improving chemotherapy scheduling at the bc cancer...
TRANSCRIPT
Improving Chemotherapy Scheduling at the
BC Cancer AgencyCIHR Team in
Operations Research for Improved Cancer Care
The Challenge
Increasing Volumes of Chemotherapy Patients New therapies More cases
Antiquated scheduling methodology Burdensome on staff
• Schedulers• Pharmacy
Inconvenient for patients• Most notified one day before treatment
Goals Increase time between notification and appointment Simplify and automate process Put changes into practice
2
Optimization Problem DescriptionG
IFU
RC
RT
Nur
sing
tim
e
Cha
ir
time
Protocol name
Start of shift
Prep time
Lunch break
Break
Time CH
AIR
25
CH
AIR
26
CH
AIR
27
CH
AIR
28
8:008:158:308:459:009:159:309:4510:0010:1510:3010:4511:0011:1511:3011:4512:0012:1512:3012:4513:0013:1513:3013:4514:0014:1514:3014:4515:0015:1515:3015:4516:0016:1516:3016:4517:0017:1517:3017:4518:0018:1518:3018:45
NURSE07
CTC
HB
RA
JFE
CD
UG
IFFI
RB
BR
AJ
TR
GU
EP
BR
AJ
TR
BR
AJ
DTF
EC
Break
Break
Treatment slots
End of shift
A Daily Chemotherapy Schedule
Optimization Problem Description1 2 3 4 Nurse 5 6 7 8 Nurse 9 10 11 12 Nurse 13 14 15 16 Nurse
8:008:158:308:459:009:159:309:45
10:0010:1510:3010:4511:0011:1511:3011:4512:0012:1512:3012:4513:0013:1513:3013:4514:0014:1514:3014:4515:0015:1515:3015:4516:0016:1516:3016:4517:0017:1517:3017:4518:0018:1518:3018:45
chair: 3:30 5:00 2:30 1:00 12:00nurse: 1:15 2:00 0:30 0:15 4:00cplxity: 5 5 3 1 14
Nurse 1 Nurse 2
Prep Time
Nurse … Nurse 9Time Slot
Prep Time
Prep Time Prep Time
chai
r: 1:
30nu
rse:
0:3
0cp
lx: 2
chai
r: 5:
00nu
rse:
2:0
0cp
lx: 5
chai
r: 2:
30nu
rse:
0:3
0cp
lx: 3
chai
r: 1:
00nu
rse:
0:1
5cp
lx: 1
chai
r: 2:
00nu
rse:
0:4
5cp
lx: 3
List of Patients
Decision Variables
= 1 if patient p is scheduled at treatment slot i in chair j of nurse k, 0 otherwise
pijkx
1,0∈pijkx
Auxiliary Variables
= 1 if nurse k has patients scheduled, 0 otherwise= 1 if chair j of nurse k has patients scheduled, 0
otherwise= maximum nurse workload= shortfall in number of patients with respect to
second part of the shift for nurse k= shortfall in number of patients with respect to
first part of the shift for nurse k
Λ+kδ
−kδ
0,,
1,0,
≥Λ
∈ΩΘ−+kk
kk
δδ
kΘ
jkΩ
Parameters
: chair time for patient p: nurse time for patient p: complexity for patient p: maximum number of patients to be assigned to nurse k: maximum chair time to be assigned to nurse k: maximum nurse time to be assigned to nurse k: maximum complexity to be assigned to nurse k: maximum number of patients for nurse k that can start
treatment simultaneously (nursing constraint): maximum number of patients for all nurses that can
start treatment simultaneously (pharmacy constraint)
pξpνpχ
kΠ
kΞkΝkΧ
kΓ
Ψ
Constraints
All patients need to be scheduled
At most one patient can be scheduled in a given chair at each treatment slot
At most one patient receiving nursing care in any of the chairs of each nurse
1=∑ijk
pijkx p∀
1ˆ
ˆ ≤∑ ∑≤≤−p iii
pjki
p
xξ
kji ,,∀
1ˆ
ˆ ≤∑ ∑ ∑≤≤−p iii j
pjki
p
xν
ki,∀
Constraints
Maximum scheduled:
number of patients
chair time
nurse time
complexity per nurse Maximum number of patients to start treatment
simultaneously (same slot)
per nurse
overall
kpij
pijkx Π≤∑
kpij
ppijkx Ξ≤⋅∑ ξ
kpij
ppijkx Ν≤⋅∑ ν
kpij
ppijkx Χ≤⋅∑ χ
k∀
k∀
k∀
k∀
kpj
pijkx Γ≤∑
Ψ≤∑pjk
pijkx
ki,∀
i∀
Constraints
Definition of maximum nurse workload in terms of:(only one of the following active at a time)
number of patients
chair time
nurse time
complexity
Λ≤∑pij
pijkx
Λ≤⋅∑pij
ppijkx ξ
Λ≤⋅∑pij
ppijkx ν
Λ≤⋅∑pij
ppijkx χ
k∀
k∀
k∀
k∀
Constraints
Relationship between scheduling variables and indicators per nurse
Relationship between scheduling variables and indicators per chair-nurse
Balance between number of patients in first and second part of nursing shift
kpijkx Θ≤
kpijkx Ω≤
−
∈
+
∈
+=+ ∑∑ kPMpij
pijkk
AMpij
pijk xx δδ
)()(
kjip ,,,∀
kjip ,,,∀
k∀
Objective Function
Where Ф can be chosen as: Maximum nurse workload (defined as number of
patients, chair time, nurse time or complexity)
Number of nurses scheduled
Number of chairs used
( )∑ −+ ++Φk
kkMin δδ
Λ=Φ
∑Θ=Φk
k
∑Ω=Φjk
jk
The Model in Numbers
Typical scenario: 40 to 50 patients to be scheduled 8-9 nurses (8-hr shift; 6-hr of effective treatment) 4 chairs per nurse 11-hr workday (8:00 to 19:00) divided in 15’ slots
Resolution: Decision variables: over 100,000 (most binary) Constraints: over 40,000 Solution time: ~1-5’ to optimality
(30” for gap < 5%) Modeling platform: GAMS/CPLEX Database platform: MS Access Tool platform: VBA
Output Report – Nursing ScheduleNURSE01 NURSE02 NURSE03 NURSE04 NURSE05 NURSE06 NURSE07
Time8:00 08:15 08:30 08:45 09:00 29:15 09:30 19:45 110:00 110:15 110:30 110:45 311:00 211:15 011:30 011:45 112:00 112:15 112:30 212:45 113:00 113:15 213:30 013:45 214:00 114:15 214:30 214:45 115:00 215:15 415:30 115:45 216:00 116:15 016:30 016:45 017:00 117:15 017:30 017:45 018:00 018:15 018:30 018:45 0
# Patients 40 2 6 7 6 6 6 7Total Time 39:0 3:0 6:0 6:15 5:45 6:15 6:0 5:45Booked Time 30:45 2:30 4:45 4:45 4:45 4:30 4:45 4:45% Booked 79% 83% 79% 76% 83% 72% 79% 83%Chair Time 72:15 4:30 11:0 12:0 11:0 8:45 13:0 12:0
# Patient Starts
UGIAJFFOX (45min)
GIIR (45min)
GUEP (30min)
CTCH (60min)
UGIFFIRB (45min)
BRAJTR (30min)
BRAJDTFEC (30min)
BRAJFECD (60min)
BRAJTR (30min)
BRAJTR (30min)
BRAJACTT (90min)
UBRAVGEMP (45min)
BRAJTR (30min)
BRCH (75min)
BRCH (75min)
BRAJTR (30min)
BRLAACDT (75min)
UGIFFIRB (45min)
BRAJTR (30min)
GIPGEM (30min)
GIFOLFIRI (45min)
BRAVTR (45min)
GUEP (30min)
BRAVTR (45min)
UGIFFIRB (45min)
UGIFFIRB (45min)
BRAJTR (30min)
GUAVPG (30min)
CTCH (60min)
BRAVTR (45min)
BRAJTR (30min)
Nurse
GIFURCRT (60min)
BRLAACD (90min)
LYFLUDR (45min)
USA0 (90min)
BRAJTR (30min)
UGIFOLFOX (45min)
BRAJTR (30min)
GIFOLFIRI (45min)
ULYRMTN (30min)
B r e a k
B r e a k
Brea
k
Brea
k
Bre
akBr
eak
Brea
k
Brea
kBr
eak
B r e a k
Brea
k
Output Report – Chair Schedule
Time# Patient
Starts CH
AIR
01
CH
AIR
02
CH
AIR
03
CH
AIR
04
CH
AIR
05
CH
AIR
06
CH
AIR
07
CH
AIR
08
CH
AIR
09
CH
AIR
10
CH
AIR
11
CH
AIR
12
CH
AIR
13
CH
AIR
14
CH
AIR
15
CH
AIR
16
CH
AIR
17
CH
AIR
18
CH
AIR
19
CH
AIR
20
CH
AIR
21
CH
AIR
22
CH
AIR
23
CH
AIR
24
CH
AIR
25
CH
AIR
26
CH
AIR
27
CH
AIR
28
8:00 08:15 08:30 08:45 09:00 29:15 09:30 19:45 110:00 110:15 110:30 110:45 311:00 211:15 011:30 011:45 112:00 112:15 112:30 212:45 113:00 113:15 213:30 013:45 214:00 114:15 214:30 214:45 115:00 215:15 415:30 115:45 216:00 116:15 016:30 016:45 017:00 117:15 017:30 017:45 018:00 018:15 018:30 018:45 0
1 1 0 0 2 0 3 1 2 1 2 2 3 1 0 2 3 1 1 1 3 1 1 1 2 2 3 01:00 1:30 0 0 2:15 0 2:00 0:30 1:15 0:30 1:15 1:45 2:30 1:15 0 1:00 2:30 0:45 0:30 0:45 1:45 0:45 0:45 1:30 2:00 1:15 1:30 03:00 1:30 0 0 6:00 0 3:15 1:45 5:00 0:45 1:30 4:45 6:00 2:30 0 2:30 4:30 0:45 0:45 2:45 3:00 1:45 3:00 5:15 4:30 3:00 4:30 0
UG
IFFI
RB
BR
AJ
TR
GU
EP
BR
AJ
TR
BR
AJ
DTF
EC
BR
AJA
CTT
CTC
HB
RA
JFE
CD
GIIR
UG
IAJF
FO
X
BR
AJ
TRU
BR
AV
GE
MP
BR
AJ
TR
BR
AV
TR
BR
AJ
TR
GIF
OLF
IR
I
BR
LAA
CD
TU
GIF
FIR
BG
IPG
EM
BR
CH
ULY
RM
TNB
RA
JTR
BR
AJ
TRG
IFO
LFI
RI
BR
CH
BR
AJ
TR
BR
AJ
TRB
RA
VTR
UG
IFFI
RB
CTC
H
GU
EP
UG
IFFI
RB
GU
AV
PG
LYFL
UD
RB
RA
JTR
BR
AV
TR
NURSE06 NURSE07NURSE01 NURSE02 NURSE03 NURSE04
Chair time
# PatientsNursing time
NURSE05
GIF
UR
CR
T
BR
LAA
CD
US
A0
UG
IFO
LFO
X
Break
Break
Brea
k
Brea
k
Bre
akBr
eak
Brea
k
Brea
kBr
eak
Break
Brea
k
Scheduling Tool Concept
Visual interface for managers to review and modify schedule
Initial schedule loaded from optimization model
Patients can be re-scheduled across nurses Constraints are checked “on the fly” and
warnings/error messages reported to user Workload stats are updated after every change Web-based platform
Schedule Modifier Tool ConceptPatients can be re-scheduled (drag & drop)
Stats are updated after changes in
schedule
Initial schedule loaded from
optimization model
Discussion
Implementation plan being developed Daily optimization model Requires expert involvement Must interface with system• Inputs•Outputs•Standalone tool
Ignores variability Many inputs are arbitrary now
18
Overall Concluding Comments
Concluding Remarks Just touched the tip of the iceberg of potential applications Many challenging optimization (OR) problems arise in
healthcare They can significantly effect practice both at the operational
and strategic levels They offer great research potential Often solutions cannot be delivered easily so guidelines are
necessary Researchers must be cognizant of needs of staff and work
collaboratively to develop applications