Linear Programming Tools for Scheduling Trainees in Healthcare
Transcript of Linear Programming Tools for Scheduling Trainees in Healthcare
Linear Programming Tools for
Scheduling Trainees in Healthcare
William Pozehl
Rishindra Reddy MD
F. Jacob Seagull PhD
Mark Daskin PhD
Amy Cohn PhD
Janice Davis
Nate Janes
Yicong Zhang
Presentation Outline
β’ Background
β’ Motivation
β’ Model Formulation
β’ Model Implementation
β’ Conclusions and Future Work
2
Presentation Outline
β’ Background
β’ Motivation
β’ Model Formulation
β’ Model Implementation
β’ Conclusions and Future Work
3
Attending
Fellowship
Residency
Internship
Medical School
Undergraduate Program
Attending
Fellowship
Residency
Internship
Medical School
Undergraduate Program
Healthcare Training Basics
M4
M3
M2
M1
PGY1
PGY4
PGY3
PGY2
4
M4
M3
M2
M1
Healthcare Training at Michigan
105 training
programs
1,199 trainees
80 fellowships
25 residencies
5
Importance of Scheduling
6
Who does the Scheduling?
β’ Program dependent
β Chief Resident
β Faculty (Program Director)
β Senior Administrative Staff
7
Presentation Outline
β’ Background
β’ Motivation
β’ Model Formulation
β’ Model Implementation
β’ Conclusions and Future Work
8
Challenges in Scheduling
β’ Time-intensive process
β’ Numerous stakeholders
β’ Complex rules and legal requirements
β’ Conflicting goals
β’ Varying strategies and interdependencies
β’ βGood enoughβ mentality
9
PGY1 PGY2 PGY3 PGY4 PGY5 PGY6 PGY7
1 RED GOLD BLUE
VA
A
C
A
D
E
M
I
C
BLUE
RED 2
3 BLUE HAND ENT GOLD
4 BLUE
5 HAND SJMH SJMH SJMH RED
BLUE 6
7 TBE DAY CSLT GOLD
OP SJMH
8 GI SURG OMFS
9 VA GS-VASC RED DAY CSLT NIGHT CSLT HAND
GOLD 10 VASC ORTHO NIGHT CSLT
11 STX TBE VASC RED HAND
12 PED SURG SICU SURG ONC
Plastic Surgery (4 residents per level per year)
PGY1 PGY2 PGY3 PGY4 PGY5
1 MAIZE MAIZE MAIZE MAIZE MAIZE
2 BLUE BLUE BLUE BLUE
3 WHITE WHITE WHITE WHITE BLUE
4 MAIZE/BLUE/WHITE MAIZE/BLUE/WHITE
ACS
ACS
5 ACS
ACS WHITE
6 DSP
7 PLA DSP ACS
8 SICU ANES FLOAT
9 STX DSP SICU STX
10 SVA FLOAT THS ELECTIVE SVA
11 VA CT SICU STX/SVA FOOTE VA GS
12 VA GS-VASC VA GS-VASC VA GS VA GS-VASC VA GS-VASC
General Surgery (6 residents + 12 fellows per level per year)
Resident Education Requirements
β’ Each program has unique educational requirements (operative and disease exposure)
10
Service Coverage Requirements
11
β’ Each service requires a resident complement comprised of varying skillsets and disciplines
SICU STX WHITE
1ST YEAR GENERAL SURG
1ST YEAR VASCULAR SURG
1ST YEAR PLASTIC SURG
PLA
Traditional Scheduling Approach
1. Build rotation templates
2. Adjust for coverage and educational needs
3. Renegotiate after reaching a dead-end
BLUE MAIZE ACS SICU STX SVA VA CT VA
G&V BLUE
BLUE MAIZE WHITE ACS SICU DSP PLA STX SVA VA CT VA
G&V BLUE
JULY AUG SEPT OCT NOV DEC JAN FEB MAR APRIL MAY JUNE
BLUE MAIZE WHITE ACS SICU DSP PLA STX SVA VA CT VA
G&V BLUE
BLUE MAIZE WHITE ACS SICU DSP PLA STX VA CT VA
G&V BLUE
12
WHITE DSP
SVA
WHITE
BLUE
PLA PLA BLUE MAIZE SVA SICU STX VA
G&V VA CT DSP BLUE
PLA MAIZE WHITE ACS SICU BLUE PLA STX
SVA
VA CT VA
G&V
BLUE DSP VA
G&V ACS SICU MAIZE WHITE
STX
SVA VA CT PLA BLUE
BLUE MAIZE WHITE ACS SICU DSP PLA STX SVA VA CT VA
G&V BLUE
Project Goal
Design a linear program which automates creation of a block schedule that satisfies the needs of the residents
and services.
13
Presentation Outline
β’ Background
β’ Motivation
β’ Model Formulation
β’ Model Implementation
β’ Conclusions and Future Work
14
β’ A technique to solve complicated story problems
β’ Four basic parts
β Sets and parameters
β Decision variables
β Objective function
β Constraints
Linear Programming Basics
15
min 2π₯1 + π₯2
subject to π₯1 + π₯2 β₯ 5
2π₯1 + 3π₯2 β€ 11
π₯1, π₯2 β₯ 0
Optimal Solution: (1, 4) Objective Value = 6
Sets
π : residents
πΆ: resident categories
π: services
π: months
16
Parameters
πππ β 0, 1 : indicates if resident π fits category π
βππ π: lower bound on number of residents fitting category π in service π during month π
π°ππ π: upper bound on number of residents fitting category π in service π during month π
πππ : lower bound on number of months resident π must spend on service π
πππ : upper bound on number of months resident π must spend on service π
17
Decision Variables
π₯ππ π β 0, 1 : whether resident π is
assigned to service π in month π
β π β π , π β π,π β π
The base model does not have an objective function.
18
Constraints
1. Every resident gets assigned to one service every month
πππππβπΊ
= π, β π β πΉ,π β π΄
2. Every resident satisfies their educational requirements
πππ β€ πππππβπ΄
β€ πππ, β π β πΉ, π β πΊ
3. Every service satisfies their service coverage needs
ππππ β€ ππππππππβπΉ
β€ π€πππ, β π β πͺ, π β πΊ,π β π΄
. . .
19
Constraints
1. Every resident gets assigned to one service every month
xSmith,Maize,July Is Dr. Smith assigned to the Maize service in July?
If yes, xSmith,Maize,July = 1. If no, xSmith,Maize,July = 0.
xSmith,Blue,July Is Dr. Smith assigned to the Blue service in July?
xSmith,White,July Is Dr. Smith assigned to the White service in July?
π₯ππππ‘β,ππππ§π,π½π’ππ¦ + π₯ππππ‘β,π΅ππ’π,π½π’ππ¦ + π₯ππππ‘β,πβππ‘π,π½π’ππ¦ = 1
20
Constraints
1. Every resident gets assigned to one service every month
π₯ππππ‘β,ππππ§π,π½π’ππ¦ + π₯ππππ‘β,π΅ππ’π,π½π’ππ¦ + π₯ππππ‘β,πβππ‘π,π½π’ππ¦ = 1
π₯ππππ‘β,ππππ§π,π΄π’π + π₯ππππ‘β,π΅ππ’π,π΄π’π + π₯ππππ‘β,πβππ‘π,π΄π’π = 1 . . . π₯ππππ‘β,ππππ§π,π½π’ππ + π₯ππππ‘β,π΅ππ’π,π½π’ππ + π₯ππππ‘β,πβππ‘π,π½π’ππ = 1
π₯π½ππππ ,ππππ§π,π½π’ππ¦ + π₯π½ππππ ,π΅ππ’π,π½π’ππ¦ + π₯π½ππππ ,πβππ‘π,π½π’ππ¦ = 1 . . . π₯π½ππππ ,ππππ§π,π½π’ππ + π₯π½ππππ ,π΅ππ’π,π½π’ππ + π₯π½ππππ ,πβππ‘π,π½π’ππ = 1
πππππβπΊ
= π, β π β πΉ,π β π΄
21
Constraints
2. Every resident satisfies their educational requirements
xSmith,Maize,July Is Dr. Smith assigned to the Maize service in July?
If yes, xSmith,Maize,July = 1. If no, xSmith,Maize,July = 0.
xSmith,Maize,Aug Is Dr. Smith assigned to the Maize service in August? . . . xSmith,Maize,June Is Dr. Smith assigned to the Maize service in June?
1 β€ π₯ππππ‘β,ππππ§π,π½π’ππ¦ + π₯ππππ‘β,ππππ§π,π΄π’π+. . . +π₯ππππ‘β,ππππ§π,π½π’ππ β€ 2
22
Constraints
2. Every resident satisfies their educational requirements
1 β€ π₯ππππ‘β,ππππ§π,π½π’ππ¦ + π₯ππππ‘β,ππππ§π,π΄π’π+. . . +π₯ππππ‘β,ππππ§π,π½π’ππ β€ 2
1 β€ π₯ππππ‘β,π΅ππ’π,π½π’ππ¦ + π₯ππππ‘β,π΅ππ’π,π΄π’π+. . . +π₯ππππ‘β,π΅ππ’π,π½π’ππ β€ 2
1 β€ π₯ππππ‘β,πβππ‘π,π½π’ππ¦ + π₯ππππ‘β,πβππ‘π,π΄π’π+. . . +π₯ππππ‘β,πβππ‘π,π½π’ππ β€ 2
1 β€ π₯π½ππππ ,ππππ§π,π½π’ππ¦ + π₯π½ππππ ,ππππ§π,π΄π’π+. . . +π₯π½ππππ ,ππππ§π,π½π’ππ β€ 2 . . . 1 β€ π₯π½ππππ ,π΅ππ’π,π½π’ππ¦ + π₯π½ππππ ,π΅ππ’π,π΄π’π+. . . +π₯π½ππππ ,π΅ππ’π,π½π’ππ β€ 2
πππ β€ πππππβπ΄
β€ πππ, β π β πΉ, π β πΊ
23
Constraints
3. Every service satisfies their service coverage needs
xSmith,Maize,July Is Dr. Smith assigned to the Maize service in July?
If yes, xSmith,Maize,July = 1. If no, xSmith,Maize,July = 0.
aSmith,GS Is Dr. Smith a General Surgery resident?
If yes, aSmith,GS = 1.If no, aSmith,GS = 0.
aSmith,PGY1 Is Dr. Smith a PGY1 resident?
If yes, aSmith,PGY1 = 1. If no, aSmith,PGY1 = 0.
aSmith,GS_PGY1 Is Dr. Smith a General Surgery PGY1 resident?
If yes, aSmith,GS_PGY1 = 1. If no, aSmith,GS_PGY1 = 0.
24
Constraints
3. Every service satisfies their service coverage needs
3 β€ πππππ‘β,πΊππ₯ππππ‘β,ππππ§π,π½π’ππ¦ + ππ½ππππ ,πΊππ₯π½ππππ ,ππππ§π,π½π’ππ¦+ ππΆβππ,πΊππ₯πΆβππ,ππππ§π,π½π’ππ¦+. . . +ππΊπ’ππ‘π,πΊππ₯πΊπ’ππ‘π,ππππ§π,π½π’ππ¦β€ 4
1 β€ πππππ‘β,ππΊπ1π₯ππππ‘β,ππππ§π,π½π’ππ¦ + ππ½ππππ ,ππΊπ1π₯π½ππππ ,ππππ§π,π½π’ππ¦+ ππΆβππ,ππΊπ1π₯πΆβππ,ππππ§π,π½π’ππ¦+. . . +ππΊπ’ππ‘π,ππΊπ1π₯πΊπ’ππ‘π,ππππ§π,π½π’ππ¦ β€ 2
ππππ β€ ππππππππβπΉ
β€ π€πππ, β π β πͺ, π β πΊ,π β π΄
25
Expanded Model
β’ Distributed Educational Requirements
β’ Distributed Coverage Needs
β’ Extended Rotations
β’ Service Sequencing
β’ Service Spacing
β’ Resident Pairing
26
Presentation Outline
β’ Background
β’ Motivation
β’ Model Formulation
β’ Model Implementation
β’ Conclusions and Future Work
27
Implementation Process
28
Presentation Outline
β’ Background
β’ Motivation
β’ Model Formulation
β’ Model Implementation
β’ Conclusions and Future Work
29
Recap
β’ Scheduling issues affect hospital workflow, training quality, and patient safety
β’ Scheduling residency programs at UMHS is highly interdependent, complex, and poorly executed
β’ We can address these scheduling needs using a linear programming formulation
30
Future Work
β’ Define metrics for schedule optimality
β Minimize deviation from desired resident complement by service
β Maximize satisfied requests for educational customization
β’ Apply model to improve scheduling for other training programs
31
Related Applications
β’ Pediatric Medicine rotation schedule
β’ C.S. Mott Emergency Department shift schedule
β’ Chemotherapy infusion patient schedule
β’ Physician clinic/OR schedule
β’ Master surgical schedule problem
β’ Nurse staff scheduling
32
Acknowledgements
β’ Center for Healthcare Engineering and Patient Safety
β’ University of Michigan Department of Surgery
β’ The Seth Bonder Foundation
β’ The Doctors Company Foundation
33
Questions [ ? ] and Comments [ ! ]
Billy Pozehl
Dr. Rishi Reddy
Prof. Jake Seagull
Prof. Amy Cohn
Prof. Mark Daskin
Janice Davis
34