Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
1
1
Healthcare Operation managementProfessor Xiaolan Xie
Schedule1. 13/09 Friday 13:30-16:45
2. 23/09 Monday 09:00-12h:15
3. 30/09 Monday 13:30-16:45
4. 07/10 Monday 13:30-16:45
5. 04/11 Monday 13:30-16:45
Lecture room: 158 (Session 1) + CIS Building (others)
http://www.emse.fr/~xie/DefiSanteFor info : [email protected]
2
Chapter 1. Healthcare delivery and its operation
3
Goal
present quantitative techniques from the perspective of health care organisations’ delivery of care, rather than their traditional manufacturing applications.
4
Context and trends
Rising health expenditures (17.2% in USA, 11.0% in France, OECD35 9% for 2016)
5
Context and trends
6
Context and trends
• Increasing demand due to demographic change and agingpopulation.
• Shift from offer-driven to patient-centered health care with more active role of patients in health care and better informed patients
• Growing concern of health care safety and quality -> Need of traceability of health care delivery
• Arrivals of new ICT technologies (IA, telemedicine, delivery robots, RFID, HIS, PDA, e-prescription, POS, EDI, online appointment, eVisit, …)
• Hospitals are bigger and bigger and more complex(CHU-StE 2000 beds – 7259 employes, Ruijin Hospital –2000 beds + 12000 outpatients/day, Huaxi hospital – 4000+ beds, Zhengzuo hospital 10000+ beds
2
7
French situations
• Mostly public hospitals
• Culture change on going. Words (clients, productivity, efficiency, service level, competition) better understood.
• Government responses:
• New healthcare financing: T2A, new governance, referee physician, payment by pathway, GHT (groupement hospitalier territorial)
• New health Info. Syst: carte vitale, personal health record
• Diversification of healthcare organizations to meet the diversity of healthcare demand (CHU, mid-size hospitals, clinics, home care, …)
• Better regional regulation by ARS
• Hospital responses :
• merging, reorganization,
• lean health care,
• Integrated Hospital Information Systems everywhere8
French situations
• Many obstacles to change.
• Healthcare costs hardly known,
• Increasing bigger hospitals with people used to work in isolated isles
• Lack of system thinking and spaghetti-like organization
• poor management skills and incentives of health professionals
• A labor-intensive industry facing quality human resourcesshortage
• about 10% of jobs in France and 40% in hospitals,
• feminisation and aging health professionals,
• working time regulation,
• Increasing importance of working conditions, …
9
Context and trends
The bad side
• long waiting list,
• long overtime,
• poor resourceutilization,
• ...
Wild demand fluctuation
Extra-beds at ED, 2013.07 Outpatient queue, 6h AM,11/15/2011
Poor demand-supply match Poor quality of services
10
Perspectives and evolution of French hospitals
• Montaigne report (2004),
l’hôpital de demain sera polymorphe et il n’y aura pas un modèle unique d’hôpital mais une variété d’établissements, recouvrant des organisations diversifiées, assurant des missions variées en fonction du contexte dans lequel ils se situent:
• Network organization - des établissements parties prenantes de réseaux, en liaison étroite avec la médecine de ville (GHT),
• Hospitals without wall des hôpitaux sans murs ou quasiment, pour gérer l’hospitalisation à domicile (HAD, SIAD, …),
• Highly specialized hospitals - des établissements organisés autour d’un plateau technique très spécialisé (bio, image, …),
• Others for hotelling & long term care - d’autres centrés sur l’hébergement et la dispensation de soins à des personnes âgées dépendantes (EPAHD, …),
• Others dedicated to emergency cares - des établissements privilégiant l’urgence et les soins de premiers recours.
11
Perspectives and evolution of French hospitals
Six important trends:
• Health expenditure regulation and healthcare cost transparency
• Increasing productivity by optimization of healthcare deliveryorganization and management.
• Developing better relations with upstream and downstream parties in the healthcare value chain
• Adapting the healthcare offers
• Transforming healthcare delivery and hospital organization by new medical, technological and scientific progresses (new healthcaremodes: HAD, SAD, ...)
• Increasing regional control of healthcare offers.
But also, informatization of hospitals and the importance of ICT as drivers for healthcare delivery improvement.
12
Distinctive characteristics of health care services
•Patient participation : interaction between the health care organisation and patient throughout the delivery of care
•Simultaneous production and consumption : product cannot be inspected and challenge for quality control
•Perisable capacity : operating rooms, physicians, ...
•Intangible nature of health care outputs : cannot be tested or handled before deciding on it.
•High levels of judgement and heterogeneous nature of health care: However, standardization (diagnosis and treatment process, T2A) is in process.
3
13
A Four-level model of health care system
14
A healthcare delivery system: function
Healthcare services or modes
• Medical cares
•Elective surgery
•Emergency surgery
•Day surgery
•Surgery at home
•Hospitalisation at home
•Rehabilitation
•...
Healthcare delivery system
PatientsCuredpatients
evironnementregulation, insurance,
competition
resources
The French GHT reform makes the functional design crucial
15
A healthcare delivery system: human resources
A hospital is a lot of highly skilled human resources :
Surgeons
Anaesthetists
Nurses - Infirmier Anesthésiste Diplômé d’Etat (IADE)
Nurses - Infirmier de Bloc Opératoire Diplômé d’Etat (IBODE)
Nurses - Infirmier diplômé d’état (IDE)
Caregivers - Aides soignants
Stretchers - Brancardiers
Hospital attendants - Agents de Service Hospitalier (ASH)
Also: radiologists, biologists, technicians, secretaries, ...
CHU-St Etienne = 7000+ employees16
A healthcare delivery system: material resources
• Expensive technical facilities (Plateaux Techniques Medicaux):
• Operating theatres (operating rooms, induction rooms, recovery rooms)
• ICU, NICU (Intensive Care Units)
• Imaging equipment (MRI, CT scan, X-rays, …)
• Biology laboratories
• Pharmacies
• Sterilization facilities
• Hospitalization beds
• Consultation rooms
• ...
17
A healthcare delivery system: material resources
18
A healthcare delivery system: organisation
• Wards
• Medical services or units
• Specialties (Medecine, Surgery, Obstetrics)
• Clusters of competencies
• Hospitals
• Healthcare networks
• Logistic units
• Technical centers
• Administration
Usually with a funtional organisation.
4
19
French health authorities
•Haute autorité de la santé (HAS)• Evaluation : évaluer l'utilité médicale de l'ensemble des actes, prestations et
produits de santé pris en charge par l'assurance maladie, • Certification : mettre en oeuvre la certification des établissements de santé • Best practice : promouvoir les bonnes pratiques et le bon usage des soins
auprès des professionnels de la santé et du grand public.
•Direction de l’Hospitalisation et de l’Organisation des Soins (DHOS)• National Healthcare organization (MOH) - Organisation de l’offre de soins à
la fois en ville et en établissement. • Décliner les priorités de santé publique en les traduisant en priorités pour le
secteur hospitalier.
•Agences Régionales de Santé (ARS)• Regional organisation of health : assurer un pilotage unifié de la santé en
région, de mieux répondre aux besoins de la population et d’accroître l’efficacité du système
• Agit dans le cadre d’un Projet Régional de Santé (PRS).20
French healthcare performance improvement agency
MeaH: Mission Nationale d’Expertise et d’Audits Hospitaliers
www.anap.fr (devenu ANAP)
• Objectifs :
– Faire émerger une meilleure organisation des activités hospitalières en conciliant (Best practice of hospital organisation):
• Service rendu au patient (patient service)
• Efficience économique (economic efficiency)
• Conditions de travail satisfaisantes pour le personnel (workingcondition)
• Exemples de chantier menés par la MEAH• Gestion des organisation des blocs opératoires (Operating theatre), • Circuit du médicament (drug distribution chain), • Radiothérapie, • Restauration, • Temps médical
21
Six Quality Aims for the 21st Century Health Care System
• Safe—avoiding injuries to patients from the care that is intended to help them.
• Effective—providing services based on scientific knowledge to all whocould benefit and refraining from providing services to those not likely to benefit (avoiding underuse and overuse, respectively).
• Patient-centered—providing care that is respectful of and responsive to individual patient preferences, needs, and values and ensuring thatpatient values guide all clinical decisions.
• Timely—reducing waits and sometimes harmful delays for both thosewho receive and those who give care.
• Efficient—avoiding waste, including waste of equipment, supplies, ideas, and energy.
• Equitable—providing care that does not vary in quality because of personal characteristics such as gender, ethnicity, geographic location, and socioeconomic status.
22
Six key healthcare performance dimensions OMS Europe- projet PATH
Clinical effectivenes Qualité technique, organisation et pratiques basées sur la preuve, gain en santé, résultat (individuel et global)
Patient centeredness Réactivité envers les patients: orientation du patient (rapidité de prise en charge, accès aux moyens d’aide sociale, qualité de l’accueil), satisfaction du patient
Efficiency Ressources, financière (syst. Financiers, continuité, gaspillage de ressources), taux d’encadrement, expérience (dignité, confidentialité, autonomie, communication)
safety Patient et soignants, environnement, structure, utilisation des technologies nécessaires à l’éfficience clinique
staff orientation Santé, bien être, satisfaction, développement (taux de renouvellement, emploi, absentéisme)
responsive governance Orientation vers la communauté (réponse au besoins et demandes), accessiblité, continuité des soins, promotion de la santé, équité, capacité d’adaptation à l’évolution de la demande de la population
23
Health Operation Management
General OM decision:
The planning and control of the processes that transform inputs intooutputs
Example
(Individual doctor/patient consultation. Input = a patient with a request for healthcare, output = patient diagnosed or cared or cured. Resource to be managed: their time, diagnositic or therapeutic services needed
Extensions:
Individual doctor -> individual provider (a hospital dept, a hospital, a network of hospitals, …)
Scale and scope of the resources to be planned increase and the complexity of OM
24
Health Operation Management
A meta-model of healthcare delivery system
INPUTS
PATIENT DEMAND(perceived need)
Other hospi & providers• Number• Specialty• Teaching• reputation
PURCHASERS(finances)
SUPPLIERS
TRANSFORMATIONPROCESSES
CLINICAL PROCESSES• Treatment modality• Treatment protocol• Provider-patient
encounters
MANAGEMENT PROCESSES• Infrastructure• Structure• Provider-patient
encounters
AN
CILA
RY
P
RO
CE
SS
ES
OUTPUTS
HEALTH STATUS
CLIENT PERCEPTION
USE OF RESOURCES
5
25
Health Operation Management
Health OM can be defined as the analysis, design, planning
and control of the steps necessary to provide a service for a
client.
26
Decision Layers
Strategic decisions:
Decisions having long term impacts on the hospital
Horizon: year or multiple years
Made by top management
On hospital-wide long-term vision, types of services and directions on material and human resources investments (hospital strategic plan, contrat d’objectif et de moyens, organisation de l’établissement, ...)
Major decisions:
Services: catchment areas, target groups/markets, specialities, case-mix
Investment: new hospital construction, new specialties, expansion
Partnership: shared resources, outsourcing, collaboration
Organisaton: units merging, mutualisation, polyvalence, working time regulation
27
Decision Layers
Operational decisions:
Short term decisions
Horizon: day, week or month
Made by each operational unit.
Ensure the smooth execution of the all activities
Examples:
Surgery planning/scheduling, Admission/discharging control, nurse scheduling, inventory control, supplier relation management, …
Reactive controls: emergency add-on, surgery cancelation, …
28
Decision Layers
Tactical decisions:
• Addresses the organization of the operations / execution of the health care delivery process ( i.e. the “what , where, how, when and who” ) .
• Similar to operational planning but on a longer planning horizon
Horizon: trimestre or year
Ensure the right match between available resources and activitiesof the strategic plan
Without profound changes of the structure and organisation
Examples:
• Surgery time allocation to specialties, block scheduling
• Bed allocations
• Tentering, supplier selection
29
Decision Layers
Usual confusions between
• tactical & operational layers
• tactical & strategical layers
• advance operational sched & real-time execution
Decision frameworkVissers, Bertrand, De Vries (2001). A framework for production control in health care organizations.
STRATEGIC PLANNING(range of services, long-term resource requirement, shared resources, annual patient volumes, service & efficiency levels)
patient flows resources2-5 years
PATIENT VOLUME PLANNING AND CONTROL(case-mix, rough-cut resource capacity allocation)
patient flows resources1-2 years
RESOURCE PLANNING AND CONTROL(time-phased resource allocation (specialist time, # of patients per period)
patient flows resources3 months - 1 year
PATIENT GROUP PLANNING AND CONTROL(service requirement and planning guidelines per patient group))
patient flows resourcesWeeks – 3 months
PATIENT PLANNING AND CONTROL(scheduling individual patients)
patient flows resourcesDays - weeks
restrictions restrictions
Demand-supply specialty
Demand-supply season
Demand-supply peak hour
Demand-supply match
6
31
Decision framework
Key resources
• Operating rooms
• Beds
• Diagnostic equipments
• Specialist time
• ICU
Patient groups
• By specialty (orthopaedic patients, general surgery, trauma, oncology, internal medicine, diabetics)
• By ages (services for older people, …)
• By DRG (diagnostic related groups)
• By health resources groups
• …
STRATEGIC PLANNINGpatient flowsSpecialties & production range
Patient groups as business units
ResourcesCollaboration & outsourcing
Shared resources2-5 years
PATIENT VOLUME PLANNING & CONTROL
patient flowsVolume contracts (insurance)# patients per patient group
Service levels
ResourcesRough-cut capacity checkTarget occupancy levels1-2 years
RESOURCE PLANNING & CONTROL
3 months - 1 year
PATIENT GROUPPLANNING & CONTROL
Weeks – 3 months
PATIENTPLANNING & CONTROL
Days - weeks
Restrictions on types of patients Restrictions on types of resources
patient flowsExpected patient # per group
Capacity requirement per group
ResourcesAllocation of leading resources
Batching rules for shared resources
patient flowsProject patient # per period
ResourcesAvailability of specialist capacity
patient flowsScheduling of patients for visits,
admission, exam
ResourcesAllocation of capacity to individual
patients
Restrictions on total patient vol Restrictions on amount of resources
Restrictions on detailed patient vol Restrictions on resource availability
Restrictions on timing of patient flows Restrictions on timing of resources
33
A Wider Decision framework
Hans, Van Houdenhoven, Hulshof, A Framework for Health Care Planning and Control, 2012
Four areas by Four hierarchy levels
34
A Wider Decision framework
Medical planning
• health care’s version of “technological planning”
• decision made by clinicians, medical protocols, treatments, diagnoses,
• Triage, RTD new treatments,
• The more complex and unpredictable the health care processes, the more autonomy is required for clinicians.
• Ex: acute care planned by clinicians, standard elective care planned centrally
• Ex: Benefits of early imaging test by triage nurse (prohibited by the current French law)
35
A Wider Decision framework
Resource capacity planning
• Resource dimensioning, planning, scheduling, monitoring, and control
• On renewable resources : staff + equipment and facilities (e.g. MRI s, physical therapy equipment , bed linen, sterile instruments, operating theatres, rehabilitation rooms)
36
A Wider Decision framework
Materials planning or Logistics
• Management of all consumable resources/materials, such as suture materials, prostheses, blood, bandages, food, etc
• Materials planning = the acquisition, storage, distribution and retrieval
• Typically encompasses functions like warehouse design, inventory management and purchasing.
7
37
A Wider Decision framework
Financial planning
• Management of costs and revenues to organizational strategical objectives.
• Financial planning = investment planning, contracting (with e.g. health care insurers) , budget and cost allocation, accounting, cost price calculation, and billing.
38
A Wider Decision framework
Hans, Van Houdenhoven, Hulshof, A Framework for Health Care Planning and Control, 2012
Application to a general hospital
39
Major operation management issues
Demand forecast (care types – geography – time)
Facility location & layout
Planning / scheduling
Capacity planning
Supply chain and inventory management
Quality control
Project management
40
Major operation management issues
Modeling
Tools: enterprise modelling, SADT, IDEF, Petri nets,
Performance assessment and diagnostic
Tools: simulation, queueing theory, Markov chains, statistics, Excel, ...
Design or re-engineering
Tools: simulation, optimisation, ...
Planning and control
Tools: planning, scheduling, optimisation, linear programing, heuristics, statistics,...
1
1
Chaptre 2. Facility Layout1. Introduction2. Product layout3. Process layout4. Fixed position layout5. Facility layout supplement
2
Introduction
Facility layout design is necessary when building a new facility or renovating an existing one in order to improve process flow and minimize waste space.
Facility layout largely depends on the shape and size of the building.
3
Facility layout is important
The overall layout of a facility will last for long time and only minor changes are possible after the building or renovation.
Layout has enormous effect on daily operations.
Layout dictate the distance a patient or staff member travels
Layout influences the interations and communications of the staff members.
4
Goals of Facility layout
Functionality:
• Placing heavily interacted departments together
• Placing apart departments that should not be close
• Ensuring space and form requirements
• Facilitating communication
Cost savings:
• Reduction of travel times
• Reduce overall space requirement
• Enabling for reduced staffing by placing similar job functionstogether
5
Three basic types of Facility layout
Product Layout
Process Layout
Fixed position layout
Can be used to either a single department or an entire facility.
But also:
Retail store layout
Warehousing and storage layout (relation of unloading and loading areas)
6
Types of manufacturing systems
• Job-Shop production
• Process layout, functionally similar machines are grouped
• Flow-shop production
•Product layout, machines are arranged along the manufacturing processes of a product
•Celluaire manufacturing systems
•Hybrid layout, similar parts and corresponding machines are grouped
• Project shop
•Product is fixed, personnel and equipment brought to it
• Continuous-flow process
•Chemical plants and flood industry
2
7
Organisation of hospitals
• Mainly a functional organisation in care units and technique facility centers to which are associated all human/materail resources
• each patient travels from one unit to anotheraccording to her clinical pathway defining the cares and tests needed
• But also with pools of shared material/humanresources and human resources seconded to otherunits, ...
8
Layout of an emergency department
Surgery
Radiology
E.R. triage room Emergency room admissions
Laboratories
E.R. beds Pharmacy Billing/exit
Patient A
Patient B
Patient A (broken leg) proceeds to ER triage, radiology, surgery, bed, pharmacy, billing. Patient B (pacemaker problem): ER triage, surgery, pharmacy, lab, bed, billing.
9
Chaptre 2. Facility Layout1. Introduction2. Product layout3. Process layout4. Fixed position layout
10
Product Layout
The product layout arranges equipment (departments) in the order of production process flow.
Generally used in mass production such as automobile assembly where the processes are standardized and there is little variation.
Product layout is generally less flexible and requires higher initial equipment costs.
But it minimizes the process cycle time and increases equipment utilisation.
Examples: hospital cafeteria, standardized biological tests.
11
Product Layout
12
Product Layout
Product layout, known as assembly line balancing problem, is generally determined by the product or service itself.
Most decisions concern
• assignment of basic operations to different workstations
• in order to balance the workloads
• such that each station has approximately the same cycle time, i.e. the time for one item to pass through that workstation (Why)
Two types of problems:
• Using a minimum number of workstations to achieve a given cycle time
• Minimizing the cycle time given the nb of workstations
3
13
Product Layout
Station#3Station#1
Station#2
5
46
10
21
20a
hd
5c j
6b
15g
35f
15i
8e k
16l
An example of 3 stations and cycle time 70.
Waste time = ?, workstation cycle time = ?
14
Product layout
• Since variability is inherent in health care, the product layout is rarely used in health care other than for supporting activities
• Although the health care process is similar for a patient group with similar diagnosis, the amount of time that patients spend in each process varies greatly. Line balancing is impossible.
15
Chaptre 2. Facility Layout1. Introduction2. Product layout3. Process layout4. Fixed position layout
16
Process layout
The process layout, known as layout by functionality, groups different types of process (departments, equipments) together to provide the maximum flexibility.
Hospital groups together functions such as intensive care, surgery, emergency medecine, and radiology as separate departments
The flexibility allows accomodate the variability of patient flows and times while preserving high utilisation of resources
The downside of a process layout is the large travel time, and high material handling costs. A good layout will reduce this negative impact.
17
Process layout
18
Process layout
4
19
Closeness-based method
Identify the desireness and undesireness of closeness by closeness rating chart.
Codes for desired closeness:
A – absolute necessary
E – very important
I – important
O – ordinary important
U – unimportant
X – undesirable.
20
Closeness-based method
Closeness rating charting
A
E
AO
U
IE
X
I
E
U
I
U
X
X
1. Nurses’ station
2. Ambulance entrance
3. Patient Room Area
4. Laundry
5. Main entrance
6. Dietary Department
40*80
40*40
remaining
40*80
80*80
40*80
area in m Department
21
Closeness-based method
Using a heuristic rule to design layout.
Step 1. Assign departments to available spaces according desired closeness relationships identified as absolutely necessary or undesirable, i.e. A and X, by starting with the most frequent department in either relationship.
A graph representation of A and X closeness can be built.
Step 2. Consider other departments with relationships E, I, O, U.
22
Closeness-based method
Ambulance entrance
Nurses’Station
Patientroom area
Main entrance
Laundry
DietaryDepartment
A A
X
X
X
A and X closeness representation
23
Closeness-based method
2 Ambulance entrance
1 Nurses’Station
3 Patientroom area
5 Main entrance
4 Laundry
6 DietaryDepartment
A A
X
X
X
A and X closeness representation
24
Closeness-based method
Final layout
5. mainentrance
Amb.entrance
4. L
aund
ry6.
Die
tary
Dep
t.
1 nurses’station
3. Patient Room area
400 m
200 m
5
25
Distance and cost-based method
This method tries to minimize the costs or repetitive distances traveled by patients and staff.
Data representing such traffic are represented in a from-to chartwhich represents the nb of trips or flows between departments.
Once the traffic information is identified, departments with most frequent traffics are assigned to adjacent locations.
Informations such as department space requirement, fixed locations, ... can be taken into account.
The problem is highly combinatorial when the nb of department is large and software tools such as CRAFT are necessary.
26
Distance and cost-based method
Formally, the objective is to
Minimize total cost TC = ij Dij Wij Cij
where
W =[Wij] is the From-To traffic matrix
D =[Dij] is the distance matrix (Manhattan distance or Euclideandistance, why)
C =[Cij] is the unit traffic cost
27
Distance and cost-based method
Example:
Consider a small hospital of 3 departments A, B, C. Three locations 1, 2, 3 of identical size are available.
Assume that a nurse can walk 100 feet in 30 seconds and earn $48 per hour.
Unit traffic cost Cij = 0.004$ / foot walk
location1
location2
location3
100 feet 100 feet
28
Distance and cost-based method
Distance matrix (in feet)
Location
From/To 1 2 3
1 - 100 200
2 100 - 100
3 200 100 -
Condensed traffic matrix
Department
From/To A B C
A - 3300 1400
B - 200
C -
Traffic matrix (in trips)
Department
From/To A B C
A - 1000 300
B 2300 - 100
C 1100 100 -
29
Distance and cost-based method
locationconfiguration 1 2 3 TC
1 A B C 2520$2 A C B 3280$3 B A C 2040$4 B C A 3280$5 C A B 2040$6 C B A 2520$
Dept.B
Dept.A
Dept.C
30
Distance measures
6
31
Complexity measures
32
Complexity measures
33
Distance and cost-based method : example 2
The Walters Company’s management wants to arrange the 6
departments of its hospital in a way that will minimise
interdepartmental material handling costs. They make an initial
assumption (to simplify the problem) that each department is
20x20 feet and that the building is 60 feet long and 40 feet wide.
The process layout procedure that they follow involves 6 steps.
34
Distance and cost-based method : example 2
Step 1. Construct a « from-to-matrix » showing the flow of patients or personnals from department to department (From hospitalinformation systems + future demand forecast).
50 100 0 0 20
30 50 10 0
20 0 100
50 0
0
35
Distance and cost-based method : example 2
Step 2. Determine the space requirements of each department.
Department
1
Department
2
Department
3
Department
4
Department
5
Department
6
room1 room2 room3
room4 room5 room6
Building dimensions and a possible department layout
36
Distance and cost-based method : example 2
Step 3. Develop an intial schematic diagram (qualitative RELationDiagram) showing the sequence of departments through whichpatients will have to move. Try to place departments of heavy flow next to one another.
1 2 3
45
6
50 30
100
20
50
10 20100
50
7
37
Distance and cost-based method : example 2
Step 4. Determine the cost of this layout with traffic cost of 1€ between adjacent departments and 2€ between non adjacent departments.
Cost = 570 €.
1 2 3
45
6
50 30
100
20
50
10 20100
50
38
Distance and cost-based method : example 2
Step 5. Try to improve this layout by trial and error to establish a reasonable good arrangement.
Switch departments 1-2 as there is a heavy traffic between dept. 1-3. Cost = 480 €.
Department
2
Department
1
Department
3
Department
4
Department
5
Department
6
room1 room2 room3
room4 room5 room6
39
Distance and cost-based method : example 2
Step 6. Prepare a detailed plan (Space Rel. Diagram) consideringspace or size requirements of each department; that is, arrange the departments to fit the shape of the building and obstables.
M1
M2
M3
M4
M5
M6
M1
M2
M3
M4
M5
M6
Rel. Diagram Space Rel. Diagram
40
Quadratic Assignment Approach for identical dept.
Problem: Locate N identical departments on N locations in order to minimize total traffic cost.
QAP model (Quadratic Assignment Approach)
Decision variable : Xij = 1 if dept. i is located at location j
Minimize TC = ijkl cijkl Xik Xjl
Subject to
j Xij = i Xij =1
where cijkl = ijwijdkl
ij = unit moving cost, wij = inter-department flow,
dkl = inter-location distance
41
Quadratic Assignment Approach for identical dept.
42
Quadratic Assignment Approach for identical dept.
8
43
Systematic Layout Planning (SLP)
44
Systematic Layout Planning (SLP)
45
Systematic Layout Planning (SLP)
46
Systematic Layout Planning (SLP)
47
Systematic Layout Planning (SLP)
48
Systematic Layout Planning (SLP)
9
49
Systematic Layout Planning (SLP)
50
Systematic Layout Planning (SLP)
51
Systematic Layout Planning (SLP)
52
Systematic Layout Planning (SLP)
53
Chaptre 2. Facility Layout1. Introduction2. Product layout3. Process layout4. Fixed position layout
54
Fixed position layout
The fixed-position layout consists of the fixed service positions where personnel and materials come together to perform the service.
Generally used in industry when the product is either too large or too delicate to more such as airplane assembly.
In inpatient hospital rooms (especially in an intensive care unit), the service position is the patien bed.
The operating table in an operating room is another example.
10
55
Fixed position layout
Designing a fixed position layout entails positiioning several service positions withing a given area, each of which may require an adjacent but separate support area.
Conflicts about space constraints and timing have to be resolved (suspended x-ray machine and overhead lighting)
56
Fixed position layout
57
Chaptre 2. Facility Layout1. Introduction2. Product layout3. Process layout4. Fixed position layout5. Facility layout supplement
58
Fixed position layout
Designing a fixed position layout entails positiioning several service positions withing a given area, each of which may require an adjacent but separate support area.
Conflicts about space constraints and timing have to be resolved (suspended x-ray machine and overhead lighting)
Renovation of Mercy Family Health Center
Haji Maryam, Wang Lei, Wong Yuet, Darabi Houshang
University of Illinois at Chicago Mechanical and Industrial Department
842 W. Taylor St. Chicago, IL, 60607, USA
[email protected] [email protected] [email protected] [email protected]
Abstract: The design of healthcare facilities has to provide a comfortable environment for the patient as well as acceptable waiting and processing times to serve the patient. In this paper, we redesign the layout of a real world healthcare facility (Mercy Family Health Center) to explicitly reduce the back and forth flows of the patients in the existing area and consequently reduce the patient’s waiting time. We propose two different alternatives. We use mixed integer programming (MIP) as the basis of our design. Keywords: Facility Layout, forecast Method, MIP algorithm.
1. INTRODUCTION In mid-1990’s a few high-profile medical errors brought healthcare quality and patient safety to the fore. Using engineering tools for advance studies in transportation and financial service areas and the success of these studies bring the healthcare experts starting to take notice that engineering tools have proven effective in the service sector. Thus, the uses of engineering tools-which have long proven useful in the other service industries-, seem to be a good solution for some of the health systems’ problems. Some of these problems are related to the structure of the healthcare systems. Therefore, hospital and healthcare facility design must be sensitive and responsive to the marketplace changes. “Failure to anticipate or respond to the market spells disaster” by Miller, et al. (2002). The need for flexibility is intensified by the technological nature of the healthcare industry. Healthcare facilities must adapt to changing patient populations and changing patient needs. In Janet R. Carpmans’ addressed by Miller, et al. (2002), she calls a dynamic design “a socially responsible health facility design”. The design
approach is positive responses from the users on physical, intellectual, and emotional levels. One of the most important problems that some of the healthcare systems are facing to, is the unnecessary flows of the patient that exist in the healthcare facilities which are mostly related to the inefficient layout design of the facility. In this study, we address this type of problem and by using engineering techniques such as operation research tools, we obtain a solution. Mixed integer programming (MIP) is used to model the problem. The lessons learned from this project can be applied to other real world healthcare facilities. The audience of this paper could be the healthcare administrators and operational managers who have experienced similar problems in their facilities. The organization of this paper is as follows. In section 2, we describe our case study facility, its problems and the solutions that were generated to solve these problems. Two alternatives will be developed and evaluated. In section 3, we conclude the paper.
2. A CASE STUDY
2.1 Mercy Family Health Center Mercy Family Health center is an outpatient clinic serving with compassion, accountability, respect, excellence and quality services. It is located in Chicago and is the first chartered hospital in the state of Illinois. Its vision is to serve the sick and uneducated people in need with quality healthcare regardless of their ability to pay. Its mission is to foster an environment of healing, providing access and needed care with compassion and excellence to the diverse communities it serves. The clinic provides a wide array of primary care services including: Adult Medicine, Women Health Services, Pediatric and Genetic Counseling, and Specialties (allergy, rheumatology, cardiology, etc.). 25% of the residents near the south side of Chicago who need hospitals care come to Mercy Hospital for various medical needs. 2.2 Problem and Causes Mercy Family Health Center has a typical traditional layout (Figure 1). The east side of the building consists of the administrative offices, the pharmacy, and the Pediatrics department with its own receptions’ area and waiting area. The west side of the building consists of the Continuity Medical Clinics (General Practices), OB/GYN, and Specialties Practices with one waiting area and receptions’ area in the middle. This study concentrates on the west side of the building.
Figure 1: The original layout of the Facility The centrally located waiting area is the major problem. Although the clinic is divided into various departments, patients are not classified separately in the waiting area. This causes an uncomfortable situation for the patients who are waiting to be served. Access to the examination
rooms brings a lot of flow in the area which is absolutely unnecessary. Therefore, the patient is served more slowly and the waiting time for the patients is increased. Many patients who come to Mercy for general health care carry contagious disease, i.e. fever, flu. The packed waiting area can be the best place to spread these diseases. It is highly recommended by physicians that Obstetrics patients do not sit among others with contagious diseases since they might be more vulnerable to some viruses. Currently the departments that are closely related are located far away from each other. For example, the eye examination room is set diagonally on the opposite side of the Specialties practice area, which includes Ophthalmology. An eye patient is first served at Specialties area, and then walks across the waiting area to the eye exam room, again back to the Specialties area. It unnecessarily increases the process time and the flow over the area. While other specialties patients have to wait for a longer time to be served. Some of the spaces that are related to the other parts of the hospital facility are unnecessarily located in this area which again increases the flow in the whole clinic area. The administrative personnel of the health care center are aware that old structures not only merely fail to serve the patients adequately but fail in what even the most reluctant healthcare providers have come to recognize as a medical marketplace. The most obvious product in the medical marketplace is excellence in healthcare, and a facility’s reputation for excellence is a strong incentive to healthcare consumers to select that institution over another. The most important factors that influence the potential consumers are the design of the facility and the patient amenities that the design offers. 2.3 Solutions Our first step is to maximize the space in order to enhance flexibility and the capability of handling more patients. This can be reached by removing redundant departments. It also eliminates the unnecessary flows in the clinic area. The second step is to reorganize all the departments. Each department will have its own exam rooms and waiting area. The reason for breaking the waiting area into smaller units is to reduce the patients’ flows and therefore make the clinic operations more efficient. Also, if the related areas become closer, the physicians’ travel time between units
O.B. Specialty
C.M.C I C.M.C II
Waiting area
will be shorter and this will improve the service. A modern healthcare facility is no longer a warehouse for the sick. According to the location of each practice area, separate waiting areas are created for each. Since the patients should stay in the waiting area as brief as possible, the waiting area should be comfortable and cheerful. In this layout, patients are classified and directed to the corresponding waiting area. This is especially desired for the obstetrics clinic to provide patients with privacy and direct attention. Besides decreasing the flow of patients from the waiting area to the exam rooms, this creates a small private healthcare atmosphere. Comfort level increases. Small waiting areas are more accessible for clinicians. It is as if the healthcare center was several private practices combined. Patients are more likely to choose up-to-date with stat-of-the-art clinics over the traditional clinics. The new design emphasizes on creating a friendly, non-threatening, yet functional hospital environment. The new facility layout will be de-institutionalized. According to the relationship between departments and the importance of their closeness a from-to-chart for the flow of the patient (Table 1) was generated. Departments are defined by the following numbers in the chart.
1. COPO room 2. Nursing room 3. O.B. department 4. Specialty department 5. Dummy area 6. Eye exam room 7. E.R. 8. Hallway I 9. Reception area 10. Waiting area for O.B. department 11. Bathroom 12. Waiting area for Specialty department 13. Hallway II 14. C.M.C department I 15. Waiting area for C.M.C 16. C.M.C department II
Flows can be measured quantitatively in terms of the amount of the patients moved between departments. The chart most often used to record these flows is a from-to-chart (Tompkins, et al. 2003). The from-to-chart is a square matrix which lists all departments down the row and across the column following the overall flow pattern. The numbers in table 1 show the average number of the patients between departments per day. This average is calculated regarding to
available information of the number of visitors per day in one year.
Table 1: From-to-Chart 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 0 30 240 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 30 30 0 0 0 30 0 0 30 0 30 30 0 30 3 0 0 0 0 0 0 0 719 719 719 240 0 0 0 0 0 4 0 0 0 0 0 40 0 476 476 0 159 476 0 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6 0 0 0 0 0 0 0 40 40 0 0 40 0 0 0 0 7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 8 0 0 0 0 0 0 0 0 1195 719 1195 476 0 0 0 0 9 0 0 0 0 0 0 0 0 0 719 5 476 759 0 759 0 10 0 0 0 0 0 0 0 0 0 0 180 0 0 0 0 0 11 0 0 0 0 0 0 0 0 0 0 0 120 180 0 180 0 12 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 759 0 14 0 0 0 0 0 0 0 0 0 0 0 0 0 0 759 0 15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 75916 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2.4 Alternative I The first alternative is focused on the renovation of the existing facility in terms of existing number of patients in recent year assuming that the number of the patient will not change over years (fixed number of visits in future). Based on the number of the patients who visits each department in one working day, a value is allocated to the area of their waiting area. It is assumed that each waiting area has to be close to their related department because of their strong relationships. All the operation areas that have interaction with each other should be close enough. Due to the management restrictions, some of the current areas are fixed and cannot be moved. Considering this as assumption, to an MIP model is generated (Tompkins, et al. 2003) to redesign the layout of the clinic. The objective is to minimize the patients’ movements between departments. Layout algorithms can be classified according to their objective functions. There are two basic objectives: one aims at minimizing the sum of flows times distances while the other aims at maximizing an adjacency score. Generally speaking, the former, that is, the “distance-based” objective is more suitable when the input data is expressed as a from-to-chart. In this study, we consider the distance-based objective. Mathematically, the objective function can be written as
ijij
m
i
m
jij dcfz ∑∑
= =
=1 1
min (1)
Where m denotes the number of departments,
ijf denotes the flow from department i to
department j (expressed in average number of patients moved per day between departments),
ijd is the distance from department i to j . In the MIP model distance is measured rectilinearly between the centers of department i and j , and
ijc denotes the relative weight of a unit of
patient moved from department i to department j . We assume that in this specific
problem ijc =1 which means the weights of the flows between the departments are equal. Therefore, the objective function minimizes the total flow of the patients between departments in the facility. Since it is assumed that all the departments of this problem are rectangular, a mixed integer programming (MIP) can be used for solving the addressed problems in this paper. Treating the department dimensions as decision variables, the layout problem can be formulated as follows. The decision variables and the parameters that are used in this model are: Decision variables
i α : The x-coordinate of the center of department i
i β : The y-coordinate of the center of department i
ix ′ : The x-coordinate of the left (or west) side of department i
ix ′′ : The x-coordinate of the right (or east) side of department i
iy ′ : The y-coordinate of the bottom (or south) side of department i
iy ′′ : The y-coordinate of the top (or north) side of department i
xijz =1: If department i strictly to the east of
department j (zero otherwise). yijz =1: If department i strictly to the north of
department j (zero otherwise) +
ijα : If x-coordinate of the center of department
i is in the east of department j
− ijα : If x-coordinate of the center of department
i is in the west of department j +
ijβ : If y-coordinate of the center of department
i is in the north of department j −
ijβ : If y-coordinate of the center of department
i is in the south of department j Parameters
xB : The building length (measured along x-coordinate)
yB : The building width (measured along y-coordinate) M : A large number
liL : The lower limit on the length of department
i uiL : The upper limit on the length of department
i l
iW : The lower limit on the width of department
i u
iW : The upper limit on the width of department i The above parameter and variable definitions lead to the following model:
)( −+−+ +++=∑∑ ijijijijiji j
ijcfzMin ββαα (2)
Subject to:
)(
uiii
li LxxL ≤′−′′≤ for all i (3)
)(
uiii
li WyyW ≤′−′′≤ for all i (4)
)(
uiiiii
li PyyxxP ≤′−′′+′−′′≤ for all i (5)
xii Bxx ≤′′≤′≤0 for all i (6)
yii Byy ≤′′≤′≤0 for all i (7)
iii xx ′′+′= 5.0 5.0α for all i (8)
iii yy ′′+′= 5.0 5.0β for all i (9) −+ −=− ijijji αααα for all ji ≠ (10) −+ −=− ijijji ββββ for all ji ≠ (11)
)1( xijij zMxx −+′≤′′ for all ji ≠ (12)
)1( yijij zMyy −+′≤′′ for all ji ≠ (13)
1≥+++ yji
yij
xji
xij zzzz for all ji < (14)
0, ≥ii βα for all i (15)
0 , , , ≥′′′′′′ iiii yyxx for all i (16)
0 , , , ≥−+−+ijijijij ββαα for all ji ≠ (17)
, yij
xij zz 0/1 integer for all ji ≠ (18)
In this model, we assumed 6 out of 16 departments (Hallway I, Hallway II, Nursing room, COPO, Reception area, and E.R.) are fixed and can not be moved. The objective function includes 148 variables and the constraints are more than 1000 (exactly 1088). It is also assumed that x
ijz and y
ijz variables must
be an integer and binary number. The result for the objective function value (the optimal amount of patient flow) in this model is equal to 876,806.2. The produced layout is shown in figure 2.
E.R.
eye exam rm
Reception BathroomWaiting area for Specialty
Waiting area for O.B.
Waiting area for C.M.C
C.M.C 2C.M.C 1
Hallway
Hallway
SpecialtyO.B.Nursing rm
COPO
Figure 2: New layout result for alternative I 2.5 Alternative II The second alternative is based on the number of the patients that are going to visit the Mercy Family Health Center in the future. The number of the visitors to this center in the future can be calculated based on one of the forecasting methods. Statistical studies in the previous years shows that the number of the visitors of this healthcare clinic behaves as a linear trend. Therefore based on the linear behavior of the flow, double exponential smoothing method (Holts’ Method) (Nahmias, et al. 2005) is used as a forecast method. Table 2 shows the increase in the number of the visitors in general based on the calculated forecast (Table 2).
Table 2: Total number of forecasted future patients visits
Year Demand Forecast 2000 8372 8130 2001 7914 7946 2002 6956 7860 2003 7335 7849 2004 7951 7953 2005 8058 2006 8162 2007 8267 2008 8685
The other three tables show the increases in the number of the visitors in each main department (O.B. Department (Table 3), Specialty Department (Table 4), and C.M.C department (Table 5)). Table 3: Forecasted future Patient visits of O.B.
Department
Year Demand Forecast 2000 5317 5223 2001 5027 5102 2002 4418 5029 2003 4659 5007 2004 5050 5060 2005 5116 2006 5172 2007 5228 2008 5451
Table 4: Forecasted future Patient visits of
Specialties Department
Year Demand Forecast 2000 5037 5079 2001 4762 4945 2002 4185 4846 2003 4413 4799 2004 4784 4829 2005 4868 2006 4907 2007 4947 2008 5104
Table 5: Forecasted future patient visits of
C.M.C. Department
Year Demand Forecast 2000 3335 3355 2001 3152 3267 2002 2771 3203 2003 2922 3174 2004 3167 3195 2005 3222 2006 3249 2007 3276 2008 3383
Based on the forecast numbers of future visits, new space requirements were defined for the areas of different departments and their corresponding waiting areas. All other assumptions are similar to those of Alternative I. Accordingly a new MIP model is generated and solved. The definition of the decision variables and the parameters are similar to those of the first model. The model minimizes the objective function to find the optimal solution for the total flow of the
patients moving between the depatments. Objective function for this alternative includes 144 variables and the constraints are more than 1000 (exactly 1084). The result for the objective function value in this model is equal to 725,473.5. The layout is shown in figure 3.
CO
POO.B.Waiting area for O.B.
C.M.C. 1
Wai
thin
g ar
ea fo
r Sp
ecia
lty
Nursing rm
BathroomWaiting area for C.M.C
Reception
E.R.
Hallway
Hallway
Eye exam rmSpecialtyC.M.C 2
Figure 3: New layout result for alternative II
3. CONCLUSION
In this Study, some of the existing problems in the Mercy Family Health Center are addressed (the large and disorganized waiting area, far distance between the departments that are strongly related and have strong relationships with each other, and finally utilizing the unused space). All these problems increase the flow of the patients between the departments. Therefore, process and waiting time for each patient increase. Consequently, the clinic space becomes an uncomfortable place for patients who are suffering from illness. As we know patient discontent is against the hospital beliefs and policies. To solve the problem, MIP model is used as a tool to redesign the layout of the clinic. Explicitly, we used the objective of minimizing the patient flow. In the first alternative the current flow of the visitors in the recent year is used and based on that the new layout is generated which gives the optimal solution related to the assumptions. In the second alternative, the forecast data is used and with the same assumptions as of the first alternative, another layout is generated with corresponding optimal solution for the total flow of the patient. The result of the first alternative satisfies the hospital request comparing to the budget that they have allocated to renovation of the clinic. Both alternatives explicitly minimize the total flow of the patients as well as implicitly decrease the waiting and service time. In this way the physicians are more available to give service to the patients. However the result of the first alternative satisfied the hospital request,
with a comparison of the two alternatives, we suggested them to use the second alternative instead of the first one. The reason is that the second alternative result can be used in long term and will be less as it considers the growth in the future number of patients.
4. ACKNOWLEDGEMENT We would like to thank Barbara Townsend-Vice President of Business Development, Katherine Freidl-Director of Mercy Family Health Center, and Daniel Vicencio, M.D.-Medical Director of Mercy Family Health Center, for their collaboration with this research study. We appreciate their valuable guidance and helpful suggestions.
REFERENCES Dettenkofer M., Seegers S., Antes G., Motschall E., Schumacher M., and F.D. Daschner, “Does the architecture of hospital facilities influence nosocomial infection rates? A systematic review”, Infection Control and Hospital Epidemiology, 25 (1), 21-25, Jan 2004 Douglas C.H., and M.R. Douglas, “Patient-friendly hospital environments: exploring the patients’ perspective”, Health Expectations, 7 (1), 61-73, Mar 2004. Douglas C.H., and M.R. Douglas, “Patient-centered improvements in healthcare built environments: perspectives and design indicators”, Health Expectations, 8 (3), 264-276, Sep 2005. HicK J.L., Hanfling D., Burstein J.L., DeAtley C., Barbisch D., Bogdan G.M., and S. Cantrill, “Healthcare facility and community strategies for patient care surge”, Annals of emergency Medicine, 44 (3), 253-261, Sep 2004. Mercy Hospital and medical Center, http://www.mercy-chicago.org/ Miller R.L., E.S. Swensson, “Hospital and Healthcare Facility Design”, Second Edition, 2002, W.W. Norton and Company, Inc Nahmias S., “Production and Operations Analysis”, Fifth Edition, 2005, McGraw-Hill Irwin Stevenson W.J., “Operation Management”, Eighth Edition, 2005, McGraw-Hill Irwin Tompkins J.A. White, Y.A. Bozer, and J.M.A. Tanchoco, “Facilities Planning”, Third Edition, 2003, John Wiley & Sons, Inc
1
Chapter 3Healthcare human resource
management
230 septembre 2015
Plan
Introduction
Planning/Scheduling
Resource dimensionning or staffing
330 septembre 2015
Introduction
Hospital
Authorities
Incentives for better healthcare cost control
Users
Increade healthcare demandAging population
Needs to rethink the models and organisations
Quality of care, reduced costs, improved working condition
Merging to benefit from the scale economy
An envisioned strategy
430 septembre 2015
Introduction
Plateau Médico-Technique (PMT)Center of technical facilities
EndoscopiaObstetrics
Interventional
radiology
Surgery Anaesthesia
10+ % of hospital budget
Aim of a regional research
project HRP²
Mutualisation of human resources
530 septembre 2015
Sharing human resources
Serviced’ORL
Serviced’orthopédie
Servicede chirurgie
digestive
Serviced’ORL
Unités d’hospitalisa-
tion
Salles d’intervention SSPI
Serviced’orthopédie
Unités d’hospitalisa-
tion
Bloc d’orthopédie
Salles d’intervention SSPI
Servicede chirurgie
digestive
Unités d’hospitalisa-
tion
Bloc de chirurgie digestive
Bloc d’ORL
Salles d’intervention SSPI
Salles d’intervention SSPI
Chirurgie ambulatoir
e
Services de
Chirurgie
Urgences chirurgicale
s Unités d’hospitalisatio
n
Soins intensifs
Réanimation
Plateau médico-technique
Retour au domicile
• ORL• Orthopédie• Ophtalmologie
• Urologie• Obstétrique • …
Services de
médecine
• Radiologie• Gastro-entérologie• …
Salles d’intervention SSPI
Chirurgie ambulatoir
e
Services de
Chirurgie
Urgences chirurgicale
s Unités d’hospitalisatio
n
Soins intensifs
Réanimation
Plateau médico-technique
Retour au domicile
• ORL• Orthopédie• Ophtalmologie
• Urologie• Obstétrique • …
Services de
médecine
• Radiologie• Gastro-entérologie• …
630 septembre 2015
Sharing human resources
RE-ORGANISATION
Monodisciplinary organization Integrated multidisciplinaryorganization
2
730 septembre 2015
Approach
Generation of
simulation model
Simulation with Infinite capacity
Characteristicsof PMT
of Processesof organisations
Data collection
Extrapolation
Workload profileFor each resource
Performance evaluation
Proposition of improvement
actions
Ajustment of simulationparameters
Dimensioning Human Resources (workforce per time
slot, working time, start time)
Simulation with Finite capacity
Modifying the model
Decision-aid for resource dimensioning and organization
830 septembre 2015
Objectives
Mutualisation of human resources
Design
Accompagner la conception de la nouvelle organisation:
• Dimensionner les ressources humaines
• Objectiver les choix d’organisation
Objectives
Control
Aider à la gestion des pools de personnel mutualisés
• Piloter la performance
• Aider à la planification des ressources
humaines
930 septembre 2015
Sommaire
Introduction
Planning/scheduling
Resource dimensionning or staffing
1030 septembre 2015
Dimensioning human resources
Workforce requirement
Time slots
Workload coverage
Shifts
Workload profile
Phase 1
Evaluate workforce requirement by the
workload profile
Phase 2
Determine a set of shifts covering the workload
profile
1130 septembre 2015
Phase 1: Deriving workload profilePhase 1
Evaluate workforce requirement by the
workload profile
Prepare the treatment of demand
Forecast demand arrivals
Simulate the system with infinite capacity
Modélisation des processus
Valueing the processes
Organisation des ressources
Workload profile
Modélisation des processus
Organisation des ressources
Generic model
Level of mutualisation
1230 septembre 2015
Phase 1: Deriving workload profile
TriageConsultation
llllll llllll
Exam
llllll
1st consulation
2nd consulation
Examples of process models
Emergency department
Birth deliveryStep 1. Birth delivery in an Operating Room by an obstetric physicianStep 2.1 Recovery in a ward for the womanStep 2.2 If type-2 patient, neonatal care for the newbornStep 2.3 If type-3 patient, NICU and then neonatal care for the newborn
3
1330 septembre 2015
Phase 1: Deriving workload profile
Deriving human resource requirement with a determinstic model
TriageConsultation
llllll llllll
Exam
llllll
1st consulation
2nd consulation
Emergency department1. Activities for each ED patient5 min (triage nurse)15 min (ED physician)0.5 to 2h later5 min with proba. 20% (ED physician)2. Average physician workload per ED patient15 + 5*20% min = 16 min3. Determine arrival rate8h-9h : on average 1112-13 : on average 64. Determine workload profile 8h-9h : 176 min12-13 : 96 min
5 min
15 min
5 min
80%
30 min – 2h
5. Workforce requirement8h-9h : 3 physicians12-13 : 2 physicians
Issues not captured by the simple model1. Uncertainty, 2. Queueing effect
Patients arriving 8-9 are likely to waitmuch longer and even beyond 9h
1430 septembre 2015
Organisation of the resources
Vertical polyvalence
horizontal polyvalence
… … …
Patient ofOR 1
Patient ofOR 2
Patient ofOR 3
Reception
Transfert
Induction
Intervention
Duty of a personal
1530 septembre 2015
Phase 2: Shift constructionPhase 2
Determine a set of shifts covering the
workload profile
Explicit approach
Énumération des vacations
Sélection des vacations
Multiple approachs are available [Partouche, 1998]
Enumerate all shifts
Selection of shifts
Enumeration algorithm of shift patterns
Set covering model
[Dantzig, 1954]
aij {0,1}
1630 septembre 2015
Shift pattern enumeration
Determine the set of shift patterns fulfilling all labor regulation constraints:
• min and max duration of the shift
•Earlist and latest starting time of the shift
•Duration of a break
•Time window of the break
•Number of hours before and after the break
1730 septembre 2015
Shift pattern enumeration
•Min and max duration (7-8h)
• Earlist and latest date of the start (7-11h)
•Duration of the break (1h)
•Time window of the break (11-14h)
•Number of hours before the break (2h) and after (1h)
avj 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 211 0 0 1 1 1 1 0 1 1 1 0 0 0 0 0 02 0 0 1 1 1 1 1 0 1 1 0 0 0 0 0 03 0 0 1 1 1 1 0 1 1 1 1 0 0 0 0 04 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 05 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 06 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 07 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 08 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 09 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 012 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 013 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 014 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 016 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 017 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 018 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 019 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 020 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 021 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 022 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 023 0 0 0 0 0 1 1 0 1 1 1 1 1 1 0 0
1830 septembre 2015
Set covering model
P = 100% P = 80%
Integer linear programming model
Coverage contraintsCoverage contraints
Number of employees of shift i
Cost of an employee working shift i
Mean number of employees needed for period j
Min % of the workload to cover
4
1930 septembre 2015
A hybrid approach
Observations: The workload of the personals is random Covering the mean workload does not garantee the
avoidance of:• Under-capacity due to arrivals greater than average• Over-capacity due to arrivals less than average
Interest of a hybrid approach: Evaluating the real coverage by simulation Integration of two types of costs:
• Personal cost Overtime cost
Determine the right value of P
2030 septembre 2015
Principle of the hybrid approach
Model parameters
Simulation model
Definition of the workload profile
Evaluation of the shifts
Optimization model
PerformanceModification of the
workload profile
Workload coverage optimisation model
1st P-workload profile
Set of optimal shifts
Modified P-
workload profile
1
2
3
4
2130 septembre 2015
Results
250
270
290
310
330
350
60% 65% 70% 75% 80% 85% 90% 95% 100%
P (in %)
Total cost (in K€)
Iteration n°1
Iteration n°2
Iteration n°3
Cost of the solution of the optimisation model with
P = 100 %
Cost of the hybrid approach with
P = 82,5 %
Cost saving 21%
Cleaning personal
CHU de Saint Etienne
2230 septembre 2015
Sommaire
Introduction
Planning human resources
Dimensioning human resources
2330 septembre 2015
Planning shared resourcesHuman resources
Better HR planning
Better operations PMT
Personal satisfaction
Planning Anaesthesia nursesIADE - Infirmiers Anesthésistes
ASH
IADEASIBODE
IADE
MAR
MAR
Planning Anaesthesists MAR - Médecins Anesthésistes
Planning pharmacy personals
2430 septembre 2015
<
Problems of hospital personal planning
Staff planning
Determine working days & working time
Meet contraints
Repeat the same weekly or monthly sift pattern
Easy implementation
Rigid & weak adaptabily to changes
New planning for each period
Flexible
Time consuming
Noncyclic PlanningCyclic planning
Cost Soft constraints violated Equity
[Blöchliger, 2004]
[Valouxis, 2000]
MARIADE
5
2530 septembre 2015
Planning anaesthesia nurses IADE
Day-regular (DR)
Day-urgent ( DU)
Night-urgent (NU)
Supervision recovery (SR)
8 H – 16 H
8 H – 20 H
20 H – 8 H
9 H – 17 H
Demand coverage
Working time regulation
Work on night & weekend
Succession of activities
ContraintsShifts
Assign IADE to all day and night activities of a week
Maximise the equity among employees
Shared ressources In operating rooms In recovery rooms
Both urgent and elective surgeries Work of the day and night Polyvalent personal running on all duties
CH de Valence
2630 septembre 2015
Planning anaesthesia nurses
Criteria Meet working time regulation and personal preferences
(vacations, ...) Maximise the equity.
Penality score (pénibilité or arduousness perceived by staff):
Lun Mar Mer Jeu Ven Sam DimDay regular 1 1 1 1 1Day Urgency 1,2 1,2 1,2 1,2 1,2 1,4 1,4
Ning Urgency 1,4 1,4 1,4 1,4 1,4 1,6 1,6Recovery 1,6 1,6 1,6 1,6 1,6
IADE
Minimise the total penality score deviation
max minZ P P Minimise
2730 septembre 2015
Planning anaesthesia nurses
ContraintsHard
C1: Nb IADE needed per time slot C2: working time per day less than 12h C3: weekly working time around 38h but less than 48h C4: no more than 3 nights per employee per week.
Soft C5: Saturday DU (resp. NU) implies Sunday DU (resp. NU) and no
work on Monday and Tuesday. C6: shift succession constraints to ensure at least 11h rest per day :
• NU during the week implies no working the next day • DU during the week implies NU or no working the next day (due to twice more
DU shift demand than NU)
IADE
2830 septembre 2015
Planning anaesthesia nurses
VariablesXijk = 1 if nurse i assign to shift k on day j
Contraints C1: # of nurses per shift
per day
C2: daily working time less than 12h
C3: weekly working time less than 48h
C4: no more than 3 nights a week.
IADE
7
( )7 6
3w
ij NUj w
X
11
K
ijkk
X
1
N
ijk jki
X b
7
max7 6 1
max( 48 , 1, )
w K
k ijk ij w k
i
n X T R
T h R regime
2930 septembre 2015
Planning anaesthesia nurses
Contraints
C5: Sat. DU (resp. NU) implies Sun. DU (resp. NU) and no work on Mon. and Tue.
C6 Shift succession
DU-NU followed by no working
DU followed by NU
(7 )( ) (7 )( ) (7 1) (7 2)1
1K
i w JU i w NU i w k i w kk
X X X X
IADE
(7 1) (7 ) 0i w k i w kX X
( ) ( ) ( 1)1
1K
ij JU ij NU i j kk
X X X
( ) 1 ( ) 0ij JU i j NUX X
3030 septembre 2015
Planning anaesthesia nurses
Criterion
IADE
max min
max1 1
min1 1
min
J Kk ijk
j k i
J Kk ijk
j k i
P P
p XP
R
p XP
R
6
3130 septembre 2015
Planning anaesthesia nurses IADE
JR = Day RegularJU = Day UrgencyNU = Night UrgencySS = Supervision Recovery
3230 septembre 2015
Contraintes obligatoires
Contraintes souples
Planning anaesthesists MAR MAR
Assign MAR to activities and half-day of a week
Pre-operation: Consultation
Per-operation: Anesthesia
Post-operation: Visit
Minimum demand coverage
Daily and weekly working time regulation
No working post-duty
No isolated half day working
ContraintsActivities
Minimise the number of soft constraints violated
An integer programming model
Extension of the scope: Activitiess pre, per and post operations
Assignment by half-day Need to take into account the competencies
Competency requirment of demands
Maximum demande coverage
Continuity of post-operation visits
3330 septembre 2015
Objective MAR
# MAR assignment outside their specialty
# of isolated half working day
# of post-visit continuity violated
Deviation from the maximum demand coverage
Minimise
Weighting factors
3430 septembre 2015
Experimentation: CH de Valence
Resolution of 5 problems, over 7, 14 and 28 days
By two solvers CPLEX GLPK
5 « specialties »: 4 specialist groups 1 covering all other specialties
Demands: Pre-operation: min and max Per-operation: according to the surgery planning (rule: one MAR
for 2 operating rooms) Post-operation: fixed according to the workload profile
List of duties
MAR
Weights λ1 λ2 λ3 λ4
Problem 1 1 1 1 1
Problem 2 1 2 1 1
Problem 3 2 2 1 1
Problem 4 1 1 1 2
Problem 5 1 4 1 1
3530 septembre 2015
Example of results: Problem 1
Lun Mar Mer Jeu Ven Lun Mar Mer Jeu VenSpécialité MAR am pm am pm am pm am pm am pm Spécialité MAR am pm am pm am pm am pm am pmSpécialité 1 2 per 0 0 0 0 0 0 0 0 0 0 Spécialité 5 1 pre 0 0 0 0 1 1 0 0 0 0
post 0 1 0 1 0 1 0 1 0 1 per 0 0 0 0 0 0 1 1 0 0Chirurgie 3 per 0 1 0 0 0 0 1 1 0 0 Toutes 2 pre 1 0 0 0 1 0 1 0 1 0viscérale post 0 0 0 0 0 0 0 0 0 0 spécialités per 0 0 1 0 0 0 0 0 0 0et urologie 4 per 1 0 0 0 1 1 0 0 0 0 3 pre 1 0 1 1 1 0 0 0 1 0
post 0 0 0 0 0 0 0 0 0 0 per 0 0 0 0 0 1 0 0 0 15 per 0 0 0 0 0 0 0 0 0 0 4 pre 0 1 1 1 0 0 0 0 0 0
post 0 1 0 1 0 1 0 1 0 1 per 0 0 0 0 0 0 0 0 0 1Spécialité 2 6 per 0 0 1 0 1 0 1 0 0 0 5 pre 0 0 0 0 0 0 1 0 0 0
post 0 1 0 1 0 1 0 1 0 1 per 0 0 0 0 0 0 0 0 0 0Orthopédie et 7 per 0 0 0 1 0 1 0 0 0 0 6 pre 0 0 0 0 0 0 0 0 0 0neurochirugie post 0 0 0 0 0 0 0 0 0 0 per 1 0 0 0 0 0 0 0 1 0
8 per 0 0 0 0 0 0 0 1 0 0 7 pre 0 0 0 0 0 0 0 0 0 0post 0 0 0 0 0 0 0 0 0 0 per 1 1 1 0 1 0 0 1 1 1
Spécialité 3 10 per 1 0 0 0 0 0 0 0 0 0 8 pre 0 1 0 0 0 0 0 0 0 0post 0 1 0 1 0 1 0 1 0 1 per 1 0 0 1 1 1 1 0 1 0
ORL, 11 per 0 0 0 0 0 1 0 0 0 0 9 pre 0 0 0 0 0 0 0 1 1 1Ophtalmologie post 0 0 0 0 0 0 0 0 0 0 per 1 1 1 1 0 0 1 0 0 0et chir. Ambu 12 per 0 1 0 0 1 0 0 0 0 0 10 pre 0 0 0 0 0 0 0 0 0 0
post 0 0 0 0 0 0 0 0 0 0 per 0 0 1 0 1 0 1 0 1 0Spécialité 4 4 per 0 0 0 0 0 0 0 0 1 0 11 pre 0 0 1 0 0 0 0 0 0 0
post 0 0 0 0 0 0 0 0 0 0 per 0 1 0 1 0 0 0 0 0 1Maternité 8 per 0 0 1 0 0 0 0 0 0 1 12 pre 1 0 0 0 0 1 0 0 0 0gynécologie post 0 0 0 0 0 0 0 0 0 0 per 0 0 0 0 0 0 0 0 0 0obstétrique 11 per 0 0 0 0 0 0 0 0 1 0 13 pre 0 0 0 0 0 0 1 0 0 0et pédiatrie post 0 0 0 0 0 0 0 0 0 0 per 0 0 0 0 0 0 0 0 0 0
13 per 1 0 1 0 1 0 0 0 1 0 14 pre 0 0 0 0 0 0 0 0 0 0post 0 1 0 1 0 1 0 1 0 1 per 0 0 0 0 0 0 0 1 0 0
14 per 1 1 1 1 1 1 1 0 0 0post 0 0 0 0 0 0 0 0 0 0
15 per 0 1 1 1 0 0 1 1 1 1post 0 0 0 0 0 0 0 0 0 0
Objective = 21
MAR
3630 septembre 2015
Planning pharmacy personal
Motivated by the restructuring of CH Villefranche (2 times bigger)
Need of a decision aid tool to generate pharmacy personal planning
Personals: 21 employees (4 pharmacists, 7 pharmacy assistants for preparation)
Various duties : • gestion, appro et distribution des médicaments
(armoires informatisées ou non)• Guichet• Préparation chimio• Gestion de gaz médicaux
Objectives: robust planning, reactivity to perturbations, equity between personals (rotation on all duties)
7
3730 septembre 2015
Planning pharmacy personal
A set of n tasks on a H days horizon Parameters of a task:
Task duration pi Frequency Ti Contraints of the days Date of execution ti (if fixed) Earliest Date ri Latest Date di Min delay between two executions
A set of m resources of different competencies Bij = 1 if resource j can execute task i Soft contraints: breaks, workload balancing Decisions :
Assign tasks to resources Planning the execution scheduling (day-date)
3830 septembre 2015
Planning pharmacy personal
Example of data
1
Chapter 3
Workforce scheduling
- 2 -
Plan
• Introduction
• Days-off scheduling
• Shift scheduling
• Cyclic staffing probme
• Crew scheduling
- 3 -
Intoduction
• Workforce allocation & personnel scheduling deal with the arrangement of work schedules and the assignment of personnel to shits to cover the demand for resources that vary over time.
• In service environments the operations are often prolonged and irregular, and the staff requirements fluctuate over time.
• The schedules are typically subject to various constraints dictated by equipment requirements, union rules, ...
- 4 -
Days-off scheduling
An elementary personnel assignment problem.
The problem is to find the minimum number of employees to cover a 7-days-a-week operation so that the following constraints are satisfied:
1. The demand per day, nj, j = 1, ..., 7 (Sunday to Saturday), is met
2. Each employee is given k1 out of every k2 weekends off
3. Each employee works exactly 5 out of 7 days (Sunday to Saturday)
4. Each employee works no more than 6 consecutive days.
Days j 1 2 3 4 5 6 7Sun Mon Tues Wed Thurs Fri Sat
Requirement 1 0 3 3 3 3 2 k1/k2 = 1/3
- 5 -
Days-off scheduling : lower bounds
Weekend constraint
(k2 – k1)W >= k2 max(n1, n7)
Where W is the minimum size of the workforce
Total demand constraint:
5W >= nj
Maximum daily demand constraint:
W >= max (n1, ..., n7)Days j 1 2 3 4 5 6 7
Sun Mon Tues Wed Thurs Fri SatRequirement 1 0 3 3 3 3 2
2 1 7
2 1
max ,k n nW
k k
k1/k2 = 1/3
7
15j
j
W n
- 6 -
Days-off scheduling : Algorithm
W = min workforce, n = max(n1, n7)
Step1. (Schedule the weekends off)
Assign the 1st weekend off to the first W-n employees
Assign the 2nd weekend off to the second W-n employees
This process is continued cyclically.
S S M M T W T F S S M M T W T F S S M M T W T F S1 x x x2 x x3 x x
2
- 7 -
Days-off scheduling : Algorithm
uj = W- nj, j= 2, ..., 6, uj = n – nj, j = 1, 7
Step2. (Determine the additional off-day pairs)
Construct a list of n pairs of off days, numbered 1 to n.
Choose day k such that uk = max(u1, ..., u7)
Choose day l (l k), such that ul > 0; if ul = 0 for all l k, set l = k
Add the pair (k, l) to the list and decrease uk and ul by 1.
Repeat the process n times.
Days j 1 2 3 4 5 6 7Sun Mon Tues Wed Thurs Fri Sat
uj 1 3 0 0 0 0 0
(2, 1), Sunday-Monday
(2, 2), Monday-Monday (non distinct pairs)
- 8 -
Days-off scheduling : Algorithm
Set i = 1Step3. (Categorize emplyees in week i)Type T1 : weekend i off, no days needed during week i, weekend i+1 offType T2 : weekend i off, 1 off day needed during week i, weekend i+1 onType T3 : weekend i on, 1 off day needed during week i, weekend i+1 offType T4 : weekend i on, 2 off days needed during week i, weekend i+1 on
|T3| + |T4| = n, |T2| + |T4| = n (as n people working each weekend)Pair Each employee of T2 with one of T3
Step 4 (Assign off-day pairs in week i)Assign the n pairs of days, starting from the top off the list as follows:First assign pairs of days to the employees of T4Then, to each employee of T3 and his companion of T2, assign the one of T3 the
earliest day of the pair.Set i = i+1 and return to step 3.
- 9 -
Days-off scheduling : Algorithm
S S M M T W T F S S M M T W T F S S M M T W T F S1 x x x2 x x3 x x
Week 1 : T2 = 1, T3 = 2, T4 = 3
Week 2 : T2 = 2, T3 = 3, T4 = 1
Week 3 : T2 = 3, T3 = 1, T4 = 2
S S M M T W T F S S M M T W T F S S M M T W T F S1 x x x x x x x2 x x x x x x3 x x x x x x
The schedule generated by the days-off scheduling algorithm is always feasible.
- 10 -
Shift scheduling
A cycle (one day, one or several weeks) is fixed.
Each work assignment pattern over a cycle has its own cost.
Problem:
m time intervals/periods in the predetermined cycle
bi personnel are required for period i
b different shift patterns, and each employee is assigned to one and only one pattern
(a1j, a2j, ..., amj) = shift pattern j with aij = 1 if period i is a work period.
cj = cost of patern j
Determine the number of employees of each pattern in order to minimise the total cost.
- 11 -
Shift scheduling
1 1 2 2
1 1 2 2
minimise ...subject to
... , 1,...,integer.
n n
i i in n i
i
c x c x c x
a x a x a x b i m
x
- 12 -
Shift scheduling
Pattern Hours of Work Total Hours Cost1 10 AM to 6PM 8 50 €2 1pm tp9pm 8 60 €3 12pm to 6pm 6 30 €4 10am to 1pm 3 15 €
Hours Staffing requirement10am-11am 311am-12am 412am-1pm 61pm - 2 pm 42 pm - 3pm 73pm-4pm 84pm-5pm 75pm-6pm 66pm-7pm 47pm-8pm 78pm-9pm 8
What if overtime is allowed?
3
- 13 -
Shift scheduling
• The integer programming formulation of the general personnel scheduling problem (with arbitrary 0-1 A matrix) is NP-hard
• The special case with each column containing a contiguous set of ones is easy and the solution of the LP-relaxation is always integer.
- 14 -
Cyclic staffing problem
The objective is to minimise the cost of
assigning people to an m-period cyclic schedule
so that
sufficient workers are present during time period i, in order to meet requirement bi,
and each person works a shift of k consecutive periods and is free the other m-k periods.
1 0 0 1 1 1 11 1 0 0 1 1 11 1 1 0 0 1 1
, min /1 1 1 1 0 0 11 1 1 1 1 0 00 1 1 1 1 1 00 0 1 1 1 1 1
A cX AX B
(5,7) cyclic staffing
Each column is a possible shift
- 15 -
Cyclic staffing problem : algorithm
Step 1. Solve the linear relaxation of the problem to obtain xi’
If (xi’) are integer, STOP
Step 2. Form two linear programs LP’ and LP’’ from the relaxation of the original problem by adding respectively the constraints:
1 2 1 2
1 2 1 2
... ' ' ... '
... ' ' ... ' ¨.
n n
n n
x x x x x x
and
x x x x x x
LP’’ has an optimal solution that is integer
If LP’ does not have a feaible solution, then the solution of LP’’ is the optimal solution
If LP’ has a feasible solution, then it has an optimal solution that is integer and the best of LP’ and LP’’ solutions is the optimal solution.
- 16 -
Cyclic staffing problem : algorithm
31 0 0 1 141 1 0 0 1
, 61 1 1 0 040 1 1 1 070 0 1 1 1
3.6, 4.8,5.5,3.7,5.243.3
A b
c
Optimal
- 17 -
CREW SCHEDULING
• Crew scheduling problems are very important in transportation especially in airline industry
• Consider a set of m jobs, or flight legs.
• A flight leg is characterized by a point of departure and a point of arrival, as well as an approximate time interval during which the flight has to take place.
• There is a set of n feasible and permissible combinations of flight legs that one crew can handle, round trips or tours.
• Each round trip j, has a cost cj.
• Crew schedule determines round trips to select in order to minimize the total cost under the constraint that each flight leg iscovered exactly once by one and only one round trip.
- 18 -
CREW SCHEDULING
• Each column in the A matrix is a round trip, and each row is a flight leg that must be covered exactly once by one round trip.
• Set partitioning problem.
1 1 2 2
1 1 2 2
minimise ...subject to
... , 1,...,integer.
n n
i i in n i
i
c x c x c x
a x a x a x b i m
x
4
- 19 -
CREW SCHEDULING
2 1
4
3 5
depot
52
32
3 4
2
2 12
Truck routing network
route 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
cj 8 10 4 4 2 14 10 8 8 10 11 12 6 6 5
1 0 0 0 0 1 1 1 1 0 0 0 0 0 0
0 1 0 0 0 1 0 0 0 1 1 1 0 0 0
0 0 1 0 0 0 1 0 0 1 0 0 1 1 0
0 0 0 1 0 0 0 1 0 0 1 0 1 0 1
0 0 0 0 1 0 0 0 1 0 0 1 0 1 1
1
Chapter 4Capacity planning of emergency
health services
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Queueing models of Emergency departments
• Hospital capacity planning by M/M/c models
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
ED of Ruijin Hospital
Top 1 hospital in Shanghai
+12000 outpatient visits / day
+800 emergency patients / day
An comprehensive Emergency Department (ED)
18 physicians of 11 different specialties
110 nurses
1 EICU, 3 dedicated OR, 1 observation room, X dedicated wards
Dedicated diagnostic facilities (CT, ECG, ...), pharmacy
Almost no transfers from ED to other parts of the hospital
ED organization at Ruijin Hospitalre
ssus
cita
tion
area
Exam phar
mac
y
Consultation area
Ambu
lanc
e Pay
Tria
ge
Examinations:X ray, B-mode ultason, CT, Blood test, Urine test
11 ED specialties:• Internal medecine• General surgery• Obstetrics• Neurology• Neuro-surgery• Burn
Urology• Pneumology• Ophthalmology &am
otorhinolaryngology• Dermatology• Orthopedics
ED patient flows of Ruijin HospitalEmergency Severity Index (ESI)
• 5-level triage acuity system.
• Acuity 1 : the most serious
• Acuity 5 : the least urgent and often represents office or clinic-type patients.
Frequent registration/pays
Second visit to physicians (D) after exams
f1 f2 f3 f4
f5
5% 19% 37% 22%
17%
a = 24.29 p/h, b = 0.625 p/h
Beyond ED
2
ED capacity requirement Continuity of care 24h
a day.
At least one physician in ED at anytime.
Significant fluctuation of patient arrival during a day
0
5
10
15
20
25
30
35
1 3 5 7 9 11 13 15 17 19 21 23
num
ber
of p
atie
nts
per
hour
hour of dayArrival rates from the emergency
department in Ruijin hospital
ED capacity
Current capacity plan does not take into account fluctuated demand.
Crowding during peak hours;
Low human resource utilization during low arrival periods
Question : how to better match capacity and demand by exploiing the flexibility in human resource management (Working time, organisation, polyvalence, mutualisation, …)?
Dilemma : resource utilisation vs service level
0
1
2
3
4
5
0
5
10
15
20
25
30
Arrival rate
Physician #
Question 1: How to plan ED shifts to optimize the overall service quality (waiting time)
Question 2: How many ED physicians are needed to meet waiting time targets1: less than 20 minutes for at least 80% of patients2: less than 1h for at least 95% of patients
Typical capacity mgt issues of the chapter
23
Similar issues for maternity services
Basic question: How many beds?Dilemma: occupancy vs service level?
Typical questions:
• Assuming the target occupancy level of 75%, what is the probability of delay for lack of beds for a hospital with s = 10, 20, 40, 60, 80, 100, 150, 200 beds.
• What is the size of an obsterics unit (nb of beds) necessary to achieve a probability of delayed admission not exceeding 1% while keeping the target occupancy level of 60%, 70%, 75%, 80%, 85%?
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Queueing models of Emergency departments
• Hospital capacity planning by M/M/c models
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
1210 septembre 2019
Dimensioning human resources
Workforce requirement
Time slots
Workload coverage
ShiftsWorkload profile
Phase 1
Evaluate workforce requirement by the
workload profile
Phase 2
Determine a set of shifts covering the workload
profile
3
1310 septembre 2019
Phase 1: Deriving workload profilePhase 1
Evaluate workforce requirement by the
workload profile
Prepare the treatment of demand
Forecast demand arrivals
Simulate the system with infinite capacity
Modélisation des processus
Valueing the processes
Organisation des ressources
Workload profile
Modélisation des processus
Organisation des ressources
Generic model
Level of mutualisation
1410 septembre 2019
Phase 1: Deriving workload profile
TriageConsultation
llllll llllll
Exam
llllll
1st consulation
2nd consulation
Examples of process models
Emergency department
Birth deliveryStep 1. Birth delivery in an Operating Room by an obstetric physicianStep 2.1 Recovery in a ward for the womanStep 2.2 If type-2 patient, neonatal care for the newbornStep 2.3 If type-3 patient, NICU and then neonatal care for the newborn
1510 septembre 2019
Phase 1: Deriving workload profile
Deriving human resource requirement with a determinstic model
TriageConsultation
llllll llllll
Exam
llllll
1st consulation
2nd consulation
Emergency department1. Activities for each ED patient5 min (triage nurse)15 min (ED physician)0.5 to 2h later5 min with proba. 20% (ED physician)2. Average physician workload per ED patient15 + 5*20% min = 16 min3. Determine arrival rate8h-9h : on average 1112-13 : on average 64. Determine workload profile 8h-9h : 176 min12-13 : 96 min
5 min
15 min
5 min
80%
30 min – 2h
5. Workforce requirement8h-9h : 3 physicians12-13 : 2 physicians
Issues not captured by the simple model1. Uncertainty, 2. Queueing effect
Patients arriving 8-9 are likely to waitmuch longer and even beyond 9h
1610 septembre 2019
Organisation of the resources
Vertical polyvalence
horizontal polyvalence
… … …
Patient of
OR 1Patient of
OR 2Patient of
OR 3
Reception
Transfert
Induction
Intervention
Duty of a personal
1710 septembre 2019
Phase 2: Shift constructionPhase 2
Determine a set of shifts covering the
workload profile
Explicit approach
Énumération des vacations
Sélection des vacations
Enumerate all shifts
Selection of shifts
Enumeration algorithm of shift patterns
Set covering model
[Dantzig, 1954]
aij {0,1}
1810 septembre 2019
Shift pattern enumeration
Determine the set of shift patterns fulfilling all labor regulation constraints:
• min and max duration of the shift
•Earlist and latest starting time of the shift
•Duration of a break
•Time window of the break
•Number of hours before and after the break
4
1910 septembre 2019
Shift pattern enumeration
•Min and max duration (7-8h)
• Earlist and latest date of the start (7-11h)
•Duration of the break (1h)
•Time window of the break (11-14h)
•Number of hours before the break (2h) and after (1h)
aij 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 211 0 0 1 1 1 1 0 1 1 1 0 0 0 0 0 02 0 0 1 1 1 1 1 0 1 1 0 0 0 0 0 03 0 0 1 1 1 1 0 1 1 1 1 0 0 0 0 04 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 05 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 06 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 07 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 08 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 09 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 012 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 013 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 014 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 016 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 017 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 018 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 019 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 020 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 021 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 022 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 023 0 0 0 0 0 1 1 0 1 1 1 1 1 1 0 0
2010 septembre 2019
Set covering model
P = 100% P = 80%
Integer linear programming model
Coverage contraintsCoverage contraints
Number of employees of shift i
Cost of an employee working shift i
Mean number of employees needed for period j
Min % of the workload to cover
2110 septembre 2019
A toy example
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 240,5 0,5 0,5 0,5 0,5 1 4 10 10 8 4 8 6 6 4 6 10 15 8 6 4 2 0,5 0,5
You are asked to help improving the nurse planning of an Emergency Department (ED). From the historic data, you have the following demand forecast on the number arrivals at the ED during a day.
From the statistics, 60% of the ED patients are regular patients and need only 15 minutes of the nursing care. However 40% of the ED patients are true emergency patients and require about 1h nursing care at ED before transfer to the wards.
You are asked to :
• Derive the workload profile of a typical day.• Enumerate all possible shift patterns. Shifts of 8h start either 7h-9h (20€/h),
or 15h-17h (22€/h), or 23-01h (25€/h). Shifts of 12h start either 7h-9h (21€/h) or 19h-21h (23€/h).
• Determine the optimal shifts with a coverage P = 100%, 80%, et 120%. What is the utilization ratio of the nurses
Solvable by excel solver
2210 septembre 2019
A toy example
hour arrival unit loadhourly
load1 0,5 0,55 0,2752 0,5 0,55 0,2753 0,5 0,55 0,2754 0,5 0,55 0,2755 0,5 0,55 0,2756 1 0,55 0,557 4 0,55 2,28 10 0,55 5,59 10 0,55 5,5
10 8 0,55 4,411 4 0,55 2,212 8 0,55 4,413 6 0,55 3,314 6 0,55 3,315 4 0,55 2,216 6 0,55 3,317 10 0,55 5,518 15 0,55 8,2519 8 0,55 4,420 6 0,55 3,321 4 0,55 2,222 2 0,55 1,123 0,5 0,55 0,27524 0,5 0,55 0,275
0
1
2
3
4
5
6
7
8
9
1 6 11 16 21
Workload profile
2310 septembre 2019
Observations
The workload of the personals is random
Covering the mean workload does not garantee the avoidance of:
– Under-capacity due to arrivals greater than average
– Over-capacity due to arrivals less than average
The set covering model neglects the queueing effectand is not appropriate for service level measurement
• Patients arriving in peak periods are likely to wait much longer
It neglects the patient over flow between periods
• Patients arriving in peak periods are likely to be served in later periods
Cover demand of peak periods in peak periods might lead to exagerated human resource requirement.
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Queueing models of Emergency departments
• Hospital capacity planning by M/M/c models
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
5
25
Stochastic processes
Server
Queue
N(t) : nb of customersin the queue
Customer arrival
A stochastic process {Xt, t T} : a random variable defined on the same state space E and evolving as time t goes on.
Example: the queue length N(t) at time t
26
Stochastic process
Discrete events
Continuous event
Discrete time
Continuous time
Memoryless
A CTMC is a continuous time and memoryless discrete event stochastic process.
Continuous Time Markov Chain (CTMC)
27
Homogenuous CTMC
Definition : A CTMC {X(t), t > 0} is homogeneous iff
P[X(t+s)= j X(t) = i] = P[X(t+s)= j X(t) = i] = pij(s)
Homogeneous memoryless:In reliability, we only say "a machine that does not fail at age t is as good as new"
Only homogeneous CTMC will be considered first.
28
Continuous Time Markov Chain (CTMC)
Definition : a stochastic process with discrete state space and continuous time {X(t), t > 0} is a continuous time Markov Chain (CTMC) iff
P[X(t+s)= j X(u), 0≤u≤t] = P[X(t+s)= j X(t)], t, s, j
Memoryless:In a CTMC, the past history impacts on the future evolution of the system via the current state of the system
29
Exponential distribution
T = EXP()• Probability density function (pdf) :
– fT(t) = dFT(t)/dt = e-t
• Distribution Function (cdf) : – FX(t) = P{X ≤ t} = 1 -e-t
• Mean : E[T] = 1/ • Standard deviation: T = 1/ • Coeficient of variation: cvT = / E[T] = 1
often called event rate (failure rate, repairrate, production rate, ...)
• Memoryless remaining life : P[T – t ≤ x| T ≥ t] = P[T ≤ x]
cdf
30
(Homogenuous) Poisson process
A Poisson process is a stochastic process N(t) such that• N(0) = 0• N(t) increments by +1 after a time T (called inter-arrival
time) random distributed according to an exponential distribution of parameter .
An arrival process is said Poisson if the inter-arrival times are exponentially distributed.
6
31
Key conditions for memoryless
Memoryless times
• All times (activity times, repair times, lifetimes, …) are exponentially distributed.
• X = EXP(m-1) : P(X x) = 1 – e-x/m, E[X] = m, X = m
• Memoryless remaining life : P[X – t ≤ x| X ≥ t] = P[X – t ≤ x]
Memoryless events
• All events (arrivals, machine failures, …) occur according to a POISSON process
• A POISSON event e of frequency (also called event rate)
time between occurrences of e = EXP()
• Memoryless remaining time to event at any independentobservation time = EXP()
32
A single server queue
Exponential service time at
rate
Queue
N(t) : nb of customersin the queue
Poisson customer arrival at rate
33
Markov chain representation
s1
s2
s3
s4
12Freq. of event e12
13Freq. of event e13
41Freq. of event e41
state
34
A single server queue
Exponential service time at
rate
Queue
N(t) : nb of customersin the queue
Poisson customer arrival at rate
0 1 2 3
35
Steady-state distribution
Steady-state distribution = probability distribution afterinfinite time
i = probability of being in state i in steady-state
Alternative definition (under ergodicity condition)
i = percentage of time of state i over infinite time
36
Determination of the steady-state distribution
s1
s2
s3
s4
12Probability flow
13Probability flow
41Probability flow
Probability flow of event eij=12=frequency of event eij
Flow balance equation
Total flow in = Total flow out
Holds for any state or subset of states
Normalisation equationi i = 1
41 = 12 + 13
7
37
A single server queue
0 1 2 3
Flow balance equation
Total flow in = Total flow out
Normalisation equationi i = 1
Online derivation
38
Steady state distribution of a CTMC
Conditions for existence of steady state :
1. Strongly connected Markov chain (irreducible)
2. Finite number of states or stable system (positive recurrent)
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Queueing models of Emergency departments
• Hospital capacity planning by M/M/c models
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
40
Definition of a queueing system
Customer arrivals
Departure of impatient customers
Departure of served customers
• A queueing system can be described as follows:"customers arrive for a given service, wait if the service cannot start
immediately and leave after being served"
• The term "customer" can be men, products, machines, ...
41
Notation of Kendall
Kendall notation of queueing systems
T/X/C/K/P/Z
– T: probability distribution of inter-arrival times– X: probability distribution of service times– C: Number of servers– K: Queue capacity– P: Size of the population– Z: service discipline
In this course,– K = : unlimited queue capacity– P = : infinity population– Z = FIFO: First In First Out service
42
Notation of Kendall
T/X/C– T: probability distribution of inter-arrival times– X: probability distribution of service times– C: Number of servers
•T or X can take the following values:
– M : markovian (i.e. exponential)– G : general distribution– D : deterministic
M/M/1 = Markovian arrival & service single server queueM/M/n = Markovian arrival & service n-servers queue
8
43
Little’s Laws
For any stable system,
L = TH×W(Number = Throughput Delay)
where• L : average number of customers in the system• W : average response time• TH : average throughput rate
Queueing system
LTH TH
W44
A single server queue
Exponential service time at
rate
Queue
Poisson customer arrival at rate
0 1 2 3
Stability condition :
45
M/M/1 queueStationary distribution:
n = n(1-), n≥0
where = / is called traffic intensity.
Ls = Number of customers in the system = ) = /()
Ws = Sojourn time in the system = ) = 1/()
Lq = queue length = 2/() Ls
Wq = average waiting time in the queue = /() Ws
= departure rate =
Server utilization ratio =
Server idle ratio = P0 = 1 -
P{n > k} = Probability of more than k customers = k+1
46
M/M/c queue – Erlang C system
N(t) is a birth and death process with• The birth rate .• The deadth rate is not constant and is equal to N(t) if N(t) C and C
if N(t) > C.Stability condition : < c.
N(t) : number of customers in the system
Exponentially distributed service tim
Poisson arrivals
47
11
00 ! ! 1
n cc
n
a a
n c
, 0n
n c ca
nc
Stationary probability distribution:a offered load c a/c traffic intensityn = an/n! 0, 0 < n c
M/M/c queue – Erlang C system
0 1 2 c
c
C+1
c c
48
C(c,a) = Waiting probability of an incoming customer= c + c+1 + ...
wq = random waiting time of a customer (Moment generating funct)
T = Waiting time target
(T) = Service level= P(wq ≤ T)
1
0
! 1,
1
! ! 1
c
cn cc
n
a
cC c a
a a
n c
0, with probability 1 ,
, with probability ,q
C c aw
EXP c C c a
1 , c TT C c a e
Erlang C formula
M/M/c queue – Erlang C system
9
49
M/M/c queue – Erlang C system
50
M/M/c with impatient customers –Erlang B
• Similar to M/M/C queue except the loss of customerswhich arrive when all servers are busy.
0 1 2 c
c
51
M/M/c with impatient customers –Erlang B
Steady state distribution :a offered load c traffic intensityn = an/n! 0, 0 < n c
Percentage of lost customers = C
Server utilization ratio = (1 – C) /C
Insensitivity to service time distribution:n depends on the distribution of service time T onlythrough its mean, i.e. with = E[T]
1
00 !
nC
n
a
n
52
M/M/c with impatient customers –Erlang B
0
!,
!
c
c c nn
a cB c a
a n
1 ,a B c a
Erlag loss function or Erlang B formula= Percentage of lost customers or overflow probability
Accepted load
53
Normal approximation for staffing Erlang Loss systems
Condition: high offered load (a > 4) and high targeted service level
N(t) = number of patients : approximately normally distributed
E[N(t)] a
In M/M/∞ system, N(t) =d POISSON(a), i.e. E[N(t)] = a, Var[N(t)] = a
Square-Root-Staffing-Formula for a delay probability
c a a
1N a c a
P Delay P N t c Pa a
Where is the cdf of the standard normal distribution
M/M/c with impatient customers –Erlang B
54
0
!,
!
c
c nn
a cB c a
a n
Computation issues of Erlang B and C formula
1
0
! 1,
! ! 1
c
n cc
n
a
cC c a
a a
n c
, 1 / , : the reciprocalR c a B c a
1, 1 1,R c a R c a
11, 1 1 1,C c a R c a
1, ,B c a R c a 0, 1B a
0, 1C a
!!! recursion for the same offered load !!!
a
c c
Loss proba Waiting proba
10
55
Computation issues of Erlang B and C formula
arrival rate 10service rate 3
offerred load a 3,333target wait time 0,1 Erlang C KPI Erlang B
queue length
# in system wait time syst time On time Blocking accepted
c R B C Lq Ls Wq Ws proba proba load0 1 1 1 1 0
1 3,3333 1,3000 0,7692 3,3333 - 4,76 - 1,42 - 0,47 - 0,14 - 5,71 0,7692 0,7692
2 1,6667 1,7800 0,5618 2,0833 - 5,20 - 1,87 - 0,52 - 0,18 - 2,10 0,5618 1,4607
3 1,1111 2,6020 0,3843 1,2165 - 12,16 - 8,83 - 1,21 - 0,88 - 0,34 0,3843 2,0523
4 0,8333 4,1224 0,2426 0,6577 3,2886 6,6219 0,3289 0,6622 0,4615 0,2426 2,5247
5 0,6667 7,1836 0,1392 0,3267 0,6533 3,9867 0,0653 0,3987 0,8019 0,1392 2,8693
6 0,5556 13,9305 0,0718 0,1482 0,1853 3,5186 0,0185 0,3519 0,9334 0,0718 3,0940
7 0,4762 30,2540 0,0331 0,0613 0,0557 3,3890 0,0056 0,3389 0,9796 0,0331 3,2232
8 0,4167 73,6096 0,0136 0,0231 0,0165 3,3498 0,0016 0,3350 0,9943 0,0136 3,2880
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Queueing models of Emergency departments
• Hospital capacity planning by M/M/c models
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
21
Poisson arrival
The assumption of Poisson arrivals has been shown to bereasonable for unschedulinged patients (Young 1965). Example : arrivals to Obstetrics, ICU, ED
Theoretical basis : Poisson approximation of large number of independent binomial trials
Most commonly used arrival process in modelling service systems and hown empirically to a good approximation (police, fire, EMS, bank, call centers, …
Overestimation of the waiting time for scheduled patients (surgery, …)
For outpatients, bad for scheduled ones or good for walk-ins
Young, J.P. 1965. Stabilization of Inpatient Bed Occupancy through Control of Admissions. Journalof the American Hospital Association 39: 41–48
21
Exponential distribution
The performance predicted by M/M/c model is fairly insensitveto the exponential assumption provided the coefficient of variation of the service times is close to 1.
Evidence from Kingman’s approximation LoS of obsteric patients : Av = 2.9d, Cv = 1.04 (Green &
Nguyen, 01) LoS of ICU patients : Av = 18d, Cv = 1.1 – 1.6 (Green &
Nguyen, 01). Underestimation of the congestion ED consultation: Av = 24-30min, Cv = ??? (Green et al. 2006) Outpatient: Cv = 0,35 – 0,85 (Cayirli & Veral 03) Assumptions widely used in service system operations
2 2/ / / /
2G G s M M s A T
q q
C CW W
21
Wait or abandon
M/M/c model assume patients wait till been served. Good for obsteric and ICU patients that cannot be placed off
service and so often do wait. Good also for best hospitals such as Ruijin for which patients
do wait for a bed Questionable assumption for ED services which have
significant ratio of LWBS (Left Without Been Seen)
21
Servers and service time
The servers can be : physicians, nurses, beds, operating rooms, diagnositic equiment
The service time depends on the capacity planning perspective.
Physician capacity for ED patients: Service time = consultation time (in minutes) Intra-day fluctuation of arrival is important
Bed capacity of obsteric patients: Service time = LoS (in days) Intra-day fluctuation is not important but intra-week
one is
11
21
Basic queueing models
Single queue for multiple identical servers
Drawback of the ED queue : problem with the same-doctor rule
Doct 1
Doct N
Bed 1
Bed 2
Bed N
EDobsetricICU
21
Erlang-R models
G. B. Yom-Tov, A. Mandelbaum, Erlang-R: A Time-Varying Queue with Reentrant Customers, in Support of Healthcare Staffing, MSOM, 2014Drawbacks : problem with the same-doctor rule return patients usually need different consultation time
1
2
Needy (s servers)rate
1-p
pContent (delay)rate
ArrivalPoisson t
21
Queueing network models
N. Izady, D. Worthington, Setting staffing requirements for time dependent queueing networks: The caseof accident and emergency departments, EJOR, 2012Same drawbacks as the Erlang-R model.
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Queueing models of Emergency departments
• Hospital capacity planning by M/M/c models
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
21
An introductory example
A hospital is exploring the level of staffing needed for a booth in the local mall, where they would test and provide information on the diabetes. Previous experience has shown that, on average, every 6.67 minutes a new person approaches the booth. A nurse can completetesting and answering questions, on average, in twelve minutes.
Assuming s = 2, 3, 4 nurses, a hourly cost of 40€ per nurse and a customer waiting cost of 75€ per hour waiting.
Determine the following: patient arrival rate, service rate, overallsystem utilisation, nb of patients in the system (Ls), the average queue length (Lq), average time spent in the system (Ws), average waitingtime (Wq), probability of no patient, probability of waiting C(c,a), total system costs.
66
Computation issues of Erlang B and C formula
arrival rate 9
service rate 5
offerred load a 1,800
M/M/c queue KPI
Erlang formulaqueue length
# in system wait time syst time Cost
c R B C Lq Ls Wq Ws pi_00 1 1 1
1 1,8000 1,5556 0,6429 1,8000 - 4,05 - 2,25 - 0,45 - 0,25 - 0,80 - 263,75
2 0,9000 2,7284 0,3665 0,8526 7,6737 9,4737 0,8526 1,0526 0,0526 655,5263
3 0,6000 5,5473 0,1803 0,3547 0,5321 2,3321 0,0591 0,2591 0,1460 159,9088
4 0,4500 13,3274 0,0750 0,1285 0,1052 1,9052 0,0117 0,2117 0,1616 167,8873
12
22
An introductory example
performance mesure 2 nurse 3 nurses 4 nursesPatien arrival rate 9 9 9service rate 5 5 5
Overall system utilisation 90% 60% 45%L (system) 9,47 2,33 1,91Lq 7,67 0,53 0,11w (system) - in hours 1,05 0,26 0,21Wq - in hours 0,85 0,06 0,01
no patient probability (idle) 0,05% 14,60% 16,16%patient waiting proba 85,26% 35,50% 12,85%
Total system cost € per hour 655 160 168
23
Target occupancy level of obsterics units
Obsterics is generally operated independently of otherservices, so its capacity needs can be determined separately.
Good fit of a standard M/M/s queueing model Most obsterics patients are unscheduled -> Poisson
arrivals. CV of LOS is typically very close to 1 -> exponential
service time.
How many Beds for an obstetrics unit
24
Dilemma efficiency vs service quality
« Optimal » bed occupancy level = 85% (from Green, « How many hospital beds »)Basis for bed capacity decision by government & hospitalmanagementSince obsterics patients are considered emergent, the American College of Obsterics and Gynecology (ACOG) recommends thatoccupancy levels of obsterics units not exceeding 75%. Delay targetNo standard target but Schneider suggested that the proba of delay for an obstetrics bed < 1%Schneider, D. 1981. A Methodology for the Analysis of Comparability of Services and Financial Impact of Closure of Obstetrics Services. Medical Care 19: 395–409
Bed capacity of maternity services
24
Q1 : Assuming the target occupancy level of 75%, what isthe probability of delay for lack of beds for a hospital withs = 10, 20, 40, 60, 80, 100, 150, 200 beds.
Lesson : For the same
occupancy level, the probability of delaydecreases with the size of the service.
Size matters!!!
Bed capacity of maternity services
0
0,05
0,1
0,15
0,2
0,25
0,3
0,35
0 50 100 150 200
Pdelay
Bed #
25
Q2 : What is the size of an obsterics unit (nb of beds) necessary to achieve a probability of delay not exceeding1% while keeping the target occupancy level of 60%, 70%, 75%, 80%, 85%?
Lesson : Achieving high occupancylevel while having small
probability of delay is onlypossible for obsterics unit of
large hospitals.
Capacity cut should be made with clear understanding of
the impact. Simple and naiveanalysis based on average
could lead to bad decisions.
Bed capacity of maternity services
26
Impact of seasonalityConsider an obsterics unit with 56 beds which experiences a significant degree of seasonality with occupancy level varying from a low of 68% in January to about 88% in July.
What is the probability of delay in January and in July?
If, as is likely, there are several days when actual arrivals exceed the month average by 10%, what is the probability of delay for these days in July?
Lesson : Capacity planning should not be based only on the yearly average. Extra bed capacity should be planned for predictable demand increase during
peak times.
Bed capacity and seasonality
13
27
Impact of clinical organisationConsider the possiblity of combining cardiac and thoracic surgery patients as thoracic patients are relatively few and require similar nursing skills as cardiac patients.The average arrival rate of cardiac patients is 1,91 bed requests per day and that of thoracic patients is 0,42. No additional information is available on the arrival pattern and we assume Poisson arrivals. The average LOS (Length Of Stay) is 7,7 days for cardiac patients and 3,8 days for thoracic patients.What is the number of beds for cardiac patients and thoracic patients in order to have average patient waiting time for a bed E(D) not exceeding 0,5, 1, 2, 3 days? What is the number of beds if all patients are treated in the same nursing unit?Delay in this case measures the time a patient coming out of surgery spends waiting in a recovery unit or ICU until a bed in the nursing unit is available. Long delays cause backups in operating rooms/emergency rooms, surgery cancellation and ambulance diversion.
Bed capacity reducing through merging
27
Lesson : Personal and equipment flexibility and service pooling can achieve higher
occupancy level and reduction of beds.
However, priority given to one patient group could significantly degrade the waiting time of other patients if all treated in the same nursing unit.
Bed capacity reducing through merging
27
Staffing Emergency Department under Service Level Constraints
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
0,5 0,5 0,5 0,5 0,5 1 4 10 10 8 4 8 6 6 4 6 10 15 8 6 4 2 0,5 0,5
You are asked to help improving the nurse planning of an Emergency Department (ED). From the historic data, you are able to obtain the following demand forecast on the number arrivals at the ED:
From the statistics, 60% of the ED patients are regular patients and need only 15 minutes of the nursing care. However 40% of the ED patients are true emergency patients and require about 1h nursing care at ED before transfer to the wards.
27
0
2
4
6
8
10
12
14
16
1 6 11 16 21
Patient arrivals
Staffing Emergency Department under Service Level Constraints
Goal 1: Planning min cost shifts to meet loss probability target (<5%, 1%) (Erlang B)
Preliminary goals :
• Derive the workload profile of a typical day.
• Enumerate all possible shift patterns. Shifts of 8h start either 7h-9h (20€/h), or 15h-17h (22€/h), or 23-01h (25€/h). Shifts of 12h start either 7h-9h (21€/h) or 19h-21h (23€/h).
Goal 2: Planning min cost shifts to meet waiting time targets (Erlang C)1: less than 20 minutes for at least 80% of patients2: less than 1h for at least 95% of patients
Staffing Emergency Department under Service Level Constraints
Pros :
• Simple analytical closed formula available.
• Quite robust to modelling errors
• Can be extended to network structures (ED, ICU, Wards, …)
• Adapted to time varying demand with SIPP (Stationary Independent Period by Period) or Lag SIPP of Green et al.
Cons: Not applicable when peak arrivals are significantly higher than
maximum possible capacity -> overflow
Pros and Cons
14
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Queueing models of Emergency departments
• Hospital capacity planning by M/M/c models
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
ED model
Doct 1
Doct pt
ED
0
10
20
30
40
50
60
1 3 5 7 9 11 13 15 17 19 21 23
Number of arrival patients
Nb of physicians changing according to the shift scheduling
ED model
Key variables
△ Length of each period
λt # of patients arrived in period
ut # of patients served in period
pt # of physicians in period
qt # of patients overflowed
Doct 1
Doct pt
ED
0
10
20
30
40
50
60
1 3 5 7 9 11 13 15 17 19 21 23
Number of arrival patients
Basic assumptions
A1. We simplify the ED service process as a single-stage multi-server queuing system: patients arrive, wait in a common queue, consult with a physician and then leave the ED.
A2. Patients arrive according to a Poisson process at rate λt and are served on a first come first serve (FCFS) basis.
A3. The consultation times are exponentially distributed of rate μ.
A4. The consultation time is short with respect to the period length. If the ED is not overloaded (λt<pt), steady state is reached within the period and each patient consultation begins and completed in t.
A5. No patient leaves without being served (LWBS).
A6. The ED has sufficient capacity to clear the workload at the end of the day.
Waiting time without overloaded perio
The main problem with this approach is that the stability condition might not hold for periods of peak arrival in which the limited number of physicians is not sufficient
M/M/s2,
1cW c
1
0
1
,! ! ! 1
c k cc
ck
c c c
c c k c
Use standard steady-state waiting time of stable M/M/s queues of arrival rate and c servers:
15
Basic Waiting time approximation
Rely on two simply yet efficient ideas.
Idea 1: use different waiting time estimations for patients served in a period and overflow patients.Av. waiting time of patients served in t WM/M/s(ut /Δ, pt)
Total waiting time of overflow patients to t+1 Foverflow(qt, 𝜆t)
Why ??? Inter-arrival times of last arrivals
1
2
1
11, if
2,
1,
2
t
t
qt t
t ti t toverflow
t t
tt t t
i t
q qi q
F q
q i q otherwise
Basic Waiting time approximation
Idea 2: Rather than employing sophisticated nonstationary queuing analysis, we transform this performance evaluation problem into an optimization problem by selecting ut and qt to minimize the total patient waiting time.
11
1 1 1
1
M/M/s
,...,1 min
, \{1
,
}
0
,
, ,
, ,
T
Tt t
t t t t
T
t
t
t
overflowt t t
t
t
u u
t
t
APP W
q u
q q u t T
q
u p t T
u W Fp q
t
u
T
W t T
u q
this simple idea is surprisingly efficient because the “artificial decision” variables ut and qt
closely match the relatedsimulated performance measures
Basic Waiting time approximation
11
1 1 1
1
M/M/s
,...,min
, \{1}
0
,
, ,
, ,
,
T
Tt t
t t t t
T
t t
overflowt tt t t
t
u u
t
t
W
q u
q q u t T
q
u p t T
u W F qu p
t T
W t T
u q
Key variables
△ Length of each period
λt # of patients arrived in period
ut # of patients served in period
pt # of physicians in period
qt # of patients overflowed
Improving Waiting time approximation
Drawbacks of APP1
APP1 relies on the assumption of the stationary arrival of all patients served in each period.
While this assumption is reasonable for patients arriving in the period, it is not reasonable for overflow patients.
Overflow patients are already waiting at the beginning of the period and are served consecutively by physicians
Improving Waiting time approximation Improving Waiting time approximation
1 1
11
2
t t t t
tt
u p u pW
p
16
Improving Waiting time approximation
1 1
1 M/M/s1
, ,2
t t t toverflowt
t t t t t tt
u p u puW u u W p F q
p
Basic Waiting time approximation
11
1 1 1
1
11
1 1
1 M/M/s
,...,2 min
, \{1}
0
,
min , ,
1
2
, , ,
, ,
T
Tt t
t t t t
T
t t
t t t
t t t t
tt
overflowtt t t t t
t
u u
t
APP W
q u
q q u t T
q
u p t T
u q u t T
u p u pW
p
uu u W p F q
t T
t T
u q
Goodness of the approximations
Simulation
APP1 APP2
Waiting time DevWaiting
time Dev
Day1 73.569±0.277 93.050 20.94% 80.104 8.16%
Day2 137.842±0.371 187.587 26.52% 149.687 7.91%
Day3 84.518±0.157 98.885 14.53% 86.658 2.47%
Day4 101.240±0.269 137.112 26.16% 111.346 9.08%
Day5 73.634±0.221 92.699 20.57% 78.595 6.31%
Day6 96.436±0.326 133.269 27.64% 109.196 11.69%
Day7 86.053±0.265 111.835 23.05% 95.394 9.79%
Avg. 93.327 122.062 23.54% 101.569 8.11%
Total waiting times for the actual staffing of Ruijin Hospital
Goodness of the approximationsTotal waiting times for the actual staffing of a larger ED
Simulation
APP1 APP2
Waiting time Dev Waiting time Dev
Day1 166.933±0.420 197.974 15.68% 174.926 4.57%
Day2 152.995±0.566 175.738 12.94% 155.648 1.70%
Day3 181.325±0.281 198.916 8.84% 178.834 -1.39%
Day4 109.462±0.354 131.546 16.79% 117.607 6.93%
Day5 206.368±0.729 246.464 16.27% 209.708 1.59%
Day6 163.323±0.432 197.623 17.36% 173.120 5.66%
Day7 154.806±0.539 181.155 14.55% 161.460 4.12%Avg. 162.173 189.917 14.61% 167.329 3.08%
Goodness of the approximationsNb of patients served per period
0
5
10
15
20
25
1 3 5 7 9 11 13 15 17 19 21 23
APP1
APP2
served
patients at each period
Time0
5
10
15
20
25
30
35
40
1 3 5 7 9 11 13 15 17 19 21 23
APP1
APP2
Simulation
served patien
ts at ea
ch period
Time
Goodness of the approximationsHourly patient watiting time
0
2
4
6
8
10
12
14
16
1 3 5 7 9 11 13 15 17 19 21 23
APP1
APP2
Simulation
Hourly patient waiting time
Time0
5
10
15
20
25
30
35
40
1 3 5 7 9 11 13 15 17 19 21 23
APP1
APP2
Simulation
Hou
rly pa
tien
t waiting
tim
e
Time
17
Goodness of the approximationsNumber of overflow patients
0
5
10
15
20
25
30
1 3 5 7 9 11 13 15 17 19 21 23
APP1
APP2
Simulation
Overflow patients per hour
Time0
5
10
15
20
25
30
35
40
45
1 3 5 7 9 11 13 15 17 19 21 23
APP1
APP2
Simulation
Overflow patients per hour
Time
Staffing model
A7. Each physician has only one working shift in one day, i.e., a physician’s daily working time is continuous and cannot be interrupted.
A8. The shift length is more than LBD and less than UBD but can start at any time.
A9. The total working time of all N physicians should not exceed the ED physician time budget TW
A10 (hand-shaking). For each period t with new physicians starting their shifts, there should be at least one physician working both periods t‒1 and t.
Staffing model
Decision variables:
st: number of physicians starting their shifts in period t
et: number of physicians completing their shifts at the beginning of period t
11 : min Tt tMIP W Staffing model11 : min T
t tMIP W
1
1 1 1
12
1
1
1
1
1
1
: min
1
, \{1} 2
, 3
, 4
1, 5
0, 6
, 7
, 8
, , ,
Tt t
T
t i ii
T
tt
T
tt
t t
T
t tt
t LBD
t t ii t
t UBD
t ii t
t t t
t
MIP W
p s e p
p s e p t T
s N t T
p t T
p s t T
s e t T
p s e t T
p e t T
e s
T
p t T
W
st: nb starting at t
et: nb completing at t
pt # of physicians
N: total # of servers
LBD : shift lower bound
UBD : shift upper bound
TW : physician time budget
△ Length of a period
(t) = 1 + (t mod T )
Staffing model
The model should be combined with an evaluation method.
Feasible shifts can be easily derived.
Two mathematical models MIP1 and MIP2 are obtained by combining the staffing model with the two approximation methods APP1 and APP2.
The models can be linearized.
MIP models can only be solved exactly for small-size problems.
An VSN algorithm is proposed in the paper to solve realistic size problems.
11 : min Tt tMIP W Comparison of optimal vs actual shifts11 : min T
t tMIP W
0
10
20
30
40
50
60
1 3 5 7 9 11 13 15 17 19 21 23
Number of arrival patients
0
1
2
3
4
5
6
7
8
1 3 5 7 9 11 13 15 17 19 21 23
MIP2-VNS staffingReal staffing
Nb of physicians
Actual staffing does not consider the hand shaking constraint while recognizing its necessity
18
Comparison of optimal vs actual shifts11 : min Tt tMIP W
1.6 3.0
4.9 8.4
17.6
34.7
61.3
4.0 8.2
14.0
24.0
40.3
66.7
133.6
0
20
40
60
80
100
120
140
0.8 0.9 1 1.1 1.2 1.3 1.4
VNS Staffing
Actual Staffing
Arrival rate
Mean waiting time(min.)
4.0
8.7
26.8
9.2
19.7
53.0
0
10
20
30
40
50
60
0.8 0.9 1
VNS Staffing
Actual Staffing
Mean waiting time(min.)
Arrival rate
Sensitivity wrt arrival rate
Ruijin day1 Larger H
Day4 23.4
12.0
4.9 3.2
1.7 1.1 0.7
51.5
24.8
13.8
9.7 5.0 3.5 2.3
0.0
10.0
20.0
30.0
40.0
50.0
60.0
8 9 10 11 12 13 14
VNS Staffing
Actuall Staffing
Mean waiting time(min.)
physician number
Comparison of optimal vs actual shifts11 : min Tt tMIP W
Sensitivity wrt physician number
Ruijin day1
Larger HDay4
Plan
• ED operations
• A simple shift scheduling model
• Introduction to Markov chains
• Queueing models and key results
• Hospital capacity planning by M/M/c models
• Queueing models of Emergency departments
• Physician Staffing for Emergency Departments with Time-Varying Demand
• Exact prediction of waiting time by uniformization
Uniformization
Problem of general Markov chains:
• Each Continuous Time Markov Chain ischaracterized by the transition rates ij of all possible transitions.
• The sojourn time Ti in each state i is exponentiallydistributed with rate (i) = j≠i ij, i.e. E[Ti] = 1/(i)
• Transitions different states are unpaced and asynchronuous depending on (i).
Any continuous-time Markov chain can be converted to a discrete-time chain through a process called« uniformization ».
Uniformization by example
1 0
1 0
Uniformized Markov chain by adding fictitious events
Each state has the same event rate of
Unpaced event change at rate in state 1 and in state 0
Uniformization by example
1 0
Uniformized Markov chain
Each state has the same event rate of
L = EXP(), M = EXP(),
Time in state i : Ti = min(L, M) = EXP()
Transition proba: • p10|T1 = P[L=T1|T1] = /()• p01|T2 = P[M=T2|T2] = /()
19
Uniformization by example
1 0
Governed by two independentrandom processes:
• A single Poisson event clock at rate ()
• Independent state transition
P
EXP()
If 1, real event wp
If 0, real event wp
Poisson clock
Uniformization
In order to synchronize (uniformize) the transitions at the same pace, we choose a uniformization rate
MAX{(i)}
« Uniformized » Markov chain with
•transitions occur only at instants generated by a common a Poisson process of rate (also called standard clock)
•state-transition probabilities
pij = ij /
pii = 1 - (i)/
where the self-loop transitions correspond to fictitious events.
Uniformization
S1 S2
a
b
S1 S2
a/1-a/
b/
1-b/
CTMC
DTMC by uniformization
Step1: Determine rate of the states
(S1) = a, (S2) = b
Step 2: Select an uniformization rate
≥ max{(i)}
Step 3: Add self-loop transitions to states of CTMC.
Step 4: Derive the corresponding uniformized DTMC
S1 S2
a
b
Uniformized CTMC
-a -b
Uniformization
Rates associated to states
Uniformizatin Rate = +
Uniformized ED queue of ct servers
Uniformization rate = uniformized event clock rate
• γt =λt+ct×μ
Transition probability (state = nb of patients):
...
ct×μ
λt
μ (ct-1)×μ
0 ct... ct-1
ct×μ
ct-2
2μ μ(ct-2)×μct×μ
λt λt λt λt
, 1 , 1
min , , if 1
, if 1
1 , if
0, if 1
t t
t t
iji i i i
c i j i
j ip
p p j i
j i
Uniformized ED queue of ct servers
State at k-th uniformization event
Where ki is the probability of i patients in the system at k-th uniformization event.
...
ct×μ
λt
μ (ct-1)×μ
0 ct... ct-1
ct×μ
ct-2
2μ μ(ct-2)×μct×μ
λt λt λt λt
( )q t i
1,0
1
k i kj jij
k k
p
P
Matrix form
20
Uniformized ED queue of ct servers
State at the beginning of the next period
Where N(t) = Poi(t) is the number of uniformization events in period t.
...
ct×μ
λt
μ (ct-1)×μ
0 ct... ct-1
ct×μ
ct-2
2μ μ(ct-2)×μct×μ
λt λt λt λt
( )q t i
0
0
1!
1!
t
t
n
ti niN t i
n
n
tnN t i
n
et E
n
et E
n
Matrix form
Uniformized ED queue of ct servers
Total expected waiting time in the period
Where WT is the total waiting time in period t.
...
ct×μ
λt
μ (ct-1)×μ
0 ct... ct-1
ct×μ
ct-2
2μ μ(ct-2)×μct×μ
λt λt λt λt
( )q t i
0 !
tn
t
n
eW t E WT N t n
n
Uniformized ED queue of ct servers
Poisson process : given n events in period t, the event times are order statistics of n i.i.d. rv Xi = U(0, )
( )q t i
0 0
0 0 0
0 0
1
1 !
where 1 !
t
t
n
kj tk j
nnt
kj tn k j
k t kjk j
n
tk
n k
E WT N t n j cn
eW t j c
n n
B j c
eB
n n
Uniformized ED queue of ct servers
...
ct×μ
λt
μ (ct-1)×μ
0 ct... ct-1
ct×μ
ct-2
2μ μ(ct-2)×μct×μ
λt λt λt λt
( )q t i
max max
0 0
N q
k t kjk j
W t B j c
Uniformized ED queue of ct servers ( )q t i
max
10 0
0
Algorithm A1: Probability distribution and waiting time evaluation
Step 1. Initialization : , ( ), 0, ( 1) 0
Step 2. For n=0 to Nmax,
Step 2.1: , ( 1) (
tt
q
n t njj
B e t W t t
W t B j c t t
1
11 1
1)!
Step 2.2: ,1 !
t
t
n
tn
n
tn n n n t
e
n
eP B B
n
max max max
0 0 0
1 ,!
1 !
t
t
nN N qt
n k t kjn k j
n
tk
n k
et W t B j c
n
eB
n n
Conclusions
The uniformization method has been combined with a meta-heuristic for weekly physician shift scheduling.
It has also been extended to evaluate the probability of meeting a waiting time target.
1
Chapter 5
Surgery planning
- 2 -
Agenda
• Introduction
• Operating theatre
• Surgery planning
• Open scheduling
• Bloc scheduling
• Daily Surgery scheduling & real-time control
- 3 -
Introduction
• Surgery interventions are organised around some expensive technical facilities (plateau médico-technique) : operating theatres (operating rooms, recovery rooms), imaging services, biology labs, sterilization facilities.
• The central part is the operating theatre where the efficiency in terms of Cost-Quality-Delay is a must.
• Mutation from a monospecialty with ad hoc organization to a multi-specialities with better organisation due to budget constraints and more strict safety regulations.
• Operating theatre consumes about 10% of hospital budget and constitutes a melting pot between different systems and different actors with different visions of its operation.
• The reforms of French health system imposes operating theatres to meet some criteria of efficient management that the actors are not prepared to.
- 4 -
Context
The operating theatre is the heart of a hospital:
Consultations of surgeons, anaesthesists, medical units, imaging facilities, biology labs, stretchers, sterilization, emergercy departments, …
A flexible and rigorous organization
A management of « patient flow »
- 5 -
Context
Patient arrivals
Waitinglists
Transfer
Leave the hospital
Surgery & Recovery
- 6 -
Context
Five-level planning framework(Vissers et al. 2001. A framework for production control in health care
organizations. Production planning & control, 12(6), 591-604)
o Strategic planning (2-5 years)
o Patient volumes planning & control (1-2 years)
o Resources planning & control (3 months - 1 year)
o Patient group planning & control (weeks - 3 months)
o Patient planning & control (days - weeks)
2
- 7 -
Context
Hospital: a profit maximizer or a cost minimizer
o Department
Surgeon group: surgeons, pathologies
Ward: a fixed number of beds, accommodating its patients for recovery
o Surgical center
Operating rooms: shared by all surgeon groups in a time-phased pattern
Operating block: the smallest time unit for which an operating room can be assigned to a
surgeon group for performing surgeries
o Patient: elective inpatients
Patient grouping: an iso-process grouping procedure
pathology, surgeon group, surgery duration, the length of stay (LOS), rewards, etc.
Patient case mix:
the number of patients of every patient group that can be treated annually
- 8 -
Context
Case mix planning problem
S: surgeon groupP: patient groupORs: operating rooms
How many beds areallocated to each ward?
How ORs are assigned tovarious surgeon groups?
How many patients fromeach patient group can be treated annually?
- 9 -
Context
Master Surgery Scheduling
- 10 -
Context
Master Surgery Scheduling
- 11 -
..h ..
17h 00
16h 00
15h 00
14h 00
13h 00
12h 00
11h 00
10h 00
9h 00Monday
Patient 7
OR 1
Patient 4
Patient 11
Patient 3
Patient 1
Patient 5
Patient 19
Patient 2
Patient 20
Patient 17
Patient 8
Patient 15
Patient 6
Patient 12
Tuesday Wednesday
OR 2OR 1 OR 2
Surgery planning and scheduling
Context
- 12 -
Plan
• Introduction
• Operating theatre
• Surgery planning
• Open scheduling
• Bloc scheduling
• Daily Surgery scheduling & real-time control
3
- 13 -
Operating theatre : macros processes
• Capacity Planning/Allocation consists in determining (over a semestre to 1 year):
• Number of surgery blocs per surgeon,• OR sessions of the operating rooms,• Modes of control for periods of over- or under- capacity
• Surgery Programming includes:• Determination of a preliminary surgey program,• Determination of the modes of anaesthesia,• Management of the operating rooms, equipments, materials and
ancillaries,• Assignment of personals,• Management of surgery program changes till day D-1,• Determine the « final » surgery program
• Regulation consists in managing the adjustments during the day D concerning the over-usages and the exchanges of surgery blocs, personal assignment, allocation of material resources.
- 14 -
• Registration allows to record the set of interventions performed a day D (name of the operators, durations of the interventions, natures of the incidences, … ).
• Feedback loop a deux objectifs :• réajuster la planification en fonction de l'activité constatée, des files
d'attente estimées, des capacités des unités de soins, des effectifs non médicaux disponibles, …,
• réévaluer les durées d'intervention par praticien et type d'intervention (nécessaires à la programmation)
Operating theatre : macros processes
- 15 -
Allocation programming supervision
D-7 / D-1Year NSemestre S
D
registration
D
Feedback loop
Year N+1Semestre S+1
Monthly
Mission : superviser les différentes catégories intervenants pour optimiser et sécuriser la prise en chargeFréquence : quotidienne
Mission : adaptation des processus et rappels formalisés pour le respect des processus validésFréquence : Mensuelle
Weekly
« operating theatreboard »
« Commission des utilisateurs
du bloc »Mission : adaptation des vacationsFréquence : Annuelle (voire semestriel)
« Supervision cell »
« Coordonnateur »
Missions : arbitrer les priorités médicale relatives à la prise en charge immédiate (= du jour)
« Chef du bloc »
Missions : veiller à l’application des règles, contribuer à l’évolution des règles pour améliorer l’efficience, la sécurité et les conditions de travail.
Operating theatre : macros processes
- 16 -
Operating theatre: Components
Operating theatres (mono disciplinary or shared)
ED operating theatres
Recovery rooms
Induction rooms
Endoscopia operating rooms
Obstetric operating theatre
Stretchers
Obstetric labor rooms
Interventional radiology ORs
ED
Sterilisations
Wards
Two scopes:
- 17 -
WardsOperating theatre
OR1
OR2
ORn
Recovery
ICU
Operating theatre: trajectory
- 18 -
WardOperating theatre
OR
OR
OR
ICU
Operating theatre: trajectory
Recovery
4
- 19 -
WardsOperating theatre
OR
OR
OR
Recovery
ICU
Operating theatre: trajectory
- 20 -
Operating theatre: some observations
Turn dedicated operating rooms (OR) into shared polyvalent ones improves the economic efficiency.
Induction outside the OR (in an induction room) improves the OR usage
Keep the right balance between the surgery volume target and all human and material resources.
SFAR + ASA recommendation: 1.5-2 recovery beds per OR
- 21 -
Operating theatre: performance indicators
Three types of patients concerning the surgery programming:
• Elective patients programmed at D-8 : elective surgery patients of week W are known the week W-1 (in general, before Thursday noon) to allow the validation of the surgery program of the operating roomsand the personal planning,
• Semi-urgent patients programmed at D-1: patients known the daybefore their intervention
• Patients not yet known at the beginning of D.
A good surgery program depends directly on the capacity to anticipate the demands of different operators. This anticipation requires the knowledgeof the patients to be operated at least one week before.
Health professionals agree that a ratio of 80 to 85% of patients known at D-8 is signal of efficient surgery programming.
Of course, one has to take into account emergency surgeries and the catchment area and target market of the hospital.
- 22 -
Operating Room time usage:
• Hospital OR time provision - « temps de mise à disposition » (TMD), a decision of hospital management based on the openingtimes of each room – OR sessions,
• OR timed used - « temps réel d'occupation des salles » (TROS), part of the TMD actualy used for an intervention, i.e. interval frompatient arrival in the room to end of cleaning of the room
• Conventional OR time « temps conventionnel MeaH », base OR times defined by Agency MeaH to benchmark different hospitals. For each OR :
• Conventional Day OR time of 10 hours (08h30 -18h30)• Conventional «continuity of care duty » of 14 hours (18h30 - 08h30)
Operating theatre: performance indicators
- 23 -
Ratio 1 (allocation) : Hospital OR time provision / MeaH convention68 - 90% with one at 111%.
Ratio 2 (programing & regulation) : Time used / Time programmed45% - 77%, mean 62%. National objective (CTN) : 75 - 80%.
Ratio 3 (productivity) : Timed used / MeaH convention31% - 63%.
Operating theatre: performance indicators
- 24 -
Surgery occupation time - TROS :• T1 patient preparation time : patient arrival to induction,• T2 induction time: induction to incision,• T3 surgery intervention time,• T4 duration of bandage,• T5 : cleaning.TUC = time needing a medical specialist (operator and anaesthestist).
Operating theatre: performance indicators
5
- 25 -
For each « Day » conventional surgery:
• T1 – close to mean12mn.• T2 - 3 hospital groups: < 20mn [3], mean (20mn) [4], >30mn [1]• T3 - 3 hospital groups: < 50mn [3], mean (55mn) [4], >100mn [1]• T4 - 3 hospital groups: < 11mn [4], mean (11mn) [3], 18mn [1]• T5 - 3 hospital groups: < 10mn [1], mean (14mn) [4], >14mn [3]
• Total TROS - 3 hospital groups : 1h15 - 1h30 [3], 1h30 - 2h [4], 3h [1]
• Time with medical personal TUC (T2+T3) =66% TROS• Time with surgeon T3 = 48% TROS
This explains the general feeling of « waste time » of the operators and their wish to operate in 2 ORs simultaneously.
Operating theatre: performance indicators
- 26 -
OR time provision (TMD) and overtime
Ratio of overtime is a significant capacity regulation issue.
General agreement (CTN), 2% overtime seems unavoidable
Most hospitals visited by MeaH found ways to improve their overtime ratio and hence reducing overtime cost.
Operating theatre: performance indicators
- 27 -
OR time allocated to physicians - Temps de vacation (TVO)• TVO = arrival of the first patient to departure of the last patient• TVO does not include OR closing (at the end of the day) or bio cleaning
between morning and afternoon surgery sessions.• This concept is « surgeon oriented » and corresponds to time during which
his/her activities are possible
Allocated OR time usage - Temps réel d’occupation des vacations (TROV)TROV = actual TVO used + standard bio cleaning after each intervention
Operating theatre: performance indicators
- 28 -
Analysis of OR staff occupancy (not including physicians)
Time in hospital of a personal : Temps de Présence d'un Professionnel (TPP)
TTP =
Time in OR (TTPS : Temps de Présence du Personnel en Salle)
+ waiting between interventions
+ working on other tasks (administrative, storing devices, checking materials, …).
Profession agreed goal : TTPS = 80% TTP
Operating theatre: performance indicators
- 29 -
Human resources: • Operators (surgeons) assurant l'acte opératoire (leur nombre est un
indicateur de la complexité de la planification et de la programmation),
• anaesthesia by anaesthesists and anaesthesia nurses (IADE)• Bandage, instrumentist and operation attendance by specialised
nurses of operating rooms (IBODE),• Cleaning and patient transportation by hospital attendants and
stretchers,• Team supervision, programming regulation and secretary by lead
nurse.
Personal Efficiency or productivity : • Staff time in OR or in the hospital per 100h of TROS
• Number of staff hours during T2+T3 of TROS with respect to standard mean useful time.
Operating theatre: performance indicators
- 30 -
analyse en fonction du type de programmation
analyse en fonction de la période (semaine, week-end, de chaque
jour de la semaine)
Taux d’occupation moyen sur la période observée des salles par
quart d’heureTemps Réel d’Occupation de
la Salle (= temps pris en compte dans la T2A)
Operating theatre: performance indicators
6
- 31 -
Plan
• Introduction
• Operating theatre
• Surgery planning
• Open scheduling
• Bloc scheduling
• Modified Bloc scheduling
• Surgery scheduling
- 32 -
Surgery programming models
• Surgery programming is an efficient tool of operating theatre management.
• It consists in building a « provisional » planning of surgery interventions to be realised during a period based on surgery demand forecast.
• It determines the activities of the operating theatre (surgeons, anaesthesists, nurses IADE – IBODE), and also the wards, ICUs, pharmacy, radiology, etc.
- 33 -
• Surgery program built various ways: (i) in chronologic order as information arrives, (ii) periodically by allowing rescheduling of earlier decisions.
• Open scheduling is always centralized and is simple to organise.
• Cons: dysfunction such as under-usage of resources, overtimes, and surgery cancelations if it is realised in an inappropriate way.
• Open scheduling in two phases: intervention planning and scheduling.
Surgery programming models
Open scheduling model : propose for each period (week or month) a surgery program that is completely independent of previous decisions, i.e. starts from an empty sheet.
..h ..
17h 00
16h 00
15h 00
14h 00
13h 00
12h 00
11h 00
10h 00
9h 00
Lundi
Patient 7
Salle 1
Patient 4
Patient 11
Patient 3
Patient 1
Patient 5
Patient 19
Patient 2
Patient 20
Patient 17
Patient 8
Patient 15
Patient 6
Patient 12
Mardi Mercredi
Salle 2Salle 1 Salle 2
- 34 -
• Widely used especially in North America
• Efficiency strongly depends on the design of a good surgery program pattern, in order to adapt to actual activities.
• Pro: no need of centralised information.
• Head of the operating theatre can fill the pattern with its planning and does not manage the detailed scheduling.
• Caution: difficulty of building a good surgery program pattern.
Surgery programming models
..h ..
17h 00
16h 00
15h 00
14h 00
13h 00
12h 00
11h 00
10h 00
9h 00
Lundi
Salle 1
Mardi
Salle 1
Chirurgien
Dr. Dupont
Service
orthopédie
Chirurgie
générale
Groupe chirurgical
DR. Durant
Dr. Martin
Service ORL
Block scheduling model :
relies on a fixed surgery program pattern to allocate OR-blocks (time slots of an Operating Room) to surgeon groups or specialties.
Each surgeon group places its interventions within its OR-blocks.
A weekly pattern
- 35 -
Modified Block scheduling similar to block scheduling but withadaptation mechanisms.
• The concept of OR-blocks becomes elastic with the possibility of lengthening and shortening.
• It allows greater flexibility in surgery programming.
• Based on the evolution of the surgery program, the head of the operating theatre has the possiblity to adjust the lengths of OR-blocks or to allow general access to some OR-blocks.
Surgery programming models
- 36 -
Modified Bloc scheduling
Improving the Block Scheduling model by the following:
• OR-blocks of elastic length. The length can vary according to an accepted ratio as a percentage deviation from a nominal length
• Insertion of unassigned OR-blocks in the surgery program pattern
• Possiblity of the lead nurse to adjust the length of or remove some OR blocks that are poorly filled or abandonned.
Unassigned blocks allow absorbing occasional demand surges of surgeon groups SP.
Block release time:
• This practice monitors the fill rate of OR-blocks. If it is below some threshold of economic breakeven, the OR-block is adjusted or turned unassigned.
• The adjustment occurs at block release time usually 120 to 48h before the validation of the surgery program. The lead nurse then invites SP to plan new interventions to improve the efficiency.
7
- 37 -
Modified Bloc scheduling
..h ..
17h 00
16h 00
15h 00
14h 00
13h 00
12h 00
11h 00
10h 00
9h 00Lundi
Salle 1
Mardi
Salle 1
SurgeonDr. Dupont
Serviceorthopedics
General surgery
Surgeon GroupDr. DurantDr. Dupont
Service ORL
Demand of surgeons
Drs. Dupont & Durant
Demand
orthopedics
Demand
General surgery
..h ..
17h 00
16h 00
15h 00
14h 00
13h 00
12h 00
11h 00
10h 00
9h 00Lundi
Salle 1
Mardi
Salle 1
SurgeonDr. Dupont
Serviceorthopedics
General surgery
Surgeon GroupDr. DurantDr. Dupont
Unassigned
2 mois - 120 or 48 h During 120 or 48 h
- 38 -
Surgery programming rules in practice
• Fixed hours system, aims at optimizing the resource usage. Interventions are programmed if they finish before the fixed closing time, leading to potential under-utilization and intervention postponement or cancelation.
• Any workday. It focuses on the satisfaction of patients and surgeons and allows them to choose the date of intervention, leading to potential over-utilisation of the operating rooms.
• Reasonnable time system, allows higher occupation of operating rooms with reasonable surgery intervention delays. The programming is realised by the head nurse in order to minimise the waiting times of interventions.
- 39 -
Key decision phases
• Trajectory of a surgery patient : pre-operation, per-operation, post-operation.
• During the pre-operation phase, after the surgery and anaesthesia consultations, a hospitalization date is given to the patient.
• Depending on the strategy, this date can be changeable or not.
- 40 -
Key decision phases
• During the per-operation phase,
• Decisions concerning the organization of the operating theatre are to be made: (i) assignment of each invention to an OR and a time slot, (ii) organization of relevant logistics (stretchers, recovery beds, sterilized medical devises, consumables, …).
• These decisions are under the responsibility of the head nurse of the operating theatre.
• She is in charge of
• planning the surgery activities over several weeks by partitioning the interventions in order to minimise the deviationof intervention dates;
• daily scheduling of patient flows by proposing detailedschedules for each operating room, recovery beds and relevant logistics.
- 41 -
Key decision phases
• During the post-operation phase,
• Only decisions on booking resources for cares during the stay, according to the evolution of the health condition.
• In most cases, these decisions do not need to be planned.
• Among the three phases, the per-operation phase has the greatest potential of performance improvement.
• The two others are more or less hotel-like activities.
- 42 -
Plan
• Introduction
• Operating theatre
• Surgery planning
• Open scheduling
• Bloc scheduling
• Daily Surgery scheduling & real-time control
8
- 43 -
Open scheduling
• Open scheduling consists in proposing, for each period, a surgery planning without any pre-assignment.
• To reduce complexity of open planning, most hospitals (i) pre-assign operating rooms (OR) to surgery specialties, and (ii) forces the surgeons to realise their interventions of a day in the same OR.
• Pros: more flexible adaptable to perturbations ad it does not fix a priori the dates, hours and the number of interventions.
• Cons: (i) efficiency strongly depends to the expertise of the head nurse and her ability to negotiate with surgeons; (ii) heavy workload of the head nurse for surgery program consolidation, coordination of different specialties, and decision-making of difficult situations.
• Condition: Software tools to aid surgery programming of a large operating theatre.
- 44 -
Open scheduling
• The problem arises after the surgery and anaesthesia consultation. The nurse proposes a hospitalisiation date or take note of patient’s preferences.
• Based on patient’s preferences and the availability of the surgeon, a date of intervention is proposed later on.
• Determination of surgery intervention dates should take into acount the following constraints:
Patients is admitted the day before their intervention and a bed is needed during the whole Length of Stay (LoS)
Each patient needs a time slot in an OR for her intervention
Each intervention should be performed in a given time window
Surgeons are not always available and can perform a limited number of intervention per day.
Patients might also have availability constraint.
- 45 -
Mathematical model for open scheduling
Decision variable :
Xisj = 1 if intervention i planned in room s on day j
Xisj = 0 if not.
Criterion (minimisation of deviation from most preferred date) :
1 1 1
N T S
isj isji j s
Minimise Z C X
• Cisj = number of days in advance or after patient’s preferred date if Xisj = 1.
- 46 -
Constraints:
1) Realise all interventions
1 1
1, 1,...,T S
isjj s
X i N
2) Capacity Rcapsj of OR-day (s, j)
1
N
i isj sji
Dur X Rcap
Mathematical model for open scheduling
- 47 -
3) Capacity Maxjm of Surgeon-day (m, j)
with Surgi = the set of surgeries of surgeon m.
5) Bed capacity Bcapj on day j
1
1 1 1
, 1,..., 1i
jN S
isk ji s k j LoS
X Bcap j T
1m
S
i isj jmi Surg s
Dur X Max
with LoSi = Length of Stay of patient i
Mathematical model for open scheduling
- 48 -
6) Time window [DMINi, DMAXi] of intervention i
Xisj = 0, if j < DMINi or j > DMAXi
7) Integrity constraints
0,1 , , ,isjX i s j
Mathematical model for open scheduling
9
- 49 -
Xisj = 0, if j < DMINi or j > DMAXi
0,1 , , ,isjX i s j
1
1 1 1
, 1,..., 1i
jN S
isj ji s k j LoS
X Bcap j T
1m
N S
i isj jmi Surg s
Dur X Max
1 1
1, 1,...,T S
isjj s
X i N
1
N
i ipj sji
Dur X Rcap
1 1 1
N T S
isj isji j s
Minimise Z C X
Mathematical model for open scheduling
- 50 -
Mathematical model = Integer Linear Programming
Solvable
by commercial solvers such as CPLEX, XPRESS
or
by heuristic methods
• Bin-packing-based heuristics (First fit, best fit, worst fit, decreasing fit)
• Local optimization (pairwise echange, patient relocation, OR room rescheduling).
Mathematical model for open scheduling
- 51 -
Problem : How to plan elective cases when the operating rooms capacity is shared with emergency patients
References :• M. Lamiri, X.-L. Xie, A. Dolgui and F. Grimaud (2008). "A stochastic model for
operating room planning with elective and emergency surgery demands",
European Journal of Operational Research, 185/3, 1026-1037
• Mehdi Lamiri, Xiaolan Xie and Shuguang Zhang (2008), "Column generation for
operating theatre planning with elective and emergency patients," IIE
Transactions, 40(9): 838 – 852.
• M. Lamiri, F. Grimaud, and X. Xie (2009). “Optimization methods for a stochastic
surgery planning problem,” International Journal of Production Economics,
120(2): 400-410.
Open scheduling under both elective & emergency demands
- 52 -
Plan
• Introduction
• Operating theatre
• Surgery planning
• Open scheduling
• Bloc scheduling
• Daily Surgery scheduling & real-time control
- 53 -
Bloc scheduling model
• Il assigns OR-blocks to surgeon groups (SP).
• A SP can be a surgeon or several surgeons of the same specialty or not which share the same OR-blocks.
• Block scheduling consists in building a weekly surgery program pattern for regular periods or for particular periods such as holiday seasons.
• Each SP places its interventions within its OR-blocks at their convenience.
• This model relies on knowledge of detailed demand forecast of each SP in order to allocate enough OR times decomposed into OR blocks of consistent length with respect to the SP’s intervention durations.
• The surgery program pattern, reminiscent of the master schedule of MRP in production control, is called Master Surgery Schedule.
- 54 -
Bloc scheduling
..h ..
17h 00
16h 00
15h 00
14h 00
13h 00
12h 00
11h 00
10h 00
9h 00Lundi
Salle 1
Mardi
Salle 1
Chirurgien
Dr. Dupont
Service
orthopédie
Chirurgie
générale
Groupe chirurgical
DR. Durant
Dr. Martin
Service ORL
10
- 55 -
Bloc scheduling
• Pros: simplicity of implementation.
• Cons:
• Extreme difficulty for building a good MSS as over-dimensionning OR-blocks leads to low productivity and under-dimensionning OR-blocks leads to tensions among surgeons and intervention delay/cancelation.
• Rigid surgery programming imposing surgeons a constant and regular activity volume at fixed dates.
- 56 -
Building a cyclic master surgery schedule
Problem:
Ensure that each surgeon (or surgeon group) obtains a specific nb of OR blocks
With surgery-dependent OR capacity requirement and stochastic LoS
In order to minimize the expected bed shortage
o Source : Jeroen Belien é Erik Demeulemeester, EJOR, 2007
- 57 -
Building a cyclic master surgery schedule
Problem setting:
Data:
LoS distribution / surgery type (To be defined)
bi : nb of OR blocks available on day i
rs : nb of OR blocks required by surgeon s
Decision :
xis = nb of OR blocks of each surgeon s on each day i of the planning cycle (weekly or bi-weekly)
- 58 -
Building a cyclic master surgery schedule
- 59 -
Building a cyclic master surgery schedule
LoS (Length of Stay): the whole period in the hospital of each patient during which a bed is needed throughout.
Deterministic LoS :
Asd = set of hospitalization days if operated on day d
Stochastic LoS :
psd = probability of staying d days after surgery by surgeon s
Ex : appendicitis surgery: (2, 20%), (3, 50%), (4, 30%)
LOS Mon Tue Wed Thu Fri Sat Sun Mon Tue Wed Thu Fri Sat Sun
3 S
3 S
3 S
10 S
10 S
hospitalization days
- 60 -
Building a cyclic master surgery schedule
11
- 61 -
Building a cyclic master surgery schedule
Solution
Analytical expressions of mean, variance, percentile
Linearization of the non-linear model
Repetitive MIP heuristic, Quadratic MIP, Simulated annealing
- 62 -
Extension to Case mix planning
To advance the resource efficiency
o Approach Matching the patient demand and the resource provision
Coordinating the resource allocation within the whole hospital
To maximize the overall financial contribution of given resources
o Long-term decision making Patient case mix that can be treated annually
Resource capacity allocation to each specialty
o Solution method Integer linear program modeling
Branch-and-price algorithms (Ma et Demeulemeester 2013)
- 63 -
Case mix planning
Major demand data
Demand range per patient group
Resource requirement per group-p patient
• OR capacity: durp
• Bed requirement
LoSp represented by
Apd = set of hospitalization days if operated on day d
LOS Mon Tue Wed Thu Fri Sat Sun Mon Tue Wed Thu Fri Sat Sun
3 S
3 S
3 S
10 S
10 S
hospitalization days
- 64 -
Case mix planning
• Integer linear programming (ILP) model
o the number of patients of group p that receive surgery on day a
o the number of beds allocated to ward w
o the number of blocks assigned to surgeon group s on day a
o Cycle: one week
. .
, ,
,
, ,
,
, , , , , ,
w p d
s
ww W
p a wp P a A
s as S
p p a s ap P
p p a pa A
p a w s a
s t y B E D S
x y w W d D
z B L O C K S a A
d u r x z L E N G T H s S a A
L B x U B p P
x y z w W s S p P a A
max p pap P a A
r x
bed capacity
integer
admission range
OR blocks
:pax
:wy
:saz
Objective: to maximize the overall financial contribution
Constraint: resource capacity
- 65 -
Case mix planning
To improve the resource utilization
o Variability: e.g., the variable LOS the bed shortage phenomenon
o Approach Adjusting the resource capacity allocation
Building balanced master surgery schedules
To coordinate the capacity utilization of various resources at each specialty
o Decision-making Long-term: the bed capacity allocation
Medium-term: (seasonal) balanced master surgery schedules
o Solution method Mixed integer programming models
- 66 -
Master Surgery Scheduling
• Master Surgery Scheduling model
o the number of patients of group p that receive surgery on day a
o the number of beds allocated to ward w
o the number of blocks assigned to surgeon group s on day a
o Cycle: one week
. .
,
, ,
,
, , , , , ,
s
ww W
s as S
p p a s ap P
p a pa A
p a w s a
s t y B E D S
z B L O C K S a A
d u r x z L E N G T H s S a A
x T H R p P
x y z w W s S p P a A
min ww W
TEBS
b e d c a p a c i ty
integer
Planned case mix
OR blocks
:pax
:wy
:saz
Objective: to minimize the Total Expected Bed Shortage (TEBS)
12
- 67 -
Master Surgery Scheduling
,0
2, ,
0
2 2
,
,
(1 )
. .
,
, , ,
, , , ,
w
w
ww W
wd pa p kT d ap P a A k
wd pa p kT d a p kT d ap P a A k
wd w wd wd
wd w wd w
y BEDS
x prob
x prob prob
y w
s
W d
w W d
t
D
D
2min w wd w w w ww W d D
Building a balancedcyclic MSS:
bed occupancy
on each ward-day
OR assignments
patient admission
integer/nonnegative
, , ,
, , ,s
sark k ars S k K
p pa sark kp P r R k K
z Length ORday a A r R
dur x z Length s S a A
, ,
, , 0,1 , 0, , , , , , , .
pa pa A
pa w sark wd
x THR p P
x y z p s w a d r k
daily bed deficit
bed surplus bed demands variance
Approach: balancing the daily bed occupancy of each wardto minimize the total expected bed shortages (TEBSs)
- 68 -
Plan
• Introduction
• Operating theatre
• Surgery planning
• Open scheduling
• Bloc scheduling
• Daily Surgery scheduling & real-time control
- 69 -
Daily operation management of an operating theatre
1. Schedule the most critical resources: operating rooms OR (daily surgery scheduling) .
2. Schedule the working time of all relevant surgery team members
3. Schedule the logistics of patients and the working time of stretchers (brancardiers)
4. Schedule the logistics of all consumables and medical devices
- 70 -
Daily surgery scheduling
Assignment and sequencing of
surgery interventions of the day
To
ORs
- 71 -
A mathematical model
i=1
1
Variable:
1, if patient i assigned to OR m
0, if not.
Minimise Cmax
subject to:
Cmax ,
1,
0,1 .
im
N
i im
M
imm
im
X
Dur X m
X i
X
- 72 -
Bin Packing Equivalence
Daily surgery scheduling Bin Packing
Surgery interventions objects
Surgery duration Size of an object
ORs or OR-blocks Bin
OR session or block length Bin capacity
Nb of ORs to open Nb of Bins to use
Placing an intervention in an OR Placing an object in a bin
Closing time of the operating theatre Maximal bin capacity
Equivalence also with the parallel machine scheduling
13
- 73 -
Dynamic Bin Packing Rules
First Fit
• Insert the next object in the first bin (without exceeding the max bin capacity if given)
Best Fit
• Insert the next object in bin with the least remaining capacity(without exceeding the max bin capacity if given)
Worst Fit
• Insert the next object in bin with the largest remaining capacity(without exceeding the max bin capacity if given)
Objects are known dynamically at their arrival
Observations: methods easily extendable to include complicated features such as non identical ORs, patient/surgeon preferences/constraints, …
- 74 -
Static Bin Packing Rules
Step 1: Sort the objects using one of the priority rule.
• FIFO : First Come First Serve (First In First Out)
• LPT: Longest Processing Time First (good for workload balancing)
• SPT: Shortest Process Time First (good for waiting time)
• EDD: Earliest Due Date first (good for meeting due date)
Step 2: Apply one of the dynamic Bin-Packing rules
Special combinations:
First Fit Decreasing = LPT + First Fit First Fit Increasing = SPT + First Fit
Best Fit Decreasing = LPT + Best Fit Best Fit Increasing = SPT + Best Fit
Worst Fit Decreasing = LPT + Worst Fit Worst Fit Increasing = SPT + Worst Fit
- 75 -
Worst-case performances
Minimise the number of bins for a given capacity
• First Fit Decreasing, Best Fit Decreasing : B <= 11/9OPT + 1
• First Fit : B <= 17/10OPT + 2
where B = nb of bins by rules, OPT = minimal nb
Minimise the maximal bin capacity for a given number B of bins
• Best Fit Decreasing : Cmax/Cmax* <= 4/3 – 1/3B
The case of a given number B of bins can be solved exactly by a pseudo-polynomial dynamic programming algorithm.
1
Dynamic Daily Surgery Scheduling
Department of Healthcare Engineering
Centre for Health Engineering
Ecole des Mines de Saint Etienne, France
Centre for Healthcare Engineering
Dept. Industrial Engr. & Management
Shanghai Jiao Tong University, China
Xiaolan XIE
- 2 -
Field observations of surgery scheduling
- 3 -
Ruijin Hospital (since 1907 by French missionaries)
Teaching hospital of the medical school of the Shanghai Jiao Tong University
Top 1 hospital in Shanghai
+12000 outpatient visits / day
A 23-floor outpatient consultation building
- 4 -
Field observation of the operating theatre of Ruijin Hospital
An integrated operating theatre of 21 OR and a second one recently constructed
60-70 elective surgery interventions + 10 emergency surgeries / day
No integrated surgery planning but each surgery speciality is given an amount of total OR time
Each speciality decides the surgeries to perform the next day
The operating theatre (OT) is responsible for daily OR assignment and the OR program execution.
- 5 -
Field observation of the operating theatre of Ruijin Hospital
Special features of the Ruijin Hospital
Queue of elective patients never empty
Availability of patients to be operated in short notice
Availability of surgeons to operate each day
Large variety of surgeons : top surgeons, senior surgeons, ordinary surgeons
Strong demand to operate at the OT opening in the morning to avoid endless waiting
Strong concern of OT personal overtime
- 6 -
Field observation of the operating theatre of Ruijin Hospital
Issues to be addressed
Promising surgery starting times to meet surgeon's demand for reliable surgery starting
(Tell me early enough when I start my surgery)
Surgery team overtime management
(How to guarantee the on-time end of duty of surgery teams?)
Outpatient surgery appointment when servers respond to congestion
2
- 7 -
Managing surgeon appointment times
- 8 -
Why surgeon appointments not used in practice
• Not used in practice to avoid potential OR capacity loss
Research question
How to provide surgeon appointment guarantee while ensuring appropriate OR capacity usage?
Observed Daily OR utilization
• But OR capacity usage is not always high over the day
- 9 -
Related work
Static scheduling for a single OR
Surgeon appointment scheduling (AS):
Two surgeries: AS solved by a newsvendor model (Weiss, 1990)
A fixed sequence of surgeries: stochastic linear program solved by SAA
and L-shape algo to determine the allowance of each surgery, or
equivalently, the arrival time (Denton 2003).
Others: discrete appointment (Begen et al, 2011), robust appointment
(Kong et al, 2011)
Sequence scheduling: The problem is to jointly determine the position and
arrival time of each surgery (Denton 2007; Mancilla 2012).
- 10 -
Related work
Dynamic scheduling for a single OR
Arrival scheduling: The demand of surgeries is uncertain, surgeries are
processed as FCFS rule. The problem is to dynamically determine the
arrival time upon each application(Erdogan 2011).
Sequence scheduling: The demand of surgeries is also uncertain. The
problem is to jointly determine the position and arrival time of each
surgery upon each application (Erdogan 2012).
- 11 -
Our focus
Multi-OR setting
- 12 -
Our focus
Multi-OR setting
Single-OR
Multi-OR
A1 A2 A3 An
A1/A2 A3 A4 An
No OR assignment
Dynamic OR assignment
3
- 13 -
Our focus
Two inter-related problems:
• Determining surgeon arrival times by taking into
account OR capacities and random surgery
durations.
• Dynamic surgeon-to-OR assignment of during the
course of a day as surgeries progress by taking
into account planned surgeon arrival times.
- 14 -
Assumptions of our work
A1: Emergency surgeries in dedicated ORs and hence neglected.
A2: Identical ORs and surgeries assignable to any OR.
A3: At most one surgery per surgeon each day.
A4: Promised starting or appointment time informed at the end of day D-1 (Surgeon appointment scheduling or proactive problem).
A5: Surgeons not available before the promised times.
A6: Dynamic surgery-to-OR assignment during the course of the day upon the surgery completion events.
- 15 -
Dilemma of promising surgery starting time
Promise too early
Surgery 1
promised start of surgeon 2
Surgery 2
Surgery 1
promised start of surgeon 2
Surgery 2
Promise too late
surgeon waiting
OR idleOR overtime
Easy if known OR time but OR times are uncerain
- 16 -
Data
J set of surgery interventions or surgeons
N number of identical ORs
T length of OR session
pi() random duration of surgery i in scenario
bi unit time waiting cost of surgeon i
c1 unit OR idle time cost
c2 unit OR overtime cost
Similar to parallel machine scheduling but with planned job release dates and random service time.
- 17 -
Dynamic Surgery Assignment of Multiple Operating Rooms with Planned Surgeon Arrival Times
Zheng Zhang, Xiaolan Xie, Na Geng
In IEEE Trans. Automation Science and Engineering
- 18 -
Plan
Promising surgery starting times
Real time OR assignment strategies
Some numerical results
Conclusion and perspective
4
- 19 -
Decision variables
si promised surgery starting time of surgeon i
Approximation assumption: fixed assignment & sequencing
xir = 1/0 assignment of surgery i to OR r
yij = 1 if surgery i precedes j in the same OR
= 0 if not
Auxiliary scenario-based random variables
Cir() completion time of surgery i on OR r
Ir() idle time of OR r
Or() overtime of OR r
Wi() waiting time of surgeon i
- 20 -
Model for promising surgery starting times
Assign each surgery to an OR ∑r xir = 1
Relation between assignment & sequencing yij + yji ≥ xir + xjr -1
Promised start before the end of the session si ≤ T
Scenario-dependent completion time xir pi() ≤ Cir ()
Cir () ≤ M xir
Cjr () Cir () + pj() - M (1- yij) - M(2- xir - xjr )
Scenario-dependent OR idle time Cir () ≤ Ir () + iJ xir pi()
Scenario-dependent OR overtime Or () Cir () - T
Scenario-dependent surgeon waiting time rE Cir() = si + Wi() + pi()
OR idle costOR overtime
costsurgeon
waiting cost
min E{c1 ∑r Ir() + c2 ∑r Or() + ∑i biIi()}
- 21 -
Proposed solution
1. Convertion into mixed-integer linear programming model by Sample Average Approximation by using a given number of randomly generated samples
2. Heuristic for large size problem based on
a) Local search for surgery-to-OR assignment optimization
b) Surgery sequencing rule based on optimal sequencing of the two-surgery case
c) Optimal promised start time by SAA and MIP
- 22 -
Plan
Promising surgery starting times
Real time OR assignment strategies
Some numerical results
Conclusion and perspective
- 23 -
Dynamic surgery assignment optimization
At time 0, start surgeries planned at time 0
At the completion time t* of a surgery in OR r*,
select a surgery i* to be the next surgery in OR r*
among all remaining ones J*
Surgery i* starts at time max{ t*, si* } in OR r* after the arrival of the surgeon at time si*
An Event-Based Framework
- 24 -
Dynamic surgery assignment optimization
Surgery i* is selected in order to minimize E[ TC(t*, i*, J*)]
where
E[ TC(t*, i*, J*)] is the minimal total cost similar to promised time planning model
by conditioning on all completed surgeries and ages of all on-going surgeries
by scheduling i* as the next surgery on OR r*
5
- 25 -
Two-stage stochastic programming approximation
• At k-th surgery completion event at time tk
where J\J(k-1) is the set of remaining surgeries
• The first stage cost is the OR-idle
or surgeon waiting cost induced by surgery l
• lk is the second stage cost, i.e. the total cost induced by
remaining surgeries plus OR overtimes.
\ 1mink lk
l J J klkV g
ˆlk l k l k ls t t sg
- 26 -
The second stage cost
\ 1 \minlk jlk
j J J k l
where• jlk is the expected stage cost induced by surgery j
• if surgery l is selected at event k and surgery j at event k+1
Jensen's inequality is used to speedup the OPLA rule.
One-period look-ahead (OPLA) approximation
- 27 -
The second stage cost (cont'd)
Min. cost of two dynamic assignment rules:
• Rule 1 (minimal stage cost first): Remaining surgeries assigned in the scenario-independent order of minimal expected first stage cost, i.e. the surgery in selected at event n > k minimizes the stage n cost induced by in.
• Rule 2 (FCFS): Remaining surgeries are selected in non-decreasing order of their surgeon arrival times si
Jensen's inequality and another valide inequality are used to speedup the MPLA rule.
Multi-period look-ahead (MPLA) approximation
- 28 -
Lower bound of the dynamic surgery assignment
• Based on perfect information, i.e. all surgery duration
realizations pj() are known at the beginning of the day, i.e.
randomness known at time 0+
• The lower bound problem is similar to the proactive
problem but with
o given promised surgery start times
o scenario-dependent surgery assignment xir() and
sequencing yij()
- 29 -
Dynamic surgery assignment policies
Policy Static:
No real time rescheduling
OR assignment / sequencing decisions of promised time planning model are followed
Policy FIFO:
Dynamic surgery assignment in FIFO order of surgeon arrival times
Policy I:
Dynamic surgery assignment optimization with OPLA
Policy II:
Dynamic surgery assignment optimization with MPLA
- 30 -
Plan
Background and motivation
Problem setting
Promising surgery starting times
Real time OR assignment strategies
Some numerical results
Conclusion and perspective
6
- 31 -
Optimality gap
Observations
• Optimality gap is relatively small
• High surgery duration variation degrades the optimality gap
• High workload reduces the optimality gap
• MPLA better than OPLA
GAP = (costX- LB) / LB
(,)GAPI(%) GAPII(%)
Ave. Min. Max. Ave. Min. Max.(0.3,0.75) 7.4 0.1 14.7 6.3 0.1 12.8(0.7,0.75) 8.5 5.1 14.8 7.7 3.8 18.4(0.3,1.25) 5.6 1.3 11.2 4.1 1.0 8.3(0.7,1.25) 7.8 1.9 17.3 6.0 1.6 9.6
(80 3-OR instances)
- 32 -
Value of dynamic scheduling
OR# (,)VDS (%)
Ave. Min. Max.
3 (0.3,75) 10.6 2.6 22.9
(0.7,75) 14.8 5.5 26.9
(0.3,125) 7.4 3.9 14.1
(0.7,125) 11.1 5.7 15.5
Ave. 11.0 4.4 19.9
6 (0.3,75) 25.4 18.7 31.6
(0.7,75) 29.2 24.7 39.9
(0.3,125) 11.1 7.1 15.5
(0.7,125) 19.1 12.8 24.1
Ave. 21.2 15.8 27.8
12 (0.3,75) 33.6 30.1 37.9
(0.7,75) 36.0 28.9 42.1
(0.3,125) 18.6 17.2 20.4
(0.7,125) 26.1 23.9 30.1
Ave. 28.6 25.0 32.6
Observations
• Dynamic surgery scheduling always helps.
• The benefit is more important for larger OT.
• Dynamic surgery scheduling is able to cope efficiently with surgery uncertainties.
• VDS decreases as the workload of OT increases.
: variation parameter of surgery time: workload
VDS = (costStatic - costDyna) / costStatic
- 33 -
Value of dynamic scheduling optimization
Observations
• VOS increases as OR# increases.
• VOS increases as increases, i.e. the variance of surgery durations increases.
• VOS decreases as increases, i.e. the workload of OT increases.
OR# (,)VOS (%)
Ave. Min. Max.
3 (0.3,75) 2.8 0.0 14.4
(0.7,75) 5.4 0.0 26.5
(0.3,125) 2.3 0.0 7.0
(0.7,125) 3.1 0.0 10.2
Ave. 3.4 0.0 14.5
6 (0.3,75) 5.4 -0.1 13.6
(0.7,75) 6.0 -0.1 11.3
(0.3,125) 2.9 0.0 5.0
(0.7,125) 5.0 0.6 8.7
Ave. 4.8 0.1 9.6
12 (0.3,75) 7.0 5.8 7.8
(0.7,75) 9.3 6.1 11.8
(0.3,125) 5.0 3.4 6.8
(0.7,125) 6.4 4.7 9.2
Ave. 6.9 5.0 8.9
: variation parameter of surgery time: workload
VOS = (costFIFO - costDynaOpt) / costFIFO
- 34 -
Value of proactive decisions
Observations
• Proactive decision is very important to dynamic assignment scheduling.
• The arrival times that optimize the proactive model may not be adjustable to the dynamic assignment scheduling.
• Joint optimization of promised start times and dynamic assignment policies is an open research issue.
VOS = (costX - costX) / costX
where costX is the average cost of the strategy X but with promised start times determined with deterministic surgery duration.
(,)VPSI(%) VPSII(%)
Ave. Min. Max. Ave. Min. Max.(0.3,0.75) 7.2 -15.2 23.3 7.0 -20.9 22.6(0.7,0.75) 6.8 -11.1 20.4 6.4 -14.4 20.4(0.3,1.25) 9.8 1.1 23.1 10.0 0.9 21.6(0.7,1.25) 10.1 1.1 19.2 10.1 3.2 17.9
- 35 -
Plan
Promising surgery starting times
Real time OR assignment strategies
Some numerical results
Conclusion and perspective
- 36 -
Optimal surgery promised starting times for a given OR assignment / sequencing?
Features of surgeries planned to start at OR opening?
Time slacks in promised times vs surgery OR time and waiting cost?
Design of efficient optimization algorithms for promised time planning and real time rescheduling?
Promising time planning under starting time reliability constraints?
Open issues
7
- 37 -
Simulation-based Optimization of Surgery Appointment Scheduling
Zheng Zhang, Xiaolan Xie
To appear in IIE Transactions
- 38 -
Outline
• BACKGROUND AND MOTIVATION
• SURGERY APPOINTMENT SCHEDULING PROBLEM
• SAMPLE PATH ANALYSIS
• STOCHASTIC APPROXIMATION
• NUMERICAL EXPERIMENTS
• CONCLUSION AND PERSPECTIVE
- 39 -
Our focus
Example :
1st released OR allocated to surgeon 3,
2nd released OR to surgeon 4, ....
Multi-OR
A1/A2 A3 r1An
FCFS assignment
r2 A4
Surgeon appointment optimization for a given sequence of
surgeries assigned to ORs on a FIFO basis.
- 40 -
Outline
• BACKGROUND AND MOTIVATION
• SURGERY APPOINTMENT SCHEDULING PROBLEM
• SAMPLE PATH ANALYSIS
• STOCHASTIC APPROXIMATION
• NUMERICAL EXPERIMENTS
• CONCLUSION AND PERSPECTIVE
- 41 -
Modeling
Parameters
n surgeries\surgeons
m ORs with regular capacity T for each OR
pi(): surgery duration with known distribution
/ /i: unit OR idling cost / overtime cost / surgeon waiting cost
Decisions
Surgeon arrival time A = [Ai] such that:
A1 = A2 = … Am = 0 ≤ Am+1 ≤ Am+2 ≤ … ≤ An
- 42 -
Modeling
Sample-path cost function
C[i](): i-th surgery completion event time.
C[i]() depends on A and and can be solved using a simple recursion.
1
1 0
( , )n m
i i ii m i m n pi m p
f A C A A C C T
Waiting cost Idling cost Overtime cost
8
- 43 -
Modeling
Expected cost function
Objective
( ) ,g A E f A
1
min ( )
0, 1, ...,
, , ..., 1
A
i
i i
g A
A i mA
A A i m n
- 44 -
Outline
• BACKGROUND AND MOTIVATION
• SURGERY APPOINTMENT SCHEDULING PROBLEM
• SAMPLE PATH ANALYSIS
• STOCHASTIC APPROXIMATION
• NUMERICAL EXPERIMENTS
• CONCLUSION AND PERSPECTIVE
- 45 -
Sample path analysis
LEMMA . The sample path cost function f(A,) is
• differentiable with probability 1 and
• Lipschitz-continuous throughout with finite Lipschitz constant
Why?
1 2 1 2 1 2( , ) ( , ) , ,f A f A K A A A A
- 46 -
Sample path analysis
Sample path cost function differentiable when
• There is no simultaneuous event occurrences (no
simultaneous departure, no simultaneous arrival-
departure),
• i.e. one event at a time (set of sample paths of proba 1
1
1 0
[i]
( , )
C ( ): i-th surgery completion event time
n m
i i ii m i m n pi m p
f A C A A C C T
- 47 -
Sample path analysis
Proof of Lipschitz differentiable
: min min max
min min min max min max
min min min max min max
th thk k k k
k k k
th th th thk k k l l k l l
k k k l k l
th th th thk k k l l k l l
k k k l k l
Lemma i a i b a b
i a i b i b a b i b a b
i b i a i a b a i b b a
If , (Lypshitz continue), then
min min max
k k
th thk k k k
k k k
x x K k
i x i x x x K
- 48 -
Sample path analysis
THEOREM 1 (unbiasednes of sample path gradient). The
objective function g(A) is continuously differentiable on , and the
gradient of g(A) exists for all A∈ with
, ,A AE f A E f A
The noisy sample-path gradient is on average correct!
Why?
9
- 49 -
Sample path analysis
- 50 -
Sample path analysis : partial derivative at interior point
\{ }
\{ }
\{ }
A:
B:
C: 1
D: 1
i
i
i
i
jj BP i
ji j BP i
jj BP i
f
A
Ai
i BP2(i) j
A.
B.
i
Ai waiting
i BP2(i) BP3(i)C.
Ai
i BP2(i) BP3(i)D.
Ai
waiting
waiting waiting
waiting waiting overtime
[i-m]
[i-m]
[i-m]
[i-m] BP4(i)
waiting
= unit OR idling cost
overtime cost
i = surgeon waiting cost
Busy Period approach
A. i does not initiate BP(i)
B. i initiates BP(i) but not the last BP of the OR
C. i initiates the last BP of the OR without overtime
D. i initiates the last BP of the OR with overtime
- 51 -
Sample path analysis : directional derivative at boundary point
Boundary point A with Ak = Ak+1 = … = Al
0
0
... , ,, lim
... , ,, lim
, if 0
, if 0
1 1 , if
1 0 , if
i
i
lk i
v jj i
ii l
u jj k
i i i
i
i i
j j
j
j m j mj
f A e e f Af A
f A e e f Af A
x W
W
C T j n mx
W x j n m
Left-hand directional derivative
Right-hand directional derivative
- 52 -
Sample path analysis : improving direction
At an interior point, i.e. Ai-1 < Ai < Ai+1
At a boundary point A with Ak = Ak+1 = … = Al
Select two surgeries i < j such that
Determine the improving direction
,f A d
, 0, , 0i jv uf A f A
,..., ,0,...,0, ,...,i i j jv v u ud f f f f
- 53 -
Outline
• BACKGROUND AND MOTIVATION
• SURGERY APPOINTMENT SCHEDULING PROBLEM
• SAMPLE PATH ANALYSIS
• STOCHASTIC APPROXIMATION
• NUMERICAL EXPERIMENTS
• CONCLUSION AND PERSPECTIVE
- 54 -
Stochastic approximation
1k k k kA A s d
where
is an improving direction according to sample-path gradient ,
= is a converging step-size
min is the orthogonal projection into the feasible set
k k
k
d f A
as
k
y
x y x
Hill-climbing with noisy sample-path gradient
10
- 55 -
Outline
• BACKGROUND AND MOTIVATION
• SURGERY APPOINTMENT SCHEDULING PROBLEM
• SAMPLE PATH ANALYSIS
• STOCHASTIC APPROXIMATION
• NUMERICAL EXPERIMENTS
• CONCLUSION AND PERSPECTIVE
- 56 -
Convergence of stochastic approximation
BAD NEWS:
The sample path cost function is not quasiconvex.
Counter-example: p() = {9, 4, 4, 1};
2 ORs, OR session T=10;
idle time cost = 1; no overtime cost; Unit waiting cost 3=1, 4=3.
Three arrival time vectors:
A1=(0, 0, 4, 7.5)
A2=(0, 0, 6, 8.5)
A = A1 + (1-)A2
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
0 0,2 0,4 0,6 0,8 1
f(A, )
- 57 -
Convergence of stochastic approximation
By randomly perturbing p around {9, 4, 4, 1}, we implement the stochastic
approximation algorithm.
Evolution of arrival times visited by the stochastic approximation algorithm in
Example 1, when applying it over 200 sample paths.
- 58 -
Convergence of stochastic approximation
Hopeful news: The sample path cost fuction f(A,) is strongly unimodal.
Properties verified experimentally:
• Unimodality of the expected cost function
• Convergence of the stochastic approximation algorithm.
- 59 -
Convergence of stochastic approximation: numerical evidence
Log normal distribution Uniform distribution
var, wkload 0.3,0.75 0.7,0.75 0.3,1.25 0.7,1.25 0.3,0.75 0.7,0.75 0.3,1.25 0.7,1.25
Initial dispersion
3-OR 5.0 4.9 6.5 7.0 5.4 4.8 6.6 6.8
6-OR 6.5 6.7 8.5 9.5 6.5 6.6 10.3 9.8
9-OR 8.0 7.4 11.2 10.5 7.9 7.7 10.5 10.5
Final dispersion
3-OR 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
6-OR 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
9-OR 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Final grad
3-OR 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
6-OR 0.0 0.0 0.1 0.1 0.0 0.0 0.1 0.1
9-OR 0.0 0.2 0.1 0.3 0.0 0.2 0.2 0.3
- 60 -
Allowances of Multi-OR vs single OR settings
Optimal allowance shapedome shape in 1-OR, zigzag shape in 2-OR
2-OR vs 1-ORsmaller allowances, half total allowance, highly uneven
Increasing surgery duration
variability (o vs o)smoothing 2-OR allowances, increasing 1-OR allowance variability
Higher waiting cost (o vs o)larger allowances in both settings but rather insensitive in the 2-OR setting
11
- 61 -
Allowances vs OR#
Zigzag shape
1 large allowance followed by m-1 small allowances
Total m-OR allowance = 1/m of total-1-OR allowance
Higher OR# and higher duration variation smoother allowances
- 62 -
Allowances vs OR#
Two-parameter heuristic
Larger 1st allowance followed by constant allowances
- 63 -
Value of dynamic assignment and proactive solution
Three strategies
Strategy I : no dynamic surgery-to-OR assignment
Strategy II : same appointment times, FIFO surgery-to-OR assignment
Strategy III : same surgeon arrival sequence, FIFO surgery-to-OR assignment, simulation-based optimized appointment times
Value of dynamic assignment (VDA)percentage improvement of strategy II over strategy I
Value of proactive anticipation and dynamic assignment (VPD)percentage improvement of strategy III over strategy I - 64 -
Value of dynamic assignment and proactive solution
VDA > 0, VPD > 0 , VPD > VDA : dynamic assignment and the proactive anticipation of dynamic assignments always pay
Higher OR number : increasing VDA and VPD due to scale effect and benefit of well planned arrivals.
Higher duration variability: increasing VDA and VPD implying the importance of careful appointment planning and dynamic scheduling.
Higher waiting costs: higher VPD but smaller VDA implying the importance of appointment time optimization.
Higher workload: smaller VPD and VDA due to unimprovability of overloaded syst
Impact of case-mix: • larger VPD when surgeries are identical due to their interchangeability.
• smaller VDA when surgeries are identical due to suboptimal appointment timesValue of dynamic assignment (VDA)Value of proactive anticipation and dynamic assignment (VPD)
- 65 -
Outline
• BACKGROUND AND MOTIVATION
• SURGERY APPOINTMENT SCHEDULING PROBLEM
• SAMPLE PATH ANALYSIS
• STOCHASTIC APPROXIMATION
• NUMERICAL EXPERIMENTS
• CONCLUSION AND PERSPECTIVE
- 66 -
Summary
A more realistic model of AS which has m servers; patients are served
in a pre-determined order but are flexible to any server.
Our aim is to proactively optimize the arrival times under the FCFS
dynamic assignment strategy.
We formulate a simulation-based optimization model to smooth integer
assignments, and derivate a continuous and differentiable cost function.
The proposed stochastic approximation algorithm is able to solve
realistic-sized instances and significantly improve the initial solution.
12
- 67 -
Managing surgery team overtime
“Branch and Price for Chance Constrained Bin Packing”
Zhang, Denton, Xie
INFORMS Journal On Computing, to appear
- 68 -
Motivation
ORs: critical resources that require high utilization
Unpredictable overtime causes high nurse turnover rate
Nurses ask for ...
• Some ORs to have low overtime
• Predictable completion times
Challenges:
• Fixed number of ORs
• Uncertain service time
• High cost of overtime
- 69 -
A chance constrained OR scheduling setting
Chance constraint (r , ) of an OR r
The surgery team of the OR r completes its daily due before time T + r with probability
where
= regular OR session time (T)
r = allowable overtime
Chance constraint = End-of-duty guarantee
Examples: No overtime with proba 90% : r = 0, = 0.1
at most 1h overtime with proba 95% : r = 1, = 0.05
- 70 -
A chance constrained OR scheduling setting
An informal setting
Decisions:Surgeries-to-ORs assignment
Constraints: For each chance constrained OR:
P(OR overtime k) ≥ 1 -
Objective: Minimize the expected overtime
A version of chance constrained extensible bin-packing problem
- 71 -
A stochastic programming formulation
Decision variables:
(1a) = Minimize total expected ovetime
(1b) = Assign each surgery to an OR
(1c) = Determine the overtime
(1d) = Chance constraints
I, R set of surgeries and set of ORs
di() duration of surgery i under scenario
T regular OR session time
set of ORs of chance constraint k
xir binary var equal to 1 if surgery i is assigned to OR r
or() overtime of OR r under scenario
Defining elements:
CkR
- 72 -
Solving Stoch. Prog. formulation: Branch-and-Price
Master problem
Decision variables
p column containing surgeries to be allocated in the same OR
p binary var equal to 1 if the column p is selected
13
- 73 -
Solving Stoch. Prog. formulation
Key ideas of branch-and-price
1. Branch on constraints
• Select a pair: (i, j )
• Left side (in the same bin): yip = yjp
• Right side (in separate bins): yip + yjp 1
2. Enforcing the antisymmetry constraints due to identical ORs
- 74 -
Solving Stoch. Prog. formulation
Pricing problem
Decision variables
yip binary var equal to 1 if surgery i is in column p
ckp binary var equal to 1 if column p is type-k chance constrained
op() overtime of column p
Stochastic knapsack problem
- 75 -
Solving Stoch. Prog. formulation
Pricing problem solution acceleration
• Tight upperbound withprobabilisticcovers and probabilisticpackings (Song et al., 2014).
: 1
: 1
:
Probability cover
Probability pack
: 1
max
Complementary
,
1,
set
ki k
i A
k
k
k
ki i ki
i ii
i ii Q P i P
ii C
F A P d T
F C
F P
Q P i P F P i
UB Y E O
d Y T O
Y M P Y P
Y C C
- 76 -
Robust optimization formulation
Assumptions:
A1. Given first two-moments (mi, i) of surgery durations.
A2. Unknown probability distributions of surgery durations.
Chance constraints replaced by worst-case chance constraints:
where D is the set of all distributions matching the first two moments:
inf 1i ir kD
i I
P d x T
d
2 2 2,i i i i iD E d m E d m d
- 77 -
Robust optimization formulation: key result
Theorem: For any random variable X of mean m and standard deviation s, the worst-chance probability CP is reached by a three-point distribution such that
22 2
22
2 2
1, if
, if ,
, if ,
k
k k
k
k kk
m T
CP m T m m TT m
mm T m m T
T
Under the mild assumption CV (-1 – 1)0.5,
the robust optimization formulation can be converted into a deterministic mixed-integer-programming model.
- 78 -
Case study
• Experiments are based on real data of 21 surgical days.
• Number of ORs:
m = 3 + 3 + 3;
OR session time T = 10h;
Overtime threshold
dk {0.0; 2.0h; }.
• Number of surgeries:
dk [11; 37].
14
- 79 -
Performance of Branch-and-Price
Performance of different methods for the stochastic model
• Simple size: 500• Computation time limit: 15,000 seconds.• Probability guarantee: 1 - = 0.9.
- 80 -
Value of Robust Optimization
Worst-case probability
Experimental setting: 1- = 0.9 (stochastic), 1- = 0.9 (robust), n [21; 25]
• Extensive form of robust optimization can be solved by Cplex
• The unachieved probability of stochastic solution could be 0.16
• The average overtime of robust solution could be 2 times higher
Average overtime
90%
- 81 -
Value of Robust Optimization
Worst-case probability
Experimental setting: 1- = 0.9 (stochastic), 1- = 0.7 (robust), n [21; 25]
• More robust solution with slighter higher overtime
Average overtime
90%
- 82 -
Conclusions
• The Branch-and-price can effectively solve the real-size problem instances
• The robust optimization problem can be much easier to solve than the stochastic problem
• The robust optimization can provide more robust solution with slightly higher overtime
- 83 -
Accounting for congestion behavior in appointment scheduling
“Appointment Scheduling Problem When the Server Responds to Congestion”
Zhang, Berg, Denton, Xie
IISE, to appear
- 84 -
Evidences from the literature
• Outpatient clinic: physicians tend to speedup when they perceive congestion in the waiting area (Rising et al. 1973; Cayirli et al. 2008);
• Emergency department: triage-ordered testing and task reduction are used to reduce service time (Batt and Terwiesch 2012);
• ICU/ED: delays in receiving intensive care can result in longer lengths of stay in the ICU (Chan et al. 2015).
15
- 85 -
An outpatient procedure case
• Data for a one year period
• Samples are classified by surgeon and procedure type
• Specific records on patient waiting time, pre-procedure time, procedure time and post-procedure time.
We look at the impact of waiting time on different service times
- 86 -
A case in the context of outpatient procedures
• Negative correlation Between pre-procedure time and waiting time
• No correlation between procedure time, post-procedure time and waiting time
- 87 -
Related work
• Although there is a vast literature on appointment scheduling, none of the existing studies considered endogenous randomness.
• Congestion was incorporated in queuing models by Chan et al (2014), Vericourt and Jennings(2011), …
• However, appointment systems have a little number of customers and they need to determine arrival times.
- 88 -
Research questions
• Can the appointment scheduling problem be solved when the endogenous randomness is incorporated?
• How important is it to anticipate a congestion response from the server when scheduling appointments?
• Why is the dome shaped rule that is claimed "optimal", in practice, not widely implemented?
- 89 -
Problem setting
A1/A2
FCFS assignment
Appointment optimization
for a given sequence of customers
to a single server system with congestion response behaviour
in order to minimize the total cost related to
• Customer waiting (lower service quality)
• Service time reduction (lower quality service)
• Overtime.
- 90 -
Problem setting
A1/A2
FCFS assignment
Decision variables:
xi = customer-i allowance or interarrival time between i-1 and i
2 2
1 1
2
min
, ,
, , , ,
,
n nw s oi i i i i
i i
i i i i
i i i i
n
i n ni
E c w c Z d c o
w w Z x i
Z f w d i
o x w Z T
: waiting cost
: service reduction cost
: overtime cost
: normal service time
: actual service time
, ,
wi
si
o
i
i
i i i i
c
c
c
d
Z
Z f w d
16
- 91 -
Problem setting : congestion behavior models
A1/A2
FCFS assignment
Linear response model
- 92 -
Problem setting : congestion behavior models
A1/A2
FCFS assignment
Logic Regression response model
2
11 i
ii i i w
Z de
- 93 -
Problem setting : congestion behavior models
A1/A2
FCFS assignment
Linear response model with customer no-show
0, if no-show
1 , if show and
1 , if show and
ii i i i i
i
i i i i
Z d w w tt
d w t
- 94 -
Solution approaches
Under mild continuity condition of the server reponse model,
• Stochastic-optimization with unbiased sample path gradients
Under linear response model
• Stochastic linear Mixed Interger Programming
- 95 -
Computational results : Comparison of SimOpt and SMIP
• Identical customers
• 500 samples are used for the SMIP, and 107 samples for the SimOpt.
• Costs are evaluated based on 106 samples.
• The SimOpt is much more efficient than solving the SMIP
• Across all instances, the SimOpt solved the global optimum
- 96 -
Computational results : Solution
• Allowances increase with variability and waiting cost
• Congestion reduces allowances
• Congestion makes allowances more flat
17
- 97 -
Computational results : comparison with heuristics
• Our method always finds the best solution
• Mean-value solution may outperform the Dome solution when congestion occurs
- 98 -
Computational results : comparison with heuristics
• Our method always finds the best solution
• Mean-value solution may outperform the Dome solution when congestion occurs
- 99 -
Conclusions
• Simulation-based Optimization can efficiently solve the congestion anticipated AS problems
• Variability and waiting coefficient affect the allowance and cost, while congestion behavior helps to lower the cost and smooth the allowances
• Ignoring the congestion is costly; the dome-shaped solution may perform worse than the mean-value solution
- 100 -
General conclusions
- 101 -
What next?
Joint optimization of surgery sequence and surgeon
appointment times.
simulation-based discrete optimization + stochastic approximation
Chance constraints of surgery starts
Dynamic control of overtime allocation
Surgeon behavior
Joint scheduling of inpatient and day surgeries
1
Chapter 2Appointment scheduling
- 2 -
Plan
• Basis of Appointment Systems
• Individual-block/Variable-interval rule
• Variable-block/Fixed-interval rule
• A static appointment model with dynamic features
• Managing the Appointments with Waiting Time Targets and Random Walk-ins
- 3 -
Intoduction
• Rising healthcare expenditure and increasing demand lead a growing pressure on health service providers to improve efficiency.
• Appointment scheduling systems lie at the intersection of efficiency and timely access to health services.
• Timely access is important for realizing good medical outcomes. It is also an important determinant of patient satisfaction.
• The ability to provide timely access is determined by a variety of factors including both strategy designs (location, capacity, size) and operation management.
• The issue of capacity planning is addressed in another chapter and an example of resource allocation in multisite service systems can be found in Chao et al.
Chao, X., Liu, L. and Zheng, S. (2003) Resource allocation in multisite service systems with intersite customer flows. Management Science, 49(12), 1739–1752
- 4 -
Dimensions of appointment systems
Goal : to find an Appointment System (AS) for which a particular measure of performance is optimized in a clinical environment—an application of resource scheduling under uncertainty.
• Nature of decision-making
• Modeling clinic environments
• Mesures of performance
• Appointment system design
• Analysis methods
T. Cayiri & E. Veral, « Outpatient scheduling in health care : a review of literature , » Production and Operations Management, 12/4, 2003
- 5 -
Nature of decision making
Static vs dynamic appointment scheduling systems
Static systems :
• all decisions must be made prior to the beginning of a clinic session (day or half-day), which is the most common appointment system.
• Most of the literature concentrates on the static problem.
Dynamic case :
• the schedule of future arrivals revised continuously over the course of the day based on the current state of the system
• Applicable when patient arrivals can be regulated dynamically, which involves patients already admitted to a hospital or clinic.
??? Possible with on-line apps?- 6 -
Modelling clinic environment
Regarded as a queuing system characterized by the followings:
• Number of services (single or multi-stage)
• Number of doctors (single or multi-server)
• Number of appointments per clinic session
• Arrival process (deterministic or stochastic)
• Punctuality of patients
• Presence of no-shows
• Presence of regular & emergency walk-ins (preemtive or not)
• Presence of companions
• Service times (empirical or theoretical distribution)
• Lateness and interrruption of doctors
• Queue discipline
2
- 7 -
Modelling clinic environment
Number of services (single or multi-stage)
• Mostly a single queue system where patients queue for a single service (rationing: doctors = bottleneck, time to consultation is the most important criterion)
• A few simulation studies on clinic environments includingregistration, pre-examination, post-examination, x-ray, laboratory, checkout, etc
Doct 1
Doct N
Outpatient unit
- 8 -
Modelling clinic environment
Number of doctors (servers)
• Mostly a single server
• Doctors have their own waiting list
• Although pooling improves waiting times & utilisation, random doctor assignment is undesirable due to the desirableone-to-one doctor-patient relationship
• Sending patients to the first available doctor is howeverpracticed in some countries
Warning : Multi-server setting is however realistic for diagnostic equipment (IMR, CT, …)
- 9 -
Modelling clinic environment
Number of Appointments per Clinic Session
• There exists a positive relationship between waiting times and the number of appointments in a clinic session (N).
• the effect of N is mitigated by no-shows and variability of consultation times, and thus cannot be easily generalized
- 10 -
Modelling clinic environment
The arrival process
Unpunctuality of patients: • Empirical evidence : early more often than late. • patient earliness may also be undesirable, since it creates
excessive congestion in the waiting area
Presence of no-shows: • Empirical evidence : 5-30% (even 40%) no-shows depending
on the specialties. • Numerical observation : no-show proba is a major factor
affecting the performance and the choice of AS• Depending on different variables (such as age, socioeconomic
level, etc.)• Countermeasures: overbooking but also mechanism to
discouraging no-shows (blacklist?)
- 11 -
Modelling clinic environment
The arrival processUnpunctuality of patients: Presence of no-shows:
Presence of walk-ins (regular and emergency):
• Rarely in UK hospital clinics used mainly for referred patients
• But must be anticipated and planned for in some US clinicsthat are patient’s GP and respeonsible for patient’s total care, whether emergent or not.
• Common in Chinese hospitals providing on-line appointment
• Observation : walk-in varying across specialties and throughoutthe day, but not from day to day
Presence of companions• Importan for waiting room area designBalking or reneging behavior
- 12 -
Modelling clinic environment
Service times
• sum of all the times a patient is claiming the doctor’s attention
• Common model: homogenuous patients with iid service times independent of the arrival process (no response to congestion)
• Empirical data: unimodal and right-skewed, CV = 0.35 – 0.85
• Erlang or exponential distribution in analytical studies
• Observation 1: optimal solutions mostly dependent on mean and variance
• Observation 2: high variability of service times deteriorates bothpatient wait and doctor’s idle time
• Observation 3: shorter mean service time -> lower waiting time.
• Technologies help reducing the consultation time
3
- 13 -
Modelling clinic environment
- 14 -
Modelling clinic environment
Lateness and Interruption Level of Doctors
• Agreement among existing studies is that patient waiting times are highly sensitive to this factor.
• If the doctor does not start the clinic on time, a delay factor builds up from the start that ripples throughout the clinicsession.
• Another doctor-related factor is the interruption level (also called the “gap times”).
• Game theory may be useful in modeling patient and doctor arrivals by considering the conflicting interests of both parties.
• patients arrive arrive late knowing that they will have to wait.
• doctors may arrive late, being afraid that the first patient will be late.
- 15 -
Modelling clinic environment
Queueing discipline
• Common model: FIFO
• Reasonable for puctual patients
• Unrealistic in the presence of unpunctuality as doctors wouldnot keep idle waiting for the next appointment in the presence of other waiting patients
• In practice, the following priority order is used :
• Emergencies
• second consultations
• scheduled patients
• lowest priority to walk-ins seen on a FCFS basis
• Calling patients in the order of arrival is used in practice but destroy the purpose of an AS.
- 16 -
Performance mesures
Cost-based mesures
• Min E[TC] = E[W] Cp + E[I] Cd + E[O] Co
• Cp= patient waiting cost, Cd= doc idling cost, Co= overtime cost
• Common model : identical linear cost
• Klassen and Rohleder (1996) : One patient waits 40 minutes ≠ 20 patients wait 2 minutes each
• In the presence of unpunctual patients and/or walk-ins, the assumption of homogeneous waiting costs may need to be relaxed
• For regular patients, there might be a threshold over which patients’ tolerance declines steeply. Some survey results indicate that tolerance diminishes after about 30 minutes.
• UK national standard, 75% within 30 min of their appointment
- 17 -
Performance mesures
Cost-based mesures
• Min E[TC] = E[W] Cp + E[I] Cd + E[O] Co
• Cp= patient waiting cost, Cd= doc idling cost, Co= overtime cost
• In practice, Co/Cd = 1.5 (overtime cost 50% more)
• Cd/Cp = 1 to 100
• Fries & Marathe (1981):
easier to estimate the costs relative to the server, which are usually available via standard cost accounting,
but the costs of waiting involve a different type of analysis where the issues of goodwill, service, and “costs to the society” place a value on patients’ waiting time
- 18 -
Performance mesures
Time-based mesures
• True waiting time = Si – max(Ai, ai) or max(0, Si – max(Ai, ai)) Si = starting time Ai = Appointment time ai = actual arrival time
• Negative true waiting = voluntary and not due to the AS
• Flow-time = total time in the clinic.
• Flow time rarely used as patients generally do not mind the service time
• Idle time:
• Overtime
• There may be a maximal acceptable level of each. Ex: % seenwithin 30 min of their appointment
4
- 19 -
Performance mesures
Congestion mesures
• Mean & distribution of the queue length
• Mean & distribution of the number in the system
- 20 -
Performance mesures
Fairness mesures
• Fairness = uniformity of performance of an AS across patients
• Mean & variance of waiting time + queue size of patients accordingto their place in the AS
• Double penalties of patients at the end of the clinic session
waiting times tend to increase over time, whereas
Consultations time tend to decrease due to congestion response
- 21 -
Performance mesures
Other mesures
• Productivity of the doctor
• Mean doctor utilization
• Delays between requests and granted appointment
• % of urgent patients served
• % of patients receiving the requested slots
- 22 -
Appointment System Design
The AS design can be broken down into a series of decisions regarding:
1) the appointment rule,
2) the use of patient classification, if any, and
3) the adjustments made to reduce the disruptive effects of walk-ins, no-shows, and/or emergency patients
- 23 -
Appointment rules
XXXXXXXXX
Block size N
Clinical session
Single-block
X X X X X X X X X X
Individual-block/Fixed-intervalai = constantni = 1
X
X
X X X X X X X X X X
Individual-block/Fixed-interval with an initial blockai = constantn1 > 1, ni = 1
X X X X X
X X X X X
Multiple-block/Fixed-intervalni = mai = constant
Three variablesn1=initial blockni = Size of block iai = time of interval i, also called allowance
X
X
X X X X X
X X X X X
Multiple-block/Fixed-interval with an initial blockn1>ni = mai = constant
X
X X
X X X
X X X X X
X X X X X
Variable-block/Fixed-interval ni variableai = constant
X X X X X
Individual-block/ Variable-interval ni variableai = variable
- 24 -
Appointment rules
• Single-block rule : the most primitive “date only” AS ensuring doctor productivity by excessive patient waiting time.
• Individual-block/Fixed-interval rule with an initial block :
• Keep an inventory of patients to hedge against the risk of the unpunctuality or no-show of the 1st patient
• Bailey’s rule (n1 = 2, ni = 1, ai = mean service time ) -> reasonable balance between patient waiting and doctor idle (1952)
• Multiple-block/Fixed-interval rule :
• Usually (ni = m, ai = m)
• Possibly more suitable when the mean consultation time is short (Nuffield Trust (1965))
• Lack of rigorous research on the circumstances multiple block rule performs better
5
- 25 -
Appointment rules
• Individual-block/Variable-interval rule:
• The “dome” shape appointment intervals: initially increase toward the middle of the session and then decrease
• Shown to be optimal for i.i.d. service times and uniform waiting costs for all patients
- 26 -
Patient classification
• Common model : homogenuous and served in FCFA (first-call first appointment) basis
• Purposes of patient classification: sequence patients at booking, and adjust appointment intervals
• Radiology examination times depend on factors such as patient’s age, physical mobility, and type of service.
• Classification scheme: new/return, variability of service times (low/high-variance patients), and type of procedure
• Limited interest in outpatient settings in which the schedule has to be ready in advance and the arriving requests are handled dynamically
• Realistic application: the patients classified into a manageable number of groups and assigned to pre-marked slots when they call for appointments. Ex: new patients before 10:00, return patients10:00 to 12:00.
• Drawback: reduced flexibility and potential lost capacity
- 27 -
Adjustments
• Whenever relevant, no-shows, walk-ins, urgent patients, and/or emergencies need to be planned for
• No-shows cannot be entirely eliminated by administrative mechanisms
• There is a tendency of walk-in aming lower social classes and it is not fair to deny walk-in access to clinics.
• When the 2nd consultation is frequent (orthopedics), some allowance should be made for the additional demand imposed on doctors
• Suggestion from literature: the patient load (i.e., percent of available appointments filled) be adjusted based on the expected number of walk-ins and no-shows.
- 28 -
Adjustments
• Adjustment for no-shows
• None
• Overbooking extra patients to predetermined slots
• Decreasing appointment intervals proportionally
• Adjustment for walk-ins, second consultations, urgent patients, and/or emergencies
• None
• Leaving some predetermined slots open
• Increasing appointment intervals proportionally
- 29 -
Adjustments
• Adjustment for no-shows
• None
• Overbooking extra patients to predetermined slots
• Decreasing appointment intervals proportionally
• Adjustment for walk-ins, second consultations, urgent patients, and/or emergencies
• None
• Leaving some predetermined slots open
• Increasing appointment intervals proportionally
- 30 -
Analysis methodologies
• Analystical studies
• Queueing theory
• Mathematical programming
• Dynamic programming
• Nonlinear programming
• Stochastic programming
• Simulation studies
• Ho et al. (1992, 1999,1995) evaluate 50 appointment rules under various operating environments.
• No rule that performs well under all circumstances and a simple heuristic is proposed to choose an appointment rule for a clinic given , CV, N, and Cp/Cd ratio
• They find that , CV, and N affect AS performance in the order of decreasing importance.
6
- 31 -
Plan
• Basis of Appointment Systems
• Individual-block/Variable-interval rule
• Variable-block/Fixed-interval rule
• A static appointment model with dynamic features
• Managing the Appointments with Waiting Time Targets and Random Walk-ins
- 32 -
Individual-block/Variable-interval rule by Lindley equation
• AS = Individual-block/Variable-interval rule
• Single server
• N scheduled patients (with a given appointment sequence)
• Served in FAFS discipline (First Appointment First Served)
• Different general distributed service times
Not considered in this basic model :
No-shows (see extension)
Unpunctuality
X X X X X
Individual-block/ Variable-interval ni variableai = variable
- 33 -
Individual-block/Variable-interval rule by Lindley equation
1
1
1 1 1
1 1
1
Sample waiting time of patient i
Idle time of doctor waiting for patient i
Overtime
interval between i and i+1, also job c ll aa ed
i i i i
N
N N ii
i
i i i i
I x W p
O W p x T
x
W W p x
servic
llowan
e time
c
e
of iip
1 1 1
i i
i i i
W I
W p x
Why?X X X X X
Individual-block/ Variable-interval ni variableai = variable
- 34 -
Individual-block/Variable-interval rule by Lindley equation
Optimal appointment scheduling
1 1
1 1 1
1
1
min
Subject to
, 2,...,
0, 0, 0, 0
N N
p i d i oi i
i i i i i
N
N N ii
i i
E C W C I C O
W I W p x i N
O u T W p x
W I O u
Can be converted into a linear programming model by sample average over K randomly generated scenario:
1 1 1
1
1min
where ,...,
K N N
p i k d i k o kk i i
T
k k N k
C W C I C OK
p p
- 35 -
Individual-block/Variable-interval rule by Lindley equation
Case without overtime cost, i.e. Co = 0
1 1
1 1 1
min
Subject to
, 2,...,
0, 0
N N
p i d ii i
i i i i i
i i
E C W C I
W I W p x i N
W I
1 1
1 1 1
ˆ ˆmin
Subject to
ˆ ˆ ˆ ˆ ˆ , 2,...,
ˆ ˆ0, 0
N N
p i d ii i
i i i i i
i i
E C W C I
W I W p x i N
W I
ˆ ˆ,
ˆ ˆ,
i i i i i i
i i i i
p p b a x x b a
W W a I I a
Special case:
1
1
1
1
ˆ
,
ˆ 0
i i i
N
i i ii
i
b aN
E p
x ax
Optimal allowance margin independent of mean
- 36 -
Individual-block/Variable-interval rule by Lindley equation
Solution methods :
• L-shape algorithm with sequence bounding in Denton & Gupta 2003
• Bender’s decomposition in Chen & Robinson 2014
• Stochastic approximation with sample path gradient (to be addressed in surgery scheduling part for a multi-server setting)
7
- 37 -
Individual-block/Variable-interval rule by Lindley equation
Numerical results
Homogenuous patients
Unit doctor idle cost
Patient wait cost =
No overtime cost
Transformed service time distribution- 38 -
Individual-block/Variable-interval rule by Lindley equation
= 0,1N = 8
• ‘Dome’ pattern allowances with more time to patients in the middle.
• The 1st allowance much lower than the others, and varies only slightly with N, for fixed.
• The final allowance is also somewhat lower.
• The intermediate allowances are all about the same.
From Robinson & Chen, 2003
- 39 -
Individual-block/Variable-interval rule by Lindley equation
From Robinson & Chen, 2003
• Best two-parameterpolicy within 1% of the true optimum
- 40 -
Individual-block/Variable-interval rule by Lindley equation
Extension to no-shows
1 1 1 1
1 1 1 1
1
1
Sample waiting time of i-th patient
Idle time of doctor waiting for patient i
Overtime
interval between i and i+1
i i i i i
i i i i i
N
N N N ii
i
i
W W p x
I x W p
O T W p x
x
p
show-up in
service
dicator
time of i
of i, becoming known at i i iBernoulli P x
Patient specific no-show probability
- 41 -
Individual-block/Variable-interval rule by Lindley equation
Extension to no-shows
1 1
1 1 1 1
1
11
min
Subject to
, 2,...,
0, 0, 0, 0, 0
N N
p i i d i oi i
i i i i i i
N
N N i ii
i i i
E C W C I C O
W I W p x i N
O u T W p x
x W I O u
Can be converted into a linear programming model by sample average over K randomly generated scenario:
1 1 1
1 1
1min
where ,..., ; ,...,
K N N
p i i k d i k o kk i i
T
k k N k k N k
C W C I C OK
p p
- 42 -
Plan
• Basis of Appointment Systems
• Individual-block/Variable-interval rule
• Variable-block/Fixed-interval rule by queueing analysis
• A static appointment model with dynamic features
• Managing the Appointments with Waiting Time Targets and Random Walk-ins
8
- 43 -
Variable-block/Fixed-interval rule by queueing analysis
• AS = Variable-block/Fixed (or Given)-interval rule
• Single server
• N scheduled patients
• Served in FAFS discipline (First Appointment First Served)
• homogenuous patient population
• Exponential service time distribution
Not considered in this basic model :
No-shows (see extension)
Unpunctuality
Walk-ins
- 44 -
Variable-block/Fixed-interval rule by queueing analysis
Basic result:
where
• qt = queue length at the beginning of block t
• xt = block size
• Lt = Poi(t) number of departures in block t
• = service rate
1t t t tq q x L
!
k
tP L k ek
- 45 -
Variable-block/Fixed-interval rule by queueing analysis
Waiting time of patients arriving in t:
1 1 1
1
11 1
2
tx
t t t t t ti
W q i q x x x
Total wait experienced in a period
1
0 2
2 1
0
1 1! 1
11 1 0.5 2 1 1
2
,! !
t t t
k k n
k m
k kx
k x k
TW f q x
f n e m n k nk k
n n n n n n
x e x ek k
Why?
- 46 -
Variable-block/Fixed-interval rule by queueing analysis
Idle time
Overtime
1
1! 1
1
!
t t
k
t t tk q x
t t t t t t
k
k x
I e k q xk k
q x q x q x
x ek
11 1
11
2 T TO q q
- 47 -
Variable-block/Fixed-interval rule by queueing analysis
Evaluating a Variable-block/Fixed-interval rule by Markov chain
• k(t) = probability of k patients at the beginning of t
• Transition probability
• Queue distribution :
• (t+1) = (t) Pt
• State space bound
• k(t) = 0, for all k > N
1
0, if
, otherwise!
t
t
i x jijt t t
t
j i x
p P q j q ie
i x j
- 48 -
Variable-block/Fixed-interval rule by queueing analysis
Optimal Variable-block/Fixed-interval rule
• In some well-chosen neighborhood (Kaandorp & Koole, 2007),
the cost function is multimodular in block size vector x and hence
any local optimum is global optimum.
9
- 49 -
Variable-block/Fixed-interval rule by queueing analysis
Stochastic programming of Variable-block/Fixed-interval rule
1
1
1
11 1
2 1
min
0.5 1
11 1 0.5 2 1 1
2
T
p t d Ot
t t t t
t t t
T T
t
E C TW C T O N C O
q q x L
TW f q x
O q q
L Poi
f n n n n n n n
1 1, 1
1 0 0
0
, 1 , 10 0
0
min 0.5 1
0, 1,...,
, 1,...,
1, 1,..., 1
T N N
p it d O i T dt i i
N
it ti
N N
i t i t jt ti j
N
iti
E C f i z C C i i z C T N
iz x t T
iz x jz L t T
z t T
zit = 1 iff qt + xt = i
- 50 -
Extension to no-shows
Basic result:
where
• qt = queue length at the beginning of block t
• xt() = Binomial(Xt, 1-) : Xt = block size, = no show proba
• Lt = Poi(t) number of departures in block t
1t t t tq q x L
1
!
tkt X k
t
k
t
XP x k
k
P L k ek
Variable-block/Fixed-interval rule by queueing analysis
- 51 -
Extension to no-shows
The Markov chain analysis extends directly
The multi-modularity and the optimality of local search under some well-choisen neighborhood also extend.
Variable-block/Fixed-interval rule by queueing analysis
- 52 -
Variable-block/Fixed-interval rule by queueing analysis
Stochastic programming of Variable-block/Fixed-interval rule withno-shows
1 1, 1
1 0 0
0 1
, 1 , 10 1 0
0
1
min 0.5 1
0, 1,...,
, 1,...,
1, 1,..., 1
1, 1,...,
T N N
p it d O i T dt i i
N N
it i iti i
N N N
i t i i t jt ti i j
N
iti
T
itt
E C f i z C C i i z C T N
iz J x t T
iz J x jz L t T
z t T
x i N
1,1 1
, 1,..., 1 (anti-symmetry)
1
T T
it i tt t
i
tx tx i N
J Bernoulli
- 53 -
Plan
• Basis of Appointment Systems
• Individual-block/Variable-interval rule
• Variable-block/Fixed-interval rule
• A static appointment model with dynamic features
• Managing the Appointments with Waiting Time Targets and Random Walk-ins
- 54 -
From
Kumar Muthuraman, Mark Lawley, « A stochastic overbooking model for outpatient clinical scheduling with no-shows » IIE Transactions, 40, 820-837, 2008
A static appointment model with many dynamic features
10
- 55 -
Clinical booking model
• Single server
• Clinical session divided into I slots of length ti
• Patients call in sequentially before the start of the session
• These « call-ins » can be scheduled to one of the I slots or rejected.
• Rejection of a patient terminates the booking process.
• The appointment decision is made dynamically when call-in.
• Patients scheduled for each slot have a patient-specific no-show probability and arrive independently of other patients
• Patients are served in FIFO order
• Service times are exponentially distributed at rate
• Patients are categorized into J groups and the group attribute is known at the call-in time. Each group-j patient has a probability pjof showing up.
- 56 -
Performance measures
• r = reward of serving a patient
• ci = cost of each patient overflowed from slot i to slot i+1
- 57 -
Scheduling policy
• It is sequential in the sense that it assigns patients as they call
• It is myopic in the sense that it does not consider future arrivals when making the assignment.
• For each new call-in, it enumerates all possible assignments for the new patient and selects the assignment that maximizes the objective function generated by all scheduled patients.
• The algorithm rejects the patient and terminates when there is no way to schedule the patient without hurting the objective (Why ?)
- 58 -
Evaluation of a schedule with n call-ins
,n n n nim i ik
i m i k
f r mQ c kQ Q R
- 59 -
Evaluation of a schedule with n call-ins
1
number of pattients arrving for slot
number of pattients overflowing from slot
,
maximal number of departures in slot
ni
ni
n n ni i i i
n n n ni i i
i i
i
i i
X i
Y i
Y Y X L
f E r X c Y
L i
L Poi t
Q R
- 60 -
Evaluation of a schedule with n call-ins
1 1, 1
1
1
1, ''
1,
appointment of pattient
group attribute of pattient
1 , if
, if
, if
' , if , 0
, if
n n
i
n
n
n nim j i m j nn
imnim n
nim n
n n nim i i m ik n
m k
n nL i m k l ik
k l
i n
j n
Q p Q p i iQ
Q i i
R i i
R P L m k R Q i i m
f l R Q
, 0n
i i
i i m
L Poi t
11
- 61 -
Properties
1 ,i n
n
It i i
i ni i
r c e i
1min i n
nn
It i i
iii i
r c e
Unimodality
- 62 -
Numerical results
Three types of patients of no-show proba (0.9, 0.5, 0.1)Extension to include patient preference
- 63 -
Numerical results
• Unique local maximum• More appointments and
lower overflow than Round-Robin
• Similar overflow probability per patient
Total overflow
Overflow per patient
- 64 -
Plan
• Basis of Appointment Systems
• Individual-block/Variable-interval rule
• Variable-block/Fixed-interval rule
• A static appointment model with dynamic features
• Managing the Appointments with Waiting Time Targets and Random Walk-ins
- 65 -
From
Xingwei Pan, Na Geng, Xiaolan Xie, Jing Wen, « Managing appointments with waiting time targets and random walk-ins”, submitted
Exercices Layout
1. Using the flow data in the table below, measure the flow dominace and estimate complexity of the facility layout problem.
From-To-Matrix A B C D E F A - 10 12 0 40 0 B 34 - 59 0 0 12 C 0 10 - 25 0 0 D 0 23 12 - 21 0 E 9 14 0 0 - 32 F 0 27 14 0 0 - Develop a corresponding Flow Graph and try to create a reasonable layout by locating ressources with high flow close to each other.
2. Using the flow data of Exercise 1, and the added information that department C and E must be separated, create a REL chart. Solve the facility location problem for this chart, assuming those departments have equal size.
3. Consider the REL chart below and assume that all departments are similar in size.
Using quantitative method based on Total Closeness Rating, create initial layout and try to improve it by pair-wise exchange.
4. The data for four-department layout problem are given below. cost 1 2 3 4 Flow 1 2 3 4 Distance 1 2 3 4
1 5 5 10 1 10 20 5 1 10 20 52 5 20 20 2 10 15 20 2 10 15 203 5 20 10 3 20 12 10 3 20 12 104 10 10 10 4 5 20 10 4 5 20 10
1. Setup the quadratic assignment problem to find the optimal layout. 2. Estimate an initial lower bound for total location cost. 3. Improve the solution by exchanging positions of the departments.
I
X
I U
I
I E
E
U
A
U
O O
A
A
MMaacchhiinnee11
MMaacchhiinnee22
MMaacchhiinnee33
MMaacchhiinnee44
MMaacchhiinnee55
MMaacchhiinnee66
4. The Snow-Bird Hospital is a small emergency-oriented facility located in a popular ski resort area. Its new administrator, Mary Lord, decides to reorganise the hospital, using the process-layout method. The current layout is shown below.
The only physical restriction is the need to keep the entrance and initial processing room in its current location. All other departments or room (each of 10 feet square) can be moved. Mary’s first step is to analyse records in order to determine the number of trips made by patients between departments in an average month. The data are shown below. The objective Ms. Lord decides is to layout the rooms so as to minimise the total distance walked by patients who enter for treatment.
1 2 3 4 5 6 7 8 100 100 1. entrance 50 20 2. exam. 1 30 30 3. Exam. 2 20 20 4. X-ray 20 10 5. Lab 30 6. Op. room 7. Rec. Room 8. Cast-setting Given the above information, redo the layout of Snow-Bird Hospital to improve its efficiency in terms of patient flow.
Entrance/ initial
processing
Exam Room 1
Exam Room 2
X-ray
Lab tests/ EKG
Operating room
Recovery room
Cast-setting room
4400’’
1100’’
1100’’
Exo Human resources 1 (Day-off scheduling). Consider the days-off scheduling problem with the following daily requirements : Day j 1 2 3 4 5 6 7 Sun Mon Tues Wed Thurs Fri Sat requirement 3 5 5 5 7 7 3 1 weekends off over 3. 2. Consider a (5, 7)-cyclic staffing problem with b = (4, 9, 8, 8, 8, 9, 4) and c = (6, 5, 6, 7, 7, 7, 7). 3. (Employee schedule) After faithfully serving the OR profession for 50 years, you decide to retire and open a restaurant. Among the hundreds of details with opening a restaurant, you need to hire and schedule employees. Based on the foot traffic of other restaurants in the area, you expect that you will need the following number of employees each day: Day j 1 2 3 4 5 6 7 Mon Tues Wed Thurs Fri Sat Sun Employees needed 4 5 5 10 12 12 2 Your employees will work four consecutive days and then have three days off. They will be paid 100€ for each day they work. In your rush to get the restaurant started, you haphazardly hire 17 employees. Five will start on Monday, five will start on Thursday and seven will start on Friday. This schedule satisfies the above work requirements, but you have no idea how optimal this is. Questions:
1) How much money would you save each week from your current schedule if you optimized your workforce?
2) How much additional money would you save or lose each week if you switched your employees to a “five days on, two days off” schedule at 80€ per day?
4. (Planning nurse shifts) You are asked to help improving the nurse planning of an Emergency Department (ED). From the historic data, you are able to obtain the following demand forecast on the number arrivals at the ED: period (h) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24arrivals 0,5 0,5 0,5 0,5 0,5 1 4 10 10 8 4 8 6 6 4 6 10 15 8 6 4 2 0,5 0,5
From the statistics, 60% of the ED patients are regular patients and need only 15 minutes of the nursing care. However 40% of the ED patients are true emergency patients and require about 1h nursing care at ED before transfer to the wards. You are asked to :
a) Derive the workload profile of a typical day. b) Enumerate all possible shift patterns. Shifts of 8h start either 7h-9h (20€/h), or 15h-17h
(22€/h), or 23-01h (25€/h). Shifts of 12h start either 7h-9h (21€/h) or 19h-21h (23€/h). c) Determine the optimal shifts with a coverage P = 100%, 80%, et 120%. What do you think of
the solutions ? How to know which one is the best ?
9 (Staffing : number of nurses) A hospital is exploring the level of staffing needed for a booth in the local mall, where they would test and provide information on the diabetes. Previous experience has shown that, on average, every 6.67 minutes a new person approaches the booth. A nurse can complete testing and answering questions, on average, in twelve minutes. Assuming s = 2, 3, 4 nurses, a hourly cost of 40€ per nurse and a customer waiting cost of 75€ per hour. DDeetteerrmmiinnee tthhee ffoolllloowwiinngg:: ppaattiieenntt aarrrriivvaall rraattee,, sseerrvviiccee rraattee,, oovveerraallll ssyysstteemm uuttiilliizzaattiioonn,, nnbb ooff ppaattiieennttss iinn tthhee ssyysstteemm ((LLss)),, tthhee aavveerraaggee qquueeuuee lleennggtthh ((LLqq)),, aavveerraaggee ttiimmee ssppeenntt iinn tthhee ssyysstteemm ((WWss)),, aavveerraaggee wwaaiittiinngg ttiimmee ((WWqq)),, pprroobbaabbiilliittyy ooff nnoo ppaattiieenntt ((PP00)),, pprroobbaabbiilliittyy ooff wwaaiittiinngg ((PPww)),, ttoottaall ssyysstteemm ccoossttss.. 10. (Dimensioning the number of beds : Target occupancy level) CCoonnssiiddeerr oobbsstteerriiccss uunniittss iinn hhoossppiittaallss.. OObbsstteerriiccss iiss ggeenneerraallllyy ooppeerraatteedd iinnddeeppeennddeennttllyy ooff ootthheerr sseerrvviicceess,, ssoo iittss ccaappaacciittyy nneeeeddss ccaann bbee ddeetteerrmmiinneedd wwiitthhoouutt rreeggaarrdd ttoo ootthheerr sseerrvviicceess.. IItt iiss aallssoo oonnee ffoorr wwhhiicchh tthhee uussee ooff aa ssttaannddaarrdd MM//MM//ss qquueeuueeiinngg mmooddeell iiss qquuiittee ggoooodd.. MMoosstt oobbsstteerriiccss ppaattiieennttss aarree uunnsscchheedduulleedd aanndd tthhee aassssuummppttiioonn ooff PPooiissssoonn aarrrriivvaallss hhaass bbeeeenn sshhoowwnn ttoo bbee aa ggoooodd oonnee iinn ssttuuddiieess ooff uunnsscchheedduulleedd hhoossppiittaall aaddmmiissssiioonnss.. IInn aaddddiittiioonn,, tthhee ccooeeffffiicciieenntt ooff vvaarriiaattiioonn ((CCVV)) ooff tthhee lleennggtthh ooff ssttaayy ((LLOOSS)),, wwhhiicchh iiss ddeeffiinneedd aass tthhee rraattiioo ooff tthhee ssttaannddaarrdd ddeevviiaattiioonn ttoo tthhee mmeeaann,, iiss ttyyppiiccaallllyy vveerryy cclloossee ttoo 11 ssaattiissffyyiinngg tthhee sseerrvviiccee ttiimmee aassssuummppttiioonn ooff tthhee MM//MM//ss mmooddeell.. Since obsterics patients are considered emergent, the American College of Obsterics and Gynecology (ACOG) recommends that occupancy levels of obsterics units not exceeding 75%. Many hospitals have obsterics units operating below this level. However, some have eliminated beds to reduce « excess » capacity and costs and 20% of NY hospitals had obsterics units that would be considered over-utilized by this standard. Assuming the target occupancy level of 75%, what is the probability of delay for lack of beds for a hospital with s = 10, 20, 40, 60, 80, 100, 150, 200 beds. 11 (Dimensioning the number of beds) Evaluation of capacity based on a delay target leads to very important conclusion. Though there is no standard delay target, it has been suggested that the probability of delay for an obsterics bed should not exceed 1%. What is the size of an obsterics unit (nb of beds) necessary to achieve a probability of delay not exceeding 1% while keeping the target occupancy level of 60%, 70%, 75%, 80%, 85%? 12. (Dimensioning the number of beds: Impact of seasonality) Consider an obsterics unit with 56 beds which experiences a significant degree of seasonality with occupancy level varying from a low of 68% in January to about 88% in July. What is the probability of delay in January and in July? If, as is likely, there are several days when actual arrivals exceed the month average by 10%, what is the probability of delay for these days in July? 13 (Dimensioning the number of beds : Impact of clinical organization) Consider the possiblity of combining cardiac and thoracic surgery patients as thoracic patients are relatively few and require similar nursing skills as cardiac patients. The average arrival rate of cardiac patients is 1,91 bed requests per day and that of thoracic patients is 0,42. No additional information is available on the arrival pattern and we assume Poisson arrivals. The average LOS (Length Of Stay) is 7,7 days for cardiac patients and 3,8 days for thoracic patients. What is the number of beds for cardiac patients and thoracic patients in order to have average patient waiting time for a bed E(D) not exceeding 0,5, 1, 2, 3 days? What is the number of beds if all patients are treated in the same nursing unit? Delay in this case measures the time a patient coming out of surgery spends waiting in a recovery unit or ICU until a bed in the nursing unit is available. Long delays cause backups in operating rooms/emergency rooms, surgery cancellation and ambulance diversion. 14. (Planning nurse shifts) You are asked to help improving the nurse planning of an Emergency Department (ED). From the historic data, you are able to obtain the following demand forecast on the number arrivals at the ED: period (h) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24arrivals 0,5 0,5 0,5 0,5 0,5 1 4 10 10 8 4 8 6 6 4 6 10 15 8 6 4 2 0,5 0,5
From the statistics, 60% of the ED patients are regular patients and need only 15 minutes of the nursing care. However 40% of the ED patients are true emergency patients and require about 1h nursing care at ED before transfer to the wards.
Assume that the ED patients arrive according to a Poisson process with time slot dependent rate and the service time of each patient is exponentially distributed with mean equal to the previous average. You are asked to :
d) Derive the workload profile of a typical day. e) Enumerate all possible shift patterns. Shifts of 8h start either 7h-9h (20€/h), or 15h-17h
(22€/h), or 23-01h (25€/h). Shifts of 12h start either 7h-9h (21€/h) or 19h-21h (23€/h). f) Determine the optimal shifts meeting the waiting time target WT with probabilty % (WT =
30 min – = 90%, WT = 1h – = 90%, WT = 30 min – = 80%). What do you think of the solutions ? How to know which one is the best ?
Operating theatre planning/scheduling 1 (surgery planning). Vous êtes chargé de planifier un bloc opératoire. Le bloc opératoire dispose de 8h par jour sur une semaine de 5 jours. Les opérations suivantes ont été inscrites:
Case OR time (in 10 min) Pre‐assigned date (ri)
1 5 3
2 22 5
3 21 5
4 21 4
5 23 3
6 22 5
7 5 2
8 15 3
9 7 1
10 9 2
11 4 5
12 9 4
13 13 2
14 16 2
15 15 1
16 3 1
17 11 4
18 16 5
19 13 2
20 17 2 On vous demande de: 1/ vérifier si la capacité du bloc opératoire est suffisante 2/ trouver un planning permettant d'opérer un nombre maximal de patients (préciser la méthode utilisée) 3/ formuler comme un problème de programmation linéaire le problème de planification afin de minimiser le coût total sous les conditions suivantes : (i) chaque jour dispose de 2h supplémentaire au coût de 300€/h; (ii) un jour de retard coûte 500€ par patient; (iii) la non planification d'un patient i coûte (5-ri+3)*500 . 4) proposer une méthode heuristique au problème 3/.
2/ (Affectation des salles) Les patients suivants sont planifiés pour être opérés lundi dans les trois salles de votre bloc opératoire.
Case OR time (in 10 min) Spécialité
1 11 3
2 6 2
3 18 3
4 12 2
5 3 1
6 13 2
7 13 1
8 5 1
9 12 1
10 9 2
11 8 2
12 5 3
13 6 1
14 24 2
15 3 3 a/ déterminer une affectation afin de minimiser la durée d'ouverture du bloc opératoire. Expliquer votre méthode; b/ proposer une formulation mathématique afin de minimiser le coût total sous les conditions suivantes : (i) l'ouverture du bloc coûte 500 euros par heure; (ii) une salle est associée à chaque spécialité; (iii) opérer un patient dans une salle qui n'est pas de la spécialité correspondante coûte C avec C = 100€, 200€, 300€, 400€. c) proposer une méthode pour résoudre le problème b/. d) proposer une formulation mathématique et une méthode garantissant à 95% de chance la fermeture d’une salle après 8h et minimisant la somme des heures supplémentaires de autres salles. Pour cette question, chaque OR time a un écart-type de % de sa moyenne avec ( = 10%, 50%) 3/ (Ordonnancement) Les patients suivants doivent être affectés et ordonnancés dans un bloc opératoire de 2 salles d'opération et 3 lits de réveil:
Case OR time (in 10 min) Recovery time (in 10 min)
1 11 14
2 6 7
3 18 23
4 12 15
5 3 3
6 13 16
7 13 16
8 5 6
9 12 15
10 9 11 Deux cas doivent être considérés: (i) chaque patient doit être transféré dans un lit de réveil à la fin de l'opération; (ii) un patient peut commencer le réveil dans une salle d'opération en attendant la libération d'un lit de réveil. a/ proposer un affectation des salles afin de minimiser l'ouverture des salles; b/ déterminer un ordonnancement des patients dans les salles d'opération et les lits de réveil c/ proposer et tester d'autres règles d'ordonnancement.
Case mix and Master Surgery Scheduling Without loss of generality, it is assumed that each department has three surgeon groups with each handling 3 pathologies in the testing instances. Moreover, for each patient group, the lower bound on the number of admitted patients per cycle is set as one, while the upper bound is generated randomly from a discrete uniform probability distribution within the interval [3, 6]. Similarly, other data parameters are also generated from discrete uniform distribution functions. Specifically, the average LOS per pathology is distributed in the interval [4, 10], the expected surgery duration is distributed in the interval [60, 240], and the mean treatment reward is distributed in the interval [12, 30]. In addition, the total number of beds and the number of operating rooms are fixed as different amounts in various instances.
min 1 3 4 60 12 480
max 1 6 10 240 30 480
DRG group Lp Up LOSp DURp rp LENGTH
1 1 3 4 236 14 480
2 1 4 5 106 13 480
3 1 4 5 122 27 480
4 1 4 9 157 15 480
5 1 4 5 223 20 480
6 1 3 6 227 12 480
7 1 4 8 69 28 480
8 1 3 9 217 29 480
9 1 6 5 220 17 480
10 1 5 7 224 26 480
11 1 4 7 152 21 480
12 1 6 9 66 21 480
13 1 6 7 130 12 480
14 1 4 5 125 26 480
15 1 6 7 167 15 480
16 1 5 4 137 19 480
17 1 6 9 193 24 480
18 1 3 6 151 20 480
19 1 3 8 112 28 480
20 1 3 5 127 19 480
21 1 4 4 125 20 480
22 1 5 10 96 16 480
23 1 4 7 157 16 480
24 1 4 10 238 18 480
25 1 5 10 100 28 480
26 1 3 6 227 28 480
27 1 6 5 203 20 480
28 1 4 5 156 13 480
29 1 6 5 127 15 480
30 1 3 7 111 16 480
31 1 3 9 96 14 480
32 1 5 4 65 17 480
33 1 3 10 129 29 480
34 1 3 8 186 14 480
35 1 3 6 119 17 480
36 1 5 8 232 23 480
37 1 6 8 122 12 480
38 1 4 5 68 22 480
39 1 4 4 116 25 480
40 1 5 4 75 22 480
41 1 6 9 217 26 480
42 1 3 8 147 18 480
43 1 5 9 98 29 480
44 1 3 5 112 22 480
45 1 6 4 82 26 480
46 1 3 10 175 21 480
47 1 3 8 117 21 480
48 1 3 7 68 12 480
49 1 4 7 119 13 480
50 1 6 5 159 12 480
51 1 4 5 60 28 480
52 1 5 7 163 25 480
53 1 3 5 216 26 480
54 1 6 8 225 19 480
55 1 6 9 114 22 480
56 1 5 7 223 12 480
57 1 3 5 101 24 480
58 1 3 7 183 14 480
59 1 6 8 157 19 480
60 1 3 10 228 22 480
61 1 6 7 60 22 480
62 1 5 8 84 24 480
63 1 6 7 109 26 480
64 1 5 6 104 25 480
65 1 5 9 81 22 480
66 1 5 10 127 21 480
67 1 5 10 196 19 480
68 1 5 6 120 22 480
69 1 3 8 208 25 480
70 1 4 4 101 22 480
71 1 3 4 184 20 480
72 1 5 6 240 30 480
73 1 6 4 214 14 480
74 1 4 6 141 18 480
75 1 5 8 130 16 480
76 1 4 5 225 20 480
77 1 4 5 129 22 480
78 1 3 6 187 29 480
79 1 3 6 87 14 480
80 1 5 4 197 29 480
81 1 4 9 219 14 480
82 1 5 7 173 20 480
83 1 3 5 77 13 480
84 1 5 8 86 24 480
85 1 6 9 145 24 480
86 1 4 4 139 23 480
87 1 3 4 122 13 480
88 1 4 8 133 20 480
89 1 5 4 209 23 480
90 1 5 6 86 13 480
91 1 6 9 205 17 480
92 1 6 5 168 20 480
93 1 6 6 141 25 480
94 1 6 9 149 22 480
95 1 4 5 110 21 480
96 1 4 10 128 16 480
97 1 3 8 90 18 480
98 1 4 6 167 20 480
99 1 5 5 78 20 480
100 1 6 10 67 26 480
101 1 5 6 100 13 480
102 1 3 6 218 18 480
103 1 4 7 81 28 480
104 1 6 4 95 20 480
105 1 3 7 90 25 480
106 1 3 4 165 21 480
107 1 5 6 68 20 480
108 1 3 6 199 19 480
109 1 3 9 93 21 480
110 1 3 9 154 22 480
111 1 3 9 220 18 480
112 1 5 5 174 14 480
113 1 4 5 90 30 480
114 1 6 8 233 21 480
115 1 5 6 163 27 480
116 1 6 5 60 19 480
117 1 6 6 220 29 480
118 1 6 10 163 13 480
119 1 3 9 186 29 480
120 1 6 7 190 19 480
121 1 5 5 137 23 480
122 1 4 8 174 22 480
123 1 3 8 74 15 480
124 1 4 7 92 17 480
125 1 5 6 101 13 480
126 1 4 10 110 27 480
127 1 3 4 199 22 480
128 1 4 10 123 19 480
129 1 5 5 100 14 480
130 1 3 4 131 23 480
131 1 3 4 220 22 480
132 1 6 7 95 21 480
133 1 5 10 213 27 480
134 1 6 7 213 14 480
135 1 4 6 221 14 480
136 1 3 10 234 15 480
137 1 4 5 231 24 480
138 1 5 5 205 14 480
139 1 5 10 173 22 480
140 1 6 6 171 19 480
141 1 6 10 79 19 480
142 1 4 10 163 24 480
143 1 5 7 97 13 480
144 1 6 9 160 29 480
145 1 4 5 206 21 480
146 1 3 6 124 21 480
147 1 3 7 103 18 480
148 1 4 4 219 29 480
149 1 4 10 190 26 480
150 1 3 7 126 23 480
151 1 3 8 147 28 480
152 1 4 6 146 15 480
153 1 5 8 238 28 480
154 1 6 7 188 25 480
155 1 5 4 161 21 480
156 1 4 4 194 19 480
157 1 3 7 61 27 480
158 1 6 9 150 17 480
159 1 4 5 98 21 480
160 1 4 8 165 12 480
161 1 3 7 148 20 480
162 1 5 5 93 26 480
163 1 4 8 184 20 480
164 1 3 8 154 12 480
165 1 3 6 236 18 480
166 1 4 5 201 18 480
167 1 3 8 168 19 480
168 1 4 10 239 30 480
169 1 3 9 86 25 480
170 1 5 10 133 15 480
171 1 6 4 131 21 480
172 1 5 9 123 24 480
173 1 6 9 133 16 480
174 1 5 7 109 18 480
175 1 4 9 193 26 480
176 1 3 7 92 14 480
177 1 5 9 125 12 480
178 1 4 7 67 30 480
179 1 4 4 80 13 480
180 1 6 4 192 14 480
181 1 4 8 176 21 480
182 1 3 4 137 19 480
183 1 4 7 229 25 480
184 1 6 8 91 16 480
185 1 5 4 77 27 480
186 1 5 10 220 21 480
187 1 5 9 170 29 480
188 1 5 7 77 25 480
189 1 6 10 208 29 480
190 1 6 7 67 14 480
191 1 6 9 140 14 480
192 1 4 6 63 28 480
193 1 4 5 170 19 480
194 1 4 8 191 13 480
195 1 5 10 176 28 480
196 1 4 4 71 18 480
197 1 5 7 183 12 480
198 1 4 5 101 13 480
199 1 6 10 237 18 480
200 1 3 8 66 19 480
201 1 3 10 221 26 480
202 1 6 5 168 16 480
203 1 6 7 102 15 480
204 1 3 9 74 22 480
205 1 6 9 226 23 480
206 1 4 6 178 18 480
207 1 5 7 190 27 480
208 1 4 4 71 26 480
209 1 6 9 80 23 480
210 1 6 10 136 17 480
211 1 4 9 128 23 480
212 1 6 6 143 27 480
213 1 5 7 70 25 480
214 1 5 7 171 20 480
215 1 4 9 223 23 480
216 1 3 8 83 30 480
217 1 4 10 189 27 480
218 1 4 6 160 13 480
219 1 4 4 146 23 480
220 1 4 6 189 29 480
221 1 4 9 163 29 480
222 1 5 4 219 24 480
223 1 5 7 142 19 480
224 1 4 10 170 14 480
225 1 5 6 177 29 480