The Evolving Field of Healthcare Engineering: Perspectives, Examples, and Challenges

1
The Evolving Field of Healthcare Engineering
Perspectives, Examples, and Challenges

• Mark Lawley, Associate Professor
• School of Industrial Engineering
• Regenstrief Faculty Scholar
• Regenstrief Center for Healthcare Engineering
• Purdue University

2
Prior Research

• Discrete event control
• Degradation modeling / residual life
• Disaster planning for water distribution

3
Discrete event control

• Deadlock avoidance supervisory controllers for
  allocating resources in automated systems.
• Objective is to avoid deadlock.
• Controllers must be correct and computation
  must be polynomial.
• Robust control supervisory controllers for
  allocating failure-prone resources to requesting
  process in automated systems.
• Objective is to avoid failure-induced blocking.
• Controllers must be correct and computation
  must be polynomial.
• Application area Buffer space and tool sharing
  in automated manufacturing systems.
• Finished three doctoral students and four masters
  students.
• Eighteen published journal publications, three
  under submission, two in progress.
• Two best paper awards, two book chapters

4
Degradation modeling and residual life prediction

• Given and operating component subject to
  degradation and condition data collected through
  sensors, can we compute a residual life
  distribution for the component? Application
  area thrust bearings

5
Degradation modeling and residual life prediction
6
Degradation modeling and residual life prediction

• Bayesian updating and neural networks
• Exponential and linear degradation models
• Finished one doctoral student, two masters
• One current doctoral student
• Four journal publications, one in progress
• One book chapter
• One best paper nomination

7
Disaster planning water distribution

• Water distribution is one of eight key national
  infrastructures
• Water networks are vulnerable to physical
  destruction resulting from attack or natural
  events
• Given a water supply network,
• How to best allocate security resources to
  protect from attack
• If destruction occurs, how to configure a
  residual network to minimize consequences.
• Gives rise to class of network vulnerability
  problems with both topological connectedness and
  hydraulic feasibility.
• Hard combinatorial problems

8
Security allocation model

• Consequence Calculation ?(P)
• Requires Solving Non-Linear Hydraulic Equations
• Constraint Set (2)
• Constraint ? subset of edges
• Exponentially large number of constraints
• Large number of redundant constraints

9
Consequence Minimization Model
Objective function
Constraints
Conservation of flow (mass) for demand node
Conservation of flow (mass) for supply node
Conservation of energy no pump(head-loss flow
relationship)
Conservation of energy pump(head-loss flow
relationship)
Minimum pressure requirement
hmin hu - Elevu ? u?V
10
Disaster planning water distribution

• Finished one doctoral student, one masters
• One current doctoral student
• Two journal publications, one submitted,
• One book chapter
• One book chapter in progress

11
Why does healthcare delivery need our attention?

• Health care is the largest industry in the US
• 15 of the GDP
• 19 of GDP within 10 years
• Growing at 3 times inflation
• Everyone is a stakeholder
• Factors contributing to growth in health care
  complexity
• Aging population along with epidemic of chronic
  diseases
• Rapid advances in expensive treatment
  technologies and pharmaceuticals
• Increasingly complex user/provider/payer
  relationships and incentives
• Growing population of uninsured and under-insured
• Under-investment in information technology and
  engineering staffing
• NAE/IOM report advocated two macro-approaches for
  improvement
• Increased emphasis and emphasis on integrated
  information technologies
• Application of the modeling and analysis tools of
  systems engineering

12
NAE/IOM Report
13

• January 2005, Regenstrief Center for Healthcare
  Engineering (RCHE) established with funding from
  Regenstrief Foundation
• Mission Statement Catalyze transformation of
  healthcare delivery by applying principles of
  engineering, management and science.
• Core areas Efficiency and effectiveness, patient
  safety, interoperability and security
• Colleges involved include Engineering, Liberal
  Arts, Consumer Family Sciences, Management,
  Pharmacy, Nursing Health Sciences, Science,
  Technology

14

• Appointed Regenstrief Faculty Scholar by Provost
• One semester teaching buyout (Spring 05) followed
  by a sabbatical (Fall 05-Spring 06)
• Set of responsibilities included
• Developing Healthcare Technical Assistance
  Program for Indiana hospitals
• Developing research projects with strategic
  partners

15
Healthcare Technical Assistance

• Established May 2005
• Partners
• Indiana Hospital Health Association
• Purdue Schools
• Nursing
• Industrial Engineering
• Pharmacy
• Statistics
• Technology

16

• Project Locations
• 30 projects with 17
• Healthcare providers
• Cost 10,000 to 30,000

17
30 Projects in16 Months

• Continuous improvement (7)
• Process improvement (7)
• Patient flow (6)
• Facility planning (6)
• Med safety (3)
• Pandemic flu planning (1)

18
RCHE Research Collaboration

• IUMG Wishard Hospital
• clinical scheduling
• emergency room modeling and analysis
• IUMG Kenya Program
• food distribution for HIV patients in East Africa
• Roudebush VA Hospital Center of Excellence
• systems design for care of complex patients
• clinical scheduling, pharmacy analysis
• St. Vincent / Indiana State Department of Health
• surgical instrument sterilization
• evaluating emergency pandemic flu plans
• planning and design of alternate care sites
• Ascension Health
• hospital patient flow

19
Blue Print for Effective Patient Flow in
Ascension Health Hospitals Mark Lawley, Principal
Investigator Mike Criswell, Professor of
Nursing Doug McWilliams, Professor of
Technology Scott Lambert, Ascension Health Gayle
Trupiano, Ascension Health Imran Hassan, Renata
Kopach, Research Assistants
20

• Hospitals are large, complex, poorly understood
  health delivery systems.
• Can we understand the current workload and
  resource limitations (current state) and make
  some prediction about how these will evolve over
  the short term?
• Why is this important? Operational Decision
  Making
• Short-term staff scheduling
• Surge capacity deployment
• Emergency divert and accelerated patient
  discharge
• Room preparation and patient transport priorities
• OR scheduling for elective surgeries

21
Modeling Tool Petri Nets

• Mathematical modeling tools that captures
  operational dynamics of discrete event system
• Supports modularity and abstraction
• Supports system simulation
• Supports formal analysis for operational
  properties such as boundedness, liveness,
  reversibility, fairness, etc.
• Petri Nets have a rich research literature

22
What is a Petri Net?

• Pp1,p2,p3,p4 Tt1,t2,t3,t4
• F(p1,t1) (p2,t2) (p3,t3) (p4,t4)
  (t1,p2)(t2,p3)(t2 p4) (t3,p1) (t4,p2)

23
System State Tokens and Net Marking
Mo(M(p1), M(p2), M(p3), M(p4))(1,1,2,1)
24
How does the Petri Net work?

• Transitions represent executable events
• Transition Enabling Rule A transition, tj, is
  enabled if for each input place, pi, M(pi)? wij
• Transition Firing Rule To fire enabled
  transition, tj, remove wij tokens from each input
  place, pi, and place wjk tokens in each output
  place, pk.
• State Equation Mnew Mold Cet

25
Initial State (1,1,2,1)
26
Fire t2, New State (1,0,3,2)
27
Fire t3, New State (2,0,2,2)
28
Fire t1, New State (1,2,2,2)
29
Fire t4, New State (1,3,2,0)
p1
1
1
p2
p3
t1
t2
t3
2
1
1
1
p4
t4
1
2
1
Initial state (1,1,2,1) Firing sequence (t2 t3
t1 t4) Final state (1,3,2,0)
30
System State Evolution State Equation
31
Qualitative Analysis Make mathematically precise
statements about what the system will and will
not do.

• Liveness Will the system continue to run?
• Boundedness Will the system overflow?
• Reversibility Will the system return to
  desirable states?
• Reachability Can good states be reached from
  the current state? Can bad states be reached
  from the current state?

32
Example Boundedness
p1
t1
t3
p2
p4
p3
t2
t4
33
Modeling with Petri Nets

• PN support these basic modeling notions
• Event Sequencing
• Event Conflict (Choice)
• Event Concurrency
• Event Synchronization
• Locality Principle

34
Event Sequencing
35
Event Conflict
36
Event Concurrency
t1
t2
37
Event Synchronization
38
Locality Principle PN models can be locally
modified without altering the global logic of the
model.
39
Petri Net Flow Models ED
40
Petri Net Flow Models ED Fast track
41
Petri Net Flow Models Outpatient OR
42
Petri Net Flow Models Med Surg
43
Additional Petri Net Flow Models

• Additional departments to model include
• Intensive care
• Pharmacy
• OBGYN
• Pediatrics
• Radiology

44
Hospital Modeling Now What?

• Suppose by merging departmental and patient
  models we can get a valid Petri Net model of the
  hospital. Then what?
• Initial Research Question Can we use model to
  estimate likelihood hospital will go on ED divert
  in the next 24 hours? (YES!)
• Can we develop operational policies that help a
  hospital avoid divert? (YES!)

45
Clinical scheduling Mark Lawley, Professor of IE
Kumar Muthuraman, Professor of IE
46
Clinical scheduling

• Scheduling method for improving clinic revenue
  subject to sequential construction, no-show
  behavior, patient waiting costs, and staff
  overtime costs.
• Sequential schedule construction
• Scheduler waits for call
• Patient call arrives
• Scheduler assigns a slot to the patient
• We develop a method that assigns patient to a
  slot to maximize revenue of the current
  schedule
• Overbooks based on patient no-show probabilities
• Myopic in that it does not consider the arrival
  pattern of future calls

47
Slots and overflow
48
Scheduling notation
49
Scheduling algorithm
50
Unimodal revenue function
51
Example schedule evolution
52
Tentative research agenda

• Clinical scheduling
• Developing non-myopic approach
• Implementation at Wishard Clinic
• Optimal decision policies for hospitals
• State-based modeling
• Optimal divert and discharge
• Care of complex patients
• Scheduling diabetic patients
• Care coordination for complex patients
• Emergency planning for healthcare
• Surge capacity, alternative care sites
• Collaboration models for competitors

53
Challenges

• Making contacts and gaining access
• Communicating with providers
• Meeting immediate needs
• Keeping students funded
• Learning a new application area

54
Questions, comments, criticisms?
55
Surgical Instrument Processing Mark Lawley,
Principal Investigator Charles Spry, Research
Assistant
56

• Project Goals
• Redesign instrument sterilization and case cart
  delivery systems
• Analyze use of
• Equipment How many machines?
• Space How best to use storage space?
• Staff How to schedule staff hours?
• Cart / tray flow Is there sufficient time and
  space?
• Volume Can system handle expected growth?

57
St. Vincent OR Overview
58
Simulation Screenshot
ORs, case carts and instrument trays
Elevator (Up) Cartveyours (Down)
Clean room processes
Decontamination room processes
Prepared case cart storage
Instrument trays
59
Outcomes

• Research-Based Solutions
• Additional washers and autoclaves not yet needed
• Increasing autoclave size not helpful
• Planned elevator capacity is adequate
• Decon staff levels are sufficient
• Need additional clean room staff

• Intuitive Solutions
• Buy additional autoclaves and washers
• Increase autoclave size

60
Condition Based Scheduling System for a
Population of Diabetic Patients

• Diabetes is 6th leading cause of death in US
  (National Center for Health Statistics)
• 18 million people above age 20 suffered from type
  2 diabetes in year 2002
• 1.3 million new cases are diagnosed in US each
  year
• As US population ages, prevalence of diabetes
  will increase and the demand for quality care
  will intensify

61
A Condition Based Scheduling Systemfor a
Population of Diabetic Patients

• Diabetic patients periodically visit medical
  facilities for routine checkups
• No standard guidelines on the frequency of these
  visits
• Little known about how patients condition should
  affect the visit interval
• Too frequent visits- wasted clinical resources
• Too infrequent visits- deteriorating health and
  overuse of emergency and hospital resources

62
A Condition Based Scheduling for a Population of
Diabetic Patients

• Research Questions
• Can we develop an optimal scheduling interval
  based on patients condition?
• Can we develop a condition based scheduling
  system that optimally manages the health of a
  population of diabetic patients with a given set
  of clinical and medical resources?

63
Architecture of a Condition Based Scheduling
System
64
What needs to be done?

• Fundamental research problems
• Operational decision making, staffing, task
  priorities, optimal divert policies and discharge
  policies
• Tracking and sensing technologies
• Surge capacity planning
• Documenting and avoiding medical errors
• Flow-based facilities planning and design
• Outpatient clinic operations, scheduling,
  overbooking
• Streamlining financial transactions
• Integrated information systems and electronic
  medical records
• Performance measures