Vous êtes sur la page 1sur 6

IFAC Conference on Manufacturing Modelling,

IFAC
IFAC Conference
Conference on Manufacturing
Manufacturing Modelling,
Management
IFAC and on
Conference Control
on Manufacturing
Modelling,
Modelling,
Management
IFAC and
Conference
Management Control
on Manufacturing
and Control Available online at www.sciencedirect.com
Modelling,
June 28-30,
Management 2016. Troyes,
and Control
Control France
June 28-30,
Management 2016.
and Troyes, France
June 28-30, 2016. Troyes, France
June
June 28-30,
28-30, 2016.
2016. Troyes,
Troyes, France
France
ScienceDirect
IFAC-PapersOnLine 49-12 (2016) 721–726
Emergency
Emergency department
department flow:
flow: A
A new
new
Emergency
Emergency department
department flow:
flow: A
A new
new
practical
practical patients
patients classification
classification and
and
practical
practical patients classification
patients classification and
and
forecasting
forecasting daily daily attendance
attendance
forecasting
Mohamed daily
Afilal ∗,∗∗ Farouk attendance
Yalaoui
∗,∗∗

Frédéric Dugardin ∗ ∗ ∗
Mohamed
Mohamed Afilal
Afilal ∗ Farouk
∗,∗∗
Farouk Yalaoui
Yalaoui ∗∗∗Frédéric
Frédéric Dugardin
Dugardin ∗∗ ∗
Lionel
Mohamed
Mohamed
Lionel Amodeo
Afilal
Afilal
Amodeo ∗ David
∗,∗∗
∗,∗∗ Farouk
Farouk
∗ David
Laplanche
Yalaoui
Yalaoui
Laplanche
∗∗∗
∗ Philippe
Frédéric
Frédéric
∗∗ Philippe Blua
Dugardin
Dugardin
Blua ∗∗ ∗
∗∗

Lionel
Lionel Amodeo
Amodeo ∗ David

David Laplanche
Laplanche ∗∗ Philippe
∗∗
Philippe Blua
Blua ∗∗
∗∗

Lionel Amodeo David Laplanche Philippe Blua
∗ Institut Charles Delaunay, LOSI, University of Technology of Troyes
∗ Institut Charles Delaunay, LOSI, University of Technology of Troyes
∗ Institut
UMR Charles
6281, CNRS.Delaunay,
12 Rue LOSI,
Marie University
Curie, CS of Technology of Troyes
of42060, 10004 Troyes

Institut
Institut
UMR Charles
Charles
6281, CNRS.Delaunay,
Delaunay,
12 Rue LOSI,
LOSI,
Marie University
University
Curie, CS of Technology
Technology
42060, 10004 of
of Troyes
Troyes
Troyes
UMR 6281,
UMR 6281,
6281, CNRS.
cedex,
CNRS. France12
12 RueRue
(e-mail:
Rue Marie
Marie Curie, CS 42060,
mohamed.afilal@utt.fr).
Curie, CS CS 42060, 10004
42060, 10004
10004 Troyes Troyes
Troyes
∗∗ UMR CNRS.
cedex,
cedex, France
France12 (e-mail:
(e-mail: Marie Curie,
mohamed.afilal@utt.fr).
mohamed.afilal@utt.fr).
∗∗ Département cedex,
cedex, d’information
France
France (e-mail:
(e-mail: médicale, Centre Hospitalier
mohamed.afilal@utt.fr).
mohamed.afilal@utt.fr). de Troyes,
∗∗ Département d’information médicale, Centre Hospitalier de Troyes,
∗∗ Département
∗∗ 101 Avenue
Département d’information
Anatole France,
d’information médicale,
101 Avenue
médicale, Centre
Centre Hospitalier
Anatole France,
Hospitalier de
de Troyes,
10000
Troyes,
Département
101
101 Avenue
Avenue d’information
Anatole
Anatole France,
France, médicale,
101
101 Avenue
AvenueCentre Hospitalier
Anatole
Anatole France,
France, de Troyes,
10000
10000
101 Troyes,
Avenue
101 Avenue
Troyes, France
Anatole
Anatole
France (e-mail:
France,
France,
(e-mail: david-laplanche@ch-troyes.fr).
101
101 Avenue
Avenue Anatole
Anatole
david-laplanche@ch-troyes.fr). France,
France, 10000
10000
Troyes,
Troyes, France
France (e-mail:
(e-mail: david-laplanche@ch-troyes.fr).
david-laplanche@ch-troyes.fr).
Troyes, France (e-mail: david-laplanche@ch-troyes.fr).
Abstract:
Abstract: Emergency
Emergency department
department (ED)
(ED) has
has become
become the
the patient’s
patient’s main
main point
point of
of en-
en-
Abstract:
trance
Abstract: in Emergency
modern
Emergency department
hospitals causing
department (ED)
(ED)it has
frequent become
has become the
overcrowding.
become patient’s
the patient’s Thus
patient’s main
hospital
main point of
of en-
managers
pointmanagers en-
Abstract:
trance
trance in
in Emergency
modern
modern department
hospitals
hospitals causing
causing (ED)it
it has
frequent
frequent the
overcrowding.
overcrowding. Thus
Thus main
hospital
hospital point of
managers en-
are
trance
trance
are increasingly
in
in modern
modern
increasingly giving attention
hospitals
hospitals
giving causing
causing
attention to
to the
it
it
the ED
frequent
frequent
ED in
in order to
overcrowding.
overcrowding.
order to provide
Thus
Thus
provide better
hospital
hospital
better quality
managers
managers
quality ser-
ser-
are
vice increasingly
for patients. giving
One attention
of the keyto the ED for in order to provide better quality ser-
are
are
vice
vice
increasingly
increasingly
for
for patients.
patients.
giving
giving
One
One
attention
attention
of
of the
the key
keyto elements
to the
the ED
elements
elements ED for in
for
aa good
in order
order
a good
good to management
to provide
provide better
management
management betterstrategy
qualityis
quality
strategy
strategy is
is
de-
ser-
ser-
de-
de-
mand
vice
vice
mand for
forforecasting.
patients.
patients.
forecasting. In
InOne
One this
this of
of case,
the
the
case, forecasting
key
key
forecasting elements
elements patients
for
for
patients flow,
a
a good
good
flow, which will
management
management
which will help
help decision
strategy
strategy
decision makers
is
is
makers de-
de-
mand
to
mand forecasting.
optimize
forecasting. In
human(doctors,this
In this case,
this case,
case, forecasting
nurses. . .
forecasting ) and patients
material
patients flow,
(beds,
flow, which
boxs.
which will
.
will . ) help decision
resources
help decision makers
allocation.
makers
mand
to
to forecasting.
optimize
optimize In
human(doctors,
human(doctors, forecasting
nurses.
nurses. .
. .. )) and
and patients
material
material flow,
(beds,
(beds, which
boxs.
boxs. will
.
. .. )
) help decision
resources
resources makers
allocation.
allocation.
The
to
to main
optimize interest of this
human(doctors, research is
nurses. forecasting
.. .. )) and daily
and material attendance
(beds,
(beds, boxs. at an .. .. emergency
resourcesdepartment.
)) resources allocation.
Theoptimize
The
The
main
main
study
main was
human(doctors,
interest
interest
interest
of
of this
this
conducted
of this
research
research
on
nurses.
the
research
is
is forecasting
forecasting
Emergency
is forecasting
material
daily
daily
Department
daily
attendance
attendance
of
attendancethe
boxs.
at
at
at
an
an
Hospital
an
emergency
emergency
of Troyes
emergency
allocation.
department.
department.
city, France,
department.
The
The main
study
study interest
was
was of this
conducted
conducted research
on
on the
the is forecasting
Emergency
Emergency daily
Department
Department attendance
of
of the
the at an
Hospital
Hospital emergency
of
of Troyes
Troyes department.
city,
city, France,
France,
in
The
The
in which
study
study
which we
wewas
was propose
conducted
conducted
propose a
a new
newon
on practical
the
the Emergency
Emergency
practical ED
ED patients
Department
Department
patients classification
of
of the
the
classification that
Hospital
Hospital
that consolidate
of
of Troyes
Troyes
consolidate the
city,
city,
the CCMU
France,
France,
CCMU
in
and which
GEMSA
in which
which we propose
into
we propose
proposeone a new
single
asingle practical
category
new practical
practical ED
and patients
an
ED patients classification
innovative
patients time-series
classification that based
thatbased consolidate
forecasting
consolidate the
the CCMU
models
CCMU
in
and
and GEMSA
GEMSA we into
into one
one a new
single category
category ED
and
and an
an classification
innovative
innovative time-series
time-series that based consolidate
forecasting
forecastingthe CCMU
models
models
to
and
and
to predict
GEMSA
GEMSA
predict long
long and
into
into
and short
one
one
short term
single
single
term daily
category
category
daily attendance.
and
and
attendance. an
an The
innovative
innovative
The models
models we
time-series
time-series
we developed
based
based
developed for this
forecasting
forecasting
for this case
case study
models
models
study
to
to predict
shows very
predict long
good
long and
and short
performances
short term
term daily
(up
dailyto attendance.
92,29%
attendance. for The
the
The models
annual
models we
Total
we developed
flow
developed forecast) for
for this
and
this case study
robustness
case study
to predict
shows
shows very
very long
good
good and short
performances
performances term daily
(up
(up to
to attendance.
92,29%
92,29% for
for The
the
the models
annual
annual we
Total
Total developed
flow
flow forecast)
forecast) for this
and
and case study
robustness
robustness
to epidemic
shows
shows
to very
very
epidemic periods.
good
good performances
performances
periods. (up
(up to
to 92,29%
92,29% for
for the
the annual
annual Total
Total flow
flow forecast)
forecast) and
and robustness
robustness
to
to epidemic
epidemic periods.
periods.
© 2016,
to epidemicIFACperiods.
(International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords:
Keywords: Emergency
Emergency Department
Department Flow, Flow, Forecasting,
Forecasting, Patient
Patient Classification,
Classification, Time Time series.
series.
Keywords:
Keywords: Emergency
Emergency Department
Department Flow, Flow, Forecasting,
Forecasting, Patient
Patient Classification,
Classification, Time Time series.
series.
Keywords: Emergency Department Flow, Forecasting, Patient Classification, Time series.
1. INTRODUCTION to improve quality service for patients Han et al. (2007).
1.
1. INTRODUCTION
INTRODUCTION to
to improve
improve quality
quality service
service for
for patients
patients Han
Han et
et al. (2007).
Over the last 1. 1. INTRODUCTION
INTRODUCTION
decade, emergency departments (ED) Many Many
to
to research
improve
improve works
quality
quality were conducted
service
service for
for patients
patients aiming
Han
Han et
ettoal.
al.
al.
(2007).
solve this
(2007).
(2007).
Over the last decade, emergency departments (ED) Many research
research works
works were
were conducted
conducted aiming
aiming to
to solve
solve this
this
Over
have the
known last decade,
an decade,
importance emergency departments
growth departments
in patients (ED) (ED) problem
Many
flow. problem
Many research by
research optimizing
works
works were were resources
conducted
conducted assignment
aiming
aiming to and
to utiliza-
solve
solve this
this
Over
Overknown
have the
the last
last
an decade,
importance emergency
emergency
growth departments
in patients (ED)
flow. problem by
by optimizing
optimizing resources
resources assignment
assignment and
and utiliza-
utiliza-
have
In known
particular, an
the importance
“Observatoire growth
Régionalein patients
des Urgences flow. tion
problem
problem
tion in
in the
by
by
the ED in
optimizing
optimizing
ED in order
order to improve
resources
resources
to improve patient’s
assignment
assignment
patient’s experience
and
and utiliza-
utiliza-
experience
have
have
In known an
known
particular, an
the importance
importance
“Observatoire growth
growth
Régionalein patients
in patients
des Urgences flow. tion
flow. and in
to the
increaseED in
the order
serviceto to improve
throughput patient’s experience
Ajmi et experience
al. (2013).
In particular, the
Champagne-Ardennes” “Observatoire
organization Régionale
(a public des Urgences
organiza- tion
tion toin the
the ED
in increase ED thein
in order
order improve
improve patient’s
tothroughput patient’s experience
In
In particular,
particular, the
Champagne-Ardennes”
Champagne-Ardennes” the “Observatoire
“Observatoire
organization
organization
Régionale
Régionale
(a
(a
des
public
public Urgences and
Urgences
des organiza-
organiza- and to
However,
and to increase
in
increase the
order
the
service
service
to
service throughput
establish
throughputsuch
Ajmi
Ajmi
solutions,
Ajmi
et
et
et
al.
al.
al.
(2013).
(2013).
decision
(2013).
tion in charge
Champagne-Ardennes”
Champagne-Ardennes” of the collection
organization
organization and (a
(a analysis
public
public of
organiza-
organiza- emer- and to
However,
However, increase
in
in the
order
order service
to
to throughput
establish
establish such
such Ajmi
solutions,
solutions, et al. (2013).
decision
decision
tion
tion in
in charge
charge of
of the
the collection
collection and
and analysis
analysis of
of emer-
emer- makers
However, needin to have
order toa good idea
establish about
such what
solutions, to expect,
decisionin
gency
tion
tion in
indepartments
charge
charge of
of activity),
the
the announced
collection
collection and
and that
analysis
analysis the of
of numberemer-
emer- However,
makers
makers need
needin order
to
to have
have toa
a establish
good
good idea
idea such
about
about solutions,
what
what to
to decision
expect,
expect, in
in
gency
gency departments
departments activity),
activity), announced
announced that
that the
the number other words, forecasting the level of demand (patients flow)
of visits
gency
gency
of visits to the
departments
departments
to the emergency
the emergency
emergency departments
activity),
activity), announced
announced
departments in the
that
in that the number
Champagne-
the
the Champagne- number other
number
Champagne-
makers
makers
other
ahead
need forecasting
need
words,
words,
to have
to have aa good
(days, forecasting
weeks) is a crucial
good
the
idea of
idea
the level
level about
about
of
step
demand
demand
what(patients
what to expect,
to
to a good(patients
expect,
management
in
in
flow)
flow)
of
of visits
Ardennes
visits to
to state
the of France
emergency departments
has increased
departments in
inbythe
6.43%
the per
Champagne- year other
other
ahead words,
words,
(days, forecasting
forecasting
weeks) is a the
the
cruciallevel
level of
of
step demand
demand
to a good (patients
(patients
management flow)
flow)
of visits tostate
Ardennes
Ardennes the emergency
of France departments
has increased inbythe Champagne-
6.43% per year ahead
strategy. (days, weeks) is aa crucial step to a good management
from 2008 state
Ardennes
Ardennes to 2013
state
state
of
of
of
France
France
France
has
has
has
increased
ORU-Champagne-Ardenne
increased
increased
by
by
by
6.43%
6.43%
6.43%
per
(2014).
per
per
year
year
year
ahead
ahead (days,
strategy.
strategy. (days, weeks)
weeks) is is a crucial
crucial stepstep toto aa good
good management
management
from
from 2008
2008 to
to 2013
2013 ORU-Champagne-Ardenne
ORU-Champagne-Ardenne (2014).
(2014). strategy.
from 2008 to 2013 ORU-Champagne-Ardenne
from 2008 to 2013 ORU-Champagne-Ardenne (2014). (2014). strategy.
The ED overcrowding problem and it’s consequences Time-series has proved to be a well adapted tool to
The
The ED
ED overcrowding
overcrowding problem
problem and
and it’s
it’s consequences
consequences Time-series has
Time-series has proved
proved to to be be aa well
well adapted
adapted tool tool toto
drew
The the
ED attention
overcrowding of several nations
problem and as
it’sresearchers
consequences forecast
in- forecast levels of
Time-series
Time-series has
has demand
proved
provedin intodifferent
to be areasadapted
be aa well
well De Gooijer
adapted tool and
toolandto
to
The
drew
drew ED
the
the overcrowding
attention
attention of
of problem
several
several nations
nationsand it’s
as
as consequences
researchers
researchers in-
in- forecast levels
levels of
of demand
demand in different
different areas
areas De
De Gooijer
Gooijer and
spected
drew the the causes
attention and
of impacts
several of
nationsthis phenomenon
as researchers Sun
in- Hyndman
forecast
forecast (2006),
levels
levels of
of the
demand
demand core in
in of this
different
different tool relies
areas
areas De
De on exploiting
Gooijer
Gooijer and
and
drew
spected the theattention
causes of
and several
impacts nations
of this as researchers
phenomenon in-
Sun Hyndman
Hyndman (2006),
(2006), the
the core
core of
of this tool
this tool relies on exploiting
spected
et the
al. (2013)
spected the causes and
and proposed impacts of
various this phenomenon
solutions to improve Sun historical
Hyndman data
(2006),aboutthe aaacore
the phenomenon
of
of this
this tool inrelies
orderon
relies on toexploiting
establish
exploiting
spected
et
et al.
al. the causes
(2013)
(2013) causes
and
and
and
and impacts
proposed
proposed impacts of
of this
various
various this phenomenon
phenomenon
solutions
solutions to
to improve
improve
Sun Hyndman data
Sun historical
historical
a mathematical
(2006),
data about
about
model
core
phenomenon
phenomenon
that can
toolin
be in relies
order
order
used
onto
to to exploiting
establish
establish
predict it
quality
et
et al. serviceand
al. (2013)
(2013) Ajmi
and et al. (2013).
proposed
proposed various
variousThe usual causes
solutions
solutions to
to improve
improve historical
of ED ahistorical mathematicaldata about
data about
model aa phenomenon
phenomenon
that can can be inused
bein ordertoto
order topredict
establish
establish it
quality
quality service
service Ajmi
Ajmi et
et al.
al. (2013).
(2013). The
The usual
usual causes
causes of
of ED
ED abehavior
mathematical
in the model
future. that
Time-series used
analysis to
has predict
been usedit
overcrowding
quality service
quality serviceare: are:
Ajmi
Ajmi inadequate
et al.
et al. (2013).resources
(2013). The
The usual allocation
usual causes
causes Cooke Cooke
of ED
of ED behaviora
a mathematical
mathematical
behavior in the model
model
future. that
that
Time-seriescan
can be
be used
used
analysis to
to
has predict
predict
been usedit
it
overcrowding
overcrowding are: inadequate
inadequate resources
resources allocation
allocation Cooke by Kadri inetthe
al. future.
(2014) Time-series
and Sun etanalysis
al. (2009) has been
to used
develop
et al. (2004),
overcrowding
overcrowding increase
are: of
are: inadequate the
inadequate demand
resources
resourcesfor ED services
allocation
allocation ORU-
Cooke
Cooke behavior
behavior
by Kadri in
inetthe
the
al. future.
future.
(2014) Time-series
Time-series
and Sun etanalysis
analysis
al. (2009) has
has been
been
to used
used
develop
et
et al. (2004),
al. (2004), increase
increase of of the
the demand
demand for
for ED services
ED seasons
services ORU-
ORU- by Kadri et
statistical al.
models (2014)
in order and toSun et
forecast al.the
(2009)patientsto develop
flow in
Champagne-Ardenne
et
et al.
al. (2004),
(2004), increase
increase of (2014)
of the and epidemic
the demand
demand for
for ED services
ED seasons
services Schull Schull
ORU- by
by Kadri
ORU- statistical Kadri et et al.
al. (2014)
models (2014)
in order and
andto Sun
Sun et
et al.
forecast (2009)
al.the
(2009)patientsto
to develop
develop
flow in
Champagne-Ardenne (2014) and epidemic statistical models in order to forecast the patients flow
Champagne-Ardenne
et al. (2005).
Champagne-Ardenne
Champagne-Ardenne
(2014)
(2014)
(2014)
and
and
and
epidemic
epidemic
epidemic
seasons
seasons
seasons
Schull
Schull
Schull
the ED.
statistical
statistical
the ED. modelsmodels in in order
order to to forecast
forecast the the patients
patients flow flow in
in
in
et
et al.
al. (2005).
(2005). the ED.
et al. (2005).
et al. (2005). the
the ED.
ED.
One of the straight solutions to tackle this kind of The main objective of this study was to develop a
One
One of
of the
the straight
straight solutions
solutions to
to tackle
tackle this
this kind
kind of
of The
The main
main objective
objective of
of this
this study
study was to develop a
problems
One
One of
of is a
the
the greater
straight
straight number
solutions
solutions of resources,
to
to tackle
tackle which
this
this kind
kindmeans of
of forecasting
The
The main
main model to predict
objective
objective of
of this
this the dailywas
study
study was
was
to
patients
to
to
develop
flow in
develop
develop
aa
a
problems
problems is
is a
a greater
greater number
number of
of resources,
resources, which
which means
means forecasting
forecasting model
model to
to predict
predict the
the daily
daily patients
patients flow
flow in
in
expanding
problems is
problems the
isthe
a facilities
a greater
greater and/or
number
number hiring
ofhiring
of more
resources,
resources, staff,
whichthis
which thismeans
means so- the long
forecasting
forecasting and short
model
model terms
toterms
to predict
predictat the Emergency
the Emergency
the daily patients
daily patients department
flow in
flow in
expanding
expanding the facilities
facilities and/or
and/or hiring more
more staff,
staff, this so-
so- the
the long
long and
and short
short terms at
at the
the Emergency department
department
lution
expanding
expandingis lessthelikely
thelikely to beand/or
facilities
facilities adopted
and/or by any
hiring
hiring more
more health
staff,organiza-
this
this so-
staff,organiza- so- of of
the the
thethelong Troyes
and city hospital
and short
longTroyes short terms
terms at center,
the France. The
the Emergency
at center, Emergency aim of
department
department
lution
lution is
is less
less likely to
to be
be adopted
adopted by
by any
any health
health organiza- of the Troyes city
city hospital
hospital center, France.
France. The
The aim
aim of
of
tion
lution
lution(unless
is
is less
less the city
likely
likely to
to is
be
be witnessing
adopted
adopted by
bya critical
any
any demographic
health
health organiza-
organiza- the
of
of forecasting
the
the Troyes
Troyes model
city
city hospital
hospitalwas to predict
center,
center, daily
France.
France. patients
The
The aim
aimflow
of
of
tion
tion (unless
(unless the
the city
city is
is witnessing
witnessing a
a critical
critical demographic
demographic the
the forecasting
forecasting model
model was
was to
to predict
predict daily
daily patients
patients flow
flow
growth)
tion
tion (unless
(unlessbecause
the
the of
city
city the
is
is high cost
witnessing
witnessing a
aof such
critical
critical policy.
demographic
demographic More- concerning
the
the forecasting
forecasting the different
model
model was
wascategories
to
to predict
predictof CCMU
daily
daily (Classifica-
patients
patients flow
flow
growth)
growth) because
because of
of the
the high
high cost
cost of
of such
such policy. More- concerning the different categories of CCMU (Classifica-
over,
growth)
growth)
over, this kind
because
thisbecause
kind of ofofact
ofof thecould
the
act high be
high
could costan
cost
be anof such policy.
of insufficient
such policy.solution
policy.
insufficient
More-
More- concerning
More-
solution tion Clinique
concerning
concerning
tion Clinique
the
thedes
thedes
different
Malades
different
different
Malades
categories
aux
categories
categories
aux Urgences
of
of CCMU
Urgences
of
Urgences CCMU
CCMU
(Classifica-
: Clinical Emer-
(Classifica-
(Classifica-
over,
over, this kind act could be an insufficient solution tion Clinique des Malades aux ::: Clinical
Clinical Emer-
Emer-
this kind of act could be an insufficient solution
over, this kind of act could be an insufficient solution tion Clinique des Malades aux Urgences : Clinical Emer-
tion Clinique des Malades aux Urgences Clinical Emer-
2405-8963 ©
Copyright © 2016, IFAC (International Federation of Automatic Control)
2016 IFAC 721 Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©
Copyright 2016
©under IFAC
2016 responsibility
IFAC 721
of International Federation of Automatic
721Control.
Copyright
Copyright © 2016
© 2016 IFAC
IFAC
10.1016/j.ifacol.2016.07.859 721
721
IFAC MIM 2016
722
June 28-30, 2016. Troyes, France Mohamed Afilal et al. / IFAC-PapersOnLine 49-12 (2016) 721–726

gency Department Classification) and GEMSA (Groupes 3. DATA DESCRIPTION


d’Etude Multicentric des Services d’Accueil : Multicentric Troyes Hospital Center (CHT) is the second largest
Emergency Department Study Group) ORU-Champagne- health care center in the Champagne-Ardenne region of
Ardenne (2014). In the data analysis step(model building France linked with a population of 130 194. The emergency
pre-step), we proposed a new practical classification of department of the CHT is open 24/7 for the inhabitants
the ED patients that takes into account the CCMU and of Troyes urban community and receives 50 488 patients
GEMSA classifications. The objective of this latter is to a year on average. The department uses a dedicated
group patients that requires similar resources and/or have hospital information system (HIS) to manage the patient’s
similar behavior in the ED. information flow, this latter record every action of the
2. STATE OF THE ART patient from the moment of admission till the time of
release from the service (discharge or hospitalization).
Forecasting emergency department flow has been stud-
ied extensively in recent years, the first appearance of such
work was in 1988 when Milner introduced an ARIMA We collected the summary attendance to emergencies
based model to forecast total attendance of an ED Mil- (RPU) of every patient from the first January 2010 to
ner (1988). Then, many research works were interested the 31st December 2014. The RPU is a standardized
in developing long term forecasting models for ED’s pa- collection of data for each emergency attendance, it is
tients flow, most of this works used linear models such as used by all the EDs of France to summarize their activity
ARMA/ARIMA, Kadri et al. (2014); Sun et al. (2009); ORU-Champagne-Ardenne (2014). The RPU contains 20
Jones et al. (2002); Shi et al. (2011) (as they proved information field that describe the crossing of the patient
to be very effective), multiple linear regression, Boyle in the ED, those that interest us the most in our case
et al. (2008) etc. Short term forecasting models were study are: date and time of entry, GEMSA and CCMU
also investigated and developed by many research works, classification, Table 1 and 2 give a detailed description of
the techniques used involved linear models as ARIMA, these classifications. These latter are assigned and updated
Abraham et al. (2009), linear regression, Ekström et al. to the patients after being treated in the service according
(2015) and exponential smoothing, Bergs et al. (2014), to their final diagnosis.
non-linear models as Queuing theory, Stout and Tawney
(2005), Monte Carlo simulation based models, Mielczarek The presence of some records with non specified fields of
(2013). Other works were interested in developing forecast- CCMU and GEMSA categories that we denoted CX and
ing models that predict the set of a high demand period, GX respectively (see table 1 and 2 for details) indicates
for example, we can cite the work of Bouleux et al. (2014) some special cases, the CX category is attributed to a
when he presented a model that can predict the start of a patient if : he is a child, in this case he is redirected to
crowding period in a pediatric division of an ED of Lille, a special pediatric division of the ED, or if the patient left
France. Authors may also refer to some research works without being examined by a doctor, the field is left blank
that were interested in forecasting demands similar to an because the patient has not been diagnosed. As for the GX
ED’s patients flow, we consider these phenomenons to be category, it is only assigned to a patient in case of leaving
similar since they are based on an urgent and unexpected without being examined by a doctor.
human need. In this category we can find works interested Table 1. GEMSA classification of ED’s patients
in forecasting short-term emergency supplies, Mohammadi
et al. (2014), using radial basis function neural network GEMSA Admittance Description
(RBFNN). mode
G1 Unplanned Dead patient on arrival
A very recent similar study of Kadri et al. (2014) showed
G2 Unplanned Patient not convened, unexpected
the importance of predicting categorized ED’s patient arrival, returned home after emer-
flow , however they only took the GEMSA classification gency care
into account which can help to predict the number of G3 Planned Patient convened, expected ar-
beds needed to host inpatient flow, moreover the study rival, returned home after emer-
was only applied to the pediatric division of the ED. gency care
Sun et al. (2009) were also interested in forecasting ED’s G4 Unplanned Patient not convened, unexpected
patient flow by category, they divided patients by acuity arrival, hospitalized after emer-
level and provided forecasting models to predict daily gency care
attendance of each category, this approach can help ED’s G5 Planned Patient convened, expected ar-
staff to anticipate the resources needed in order to meet rival, hospitalized after emergency
the demand. care
G6 Unplanned Patient requiring immediate or
The contribution of this research work is the introduc- prolonged care (intensive care)
tion of a new practical ED’s patient classification that GX Not specified Not specified
gathers the CCMU and GEMSA categories used in the
EDs of France and propose a general and efficient fore-
casting model to predict daily attendance of each category The data collected were grouped by each category of
in the long and short term. The purpose of the new GEMSA and CCMU and were transformed to a time
classification is to group patients that require similar level series data representing the daily attendance at the ED,
of resource and/or present similar behavior in the ED. figure 1 and 2 provides the number of daily arrivals by

722
IFAC MIM 2016
June 28-30, 2016. Troyes, France Mohamed Afilal et al. / IFAC-PapersOnLine 49-12 (2016) 721–726 723

CCMU, respectively GEMSA, category at the emergency related to the amount of resources needed to treat the
department of Troyes hospital center from January 2010 patient: need for an MRI or X-rays examination, need
to December 2014. for lab blood work, duration of treatment by the doc-
tor, need for a surgery, need of a special consultation
etc. The GEMSA category distinguish between two main
categories: inpatients and outpatients, as for the mode of
admittance planned or unplanned, it has no impact regard-
ing the flow management, indeed the planned patients who
comes to the ED are either transferred from another health
care center, critical arrivals that has been informed by the
urgent medical assistance services or redirected to the ED
by an external doctor, in all of these cases, the staff of the
ED is informed by the arrival a few hours earlier, which
can only help them plan their activities in the operational
level. As a conclusion, the authors can state that there are
Fig. 1. Daily attendance at the ED from 01/01/2010 to no planned arrivals to the ED in the short (days) or the
31/12/2014 by CCMU category. long (weeks, months) term.

The crossing of a patient through the ED is categorized


by a CCMU and GEMSA code that can take 8 respectively
7 states, this means there are 56 (8 × 7) possible states for
a patient in the ED. However, some of these states are
inconsistent, for example, a patient can not be categorized
as CCMUD (died at the entrance to the ED, see table 2)
and GEMSA2 (returned home after emergency care, see
table 1) in the same time. . . Which means the number of
the actual possible states of a patient in the ED is less
than 56.
Fig. 2. Daily attendance at the ED from 01/01/2010 to 4.2 Practical patients classification
31/12/2014 by GEMSA category.
In this subsection we introduce a new practical patients
classification at the ED. This new classification consists
4. A NEW EMERGENCY DEPARTMENT PATIENTS of grouping crossed CCMU and GEMSA categories based
CLASSIFICATION on statistical tests and ED’s staff expertise. We named
4.1 Categories description this new classification by EP category that stands for
: Patient Status. By regrouping the crossed CCMU and
In the data description section, we pointed out that GEMSA categories to the EP category we are going to
the data retrieved contained both CCMU and GEMSA reduce the number of possible states of a patient in the ED
fields for each record of patient’s attendance at the ED. to a practical number of categories that have a particular
The CCMU category indicates the severity of patient’s description regarding the flow management. The objective
condition while the GEMSA category indicates the mode of introducing this new classification is to consolidate the
of admittance and the affiliation of the patient after being CCMU and GEMSA into one practical classification.
treated in the ED. In practice, the CCMU category is
Table 2. CCMU classification of ED’s patients
Figure 3 introduces the EP classification which has 8 dif-
CCMU Description ferent attributes, each category was created by regrouping
multiple crossed CCMU and GEMSA categories. Table 3
C1 Stable situation, abstention of complementary gives a description of each category.
diagnostic or therapeutic act
C2 Stable situation, perform a complementary di-
agnostic or therapeutic act
C3 Situation likely to deteriorate without involving
life-threatening
C4 Prognosis committed, no immediate resuscita-
tion maneuver
C5 Prognosis committed, perform immediate resus-
citation maneuver
CP Patient with psychological and / or psychiatric
problems dominant in the absence of any unsta-
ble somatic pathology
CD Patient dies at the entrance to the emergency Fig. 3. Practical patient status classification at the ED.
CX Not specified

723
IFAC MIM 2016
724
June 28-30, 2016. Troyes, France Mohamed Afilal et al. / IFAC-PapersOnLine 49-12 (2016) 721–726

Table 3. EP categories description over the year where the short-term ones where developed
to give a better forecasting using recent observations of
EP Description the patient’s flow. The forecasting models were applied to
the most important categories of patient’s flow in the ED.
EP1 Outpatients with moderate medical treatment In the next subsection, we present an ABC classification
EP2 Outpatients with significant medical treatment study based on the number of occurrence of each EP in
order to select the class of EP that impacts the most the
EP3 Inpatients with moderate medical treatment ED.
EP4 Inpatients with significant medical treatment 5.1 Forecasting results
EP5 Patients who needs major medical treatment In this subsection we present the results of the long and
short-term forecasting models of the EP1, EP3, EP1+EP3
EP6 Patients who needs major medical treatment + and the Total patients flow constructed using data from
resuscitation acts year 2010 to 2013 and tested on the year 2014. These
EP7 Patient dead at the entrance latter are showed in figures 5,6,7 and 8 respectively. From
this figures we can notice that the long-term forecast
EPX Others model follows the overall trend of the observation for each
category. The short-term model uses recent information
of the observation to improve the long-term forecast: we
notice that the short-term models tend to adjust the
forecast in the case of high or low demand peak (see figures
5,6,7 and 8).

Fig. 4. Daily attendance at the ED from 01/01/2010 to


31/12/2014 by EP category.
5. FORECASTNG MODEL RESULTS
The test results of the forecasting models constructed
using the arrivals data from January 2010 to December Fig. 5. EP1 forecast from 01/01/2014 to 31/12/2014
2013 that we test using the data of the year 2014 are pre-
sented in this section. We discuss the performances of the
long and short term forecasting models and the analysis of
the residuals. The performance of the models is assessed
trough the relative mean absolute performance(RMAP)
criterion given by equation 4. A residual analysis is also
performed in order to test the null hypothesis of white
noise distribution. Finally we test the assumption of cor-
relation between the residuals and epidemic cases in the
Champagne-Ardenne region of France, if the correlation
is found to be insignificant this would mean that the
forecasting model is robust to epidemic periods.
εt = Yt − Pt (1) Fig. 6. EP3 forecast from 01/01/2014 to 31/12/2014
M AE = mean|εt | (2)
M AE
RM AE = (3)
mean(Yt )
RM AP = 1 − RM AE (4)
Yt : Observed number of arrivals at day t.
Pt : Predicted number of arrivals at day t.
εt : Forecast error committed at day t.
M AE : Mean absolute error.
RM AE : Relative mean absolute error.
RM AP : Relative mean absolute performance.

The long-term forecast models were developed to give Fig. 7. EP1+EP3 forecast from 01/01/2014 to 31/12/2014
decision makers an extensive vision of the patient’s flow

724
IFAC MIM 2016
June 28-30, 2016. Troyes, France Mohamed Afilal et al. / IFAC-PapersOnLine 49-12 (2016) 721–726 725

Fig. 8. Total forecast from 01/01/2014 to 31/12/2014


Fig. 9. ACF of the residuals of the short-term forecasting
Table 4 represents the performance of the long and model of the Total patient’s flow.
short-term forecasting models of each category, we notice, convergence of a given statistic to the chi-squared distri-
as expected, that the performances of the short-term bution(see Box and Jenkins (1971)). The p-values of this
models are slightly better than the long-term ones, this test are 0.86, 0.75, 0.91 and 0.97 for the residuals of the
would help decision makers to anticipate over or under short-term forecasting models of the EP1, EP3, EP1+EP3
crowding situations. We also notice that the performances and the Total flow respectively. This indicates that the
of the models decrease from EP1+EP3 to the Total residuals of all models are to be considered as white noise,
category, this means that the EP1 and EP3 categories are it means that no further uni-variate time-series analysis
the main patient’s flow at the ED and the rest (EP2, EP4, could induce a better forecasting models for this cases.
EP5, EP6, EP7 and EPX) are more likely to be random
phenomenons. Epidemic correlation

Table 4. Forecasting models performance A correlation test between the residuals of the long-
term forecasting model of the EP1+EP3 category and
RMAP
the epidemic cases in the Champagne-Ardenne region of
Long-term Short-term France is performed in order to assess the robustness of
EP1 89.51% 90.20% the forecast to epidemic periods. The epidemic data used
EP3 80.60% 81.29% in this study where recovered from the Sentinelles web
EP1+EP3 91.63% 92.29% platform (source: réseau Sentinelles, INSERM/UPMC,
http://www.sentiweb.fr). We collected epidemic data of
Total 91.08% 91.84%
the Flu and the Acute diarrhea cases from 01/01/2010 to
31/12/2014. In order to plot these latter with the residuals
5.2 Residual analysis from the forecasting model we had to standardize these
variables since they have different order of magnitude.
In this subsection we are going to present a residual
analysis of the forecasting models for a better assessment Figure 10 presents a line plot of the standardized resid-
of the forecasting models. We also present a correlation uals of the long-term forecasting model of the EP1+EP3
test between the residuals of the long-term forecasting category along side with the standardized flu and acute
model of the EP1+EP3 (as it represents the majority of diarrhea epidemic cases, it shows that there is no visual
the ED’s patients flow: 95.1% of the total flow) category correlation between these latter which means that the
and the epidemic cases in the Champagne-Ardenne region epidemic cases does not affect the quality of the forecast.
of France to test the robustness of the model to epidemic
periods.
White noise test

In order to assess the quality of a forecasting model


the residuals of this latter should be tested for a white
noise null hypothesis Box and Jenkins (1971). One method
to test the white noise null hypothesis is to check the
number of lags out of the normal distribution bounds in
the ACF(auto correlation function) plot of the sample (see
Box and Jenkins (1971)). Figure 9 shows an example of this
test to the residuals of the short-term forecasting model of
Fig. 10. Line plot of standardized Residuals, Flu and Acute
the total patient’s flow, we can notice that almost all lags
diarrhea from 01/01/2014 to 31/12/2014
are within the normal distribution bounds which indicates
the presence of a white noise distribution.
6. CONCLUSION AND DISCUSSIONS
A more enhanced test for the white noise null hypoth- In this paper the authors studied the daily arrivals at
esis is the Ljung-Box test which consist of testing the the emergency department in Troyes city hospital center,

725
IFAC MIM 2016
726
June 28-30, 2016. Troyes, France Mohamed Afilal et al. / IFAC-PapersOnLine 49-12 (2016) 721–726

France. We introduced a new practical patients classifica- Cooke, M., Wilson, S., Halsall, J., and Roalfe, A. (2004).
tion that gathers categories of patients with similar behav- Total time in English accident and emergency depart-
ior or/and requires the same amount of resources in the ments is related to bed occupancy. Emergency Medicine
ED. Using time-series analysis we developed an efficient Journal : EMJ, 21(5), 575–576.
forecasting models for these newly introduced categories De Gooijer, J.G. and Hyndman, R.J. (2006). 25 years
to help the ED’s staff planning their activities in the long of time series forecasting. International Journal of
and the short-term. Forecasting, 22(3), 443–473.
Ekström, A., Kurland, L., Farrokhnia, N., Castrén, M.,
The new practical patients classification at the ED was and Nordberg, M. (2015). Forecasting Emergency De-
introduced based on statistical tests and the experience partment Visits Using Internet Data. Annals of Emer-
of the ED’s staff, it aims to group patients with similar gency Medicine, 65(4), 436–442.e1.
behavior in new categories (other than the ones used by Han, J.H., Zhou, C., France, D.J., Zhong, S., Jones, I.,
all EDs of France: CCMU and GEMSA). The resulting Storrow, A.B., and Aronsky, D. (2007). The effect of
classification named EP (stands for état patient : patient emergency department expansion on emergency depart-
status) reduced the ED’s patients categories number from ment overcrowding. Academic Emergency Medicine:
58 to 8 possible states. This classification will help ED’s Official Journal of the Society for Academic Emergency
staff to have a better understanding of their patient’s flow. Medicine, 14(4), 338–343.
Jones, S.A., Joy, M.P., and Pearson, J. (2002). Forecasting
Demand of Emergency Care. Health Care Management
Based on the data analyzed from January 2010 to De- Science, 5(4), 297–305.
cember 2013 we developed long and short-term forecasting Kadri, F., Harrou, F., Chaabane, S., and Tahon, C. (2014).
models for the most important EP categories. The test of Time Series Modelling and Forecasting of Emergency
resulted models on data from January 2014 to December Department Overcrowding. Journal of Medical Systems,
2014 showed very acceptable performances: 92.29%. Later 38(9), 1–20.
we tested the correlation between forecasting residuals Mielczarek, B. (2013). Estimating future demand for
models and epidemic data cases in order to access the hospital emergency services at the regional level. In Sim-
robustness of the models to epidemic periods, we found out ulation Conference (WSC), 2013 Winter, 2386–2397.
that the correlation between these latter is insignificant Milner, P.C. (1988). Forecasting the demand on accident
proving that the models are robust to epidemic periods. and emergency departments in health districts in the
The forecasting models we developed for our case study Trent region. Statistics in Medicine, 7(10), 1061–1072.
can be extended and used in similar EDs as they showed Mohammadi, R., Fatemi Ghomi, S.M.T., and Zeinali, F.
to be adaptable to all categories of patient’s flows. These (2014). A new hybrid evolutionary based RBF net-
latter are now being used in the ED of Troyes city hospital works method for forecasting time series: A case study
center to predict daily arrivals of the patients. of forecasting emergency supply demand time series.
REFERENCES Engineering Applications of Artificial Intelligence, 36,
204–214.
Abraham, G., Byrnes, G., and Bain, C. (2009). Short- ORU-Champagne-Ardenne (2014). Panorama Urgences
Term Forecasting of Emergency Inpatient Flow. 2013.
IEEE Transactions on Information Technology in Schull, M.J., Mamdani, M.M., and Fang, J. (2005). In-
Biomedicine, 13(3), 380–388. fluenza and emergency department utilization by elders.
Ajmi, I., Zgaya, H., and Hammadi, S. (2013). Optimized Academic Emergency Medicine: Official Journal of the
Workflow for the Healthcare Logistic: Case of the Pe- Society for Academic Emergency Medicine, 12(4), 338–
diatric Emergency Department. In M.S. Mohamad, 344.
L. Nanni, M.P. Rocha, and F. Fdez-Riverola (eds.), 7th Shi, H.Y., Tsai, J.T., Ho, W.H., and Lee, K.T. (2011).
International Conference on Practical Applications of Autoregressive integrated moving average model for
Computational Biology & Bioinformatics, number 222 long-term prediction of emergency department revenue
in Advances in Intelligent Systems and Computing, 77– and visitor volume. In 2011 International Conference on
84. Springer International Publishing. Machine Learning and Cybernetics (ICMLC), volume 3,
Bergs, J., Heerinckx, P., and Verelst, S. (2014). Knowing 979–982.
what to expect, forecasting monthly emergency depart- Stout, W.A., J. and Tawney, B. (2005). An Excel forecast-
ment visits: A time-series analysis. International Emer- ing model to aid in decision making that affects hospital
gency Nursing, 22(2), 112–115. resource/bed utilization - hospital capability to admit
Bouleux, G., Marcon, E., and Mory, O. (2014). Early emergency room patients. In 2005 IEEE Systems and
Index for Detection of Pediatric Emergency Department Information Engineering Design Symposium, 222–228.
Crowding. IEEE Journal of Biomedical and Health Sun, B.C., Hsia, R.Y., Weiss, R.E., Zingmond, D., Liang,
Informatics, PP(99), 1–1. L.J., Han, W., McCreath, H., and Asch, S.M. (2013).
Box, G.E.P. and Jenkins, G.M. (1971). Time Series Effect of emergency department crowding on outcomes
Analysis, Forecasting and Control. Journal of the Royal of admitted patients. Annals of Emergency Medicine,
Statistical Society. Series A (General), 134(3), 450. 61(6), 605–611.e6.
Boyle, J., Wallis, M., Jessup, M., Crilly, J., Lind, J., Miller, Sun, Y., Heng, B.H., Seow, Y.T., and Seow, E. (2009).
P., and Fitzgerald, G. (2008). Regression forecasting of Forecasting daily attendances at an emergency de-
patient admission data. In 30th Annual International partment to aid resource planning. BMC Emergency
Conference of the IEEE Engineering in Medicine and Medicine, 9(1), 1–9.
Biology Society, 2008. EMBS 2008, 3819–3822.

726

Vous aimerez peut-être aussi