High tech, human touch:

Transcription

1 High tech, human touch: Operations Research in the Operating Room and beyond Dr.ir. Erwin W. Hans Associate prof. Operations Management and Process Optimization in Healthcare dep. Operational Methods for Production and Logistics (MB) Center for Healthcare Operations Improvement & Research

2 My background Positions ( ) MSc in Applied Mathematics, OR and math. programming ( ) PhD Resource loading by branch-and-price techniques ; tactical capacity planning in discrete manufacturing ( ) Assistant prof. Oper. Methods for Production & Logistics (2008-) Associate prof. OM & process optimization in healthcare (2011-) Director of Education Industrial Engineering & Management Research ( ) Planning and scheduling in discrete manufacturing (2004-) OR/OM in healthcare Chair Center for Healthcare Operations Improvement & Research Chair OR in healthcare working group of the Netherlands OR society e.w.hans@utwente.nl 1/10/2012 2

3 Agenda Introduction O.R. in healthcare process optimization Research of the CHOIR research center O.R. in the operating rooms and beyond 1/10/2012 3

4 What is Operations Management and Operations Research? Operations Management: Part of management involved in effectively and efficiently organizing processes Operations Research: Part of mathematics involved in modeling and optimizing real life processes 1/10/2012 4

5 Fries, Operations Research, 1976 Smith-Daniels, Decision Sciences, 1988 Delesie, EJOR, 1998 Cayirli, POM, 2003 Hall et al., Handbook HC Scheduling /10/2012 5

6 In 2002: <2% of the OR/MS community actually focuses on healthcare 1/10/2012 6

7 Importance of healthcare Affects all in society Ageing population More chronically ill, co-morbidity Increasingly advanced technology Expenditures growing rapidly 1/10/2012 7

8 Healthcare expenditure (% GDP) USA France Germany Belgium NL U.K. Turkey 1/10/2012 8

9 Germany vs. Netherlands Germany Netherlands Total expenditure %GDP Pharmaceutical expenditure / capita ($) Practicing physicians (nurses) / 1000 capita 3.5 (9.9) 3.9 (8.7) # beds per 1000 capita (acute care beds) 8.2 (5.7) 4.5 (3.0) Doctor consultations per capita # CT scanners per million capita Source: OECD.ORG, data from 2008 e.w.hans@utwente.nl 1/10/2012 9

10 Despite the importance of healthcare, why so little attention? Financial system did not reward efficiency Poor education of managers in operations management Poor information systems and business intelligence software Autonomy of hospital departments Autonomy of clinicians Conflicting goals Oath of Hippocrates 1/10/

11 In 2003, somewhere in the Netherlands You don t have a waiting list?? you must be a lousy doctor!! e.w.hans@utwente.nl 1/10/

12 Logistical improvements go hand-in-hand with quality improvements: patients that have to visit the hospital less often, have shorter waiting times, and may count on more attention from nurses and physicians. Logistical quality improvements will yield some 3 to 3.5 billion EUR: almost a quarter of the entire hospital budget In other words: improved care for less money! e.w.hans@utwente.nl 1/10/

13 Patient attitude change Due to: Media attention for waiting lists, bad practices Internet Benchmarking Market mechanisms patients shop Patients become more demanding 1/10/

14 Key issues for hospital management ICT innovation Hiring OM experts / OM education of managers Market positioning Specialization? Standardization of protocols (clinical pathways) LOS reduction (minimize bed usage) Copying logistical paradigms from industry with help of consultancy firms e.w.hans@utwente.nl 1/10/

15 Logistical paradigms 1/10/

16 What they all have in common 3 basic principles of Operations Management: Reduction of waste eliminate non-value-adding activities Reduction of variability eliminate disturbances, errors, fluctuations Reduction of complexity easiest effective solution is the best e.w.hans@utwente.nl 1/10/

17 Strengths Focus on performance measurement Analyzing performance Simple principles Organization-wide involvement Organization-wide improvement 1/10/

18 Weaknesses Selection of paradigm generally not based on effectiveness, but on enthusiastic consultant Paradigm = Philosophy / strive How to attain objective? Focus on operational level Low hanging fruit e.w.hans@utwente.nl 1/10/

19 What is missing? What performance levels can theoretically be attained? 10% improvement of a lousy performance is still a lousy performance! e.w.hans@utwente.nl 1/10/

20 Research is required To develop new concepts To test these concepts prospectively Using mathematical (simulation) models Under various scenarios, and a long horizon For different types of hospitals e.w.hans@utwente.nl 1/10/

21 With a chain perspective The entire care pathway optimized

22 Research is required (cont.) Operations Research provides: Optimization techniques Meta-heuristics Mathematical programming (LP, ILP) Evaluation models Queuing models Computer simulation models (DES, MC, SD) 1/10/

23 A modern framework for health care planning & control (Hans, Houdenhoven, Hulshof, 2010) Society Strategic Tactical Operational offline Operational online Medical planning Resource capacity planning Material planning Financial planning hierarchical decomposition managerial areas 1/10/

24 A modern framework for health care planning & control (Hans, Houdenhoven, Hulshof, 2010) Society Strategic Tactical Operational offline Operational online Medical planning Research planning, introduction of new treatment methods Care pathway planning Diagnosis and planning of an individual treatment Triage, diagnosing complications Resource capacity planning Case mix planning, layout planning, capacity dimensioning Allocation of time and resources to specialties, rostering Elective patient scheduling workforce planning Monitoring, emergency rescheduling Material planning Supply chain and warehouse design Supplier selection, tendering, forming purchasing consortia Purchasing, determining order sizes Rush ordering, inventory replenishing Financial planning Agreements with insurance companies, capital investments Budget and cost allocation DRG billing, cash flow analysis Expenditure monitoring, handling billing complications hierarchical decomposition managerial areas 1/10/

25 OR/OM in health care research at University of Twente: CHOIR Center for Healthcare Operations Improvement & Research Our website: Online bibliography: 1/10/

26 : collaborations UT Academic centers Top-clinical hospitals General hospitals Specialized clinic Rehabilitation centers DSS developer, consultancy Germany Belgium 1/10/

27 Research development : Focus on single departments Operating rooms (planning, scheduling, etc.) Radiology (CT, MRI) : Focus on care pathways within hospitals STW funded project LogiDOC 12 hospitals, 6 PhD students PhD students are at hospitals 2-3 days per week : optimization of the transmural care pathway optimization of rehabilitation processes e.w.hans@utwente.nl 1/10/

28 ORAHS 2012, July th annual meeting of the EURO working group on Operations Research Applied to Health Services Enschede, the Netherlands

29 Operations Research in the Operating Room 1/10/

30 First projects were no rocket science But had a huge impact! SURGERY DURATIONS SURGEON S ESTIMATE VS. HISTORICAL AVERAGE DURATION minutes Planning based on surgeon s estimates months Shorter than expected Longer than expected Planning based on historical averages e.w.hans@utwente.nl 1/10/

31 How many surgical teams are needed during the night? A discrete event simulation study (strategic level) e.w.hans@utwente.nl 1/10/

32 How many surgical teams are needed during the night? Erasmus Medical Center: 3 teams available during the night Use of 3 teams at the same time extremely rare Financially rewarding for staff Potentially dangerous to intervene Reduction in capacity may lead to deaths Simulation necessary Heavy involvement of staff in all major project steps Intervention: 1 hospital, 1 team on call e.w.hans@utwente.nl 1/10/

33 Elective surgery scheduling and sequencing (offline operational level) 1/10/

34 Offline operational level of OR planning Assignment of elective surgeries to blocks Surgery durations based on historical average Planning of slack time based on planned surgery duration variability Elective surgery sequencing Avoid problems with limited equipment Minimize chance of delays 1/10/

35 Example (11 ORs) 1/10/

36 Introduction OR planning: offline operational level Determination of the amount of slack per OR 1/10/

37 Historical data 1/10/

38 Exploiting the portfolio-effect Capacity gain 2.3%, increase in unused capacity: 40% e.w.hans@utwente.nl 1/10/

39 Emergency OR, or NOT? (tactical level) 1/10/

40 Research motivation The arrival of emergency surgeries is the most important source of disturbances in the OR leads to: overtime, surgery cancellations, waiting time, reduced OR utilization Options to deal with emergency surgery: Dedicated emergency ORs vs. Schedule emergency surgery in elective ORs 1/10/

41 Emergency OR, or not? Concept: emergency ORs Concept: No emergency ORs 1/10/

42 Emergency OR, or not? Concept: emergency ORs Concept: No emergency ORs Result of simulation: emergency OR has worse performance w.r.t.: emergency surgery waiting time, overtime, OR utilization 1/10/

43 Robust optimization of the OR schedule to deal with emergency surgery (offline operational level) 1/10/

44 Minimize emergency waiting time by optimizing the elective sequence OR1 OR2 OR3 Before 1/10/

45 Minimize emergency waiting time by optimizing the elective sequence OR1 OR2 OR3 OR1 OR2 OR3 Before After 1/10/

46 Solution approach Goal: spread Break-In-Moments between elective surgeries as evenly as possible Problem is NP-hard in the strong sense (proof by reduction from 3-partition) Input: an elective surgery schedule for a given week Optimization: constructive + local search heuristics e.w.hans@utwente.nl 1/10/

47 Constructive heuristic First calculate λ: a lower bound to min max BII λ = 1+ E S ( j J M j 1) E: earliest OR end time S: latest OR start time M j : number of surgeries in OR j Then iteratively schedule a surgery forward or backward closest to * OR1 Backward move OR2 Forward move e.w.hans@utwente.nl 1/10/

48 Simulation results operational problem Waiting time less than: First emergency procedure No BII opt. BII opt. Second emergency procedure No BII opt. BII opt. Third emergency procedure No BII opt. BII opt. 10 minutes 28.8% 48.6% 34.9% 44.9% 40.4% 46.2% 20 minutes 53.0% 75.8% 56.9% 73.6% 63.0% 69.8% 30 minutes 70.5% 90.9% 71.8% 87.2% 76.3% 86.7% Case mix Academic Hospital 1/10/

49 Results after simulation Emergency surgery in elective program instead of emergency ORs yields: Improved OR utilization (3.1%) Less overtime (21%) Break-in-moment optimization yields: Reduced waiting time for emergency surgery, especially for the first arrival (patients helped within 10 minutes: from 28.8% 48.6%) 1/10/

50 An exact approach to calculate the ward census based on the OR block schedule 1/10/

51 Peter Vanberkel An exact approach for relating recovering surgical patient workload to the OR block schedule Problem How does opening an extra op. room affect the wards? Occupancy rate Admission & discharge rates Frequency of treatments Surgery activities dictated by OR block schedule Assigns rooms to specialties Organizes the op. room department Typically cyclical Waiting Patients OR Wards Exit 51

52 Peter Vanberkel The OR block schedule Mon Tue Wed Thu Fri OR1 SUR (KLM) SUR (VWL) SUR (vwl/rur) HIPEC SUR (Kidney) SUR (VRP) OR2 ENT SUR (RUT) Urology (hbs) RT Urology (MND) OR3 ENT Plas Sur ENT ENT Plas Sur OR4 SUR (COR) Gyne SUR Mamma Plas Sur Gyne OR5 RT SUR (SND/WOS) RT (vwl/rur) Urology (pel/bex) Urology (P&B) OR6 Urology (P&B) SUR (VWL) Gyne SUR (ODB) SUR (Cor/rur) Goal: Directly derive ward workload metrics from the block schedule 1/10/

53 Peter Vanberkel Model: ward workload as a function of the OR block schedule Conceptual Model Scheme Infinite server queue Patients do not interfere Batches of patients arrive according to block schedule Ward Discharge Surgery Recovery Data For each surgical specialty Empirical Distributions of Cases/Block (batch size) Empirical Distribution of Length of Stay (LOS) Solution approach Cyclical block schedule Evaluate steady state distribution of ward census (discrete convolutions e.w.hans@utwente.nl 1/10/

54 Peter Vanberkel Model: ward workload as a function of the OR block schedule Conceptual Model Scheme Batches of patients arrive daily according to the MSS Metrics 1) Recovering Patients in the Hospital 2) Ward occupancy 3) Rates of admissions and discharges 4) Patients in recovery day n Ward Recovery Discharge Calculations: discrete convolutions of empirical distributions e.w.hans@utwente.nl 1/10/

55 Peter Vanberkel Example Result 90 th Percentile of Demand Initial MSS 1/10 days required 61 staffed beds 4/10 days required > 54 staffed beds 2/10 days required < 50 staffed beds Other days required b/w 50 & 54 Final MSS 1/10 days required 58 staffed beds 9/10 days required b/w 50 & 54 Further discussion is ongoing to change physician schedules to eliminate peak in week 2 e.w.hans@utwente.nl 1/10/

56 Instrument tray optimization 1/10/

57 Instrument trays for surgery Each surgery requires dozens of instruments, most of which are re-used after sterilization Stochastic requirements per surgery type Instruments are expensive Diversity of instruments is enormous Sterilization is expensive (± 1 per instrument) e.w.hans@utwente.nl 1/10/

58 Instrument trays for surgery Most hospitals use instrument trays There are: surgery type-specific trays base trays add-on trays Instruments remain in their tray (are sterilized together) Rarely used instruments are kept in inventory 1/10/

59 Problems with instrument trays Instrument trays evolve Many instruments are outdated Many instruments are not used during surgery Missing instruments must be collected from a storage space (takes time another tray is opened) The more types of trays the more inventory ( ) Preparing trays to order is very hard e.w.hans@utwente.nl 1/10/

60 Instrument trays: potential savings Potential savings: Unnecessary sterilizations, repairs, replacements Unnecessary inventory Location of inventory Required instruments not in tray(s) Time required for gathering instruments Time required for counting instruments Elske Florijn (MSc student from UT): In AMC, 21% of the instruments are obsolete 2.3 million sterilization costs per year Repair costs Handling costs / year sterilization cost savings when 12 out of the 550 trays types contents are optimized Problem: data collection is very hard e.w.hans@utwente.nl 1/10/

61 Elective surgery scheduling Challenges: Optimize utilization surgeons and ORs Optimize robustness (e.g. minimize overtime) Optimize other resources (ward/icu bed, X-ray) Care chain optimization, early personnel coord. etc. Easy implementation while maintaining the autonomy of the surgeons as much as possible Promising approach: Master Surgical Scheduling 1/10/

62 Preliminary study (see: EJOR 185) Question: how much can OR-utilization be increased by optimizing the elective surgery schedule? Approach: Optimization of elective scheduling by exploiting the portfolio effect 1/10/

63 Preliminary study Portfolio-effect Capacity gain 2.3%, increase in unused capacity: 40% e.w.hans@utwente.nl 1/10/

64 Master surgical scheduling a cyclic, integral planning of ORs and ICU department (tactical planning level) OR Spectrum, 2007 (co-work Van Oostrum et al.) e.w.hans@utwente.nl 1/10/

65 Motivation of research Low OR utilization, many cancellations OR-scheduling is time-consuming, and repetitive However: many elective surgery types are recurring! Weekly optimization using mathematical techniques Leads to nervous schedules May interfere with autonomy of medical specialists Hard to implement 1/10/

66 ICU bed requirements after surgery 1/10/

67 Capacity usage for shortstay ward 1/10/

68 Master surgical scheduling: idea Idea: design a cyclic schedule of surgery types that: covers all frequent elective surgery types levels the workload of the specialties levels the workload of subsequent departments (ICU, wards) is robust against uncertainty improves OR-utilization maintains autonomy of clinicians Assign patients to the slots in the schedule e.w.hans@utwente.nl 1/10/

69 MSS: problem description Goal: Maximize the OR-utilization Level capacity usage of subsequent resources (ICU) Constraints: OR-capacity constraints (probabilistic) All surgery types must be planned i.c.w. their frequency To determine: Length of the planning cycle A list of surgery types for every OR-day ( OR-day schedule ) e.w.hans@utwente.nl 1/10/

70 Mathematical program (base model) maximizes the OR utilization levels the hospital bed usage All surgeries assigned Probabilistic constraints for wards, ORs 1/10/

71 Master surgical scheduling: approach PHASE 1: Generation of OR-day schedules Goal: capacity utilization PHASE 2: Assignment of OR-day schedules ILP, solved by column generation and then rounding Constraints: All surgeries must be planned OR-capacity (probabilistic) ILP, solved using CPLEX in AIMMS modeling language Goal: bed usage leveling 1/10/

72 OR-day schedule example 15:30h Unused capacity Planned slack Planned surgery types 08:00h 1/10/

73 Master surgical scheduling: approach PHASE 1: Generation of OR-day schedules Goal: capacity utilization PHASE 2: Assignment of OR-day schedules ILP, solved by column generation and then rounding Constraints: All surgeries must be planned OR-capacity (probabilistic) ILP, solved using CPLEX in AIMMS modeling language Goal: bed usage leveling 1/10/

74 MSS test approach 1. Statistical analysis of surgery frequencies 2. Select a cycle length (1, 2, or 4 weeks) 3. Construct an MSS (2-phase approach) Tools: AIMMS modeling language with integrated CPLEX solver 4. Discrete event simulation Schedule rare elective procedures in reserved capacity Admission of emergency surgeries (add-on and online planning) Data: historical data from 3 types of hospitals; academic hospital, regional hospital, and clinic e.w.hans@utwente.nl 1/10/

75 Master surgical scheduling: results Outcomes differ per type of hospital: Percentage of surgeries in MSS Clinic Regional hospital Academic hospital 1 year 4 weeks 2 weeks 1 week Reason: different volume and case mix range e.w.hans@utwente.nl 1/10/

76 Master surgical scheduling: results Req. number of ICU-beds without MSS: between 0 and 12 p.day Req. number of ICU-beds with MSS (4 week cycle): 74.3% of the total ICU bed requirement is planned in an MSS of four weeks. e.w.hans@utwente.nl 1/10/

77 Master surgical scheduling: results Reduction OR-capacity usage (portfolio effect): Cycle length Academic hospital Regional hospital 1 week 2 weeks 4 weeks 1.1 % 2.7 % 4.2 % 2.8 % 5.7 % 6.3 % Clinic 4.9 % 7.3 % 8.6 % e.w.hans@utwente.nl 1/10/

78 Master surgical scheduling conclusions Advantages: Easy to implement Allows personnel coordination in early stage Less overtime, higher utilization (up to 8.6%) Less surgery cancellations shorter leadtimes Improved coordination between departments Disadvantage: Does not cover all surgeries 1/10/

79 Questions? 1/10/

80 1/10/

81 Questions? 1/10/