Optimization of On-line Appointment Scheduling

Transcription

1 Optimization of On-line Appointment Scheduling Brian Denton Edward P. Fitts Department of Industrial and Systems Engineering North Carolina State University Tsinghua University, Beijing, China May, 2012 Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

2 Acknowledgements Ayca Erdogan, School of Medicine, Stanford University Alex Gose, NC State University Supported by National Science Foundation: CMMI Service Enterprise Systems Grant Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

3 Appointment Scheduling Systems Interface between healthcare providers and patients Arises in many healthcare contexts Primary care Radiation Oncology Surgery Outpatient Procedures Chemotherapy Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

4 Scheduling Challenges Competing criteria Patient waiting time Provider idle time and overtime Complicating Factors Uncertain service durations Uncertain patient demand No-shows Urgent Add-ons Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

5 Research Questions Given a probabilistic arrival process for customer appointment requests to a single server, in which appointments must be quoted on-line: What is the structure of the optimal appointment schedule? How can problems be classified into easy and hard? How important is it to find optimal schedules? Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

6 Presentation Outline Introduction Problems Static Appointment Scheduling Dynamic Appointment Scheduling Dynamic Appointment Sequencing and Scheduling Conclusions Other Research Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

7 Static Appointment Scheduling Problem Problem: Schedule n customers with uncertain service times during a fixed length of day, d Planned Available Time (d) x 1 x 2 x 3 x 4 x 5 Idling (s) Waiting (w) Overtime (l) Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

8 Common Heuristics Mean Service Times: a 1 = 0 a i = a i 1 + µ i 1, i Hedging: a 1 = 0 a i = a i 1 + µ i 1 + κσ i 1, i Ho, C., H. Lau Minimizing Total Cost in Scheduling Outpatient Appointments, Management Science 38(12). Cayirli, T., E. Veral Outpatient Scheduling in Health Care: A Review of Literature, Production and Operations Management 12. Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

9 Literature Review Queuing Analysis Bailey and Welch (1952) Jansson (1966) Sabria and Daganzo (1989) Heuristics White and Pike (1964) Soriano (1966) Ho and Lau (1992) Optimization Weiss (1990) Wang (1993) Denton and Gupta (2003) Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

10 Two-Stage Stochastic Linear Program First stage decisions x i : Time allowance for customer i Second stage decisions w i (ω): Customer i waiting time l(ω): Server overtime w.r.t. length of session d Random service durations: Z i (ω): Random service time for customer i Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

11 Model Formulation n min E ω [ ci w w i (ω) + c l l(ω)] i=2 s.t. w 2 (ω) Z 1 (ω) x 1, ω w 2 (ω) + w 3 (ω) Z 2 (ω) x 2, ω w n 1 (ω) + w n (ω) Z n 1 (ω) x n 1, ω n 1 w n (ω) + l(ω) Z n (ω) + x i d, ω x 0, w(ω), l(ω) 0, ω i=1 Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

12 Example: 6 Customers Denton, B.T. and Gupta D., 2003, A Sequential Bounding Approach for Optimal Appointment Scheduling, IIE Transactions, 35, Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

13 Dynamic Appointment Scheduling Problem: Up to n U customers are scheduled dynamically as they request appointments. Appointment requests are probabilistic. C2 C1 C3 C1 C2 C4 C1 C2 C3 C5 C1 C2 C3 C4 C1 C2 C3 C4 C5 Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

14 Multi-stage Stochastic Program Appointment requests are defined by a multi-stage scenario tree: q n u n U 1 q 4 n U -1 1-q n u n U q q 4 3 n U q 3 2 min x 1 {(1 q 3 )Q 2 (x 1 )+min{q 3 (1 q 4 )Q 3 (x 2 )+ + min {( x n U 1 x 2 n U i=3 )(q i )Q n U (x n U 1)} }} Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

15 Model Formulation: Stage j j+1 Q j (x j, ω j ) = min { ci w w j,i (ω j ) + c l l j+1 (ω j )} w,l i=2 s.t w j,2 (ω j ) Z 1 (ω j ) x 1 w j,2 (ω j ) + w j,3 (ω j ) Z 2 (ω j ) x w j,j (ω j ) + w j,j+1 (ω j ) Z j (ω j ) x j w j,j+1 (ω j ) + l j+1 (ω j ) Z j+1 (ω j ) + w j,i (ω j ) 0 i, l j (ω j ) 0. j x i d i=1 Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

16 Model Properties Motivation for first come first serve (FCFS) appointment sequence: Proposition For n U = 2 with i.i.d. service durations, and identical waiting costs, the optimal sequence is FCFS. Counter-examples exist for non i.i.d. and nonidentical waiting costs. Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

17 Solution Methods Variants of nested decomposition: Fast-forward-fast-back implementation Multi-cut method 2 variable method for master problems Valid inequalities based on relaxations of the mean value problem Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

18 Outer Linearization Outerlinearize the recourse function: min{θ θ Q(x)} Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

19 Methodology: Nested Decomposition Method Xn U -1 X n U X 1 X 2 3 q 4 n U u 1 q n -1 n U -1 u 1- q n n U -1 X U n -1 X n U n U 2 q 3 1- q 3 X 2 1- q 4 3 Forward X 1 2 Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

20 Methodology: Nested Decomposition Method Add Optimality cut Θ Eω [ π ( h Tx )] 2 Add Optimality cut q 4 q 3 1- q q 4 Add Optimality cut 3 n U -1 q n u -1 n U u 1- q n -1 n U -1 1 Add optimality cut n U Backward Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

21 Multi-Cut Method Separate cuts from master problems and subproblems (similar to multi-cut approach proposed by Birge and Louveaux (1985) Cut 1 3 q n u -1 n U 1 n U 2 q 3 1- q 3 3 Cut 2 1- q 3 q 2 Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

22 Two-variable LPs Master problems at each stage are two-variable LPs (x j and θ j ) α j x j + θ j β (α 1 x 1 + α 2 x α j 1 x j 1 ) Solve LPs with a modified version of the algorithm proposed by Dyer (1984) x 1 2 intersect x 3 4 intersect median x intersect Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

23 Valid Inequalities Proposition The optimal solution to the mean value problem is x i = µ i, i. Constraints based on mean value problem θ j Q j (x, ξ) Similar to valid inequalities proposed by Batun et al. (2011) Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

24 Solution Methods Several adaptations of nested decomposition were compared: Standard nested decomposition (ND) Multi-cut ND Two-variable ND ND with mean value valid inequalities (VI) Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

25 Comparisons of Methods Number of Iterations CPU Time (seconds) n U = 10 n U = 20 n U = 30 n U = 10 n U = 20 n U = 30 (d=200) (d=400) (d=600) (d=200) (d=400) (d=600) ND c l c = 10 Multi-cut ND w Two-variable ND ND with VIs ND c l c = 1 Multi-cut ND w Two-variable ND ND with VIs ND c l c = 0.1 Multi-cut ND w Two-variable ND ND with VIs QuadCore Intel R Xeon R Processor 2.50GHz CPU, 16GB Ram, CPLEX 11.0 Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

26 Value of Stochastic Solution (VSS) Table: VSS for test instances with Z i U(20, 40) and q i = 0.5 for add-on requests. VSS (%) Number of Customers d = 200 (Routine, Add-on) c l c c = 10 l c w c = 1 l w c = 0.1 w (0,30) (10,30) (20,30) Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

27 Example: Scheduling an Endoscopy Suite x i Patients 9-3 Patients 6-6 Patients 3-9 Patients Patients Figure: Service times based on colonoscopy times for an outpatient endoscopy practice: Z i Lognormal(23.55, 11.89), i. Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

28 Example: Multi-Procedure Room Endoscopy Practice Endoscopy Practice: 2 intake rooms 2 procedure rooms 4 recovery rooms Service timed based on empirical data Table: Expected waiting time and overtime according to different schedules Heuristic Stochastic Program Based Schedule c l c c = 10 l c w c = 1 l c w c = 0.1 l c w c = 10 l c w c = 1 l w c = 0.1 w Expected total cost Expected waiting time Expected overtime Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

29 Dynamic Appointment Sequencing and Scheduling The appointment request sequence and the appointment arrival sequence are not necessarily the same. C2 C2 C1 C1 C3 C3 C1 C2 C2 C1 C4 C4 C1 C2 C3 C2 C3 C1 C5 C5 C1 C2 C3 C4 C2 C4 C3 C1 C1 C2 C3 C4 C5 C2 C4 C3 C5 C1 (A) (B) Figure: (A) FCFS; (B) Example of the general case. Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

30 Two Stage Stochastic Integer Program Minimize {Cost of Indirect Waiting + E ω [Direct Waiting + Overtime]} First Stage Decisions: Customer sequencing (binary) Service time allowances (continuous, sequence dependent) Appointment times (continuous, sequence dependent) Second Stage Decisions: Waiting time (continuous, sequence dependent) Overtime (continuous) Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

31 Two Stage Stochastic Integer Program First Stage Decisions: o j,i,i : binary sequencing variable where o jii = 1 if customer i immediately precedes i at stage j, and o ii j = 0 otherwise x j,i,i : time allowance for customer i given that i immediately precedes i at stage j a j,i,i : appointment time of customer i, given that i immediately precedes i at stage j Second Stage Decisions: w j,i,i (ω) : waiting time of customer i given that customer i immediately precedes i at stage j under duration scenario ω s j,i,i (ω) : server idle time between customer i and i, given that i immediately precedes i at stage j l j (ω) : overtime at stage j with respect to session length d Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

32 First Stage Problem s.t. min j+1 o j,i,i = 1, i =1 n j p j [ j=1 j i=1 i =1 j+1 o j,i,i = 1 i =0 ci a a j,i,i ] + Q(o, x) j, i = 1, 2,..., j j+1 j+1 o j,i,i = j + 1 j i=0 i=0 o j,i,j + o j,j,i 2(o j 1,i,i o j,i,i ) 0 j, i, i < j x j,i,i M 1 o j,i,i, a j,i,i M 1 o j,i,i j, i, i j+1 j+1 j+1 x j,i,i = a j,i,i i =1 i =1 i =1 a j,i,i j, i x j,i,i, a j,i,i 0, o j,i,i {0, 1} j, i, i, j Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

33 Second Stage Subproblem j j Q(o, x, ω) = min E ω [ (ci w w j,i,i (ω) + cl l j (ω)] i=1 i =1 s.t. w j,i,i (ω) M 2 (ω)o j,i,i i, i, j, ω s j,i,i (ω) M 3 (ω)o j,i,i i, i, j, ω j j j j w j,i,i(ω) + w j,i,i (ω) s j,i,i (ω) = Z i (ω) x j,i,i i, j, ω i =1 j l j (ω) i =1 j s j,i,i (ω) + i =1 j Z i (ω) + i=1 i =1 i=1 i =1 i =1 j x j,0,i d j, ω w j,i,i (ω), s j,i,i (ω), l j (ω) 0, j, i, i, ω. Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

34 Model Properties The addition of indirect waiting costs results in conditions under which FCFS is not optimal: Proposition For n U = 2 with i.i.d. service durations if c a 2 cw 1 then the optimal sequence is LCFS. Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

35 Methodology Compared L-shaped method and Integer L-shaped method Fast solution to second stage subproblems Presolve Warm start Branch-and-cut vs. dynamic search MIP cuts (MIR, implied bound cuts, etc.) Mean value problem based valid inequalities Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

36 Computational Performance L-Shaped Method No. of Class Type of CPU Time # of Iterations Customers Customers Average Max Average Max Add on Customers Routine Add on Add on 15000* 15000* Customers Routine 15000* 15000* Add on Add on 15000* 15000* Customers Routine 15000* 15000* Add on Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

37 Computational Performance Table: Gap at the time of termination for the instances that are not solved to optimality Problem Instance Patient Best Gap Size No Type L-Shaped Method L-Shaped Method (mean value based cuts) Add on % optimal 7 Patients (uniform) Routine % 1.95% 3 Add on Add on % 7.26% Patients (uniform) Routine % 1.99% 3 Add on Add on % 21.99% 7 Patients (lognormal) Routine % 15.71% 3 Add on Add on % 31.53% Patients (lognormal) Routine % 13.07% 3 Add on Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

38 Example 1: Structure of the Optimal Solution Table: Examples with varying direct/indirect cost for instance 3.6 (7 routine, 3 add on, lognormal service times) parameters Instance c a c w c a c w No Routine Routine Add-on Add-on c L CPU Time # of Iterations Optimal Sequence ave max ave max R-R-R-R-R-R-R-A-A-A A-A-A-R-R-R-R-R-R-R A-A-A-R-R-R-R-R-R-R A-A-A-R-R-R-R-R-R-R A-A-A-R-R-R-R-R-R-R A-A-A-R-R-R-R-R-R-R A-A-A-R-R-R-R-R-R-R A-A-A-R-R-R-R-R-R-R Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

39 Conclusions VSS can be as high as 95% and as low as 0.5% Large instances of dynamic scheduling problem can be solved efficiently but sequencing and scheduling is much harder FCFS generally optimal when probabilities of add-on customers are low and/or indirect cost of waiting is low Placement of add on customers is frequently all or nothing Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

40 Other Research Complex service systems with multiple servers and stages of service Uncertain service time, demand, and patient/provider behavior Applications: Hospital surgery practices Outpatient procedure and treatment centers Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40

41 Questions? Brian Denton Brian Denton, NC State ISyE On-line Appointment () Scheduling May, / 40