people died
of AMI, to which must be added a majority of the 1,415 deaths assigned to the diagnosis of
cardiac arrest, death without assistance and unknown cause deaths [38].
Moreover, in [37] the clinical and angiography data in the database ofthe Hospital Puerta
de Hierro in Madrid from January 2005 to October 2007 was analyzed. The hospital covers
a local population of 635,495 inhabitants generally concentrated in the urban area nearby
and well connected by road. The goal door-to-balloon time in this study performed over
389 patients was 60 min or less. 84.7%of STEMI patients were treated with angioplasty
with a median door-to-balloon time of 79 (53-104) min, which was 30 (60-90) min lower
when the EMS gave prior warning to the hospital (p < 0.01). Reducing this time was
especially relevant in the cases seen outside working hours. Although the first-medical-
contact-to-balloon time is not mentioned, presumably it was rather long for the patients
that were transferred to hospital since the beginning-of-symptoms-to-balloon median time
was 235 (percentiles 25-75, 170-335) min. However, patients who came to the emergency
department by their own means had the longest door-to-balloon (100 min, p < 0.01).
In addition, 32%of STEMI patients in Madrid do not receive reperfusion therapy and
for patients who do, the delay times are greater than those recommended in clinical practice
guidelines. Therefore, the high frequency of this pathology, together with a greater proba-
bility of occurrence during morning hours and on days with bad climate conditions, leads
to increased patient delay times. This calls for an improved approach to the coordination of
simultaneous EMA of multiple patients assigned for angioplasty.
5 Problem formulation
In this paper, we consider the problem of the dynamic real-time assignment of multiple
simultaneously appearing urgent out-of-hospital angioplasty patients to ambulances and
consequently to angioplasty-enabled hospitals with out-of-hospital cardiology teams. The
objective is to minimize overall patient delay to angioplasty while respecting as much as
possible the limits on allowed maximal individual patient delay.
From the system’s efficiency and fairness point of view, EMS coordination should result
in as high utilitarian social welfare as possible while respecting the requirements on maximal
individual patient delay. A high utilitarian social welfare means a significant reduction of
patient delay for most of the patients, with (usually a few) worst off patients (see, e.g., [42]).
By introducing the constraint on maximum patient delay, the worst off patients are assigned
acceptable delay times. If, on the other hand, we decided to optimize the EMA worst-off
behavior or egalitarian welfare, we would deteriorate the system efficiency and thus the
utilitarian welfare, which in the EMS coordination case would mean higher patient delay for
most of the simultaneous patients (see, e.g., [13]).
Hesitation of patients to search for medical help sometimes might average several hours
and thus can prevent the early application of life-saving procedures and contribute substan-
tially to a diminished effectiveness of treatment. Since we cannot influence this hesitation
time, in the development of the coordination model, we concentrate on the minimization of
the expected patient delay intended as the time elapsed from the moment a patient contacts
the ECC to the moment (s)he is treated with angioplasty in the hospital.
The expected patient delay defined in this way is made of the following parts, Figure 1:
T1 Emergency call response and decision-making for the assignment of EMS resources;
T2 Mobilization of an idle ambulance and its transit from its momentary position to the
patient’s out-of-hospital position;
T3 Patient’s treatment in-situ by ambulance staff;
T4 Patient’s transport in the ambulance to an assigned hospital;
T5 Cardiology team’s transport from their momentary out-of-hospital location to the
hospital;
T6 Expected waiting time due to previous patients in the catheterization lab (if any).
Considering the aforementioned EMS tasks’ delays, four distinct agent sets have been
identified. Let Pbe a pending patient set. Let Cbe a set of cardiology teams and Cav C
the subset of available cardiology teams. Furthermore, let Abe the set of identical, capac-
itated ambulances to be scheduled to assist patients based on one-to-one assignment and
Aav Athe subset of available such ambulances. Moreover, let Hbe a set of hospitals
with catheterization lab and Hav Hthe subset of available such hospitals.
Availability of every catheterization lab in each hospital hHdepends on previous
patients (if any) booked for that catheterization lab with higher or the same urgency as the
patient in question. Therefore, let ρh,p represent expected delay times of free time windows
of catheterization lab(s) of hospital hfor patient p. Moreover, all agent sets are represented
Fig. 1 Diagram showing temporal sequence and relation of six medical emergency tasks (T) and objectives
(O) necessary in coordinating EMS for a patient in the need of angioplasty treatment
by points in the plane. The abbreviation posais used for the position of any kind of agent a
while t(x, y)is the expected arrival time from position xto position y.
In the case there is only one pending patient in the system, the problem is to find am-
bulance aAav, cardiology team cCav and hospital hHav that in combination
minimize patient’s delay time, ∆tp:
min ∆tp= min
aAav
t(a, p) + t(p) + min
hHav max t(p, h),min ρh,p,min
cCav
t(c, h)!(1)
subject to:
∆tptmax
p,pP , (2)
where t(p)is the expected in-situ patient assistance duration, assumed to depend on the
patient pathology but not on the assigned ambulance. It can be estimated based on the initial
ECC communication with the patient with reasonable precision. Furthermore, tmax
pis the
maximal allowed patient delay time for patient pPdepending on the patient’s pathology,
while min ρh,p represents the expected shortest delay time until hospital hwill be free for
patient p. Additionally, the expected patient’s t(p, h)and cardiology team’s arrival times to
hospital t(c, h)are considered as the delay times until the arrival to the catheterization lab in
the hospital. Then, the objective for each patient pPis to choose a triple haAav, h
Hav, c Cavioptimizing (1) subject to (2). Hospital hpchosen for patient pPis thus
hp= arg min
hHav max t(p, h),min ρh,p,min
cCav
t(c, h)!.(3)
Therefore, the optimal patient delay time for a single patient is the lowest among the
highest values of the following three times for all available ambulances and angioplasty-
enabled hospitals, Figure 1:
the expected patient delay time to hospital (the sum of times T2, T3, and T4),
the expected minimal arrival time among cardiology teams to the same hospital (T5),
and
the expected shortest waiting time until hospital hwill be free for patient p,min ρh,p,
(T6).
For simplicity, we let tphp = maxhHav (t(p, h),min ρh,p)for all patients pP. Then,
from the global point of view, considering all pending out-of-hospital patients, the problem
transforms into:
min ∆tP=X
pP
∆tp=X
pP
t(a, p)+ X
pP
t(p)+ X
pP max
hHav tphp,min
cCav
t(c, h)!(4)
subject to
∆tptmax
p,pP. (5)
The overall patient delay time ∆tPin (4) is an additive function. Since the minimum
arrival times cannot be always guaranteed for all patients due to the limited number of
EMS resources, a sum of the EMA tasks’ durations should be minimized for each patient
individually and for the system globally considering individual constraints. This gives an
underlying linear programming structure to the EMS coordination problem. Therefore, it is
possible to guarantee optimal outcomes even when the optimization is performed separately
on individual sum components, i.e., when ambulance assignments are negotiated separately
from the hospital and cardiology team assignment, e.g., [9, 14, 19]. This fact significantly
facilitates the multi-agent system’s distribution and enables a multi-level optimization.
Hence, we decompose (4) as follows. On the first level, we assign ambulances to patients
such that the expected arrival time of ambulances to patients t(a, p)is minimized. Note that
since t(p)in (4) is a constant for every patient pdepending only on the patient’s pathology
and not on the assigned ambulance, we can exclude it from the optimization. We have:
min X
pPX
aAav
t(a, p)xap (6)
subject to:
X
aAav
xap = 1,pP(7)
X
pP
xap 1,aAav (8)
X
aAav
t(a, p)xap tmax
ap ,pP(9)
xap {0,1},pP, a Aav.(10)
where (7) and (8) are constraints on the one-on-one assignment of patients to ambulances
assuming that the number of available ambulances is larger than or equal to the number of
patients, i.e., |P| |Aav|. Moreover, (10) defines a value of 0 or 1 for binary decision vari-
able xap. All of the mentioned constraints are hard constraints, while (9) is a soft constraint
on the allowed maximal patient waiting time for ambulance tmax
ap recommended by legal
requirements.
Then, on the second optimization level, we approach the second part of (4):
min X
pP max
hHav tphp,min
cCav
t(c, h)!(11)
which is an NP-hard combinatorial problem. However, by approximating (11) with a se-
quence of problems where we first decide on the assignment of hospitals to pending patients
and then assign cardiologists to patients already assigned to hospitals, we obtain two lin-
ear programs to which we can apply tractable optimal solution approaches as is the auction
algorithm [5]. Moreover, if we introduce constraints on the expected arrival times of indi-
vidual EMS actors with respect to maximal allowed patient delay time in this approximation
process and gradual relaxation of the restrictions in case of a nonexistent feasible solution,
we can achieve efficient and computationally fast solutions. While we explain the relaxation
approach in more detail in Section 6.2, in continuation we present the approximation of (11)
by the following two sequential problems. On the second level, we first optimize the arrival
of patients to hospitals, i.e.,
min X
pPX
hHav
tphpxph (12)
subject to:
X
hHav
xph = 1,pP(13)
X
pP
xph 1,hHav (14)
X
hHav
t(php)xph tmax
p,pP(15)
xph {0,1},pP, h Hav.(16)
where (13) and (14) are hard constraints on the one-on-one assignment of patients to hos-
pitals assuming that the number of available hospitals is larger than or equal to the number
of patients, i.e., |P| |Hav|. (15) is a soft constraint for overall patient delay time to hos-
pital constrained by the maximal allowed patient delay time tmax
p. Moreover, (16) is a hard
constraint on binary decision variable xph.
Then, the optimization problem on the third level is:
min X
hHas X
cCav t(c, h)tphpxch (17)
subject to:
X
hHas
xch 1,cCav (18)
X
cCav
xch = 1,hHas (19)
xch 0,(20)
where (18) ensures that no cardiology team is allocated more than once and (19) assigns
one cardiology team to each hospital in Has.Has is a solution set of hospitals assigned in
(12)–(16) while (20) is a constraint on nonnegativity of decision variable xch.
6 Proposed EMS distributed multi-agent coordination model
The EMS system can be seen as an interconnected geographically distributed system where
EMS participants are capable of processing local real-time information independently while
interacting in a cooperative way. Moreover, the EMS participants’ interactions are mostly
local in nature, thus leading to the setting where most of the participants usually have intense
interaction within some range based on the geographical position, but almost no interaction
outside this range. This is true both for ambulances as for the hospitals with out-of-hospital
cardiology teams. In this context, distributing the computation among ambulances and hos-
pitals while balancing their communication load can increase the overall throughput, ro-
bustness, and flexibility of the system. Additionally, in such a distributed setting, the ECC
is unnecessary for coordination.
For the aforementioned reasons, we propose a dynamic and distributed EMS resource
assignment model for angioplasty patients applicable to all emergency out-of-hospital pa-
tients whose time to hospital treatment is of the utmost priority. The key of our proposed
model lies in the distribution of the EMS related decisions to allow for as high autonomy as
possible regarding local decisions.
The proposed solution is founded on the collaborative multi-agent system organization
with four classes of agents described in Section 6.1. In the proposed model, patient assign-
ment decisions are performed on three levels via the proposed modification of the auction al-
gorithm presented in Section 6.2. On the first optimization level, arrival times of ambulances
to out-of-hospital patients are optimized for all emergency patients who arrive at the hospital
by ambulance. On the second level, the optimization of arrival times of out-of-hospital pa-
tients to hospitals considering the hospital availability is performed for all patients assigned
for angioplasty. Moreover, on the third level, out-of-hospital cardiology teams are assigned
to the hospitals that have been assigned to patients on the second level.
6.1 EMS multi-agent system
All agents in the proposed multi-agent system are autonomous and independent decision
makers that feature a determined sequence of steps and message exchanges in order to re-
solve each emergency case as described in the following.
Medical emergency-coordination center (ECC): ECC receives emergency calls from
patients and informs the rest of the EMS system of the new pending patients. In the case
of centralized EMS organization, it assigns an ambulance and an angioplasty-enabled
hospital for each case, thus performing the high-level management of the STEMI EMA
procedure. In the distributed case, it is not necessary for the coordination since am-
bulances and hospitals assign patients on their own in a distributed way and mutually
communicate and coordinate their actions if necessary.
Patient: Each pending patient agent pPrepresents a real out-of-hospital (angio-
plasty) emergency patient requiring ambulance EMA and angioplasty. After calling ECC
from his/her out-of-hospital location, (s)he gets assisted in-situ by an ambulance crew
and, if necessary, gets transferred to a hospital where (s)he receives angioplasty. More-
over, new patient requests continuously unfold through time and must be assigned in
real time to ambulances such that the patients get assisted with as low delay as possi-
ble while respecting the maximal individual patient delay tmax
p. Each patient pPis
described as follows:
p={posp, tin
p, tmax
p},(21)
where pospis the patient’s position and tin
pis the patient’s detection time defined as the
time when the ECC is informed about the incident.
Ambulance: Ambulance agent aArepresents an ambulance with ALS, together with
the human crew assigned to it. The task of each ambulance is, once it has arrived to a
patient, to assist him/her in-situ and, if necessary, to transfer him/her to a hospital in
the shortest time possible. Furthermore, ambulances communicate to the ECC to obtain
information on patients awaiting ambulances, to hospitals for patient transfer, and to an
assigned patient for his/her first medical treatment in-situ and transport to hospital.
Each ambulance aAis described as follows:
a={posa, Sa},(22)
where posaindicates position and Sathe ambulance’s status that defines subsets of am-
bulances Aand can be: idle Ai, moving to incident position Amip, in-situ assistance
Aisa, and ambulances moving to a hospital Amth. Moreover, the set of available ambu-
lances is denoted by Aav =AiAmip. Available ambulances aAav are considered
for the assignment to pending patients. Furthermore, each ambulance route originates
from its momentary location to the patient’s location and terminates at the assigned hos-
pital if necessary. We assume that an ambulance is committed to a given patient only
when it gets to the location of the patient. After arriving at the patient’s location, the
ambulance cannot be redirected elsewhere until transferring the patient to the hospital
if necessary. When the patient is transferred to the hospital, the ambulance returns to its
base hospital where it waits for a next patient assignment.
Angioplasty-enabled hospital: Hospital agents hHrepresent angioplasty-enabled
hospitals within the region of interest. They are responsible of managing their catheter-
ization labs while they coordinate with cardiology team(s) and ambulances’ agents to
assist patients. Each hospital hHis described as:
h={posh, ρh,p, Sh,p },(23)
where poshis the position and Sh,p is the temporal availability of a hospital hHfor
a patient pPafter assisting patients who are booked before and whose severity of the
pathology is equal to or worse than the one of the patient in question. The availability
states can be either available av or unavailable un. Each hospital has at its disposal also
updated information on cardiology team availability ρc,h and incoming patients.
Cardiology team: cCis responsible of the performance of angioplasty. Each cardi-
ology team cCis described as:
c={posc, ρc,h, Sc,h },(24)
where poscis the cardiology team’s position and ρc,h is its availability for hospital
hHthat can be either available av or unavailable un. Moreover, the team’s status
Scdefines its state that can be: idle Ci, moving to an assigned hospital Cmh, or in the
assigned hospital Cih. Then, the set of available cardiology teams Cav comprises idle
Ciand the cardiology teams moving to an assigned hospital Cmh, i.e., Cav =CiCmh .
Additionally, it is assumed that the cardiology team’s members are positioned at dif-
ferent locations out of hospital and move to hospital when needed. Their combined
expected arrival time to hospital is the maximum of the individual arrival times.
Besides the aforementioned agents, we assume the existence of additional services ac-
cessible on the Internet that may provide patients’ medical information (e.g., health records).
Such services may be provided and consulted by different medical professionals involved in
the process in order to obtain medical data of an angioplasty patient.
6.2 Proposed EMS coordination algorithm
In this Section, we present the EMS distributed coordination approach that assigns ambu-
lances and hospitals with cardiology teams to angioplasty patients and, therefore, solves
problem (4)-(5). The solution approach is presented in Algorithms 1–3.
In Algorithm 1, the overall decision for the assignment of each patient to the triple made
of ambulance, hospital, and cardiology team is divided into the assignment of patients to am-
bulances on the first level, where Resource Assignment(P, Aav )resolves (6)–(10), the as-
signment of patients to hospitals on the second level, where Resource Assignment(P, Hav )
computes (12)–(16), and the assignment of hospitals with allocated patients from the sec-
ond level to cardiology teams on the third level at which Resource Assignment(Has
p, Cav)
finds the solution to (17)–(20), (see Figure 2).
Algorithm 1: Coordination of EMS resources for urgent angioplasty assistance
Input:P;Aav;Hav ;Cav
Output:{p, ap:apAav, hp:hpHav, cp:cpCav },pP
1initialization: AS
2{p, ap} ResourceAssignment(P , Aav),pP
3AS AS S{p, ap},pP
4{p, has
p} ResourceAssignment(P , Hav),pP
5AS AS S{p, has
p},pP
6{has
p, cp} ResourceAssignment(Has
p, Cav),has
pHas
p
7AS AS S{has
p, cp},has
pHas
p
8return AS
For the distributed assignment of EMS resources to patients, at every decision-making
level, each agent included in the assignment of that level independently decides its assign-
ment based on its local information and the communication only with other relevant agents
on the same level, as can be seen in Algorithms 2–4.
Algorithm 2 is a modified version of Bertsekas’ auction algorithm [5]. The reason for
the application of the auction algorithm is its intuitive economical flavor, low computational
complexity, and the distribution of the decision-making procedure which is performed be-
tween a set of bidder agents bB, and a set of bid object agents oO. In the algorithm,
Fig. 2 Proposed three-level decomposition of the problem of EMS coordination for urgent out-of-hospital
STEMI patients awaiting angioplasty
it is important that the number of bidders |B|is equal to or less than the number of bid ob-
jects |O|, i.e., |B| |O|. Moreover, the assignment of bidders to bid objects is performed in
iterations that are composed of a bidding (Algorithm 3) and an allocation phase (Algorithm
4). In Figure 3, we present the distributed functioning of the modified auction algorithm.
In more detail, each bidder agent bBkeeps in its memory the value priceb
o(k), i.e.,
its most recent knowledge about the actual price of each object oO, and the set AS0
bof
its most recent knowledge about all the bidders’ assignments at the respective optimization
level (Algorithm 3). On the other hand, each bid object keeps in its memory the bids received
from bidders in the last two iterations Bidso(k1) and Bidso(k)and the set AS0
oof its
most recent knowledge about all the bidders’ assignments at the respective optimization
level. None of the aforementioned local copies of variables have to coincide with the actual
values; they may also differ from one agent to another due to the dynamics of their previous
communication and local interaction with the environment.
At each time period, all three levels of Algorithm 1 are performed by executing algo-
rithm 2. The first and the second level can be performed in parallel, while the third level is
performed after the second since the results of the second are the input to the third. There-
fore, when pending patients get assigned to hospitals, they are assigned to cardiology teams.
Similarly, in each iteration of Algorithm 2, the allocation phase is performed sequen-
tially after the bidding phase.
Each agent keeps a local copy of Algorithms 3 and 4, and, depending on its role, per-
forms the bidding or the allocation phase of the auction algorithm. In every phase, all partic-
ipating agents perform their individual computations in parallel (computation of a bid by all
unassigned bidders on the bidding phase and the computation of the object allocation by bid
objects at the allocation phase). After updating the local information, at the bidding phase,
each unassigned bidder independently finds the best bid object (the object with the highest
value) given the actual auction prices, and issues its bid to that object. This stage is over
when all unassigned bidders issue their respective bids. Furthermore, at the allocation phase
Fig. 3 Proposed modification of the auction algorithm for distributed EMS coordination
(Algorithm 4), each bid object with at least one received bid determines if the highest re-
ceived bid is higher than the one in the previous iteration and if the present bidder is different
from the winning one in the previous iteration (if any). If this is the case, the object cancels
the previous assignment communicating it to the previous assigned bidder (if any), assigns
itself to the present bidder with the highest bid, increments its price, and broadcasts its new
assignment. Otherwise, it does not update its assignment and remains with the previously
assigned price and bidder (if any).
If at the end of the allocation phase, a bid object received more than one bid and/or
there was a different bidder previously assigned to the object, then there is still at least one
unassigned bidder present. The bid object broadcasts the message that unassigned bidding
agents are still present, and a new iteration begins. The unassigned bidding agents update
the objects’ prices and the algorithm continues in iterations until all bidders are assigned
and all conflicting bids are resolved.
Algorithm 2: Resource Assignment(b,o) based on the auction algorithm
Input:B, O
Output:{b, ob:obO},bB
1initialization: k1
2repeat
3kk+ 1
4for each bBdo
5Bidding P hase(b, o)
6for each oOdo
7AllocationP hase(b, o)
8until {b, ob} AS0,bB;
9return AS’
Algorithm 3: BiddingPhase(b,o) executed by each bidder agent bB
Input:b, O, k
Output:{b, bido,b(k)}
1Initialization
2if k= 1 then
3AS
0b ,bido,b(k)
4for each oOdo
5priceb
o
6if b /AS
0bthen
7Local information update
8Broadcast AS
0band priceb
o(k),oO
9Received Receive{AS
0b0
, priceb0
o(k)|b06=b oO}
10 priceb
o(k)maxb∈{Received Slocal}{priceb
o,oO}
11 AS
0barg maxb∈{Received Slocal}{priceb
o(k),oO}
12 Find objects that offer the least coand the second least cost co0
13 o= arg minoO{t(o, b) + priceb
o(k)}
14 co= minoO{t(o, b) + priceb
o(k)}
15 co0= mino0Oo06=o{t(o0, b) + priceb
o0(k)}
16 Calculate bid for the object owith the least cost
17 bido,b =coco0+
18 Issue bid bido,b to object o
19 return {b, bido,b(k)}
If we assume that the cardinality of the patient set |P|is smaller than the one of ambu-
lances, i.e., |P|≤|Aav|and hospitals, |P|≤|Hav |, then pending patients bid for ambu-
lances at the first level, and for hospitals at the second level. Furthermore, if the cardinality
of the set of the available cardiology teams is smaller than the set of hospitals assigned to
patients at the second level, i.e., |Cav|≤|Has |, then on the third level, cardiology teams
bid for the hospitals has
pHas. The particularity of the third optimization level, which re-
solves problem (17)-(20) is the individual cardiology team’s bidding cost that, to minimize
the team’s tardiness with respect to the patient’s arrival to hospital, does not include the ex-
pected arrival time from the bidder to the object t(o, b), as is the case for the first two levels,
but the difference between the arrival times of cardiology teams and patients to hospitals
assigned on the second level, i.e., for each hospital has
pHas assigned to patient pP,
the bid in the bidding phase of Algorithm 2 is calculated as:
ch= min
hHas t(c, h)tphp+priceh(k),cCav.(25)
Here, if a cardiology team is assignable to more hospitals, preferably all, then the op-
timization over multiple arrival times gives a globally optimal solution. Oppositely, if each
cardiology team can work only at one or a subset of hospitals, the expected arrival time
might be significantly longer, which may jeopardize patients’ outcomes.
A global termination condition of Algorithm 2 is that all patients are assigned to EMS
resources based on a one-to-one assignment. It is decomposed into a collection of local
termination conditions, one for each bidder. Each bidder monitors its own computations and
Algorithm 4: AllocationPhase(b,o) executed by each object agent oO
Input:B, o, k
Output:{AS
0o, priceo0(k)}
1initialization
2if k= 1 then
3AS
0o ,bidmax
o , bmax
o
4Bidso(k)
5Receive bids from the bidders and save them
6Bidso(k)Bidso(k) {bido,b},bBo
7if Bidso(k)6=then
8Determine the highest bid
9bidmax
o(k)max{bid Bidso(k)}
10 Determine the corresponding highest bidder
11 bmax
o(k)arg max{bid Bidso(k)}
12 if (bidmax
o(k)bidmax
o(k1)) (bmax
o(k)6=bmax
o(k1))then
13 Assign object oto the highest bidder bmax
o(k)
14 AS
0oAS
0o {hbo, oi AS
0o}
15 AS
0oAS
0oS{bmax
o(k), o}
16 Increment the price for object o
17 priceo(k)priceo(k1) + bidmax
o(k)
18 else
19 AS
0oand priceoremain unchanged
20 Broadcast local assignment AS
0oand price priceo0(k)
21 return AS
0o, priceo0(k)
checks whether a local termination condition on the local assignment holds. Termination
occurs at some iteration kif kis the shortest time for which the local assignment holds for
all bidders and no message is in transit between any two agents [3].
For the overall system’s solution to comply with soft constraints, the matrix representing
the arrival times from bidders (e.g., patients) to bid objects (e.g., ambulances or hospitals)
reachable within the maximal allowed patient delay time tmax
pfor all pPshould be
at least triangular with at least |P|(|P|+ 1)/2actual arrival times. Otherwise, an optimal
solution to the problem may include times higher than tmax
p. In other words, if patients
decide sequentially on the EMS resource assignment, there should be at least one decision-
making ordering of patients that will give an acceptable EMS resource assignment to every
patient. Otherwise, a solution respecting the maximal delays does not exist and the EMS
resources with the arrival times higher than tmax
pshould be introduced for a minimal num-
ber of patients. This is done in a non-decreasing order of the arrival times until at least a
triangular matrix shape is achieved for actual patient-EMS resource arrival times. The EMS
resources with combined arrival times higher than tmax
pfor every patient pPare penal-
ized with weights varying depending on the distance from tmax
psuch that higher deviations
get avoided in the optimal solution as much as possible. In this way, we relax the constraints
on the maximal allowed patient delay for a minimal number of patients. Furthermore, the
delay time deviations of those patients are minimal at the system’s level.
Lastly, to assure a correct functioning of the proposed approach, there should be a co-
ordinator agent that will coordinate the sequencing of the three EMS resource assignment
levels in Algorithm 1. This algorithm is initially performed in the first task, i.e., when the
emergency call is received. It is relaunched with every new relevant event - changes of avail-
ability of the EMS resources, delay and arrival times, or of the patients’ states. By allowing
for reassignment of resources based on the adaptation to contingencies in real time, we ob-
tain a flexible EMS coordination solution.
7 Simulation experiments
In this Section, we describe the simulation settings, experiments, and results. We test the
proposed approach for the coordination of EMS resources in angioplasty patients’ assistance
focusing on the average patient delay time in the case of multiple pending patients. We
compare the performance of our approach with the FCFS method since it is applied by most
of the medical emergency-coordination centers in Western world countries (e.g., SUMMA
112 in Madrid, Spain).
Ambulances share the transport infrastructure with other vehicles. In the experiments,
we use a so-called mesoscopic traffic simulation model where traffic flow is described with
high level of detail, but at the same time flow behaviour is presented at a low level of descrip-
tion, i.e., we concentrate on the statistical description of the same. We introduce a further
simplification by substituting the function of travel time t(x,y)between two neighboring
nodes of a transport network x= (x1, x2)and y= (y1, y2)in Euclidean 2-space with
Euclidean distance d(x,y) = qP2
i=1(xiyi)2, and thus we convert the travel time mini-
mization problem to the Euclidean distance minimization problem.
7.1 Simulation settings
To demonstrate the scalability of our solution and its potential application to small, medium
and large cities and regions, in the experiments we vary the number of EMA ambulances
from 5 to 100 with increment 5 and the number of angioplasty-capable hospitals from 2
to 50 with increment 2. The number of cardiology teams |C|in each experiment equals the
number of hospitals |H|. Thus, the number of setup configurations used, combining different
numbers of ambulances and hospitals with cardiology teams, sums up to 500.
For each configuration, we perform the simulation on 3 different instances of random
EMS participants’ positions since we want to simulate a sufficiently general setting applica-
ble to any urban area that does not represent any region in particular. The EMS participants
are distributed across the environment whose dimensions are 50 ×50 kms. In each instance,
we model hospital positions and the initial positions of ambulances, out-of-hospital car-
diology teams, and patients based on a continuous uniform distribution. Therefore, each
configuration can be considered as a unique virtual city with its EMS system. Assuming
that the EMS system is placed in a highly dense urban area, this kind of modelling of the
positions of EMS participants represents a general enough real case since the election of the
hospital positions in urban areas is usually the result of a series of decisions developing over
time with certain stochasticity, influenced by multiple political and demographical factors.
In the simulations, ambulances are initially assigned to the base stations in the hospi-
tals of the region of interest. Additionally, we assume that after transferring a patient to the
hospital, an ambulance is redirected to the base station where it waits for the next patient as-
signment. Furthermore, we assume that the hospitals have at the disposal a sufficient number
of catheterization laboratories so that the only optimization factor from the hospital point of
view is the number of available cardiology teams. If there are more patients with the same
urgency already assigned waiting for treatment in the same hospital, they are put in a queue.
The simulation of each instance is run over a temporal horizon in which new patients
are generated based on a certain appearance frequency. The EMS resources are dynamically
coordinated from the appearance of a patient until the time he/she is assisted in hospital
by a cardiology team. Each instance simulation is run over the total of 300 patients whose
appearance is distributed equally along the overall time horizon based on the following two
predetermined frequency scenarios: low (1 new patient every 10 time periods) and high (1
new patient every 2 time periods). In more detail, in the first scenario, we test our solution
over minimally 3000 consecutive periods in each instance of each configuration. In the sec-
ond scenario, the experiment is performed over at least 600 consecutive periods for each
instance of each configuration.
The period between two consecutive executions of the EMS coordination algorithm is
considered here as a minimum time interval in which the assignment decisions are made;
usually it ranges from 1 to 15 minutes. In each period, the actual state of EMS resources
and pending patients is detected and the EMS coordination is performed such that the EMS
resources are (re)assigned for all patients. To achieve an efficient dynamic reassignment
of ambulances, the execution of the EMS coordination algorithm is furthermore performed
with every new significant event, i.e., any time there is a significant change in the system
due to new patients, or the significant change of travel time or state of any of the EMS
participants.
Let us assume that the update of information in such an EMS system is performed ev-
ery 5 minutes. Then at least 3000 consecutive periods of simulation in one configuration
respond to at least 15000 minutes of simulation, which is more than 10 days of simulation
per instance. Three such instances result in more than a month of simulation for each such
virtual urban area defined by every setup configuration.
7.2 Simulation results
In the experiments, we test the performance of the proposed EMS coordination method
with respect to the FCFS benchmark approach. The comparison is based on the relative
performance function Pdefined as:
P=tF CF S tOR
tF CF S
·100 ,[%],(26)
where tF CF S and tOR are average patient delay times of the benchmark FCFS approach
and the proposed model, respectively.
The simulation results of the performance function Pfor the two simulated cases of
frequency of new patient appearance of 1 and 5 new patients every 10 time periods are
presented in Figures 4, 5, 6, 7, 8, and Table 1. Figures 4 7 show the average patient delay
improvement of proposed EMS coordination solution versus the FCFS strategy considering
varying number of ambulances and hospitals in the EMS system. The performance of the
proposed approach increases as the number of angioplasty enabled hospitals increases from
almost identical average patient delay in the configuration setup with 2 hospitals up to 87,14
%with 50 hospitals, as can be seen in Figure 7 and Table 1.
Fig. 4 Average patient delay time performance of the proposed EMS coordination approach vs. the FCFS
strategy [%] for the frequency of appearance of 1 new patient every 10 time periods
Moreover, Figure 8 shows mean patient delay improvement of proposed EMS coordi-
nation solution in respect to the FCFS strategy as the number of hospitals increases. This
improvement is a mean value for each experimented number of hospitals over all the exper-
imented scenarios with the number of ambulances ranging from 5 to 100.
Observing the performance dynamics with respect to the varying number of hospitals, it
is evident from Figures 5, 7, and 8 that the performance of the proposed EMS coordination
method increases on average proportionally to the increase of the number of hospitals. With
a relatively low number of angioplasty-enabled hospitals (less than 15), our proposed EMS
coordination approach performs on average better than FCFS up to 15%. As the number of
hospitals increases, the performance improves on average up to the maximum of 39,98%for
the first case, Figure 5, and up to 87,14 %, for the second case, Figure 7. However, mean
patient delay improvement for the two cases is 35% and 45.5% respectively, Figure 8.
Looking at the EMS coordination algorithm’s performance dynamics with respect to the
varying number of ambulances, in Figure 6, two regions are evident: the first one appears
with a very low number of ambulances where the performance of the EMS coordination
algorithm with respect to the FCFS method grows significantly faster with the increase of
the number of the hospitals than the second region with a higher number of ambulances
where the performance’ improvement growth tendency is steadier. The performance values
of the first region steeply decrease to the steady values of the valley region as the number of
ambulances changes from 15 to 20. Moreover, the comparison of the overall values of Figure
4 and 6 implies that the proposed EMS coordination method’s performance improves on
average with respect to the FCFS method as the frequency of patient appearance increases,
which is also evident from Figure 8.
Fig. 5 Average patient delay time performance of the proposed EMS coordination approach vs. the FCFS
strategy [%] for the frequency of appearance of 1 new patient every 10 time periods, horizontal view with
respect to the number of hospitals
Fig. 6 Average patient delay time performance of the proposed EMS coordination approach vs. the FCFS
strategy [%] for the frequency of appearance of 1 new patient every 2 time periods
Fig. 7 Average patient delay time performance of the proposed EMS coordination approach vs. the FCFS
strategy [%] for the frequency of appearance of 1 new patient every 2 time periods, horizontal view with
respect to the number of hospitals
Fig. 8 Mean patient delay time improvement of the proposed EMS coordination approach vs. the FCFS
strategy [%] for the frequencies of appearance of 1 new patient every 2 and 10 time periods
Table 1 Minimum and maximum values of performance function Pin the simulation
Frequency of patient
appearance
1/10 5/10
P min.value, [%] 0,001 0,002
P max. value, [%] 39,98 87,14
The static assignment of the FCFS principle discriminates against patients appearing
later. Since ambulances are not equally distributed in the area, the proposed EMS coordina-
tion method compensates for the lack of EMS resources and their unequal distribution by
reassigning them dynamically to pending patients. Dynamically optimized reassignment of
EMS resources in real time is the main key to the improvement of the system’s performance.
Thus, proportional to the increase of the number of hospitals, there is a constant improve-
ment of performance up to the maximum values of the simulated experiments as seen in
Table 1.
Even though the velocity of the EMS actors is not a relevant factor in the comparison of
the performance of our proposed EMS coordination solution and the FCFS method, looking
individually at the performance of each one of these methods, it is evident that the assign-
ment cost accumulated through the time will be lower when the velocity of the EMS actors
is higher. The efficiency of the solution is limited from above by the time period duration,
i.e., the proposed EMS algorithm should be executed with every important event that might
change the system’s assignment solution.
8 Conclusions
In this paper, we proposed a distributed and optimized EMS coordination model that fa-
cilitates a seamless coordination among the emergency medical assistance participants for
the minimization of delay times of angioplasty patients. The proposed model implies the
change of the current functioning based on a manual coordination through communications
via phone calls, towards an automated coordination process where the basic decisions are
taken (or proposed) by software agents.
In order to reduce patient delay times, we proposed a distributed coordination tool based
on a three-level optimization in which the assignment of EMS resources to patients is per-
formed via iterative auctions for the minimization of ambulance arrival times to patients,
and the patients’ and cardiology teams’ arrival times to hospitals. The proposed EMS coor-
dination approach enables globally optimized EMS resource assignment even in cases when
multiple patients have to be assisted at the same time and provides an increased flexibil-
ity and responsiveness of the emergency system. Additionally, the ambulance assignment
can be optimized both in combination or independently of the hospital and cardiology team
assignment, depending on the patients’ needs and the means of transport to the hospital.
Our simulation results show the efficiency of the proposed solution approach, resulting
in significantly lower delay times for angioplasty on average. Of course, the effectiveness of
the proposed model depends on the initial classification of patients, and the related determi-
nation of the urgency of their cases, as well as on the timely availability of cardiology teams
and hospitals. Still, as the current experience in the region of Madrid shows, good quality
patient assessments and the EMS resource availabilities can be assured in practice.
To implement our approach in practice, a patient’s location needs to be known to the
system. Ideally, patients should contact the ECC through a mobile phone with GPS for
easier location. In addition, ambulances should have a GPS and a navigator for localizing the
patient and navigating the way to him/her, as well as a means of communication with the rest
of the EMS participants, and a digitalized map showing ambulances, patients and hospitals.
Moreover, hospitals should have a digitalized receptionist service to receive and process
relevant data of a patient before his/her arrival. None of these requirements go significantly
beyond the current state of affairs in major cities (such as Madrid). Moreover, there are
intrinsic uncertainties present in the EMS coordination. In our experiments, we assume that
travel times can be accurately forecasted, which, of course, is an important factor for the
performance of the proposed system. In reality, this may not always be the case, as real-
world traffic conditions are notoriously hard to predict. However, there is abundant literature
on traffic-aware vehicle route guidance systems tackling this problem, and we believe that
such systems can be easily integrated into our approach. Still, an effective proof of this
conjecture is left to future work.
There are some additional prerequisites for the practical application of the proposed
approach. The EMS provider, especially if it currently implements hierarchically structured
procedures, should be willing to move towards a patient-oriented organizational structure.
In addition, the parameters of the proposed EMS coordination approach should be chosen
with care for each particular application scenario and area, since overly fast reassignments
of EMS resources could cause instability in the EMS system. Additionally, as a future work,
we plan to develop an EMS coordination model that considers both efficiency and equity
issues among EMS actors, while forecasting future patients across a receding horizon.
Acknowledgements This work was supported in part by the Spanish Ministry of Econ-
omy and Competitiveness through the projects iHAS (grant TIN2012-36586-C03-02) and
SURF (grant TIN2015-65515-C4-4-R), and by the Community of Madrid through the pro-
gramme MOSI-AGIL-CM (grant P2013/ICE-3019, co-funded by EU Structural Funds FSE
and FEDER).
References
1. D. Bandara, M. E. Mayorga, and L. A. McLay. Optimal dispatching strategies for emergency vehicles to
increase patient survivability. International Journal of Operational Research, 15(2):195–214, 2012.
2. N. Behzadia, M. A. Salinero-Fortb, A. de Blasc, M. Taboadaa, L. P´
erez de Islad, and J. L. L´
opez-
Send´
one. Prehospital thrombolysis: Two years experience of the community of madrid emergency ser-
vices (summa 112). Rev Esp Cardiol (Engl Ed)., 65(10):960–961, 2012.
3. Dimitri P Bertsekas. Auction algorithms for network flow problems: A tutorial introduction. Computa-
tional Optimization and Applications, 1(1):7–66, 1992.
4. Dimitri P. Bertsekas. Auction algorithms. In Encyclopedia of Optimization, pages 128–132. 2009.
5. D.P. Bertsekas. The auction algorithm: A distributed relaxation method for the assignment problem.
Annals of Operations Research, 14(1):105–123, 1988.
6. Holger Billhardt, Alicia Fernandez, Lissette Lemus, Marin Lujak, Nodar Osman, Sascha Ossowski, and
Carles Sierra. Dynamic coordination in fleet management systems: Toward smart cyber fleets. Intelligent
Systems, IEEE, 29(3):70–76, 2014.
7. Holger Billhardt, Marin Lujak, Sascha Ossowski, Ralf Bruns, and J¨
urgen Dunkel. Intelligent event
processing for emergency medical assistance. In Proceedings of the 29th Annual ACM Symposium on
Applied Computing, pages 200–206. ACM, 2014.
8. Holger Billhardt, Marin Lujak, Vicente S´
anchez-Brunete, Alberto Fern´
andez, and Sascha Ossowski.
Dynamic coordination of ambulances for emergency medical assistance services. Knowledge-Based
Systems, 70:268–280, 2014.
9. Stephen Bradley, Arnoldo Hax, and Thomas Magnanti. Applied mathematical programming. 1977.
10. Steven C Brooks, Robert H Schmicker, Thomas D Rea, et al. Out-of-hospital cardiac arrest frequency
and survival: evidence for temporal variability. Resuscitation, 81(2):175–181, 2010.
11. C.P. Cannon, C.M. Gibson, C.T. Lambrew, D.A. Shoultz, D. Levy, W.J. French, J.M. Gore, W.D. Weaver,
W.J. Rogers, and A.J. Tiefenbrunn. Relationship of symptom-onset-to-balloon time and door-to-balloon
time with mortality in patients undergoing angioplasty for acute myocardial infarction. JAMA: the jour-
nal of the American Medical Association, 283(22):2941–2947, 2000.
12. Brecht Cardoen, Erik Demeulemeester, and Jeroen Beli¨
en. Operating room planning and scheduling: A
literature review. European Journal of Operational Research, 201(3):921–932, 2010.
13. Yann Chevaleyre, Paul E Dunne, Ulle Endriss, et al. Issues in multiagent resource allocation. Informat-
ica, 30(1), 2006.
14. Yann Chevaleyre, Ulle Endriss, Sylvia Estivie, and Nicolas Maudet. Multiagent resource allocation
with k-additive utility functions. In In Proc. DIMACS-LAMSADE Workshop on Computer Science and
Decision Theory, Annales du LAMSADE, pages 83–100, 2004.
15. Gianpaolo Cugola and Alessandro Margara. Processing flows of information: From data stream to com-
plex event processing. ACM Computing Surveys (CSUR), 44(3):15, 2012.
16. Giuseppe De Luca, Harry Suryapranata, Jan Paul Ottervanger, and Elliott M Antman. Time delay to
treatment and mortality in primary angioplasty for acute myocardial infarction every minute of delay
counts. Circulation, 109(10):1223–1225, 2004.
17. P Dilaveris, Andreas Synetos, Georgios Giannopoulos, et al. Climate impacts on myocardial infarction
deaths in the athens territory: the climate study. Heart, 92(12):1747–1751, 2006.
18. Elton Domnori, Giacomo Cabri, and Letizia Leonardi. Coordination issues in an agent-based approach
for territorial emergence management. In Advanced Information Networking and Applications (WAINA),
2011 IEEE Workshops of International Conference on, pages 184–189. IEEE, 2011.
19. Ulrich Endriss, Nicolas Maudet, Fariba Sadri, and Francesca Toni. On optimal outcomes of negotiations
over resources. In AAMAS, volume 3, pages 177–184, 2003.
20. Michel Gendreau, Gilbert Laporte, and Fr´
ed´
eric Semet. A dynamic model and parallel tabu search
heuristic for real-time ambulance relocation. Parallel computing, 27(12):1641–1653, 2001.
21. S. Giordani, M. Lujak, and F. Martinelli. A distributed algorithm for the multi-robot task allocation
problem. In Trends in Applied Intelligent Systems: Proc. of the 23rd internat. conf. on Ind. eng. and
other applications of applied int. systems, volume 6096 of Lecture Notes in Artificial Intelligence, pages
721–730. Springer-Verlag, 2010.
22. Francesca Guerriero and Rosita Guido. Operational research in the management of the operating theatre:
a survey. Health care management science, 14(1):89–114, 2011.
23. Shane G Henderson. Operations research tools for addressing current challenges in emergency medical
services. Wiley Encyclopedia of Operations Research and Management Science, 2011.
24. Johan Herlitz, Mikael Eek, Mikael Holmberg, and Stig Holmberg. Diurnal, weekly and seasonal rhythm
of out of hospital cardiac arrest in sweden. Resuscitation, 54(2):133–138, 2002.
25. James G Jollis, Christopher B Granger, Timothy D Henry, et al. Systems of care for st-segment–elevation
myocardial infarction: A report from the american heart associations mission: Lifeline. Circulation:
Cardiovascular Quality and Outcomes, 5(4):423–428, 2012.
26. Ildik´
o Kriszbacher, Imre Boncz, Mikl´
os Kopp´
an, and J´
ozsef B´
odis. Seasonal variations in the occurrence
of acute myocardial infarction in hungary between 2000 and 2004. International Journal of cardiology,
129(2):251–254, 2008.
27. Harold W Kuhn. The hungarian method for the assignment problem. Naval research logistics quarterly,
2(1-2):83–97, 1955.
28. Jang Hoon Lee, Shung Chull Chae, Dong Heon Yang, et al. Influence of weather on daily hospital admis-
sions for acute myocardial infarction (from the korea acute myocardial infarction registry). International
journal of cardiology, 144(1):16–21, 2010.
29. Jose Ramon Garmendia Leiza, Jesus Maria Andres de Llano, Juan Bautista Lopez Messa, et al. New
insights into the circadian rhythm of acute myocardial infarction in subgroups. Chronobiology interna-
tional, 24(1):129–141, 2007.
30. Donald Lloyd-Jones, Robert J Adams, Todd M Brown,Mercedes Carnethon, Shifan Dai, Giovanni De Si-
mone, T Bruce Ferguson, Earl Ford, Karen Furie, Cathleen Gillespie, et al. Heart disease and stroke
statistics2010 update a report from the american heart association. Circulation, 121(7):e46–e215, 2010.
31. Beatriz L´
opez, Silvana Aciar, Bianca Innocenti, and Isabel Cuevas. How multi-agent systems support
acute stroke emergency treatment. In IJCAI Workshop A2HC, pages 51–59, 2005.
32. Beatriz L´
opez, Bianca Innocenti, and D´
ıdac Busquets. A multiagent system for coordinating ambulances
for emergency medical services. Intelligent Systems, IEEE, 23(5):50–57, 2008.
33. Marin Lujak and Holger Billhardt. Coordinating emergency medical assistance. In Agreement Technolo-
gies, pages 597–609. Springer, 2013.
34. Marin Lujak, Holger Billhardt, and Sascha Ossowski. Optimizing emergency medical assistance coordi-
nation in after-hours urgent surgery patients. In Nils Bulling, editor, Multi-Agent Systems, volume 8953
of LNCS, pages 316–331. Springer, 2015.
35. Roberto Manfredini, Fabio Manfredini, Benedetta Boari, et al. Seasonal and weekly patterns of hospital
admissions for nonfatal and fatal myocardial infarction. The American journal of emergency medicine,
27(9):1097–1103, 2009.
36. PR Martens, Paul Calle, B Van den Poel, P Lewi, Belgian Cardiopulmonary Cerebral Resuscitation Study
Group, et al. Further prospective evidence of a circadian variation in the frequency of call for sudden
cardiac death. Intensive care medicine, 21(1):45–49, 1995.
37. Susana Mingo, Javier Goicolea, and Luis et al. Nombela. Angioplastia primaria en nuestro medio.
an´
alisis de los retrasos hasta la reperfusi´
on, sus condicionantes y su implicaci´
on pron´
ostica. Revista
espa˜
nola de cardiolog´
ıa, 62(1):15–22, 2009.
38. Instituto nacional de estadistica. Defuncines seg ´
un la causa de muerte, a˜
no 2009. http://www.ine.
es/prensa/np664.pdf, February 2011.
39. Abdur Rais and Ana Viana. Operations research in healthcare: a survey. International Transactions in
Operational Research, 18(1):1–31, 2011.
40. S.S. Rathore, J.P. Curtis, J. Chen, Y. Wang, B.K. Nallamothu, A.J. Epstein, H.M. Krumholz, et al. Asso-
ciation of door-to-balloon time and mortality in patients admitted to hospital with st elevation myocardial
infarction: national cohort study. BMJ: British Medical Journal, 338, 2009.
41. Joaqu´
ın ´
Alvarez Rodr´
ıguez, Isabel Casado Flores, Javier Botas Rodr´
ıguez, et al. Reperfusi ´
on del infarto
agudo de miocardio con elevaci´
on del segmento st en la comunidad de madrid.
42. Amartya Kumar Sen. Collective choice and social welfare, volume 11. Elsevier, 2014.
43. WHO Fact sheet No. 310. http://www.who.int/mediacentre/factsheets/fs310/en/,
May 2014.
44. CY Soon, WX Chan, and HC Tan. The impact of time-to-balloon on outcomes in patients undergoing
modern primary angioplasty for acute myocardial infarction. Singapore medical journal, 48(2):131,
2007.
45. P.G. Steg, S.K. James, D. Atar, et al. Esc guidelines for the management of acute myocardial infarction in
patients presenting with st-segment elevation the task force on the management of st-segment elevation
acute myocardial infarction of the european society of cardiology (esc). European Heart Journal, 2012.
46. Masashi Takahashi, Shun Kohsaka, Hiroaki Miyata, et al. Association between prehospital time interval
and short-term outcome in acute heart failure patients. Journal of Cardiac Failure, 17(9):742–747, 2011.
47. Christian Juhl Terkelsen, Jacob Thorsted Sørensen, Michael Maeng, et al. System delay and mortality
among patients with stemi treated with primary percutaneous coronary intervention. JAMA: the journal
of the American Medical Association, 304(7):763–771, 2010.
48. Frans Van de Werf, Jeroen Bax, Amadeo Betriu, et al. Management of acute myocardial infarction in
patients presenting with persistent st-segment elevation. European heart journal, 29(23):2909–2945,
2008.
... In our previous work, we studied dynamic out-of-hospital urgent patient assignment to ambulances, dynamic reallocation of idle ambulances to better respond to future incidents based on historical data, [12]- [14], and coordination between ambulances, hospitals and out-of-hospital surgery teams for the assistance of urgent surgery patients [15], [16]. This work is a continuation of our previous work and especially of [17], where we proposed a first distributed and dynamic work break scheduling algorithm considering individual crews' preferences in emergency vehicle fleets with highly stochastic and unpredictable demand. ...
... The algorithm should be applied in conjunction with a procedure to assign agents to incidents. Here, different strategies may be applied (see, e.g., [12], [14], [16]). ...
... n ← getM inRequiredAgents(τ max );16 if n ≤ |I| then 17 V =sort C by non-decreasing value of S a (t); 18 Y ← get first |I| − n elements of V ; 19 B ← B ∪ Y ; 20 I ← I \ Y ; 21 else 22 X ← {a ∈ A t |s a (t) = on break ∧ S wt a (t b,a ) > 0 ∧ t b,a + ∆ wb > t}; 23 sort X by non-increasing value of S a (t); 24 BI ← get first n − |I| elements of X; 25 B ← B \ BI; 26 I ← I ∪ BI; 27 end 28 P I ← getLocations(I); 29 return B, I, P I ; ...
... emergency services (e.g. [5,40,41,70]), taxi, hot meal home delivery and vehicle sharing. After a meticulous analysis of the available solutions, we have identified that the combination of these methods can provide a true differential value in the agriculture vehicle fleets. ...
... Then, the multi-index assignment problem is simplified to the assignment problem focusing on the one agent-one task allocation at the time (e.g. [6,41]). ...
... In [42,43], Lujak et al. proposed a modified version of Bertsekas' auction algorithm for the case of incomplete information exchange and explored the deterioration of the solution quality according to the size of the communication network and proposed strategies to overcome this problem. Responding to the task assignment in the case of the medical emergency assistance of urgent out-of-hospital patients by ambulances, Lujak et al. proposed a distributed algorithm for the simultaneous assignment of ambulances [5,40] and ambulances and hospitals to multiple simultaneous patients in [41], where the authors also proposed an ambulance vehicle Voronoi-based relocation approach. Moreover, in [45], Lujak et al. proposed the route assignment approach that considers fair and envy-free routes and improves the overall efficiency in respect to the user optimum. ...
Article
Full-text available
To date, the research on agriculture vehicles in general and Agriculture Mobile Robots (AMRs) in particular has focused on a single vehicle (robot) and its agriculture-specific capabilities. Very little work has explored the coordination of fleets of such vehicles in the daily execution of farming tasks. This is especially the case when considering overall fleet performance, its efficiency and scalability in the context of highly automated agriculture vehicles that perform tasks throughout multiple fields potentially owned by different farmers and/or enterprises. The potential impact of automating AMR fleet coordination on commercial agriculture is immense. Major conglomerates with large and heterogeneous fleets of agriculture vehicles could operate on huge land areas without human operators to effect precision farming. In this paper, we propose the Agriculture Fleet Vehicle Routing Problem (AF-VRP) which, to the best of our knowledge, differs from any other version of the Vehicle Routing Problem studied so far. We focus on the dynamic and decentralised version of this problem applicable in environments involving multiple agriculture machinery and farm owners where concepts of fairness and equity must be considered. Such a problem combines three related problems: the dynamic assignment problem, the dynamic 3-index assignment problem and the capacitated arc routing problem. We review the state-of-the-art and categorise solution approaches as centralised, distributed and decentralised, based on the underlining decision-making context. Finally, we discuss open challenges in applying distributed and decentralised coordination approaches to this problem.
... Vehicle fleets that rely on one-on-one vehicle task assignment are, for example, rescue fleets (see, e.g., [15]), ridehailing and taxi service (see, e.g., [16]), ambulance assistance of urgent out-of-hospital patients (see, e.g., [17]), and homedelivered restaurant hot meal services (see, e.g., [18]). Ridehailing and restaurant hot meal delivery services are examples of open vehicle fleets that use online on-demand service platforms (see, e.g., [19]) to allocate in real-time customers and independent private vehicle owners, drivers, or couriers, using their personal vehicles. ...
... Each agent and task is assumed to be characterized by a set of attributes that influences the cost or profit resulting from an agent-task allocation. In this way, the task allocation problem that assigns tasks to agents in time is simplified to task assignment problem focusing on the one agent-one task allocation at the time (see, e.g., [17,21]). Optimized and dynamic task (re-)assignment may considerably improve the performance of the fleet while considering individual fairness and efficiency (see, e.g., [21]). ...
... e static and deterministic AP is a computationally easy problem, which allows us (in theory) to find an optimal solution in close to real time (in the nonrenewable agent case). Dynamic AP can be solved by (suboptimal) myopic approaches that consider only the information available at the present time with no consideration for future events and possibly reassign tasks among idle and already assigned agents to improve the system's efficiency (see, e.g., [8,17,21,22]). However, in the case where tasks are not randomly appearing, this approach can be significantly improved by considering future developments. ...
Article
Full-text available
One of the major challenges in the coordination of large, open, collaborative, and commercial vehicle fleets is dynamic task allocation. Self-concerned individually rational vehicle drivers have both local and global objectives, which require coordination using some fair and efficient task allocation method. In this paper, we review the literature on scalable and dynamic task allocation focusing on deterministic and dynamic two-dimensional linear assignment problems. We focus on multiagent system representation of open vehicle fleets where dynamically appearing vehicles are represented by software agents that should be allocated to a set of dynamically appearing tasks. We give a comparison and critical analysis of recent research results focusing on centralized, distributed, and decentralized solution approaches. Moreover, we propose mathematical models for dynamic versions of the following assignment problems well known in combinatorial optimization: the assignment problem, bottleneck assignment problem, fair matching problem, dynamic minimum deviation assignment problem, Σk-assignment problem, the semiassignment problem, the assignment problem with side constraints, and the assignment problem while recognizing agent qualification; all while considering the main aspect of open vehicle fleets: random arrival of tasks and vehicles (agents) that may become available after assisting previous tasks or by participating in the fleet at times based on individual interest.
... In this context, agents strive to achieve superior performance through collective action. Instances of such coalitions can be observed in various domains, including emergency services (e.g., [26]), agricultural cooperatives (e.g., [28]), taxi and ride-sharing services (e.g., [5,27]), as well as smart grids (e.g., [36]), among others. ...
Article
Full-text available
An individually rational agent will participate in a multiagent coalition if the participation, given available information and knowledge, brings a payoff that is at least as high as the one achieved by not participating. Since agents? performance and skills may vary from task to task, the decisions about individual agent-task assignment will determine the overall performance of the coalition. Maximising the efficiency of the one-on-one assignment of tasks to agents corresponds to the conventional linear sum assignment problem, which considers efficiency as the sum of the costs or benefits of individual agent-task assignments obtained by the coalition as a whole. This approach may be unfair since it does not explicitly consider fairness and, thus, is unsuitable for individually rational agents? coalitions. In this paper, we propose two new assignment models that balance efficiency and fairness in task assignment and study the utilitarian, egalitarian, and Nash social welfare for task assignment in individually rational agents? coalitions. Since fairness is a relatively abstract term that can be difficult to quantify, we propose three new fairness measures based on equity and equality and use them to compare the newly proposed models. Through functional examples, we show that a reasonable trade-off between efficiency and fairness in task assignment is possible through the use of the proposed models.
... We are pleased to see the publication of this special issue focusing on the design and implementation of new methods, techniques, and models (see, e.g., [3]) that adapt or hybridize findings from Distributed Optimization, Multiagent Systems, Network Science, and Distributed Computing and facilitating distributed/parallel/multiagent decision-making and coordination for solving complex computational and real-life problems in large systems. e applications of DDM vary from coordination problems in groups and crowds (e.g., [4]), Internet (e.g., [5]), emergency logistics (e.g., [6]), multirobot systems (e.g., [7]) to transport (see, e.g., [8]), and beyond. ...
Article
Full-text available
We are pleased to see the publication of this special issue focusing on the design and implementation of new methods, techniques, and models that adapt or hybridize findings from Distributed Optimization, Multi-Agent Systems, Network Science, and Distributed Computing, and facilitate distributed/parallel/multi-agent decision-making and coordination for solving complex computational and real-life problems in large systems. The applications of DDM vary from coordination problems in groups and crowds, Internet, emergency logistics, multi-robot systems to transport, and beyond. The main objective of this special issue is to provide an opportunity to study different aspects of intelligent and distributed decision-making and coordination in large and complex systems, including their formal analysis, with an intention to balance between theoretical research ideas and their practicability. Overall, this special issue collects six research articles and one review article on the state-of-the-art in DDM.
... The improvements of the EMS coordination in the literature were achieved both by novel fleet real-time optimisation and communication methods, using new multiagent models (see, e.g., [44,[52][53][54][55]). Despite an exhaustive quantity of work on the optimisation of EMA, to the best of our knowledge, there is little work on optimisation models for the coordination of EMS when the arrival of multiple EMS actors needs to be coordinated for the beginning of the patients' treatment. ...
Article
Full-text available
Many challenges in today's society can be tackled by distributed open systems. This is particularly true for domains that are commonly perceived under the umbrella of smart cities, such as intelligent transportation, smart energy grids, or participative governance. When designing computer applications for these domains, it is necessary to account for the fact that the elements of such systems, often called software agents, are usually made by different designers and act on behalf of particular stakeholders. Furthermore, it is unknown at design time when such agents will enter or leave the system, and what interests new agents will represent. To instil coordination in such systems is particularly demanding, as usually only part of them can be directly controlled at runtime. Agreement technologies refer to a sandbox of tools and mechanisms for the development of such open multiagent systems, which are based on the notion of agreement. In this paper, we argue that agreement technologies are a suitable means for achieving coordination in smart city domains, and back our claim through examples of several real-world applications.
... The scheduling layer contains the task assignment and route planning module. Based on the semantically described delivery demand, each robot agent coordinates with other robot agents for the task assignment through the bi-level task assignment algorithm in [17]. While MRT is responsible of the MRT task assignment, the AmI network is responsible of updating the travel times under congestion in the network and distributively optimizing robots' routes by using the route finding algorithm in [19]. ...
Conference Paper
Full-text available
In this work, we set the bases of the integration of ambient intelligence (AmI) with mobile robot teams (MRT), aiming to enhance ambient assisted living services addressing a variety of tasks. We argue that people with reduced mobility can benefit from a synergy between AmI and MRT in various aspects. Towards this direction, we identify principal function-alities such an integrated system should provide in connection to relevant previous works and the way by which synergy could be accomplished, from low-level behavioural to higher-level task planning of a multi-layered system architecture.
Chapter
An integrated network of mobile robots, personal smart devices, and smart spaces called “Robots-Assisted Ambient Intelligence” (RAmI) can provide for a more effective user assistance than if the former resources are used individually. Additionally, with the application of distributed network optimization, not only can we improve the assistance of an individual user, but we can also minimize conflict or congestion created when multiple users in large installations use the limited resources of RAmI that are spatially and temporally constrained. The emphasis of RAmI is on the efficiency and effectiveness of multiple and simultaneous user assistance and on the influence of an individual’s actions on the desired system’s performance. In this paper, we model RAmI as a multi-agent system with AmI, user, and robot agents. Moreover, we propose a modular three-layer architecture for each robot agent and discuss its application and communication requirements to facilitate efficient usage of limited RAmI resources. Our approach is showcased by means of a case study where we focus on meal and medicine delivery to patients in large hospitals.
Article
Exploration is a fundamental problem in robotics, and multi-robot systems exploration has been extensively studied in this field. In order to overcome the problem of a non-optimal target being selected in the exploration process, a revised single linkage clustering frontier cell (CFC) algorithm is proposed to calculate the exploration benefit of all available frontier cells. Moreover, there exist unexplored islands for most of the bidding-based multi-robot coordination algorithms in the exploration of unknown environments. To deal with this problem, some rules switched by emotion states are proposed. So, the proposed bidding coordination algorithm with CFC and an emotion switch has a hierarchical architecture. The upper level is modeled as an automaton, which is used to represent emotion status, and the emotion variables decide whether a robot will participate in a bid and explore an unknown area abiding by the walking rules. In the lower level, the robots perform bidding activities with CFC and the walking rules according to the emotion variables. We tested and evaluated our approach by means of experiments both in a simulated environment and with real robots. The experiments results demonstrate that the exploration efficiency is improved, and our algorithm has a greater coverage rate than classic bidding-based coordination algorithms.
Conference Paper
Full-text available
This paper treats the coordination of Emergency Medical Assistance (EMA) and hospitals for after-hours surgeries of urgent patients arriving by ambulance. A standard hospital approach during night-shifts is to have standby surgery teams come to hospital after alert to cover urgent cases that cannot be covered by the in-house surgery teams. This approach results in a considerable decrease in staffing costs in respect to having sufficient permanent in-house staff. Therefore, coordinating EMA and the hospitals in a region with their outhouse staff with the objective to have as fast urgent surgery treatments as possible with minimized cost is a crucial parameter of the medical system efficiency and as such deserves a thorough investigation. In practice, the process is manual and the process management is case-specific, with great load on human phone communication. In this paper, we propose a decision support system for the automated coordination of hospitals, surgery teams on standby from home, and ambulances to decrease the time to surgery of urgent patients. The efficiency of the proposed model is proven over simulation experiments.
Conference Paper
Full-text available
This paper treats the coordination of Emergency Medical Assistance (EMA) and hospitals for after-hours surgeries of urgent patients arriving by ambulance. A standard hospital approach during night-shifts is to have standby surgery teams come to hospital after alert to cover urgent cases that cannot be covered by the in-house surgery teams. This approach results in a considerable decrease in staffing costs in respect to having sufficient permanent in-house staff. Therefore, coordinating EMA and the hospitals in a region with their outhouse staff with the objective to have as fast urgent surgery treatments as possible with minimized cost is a crucial parameter of the medical system efficiency and as such deserves a thorough investigation. In practice, the process is manual and the process management is case-specific, with great load on human phone communication. In this paper, we propose a decision support system for the automated coordination of hospitals, surgery teams on standby from home, and ambulances to decrease the time to surgery of urgent patients. The efficiency of the proposed model is proven over simulation experiments.
Article
Management of acute myocardial infarction in patients presenting with persistent ST-segment elevation.
Chapter
Emergency medical service (EMS) providers face a host of significant challenges, including traffic congestion, increasing call volumes, hospital diversion (i.e., emergency departments at hospitals not allowing ambulances to deliver patients to them), and increasing transfer times at emergency departments. Many of these challenges can be attacked through an application of operations research (OR) techniques in conjunction with EMS expertise. The purpose of this article is to outline some of the key challenges and potential OR-based remedies, and to discuss in detail one such remedy that has various names including “system-status management” and “move up.”Keywords:emergency medical service;OR tools;performance measure;system status management;optimization