1. Patient Journey Optimization using a Multi-agent Approach Victor Choi Supervisor: Dr. William Cheung Co-supervisor: Prof. Jiming Liu 1

2. Agenda Introduction Patient scheduling problem in Hong Kong Proposed scheduling framework Experiments Conclusions and future works 2

3. INTRODUCTION 3

4. Objective To improve patient journey by reducing undesired waiting times for patients 4

5. How to achieve our objective With limited medical resources, we need to schedule patients in a way such that the resources could be utilized in a more efficient manner 5

6. Reasons of using a multi-agent approach It is found that hospitals have a decentralized structure, a multi-agent approached is proposed since it favors geographically distributed entities to be coordinated 6

7. Related works of using a multi-agent approach for patient scheduling T. O. Paulussen, I. S. Dept, K. S. Decker, A. Heinzl, and N. R. Jennings. Distributed patient scheduling in hospitals. In Coordination and Agent Technology in Value Networks. GITO, pages 1224–1232. Morgan Kaufmann, 2003. I. Vermeulen, S. Bohte, K. Somefun, and H. La Poutre. Improving patient activity schedules by multi-agent pareto appointment exchanging. In CEC-EEE ’06: Proceedings of the The 8th IEEE International Conference on E-Commerce Technology and The 3rd IEEE International Conference on Enterprise Computing, E-Commerce, and E-Services, page 9, Washington, DC, USA, 2006. IEEE Computer Society. The use of health state as an utility function has been challenged Temporal constraints between treatment operations are not considered 7

8. PATIENT SCHEDULING PROBLEM IN HONG KONG 8

9. Seven cancer clusters in Hong Kong C = {HKE, HKW, KC, KE, KW, NTE, NTW} 9

10. Treatment operations and medical resources Treatment plan Treatment operations { Radiotherapy, Surgery, Chemotherapy } Treatment operations { Radiotherapy, Surgery, Chemotherapy } Medical resources (A) { Radiotherapy unit, Operation unit, Chemotherapy unit } 10

11. Patient journey We define patient journey as the duration between the date of diagnosis and the date of the last treatment completed Diagnosis… (K – 1) Treatment K Treatment Patient journey 11

12. PROPOSED SCHEDULING FRAMEWORK 12

13. Two types of agents Patient agent Resource agent Patient agent Resource agent 13

14. Patient agent A patient agent (Pi) is used to represent one cancer patient Each Pi stores the corresponding patient’s treatment plan Patient agent 14

15. Resource agent A resource agent is used to represent one specific medical unit, denoted as Rab a є A, b є C Resource agent 15

16. Resource agent (cont.) Cluster (HKE) Cluster (HKW) Cluster (KC) Cluster (KE) Cluster (KW) Cluster (NTE) Cluster (NTW) Radiotherapy unit Operation unit Chemotherapy unit 16

17. Scheduling algorithm 17

18. Coordination framework 18

19. Coordination framework (cont.) For each request, it includes: Earliest Possible Start Date (EPS) ◦ The earliest date on which a treatment operation could start Latest Possible Start Date (LPS) ◦ The latest date on which a treatment operation should start such that the treatment operation could be performed earlier 19

20. Coordination framework (cont.) 20

21. Coordination framework (cont.) For each Target patient agent P G : Last = 0 if the involving treatment operation is not the last one for P G ; otherwise Temp = 0 if no temporal constraints are violated for P G ; otherwise Noti = 0 if there is a week’s time of notification for P G ; otherwise 21

22. EXPERIMENTS 22

23. Dataset A dataset provided by the Hospital Authority in Hong Kong (containing 4720 cancer patient journeys) is used for performing the simulation The diagnosis period of these 4720 patient journeys spanned across six months (1/7/2007 – 31/12/2007) 23

24. 4 experiment settings Setting 1: Patient agents are willing to exchange timeslots with others whenever none of their overall schedules would be lengthened as a result Setting 2: Only 20% of patients from each cancer cluster are allowed to exchange their timeslots Setting 3: Patients are only be swapped to a nearby cancer cluster Setting 4: Timeslots released by deceased patients are allocated to the patient agents with the longest patient journey 24

25. Results Average length of patient journey Maximum length of patient journey 25

26. Simulations revealing the impacts of varying the unit capacities To study the cost-effectiveness of increasing the capacities of medical units, 3 different timeslot allocation strategies were used: 1) 2 timeslots were added to each medical unit on a daily-basis 2) 14 timeslots were added to each medical unit on a weekly-basis 3) 60 timeslots were added to each medical unit on a monthly-basis 26

27. Simulations revealing the impacts of varying the unit capacities - Results Average length of patient journey Maximum length of patient journey 27

28. CONCLUSIONS AND FUTURE WORKS 28

29. Conclusions and future works A multi-agent framework had been proposed for patient scheduling While no temporal constraints are violated for any single patient, no patients will get a lengthened overall schedule 29

30. Conclusions and future works (cont.) Experiments showed that even with a fixed amount of medical resources, the average length of patient journey could be shortened by about a week’s time In the near future, rather than routinely allocate a fixed amount of additional timeslots to each cancer cluster, we are going to assess how resources (or timeslots) should be allocated to cancer clusters in a more sophisticated way such that the overall patient journey could be shortened in a greater extent. 30

31. THE END 31

32. Q & A 32