4
Examples of Operational Systems Engineering Applications Relevant to Traumatic Brain Injury Care1

William P. Pierskalla


Operational systems engineering (OSE), which combines science and mathematics to improve the operations of systems and enterprises that provide goods and services, entails describing, analyzing, planning, designing, and integrating complex systems. Operational systems engineers use deterministic and probabilistic mathematics (called stochastic processes) to describe how systems operate, design systems based on those descriptions, and integrate all elements of systems operations, including people, processes, materials, equipment, and facilities, to improve efficiency and effectiveness.

The tools, techniques, and concepts of OSE have been applied to many different areas of health care (NAE/IOM, 2005):

  • Epidemiology, health promotion, disease prevention, and predictions of the incidence, prevalence, and mortality of diseases. OSE has been used to evaluate intervention strategies, disease-control programs, and screening programs.

  • Health-care and health-systems design, including estimates of future resource needs and the deployment of those resources.

1

This chapter is based on the author’s presentation and responses to questions during the plenary session of the NAE-IOM workshop on Harnessing Operational Systems Engineering to Improve Traumatic Brain Injury Care in the Military Health System on June 11, 2008. The author would like to thank NAE staff member Proctor Reid for his assistance in preparing this material for publication.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 49
4 Examples of Operational Systems Engineering Applications Relevant to Traumatic Brain Injury Care1 William P. Pierskalla Operational systems engineering (OSE), which combines science and mathematics to improve the operations of systems and enterprises that provide goods and services, entails describing, analyzing, planning, designing, and integrating complex systems. Operational systems engi- neers use deterministic and probabilistic mathematics (called stochastic processes) to describe how systems operate, design systems based on those descriptions, and integrate all elements of systems operations, including people, processes, materials, equipment, and facilities, to improve efficiency and effectiveness. The tools, techniques, and concepts of OSE have been applied to many different areas of health care (NAE/IOM, 2005): • Epidemiology, health promotion, disease prevention, and predic- tions of the incidence, prevalence, and mortality of diseases. OSE has been used to evaluate intervention strategies, disease-control programs, and screening programs. • Health-care and health-systems design, including estimates of future resource needs and the deployment of those resources. This chapter is based on the author’s presentation and responses to questions during 1 the plenary session of the NAE-IOM workshop on Harnessing Operational Systems Engi- neering to Improve Traumatic Brain Injury Care in the Military Health System on June 11, 2008. The author would like to thank NAE staff member Proctor Reid for his assistance in preparing this material for publication. 

OCR for page 49
0 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE OSE has been used to design service systems (e.g., kinds of services offered, technologies, site/location selection); optimize capacity planning for health-delivery facilities; select approaches to business/enterprise planning; and design emergency services. • Support of medical decision making. OSE has been used for pre- vention programs, modeling of diseases, optimizing diagnostic tools, optimizing therapy programs and chronic care programs, and genetic modeling. • Operation of health care systems. OSE has been used to develop and implement appointment systems and waiting times for outpatients/inpatients; determining staff levels and scheduling; conducting inventories, material requirements planning, supply chains; forecasting short- and long-term demand; planning aux- iliary services; and evaluating medical technologies. In this chapter, we review five examples of OSE modeling used to address challenges to health care and health care systems relevant to the major issues in traumatic brain injury (TBI) care and management. Four of these examples were presented and discussed during the workshop; the fifth, which was mentioned briefly in the plenary discussion, is included because of its direct relevance to TBI care. The first example is based on research by Hazen (2004) on using dynamic influence diagrams in medical decision making. The term “dynamic influence diagram” applies to a process with a stochastic (i.e., random or probabilistic) component. Bayesian networks, a subset of dynamic influence diagrams, are a means of graphically representing the relationship between a set of variables and their varying probabilistic interdependencies. The second example involves strategies for screening blood for the presence of the human immunodeficiency virus (HIV) antibody (Schwartz et al., 1990). This example shows how OSE can address the conditional sensitivities and specificities of testing, which are directly applicable to diagnosing and treating patients with mild TBI (mTBI). The third example illustrates an approach to policy decision model- ing of the costs and outputs of medical-school education, a large com- plex problem with both economic and policy ramifications, as well as strategic and operational issues (Lee et al., 1987). The fourth example illustrates the large-scale application of an OSE simulation to a geographically dispersed health care delivery enterprise

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS (Bonder, 2005). The model was built for the Military Health System (MHS) during the late 1990s to address a large number of issues related to capacity, organization, resource allocation, and process change. The final example illustrates the direct application of OSE model- ing to the management of TBI patients in the Department of Veterans Affairs (VA) medical system. ExAMPLE 1 DyNAMIC INFLuENCE DIAgRAMS FOR MEDICAL DECISION MAkINg Influence diagrams, which have been used for modeling in decision analysis for some time, facilitate the analysis of sequential decisions. In Hazen’s paper, “Dynamic Influence Diagrams: Applications to Medical Decision Making,” he uses dynamic influence diagrams to structure and analyze a continuous chain of decisions related to whether or not a patient should proceed with total hip replacement surgery in a context in which back-stepping loops are possible (Hazen, 2004). In other words, once a decision is made, things might happen that require revisiting that deci- sion and then moving forward again. Thus a dynamic influence diagram models a looping, continuous, recycling decision process (Figure 4-1). In this example, Hazen was dealing with a simple decision between two choices—whether a patient should elect to have a total hip arthro- plasty (THA) (i.e., total hip replacement) or should opt for conservative management (i.e., no medical intervention). The purpose of the model is to calculate the optimal expected quality-adjusted lifetime for both choices. Although many variables can be included in the model, such as age, sex, mobility, and/or other functional, social, demographic, and racial characteristics of the patient, the main variables in this example are race, age, and sex. The operation of the model begins when the patient is given a diagnosis of American College of Rheumatology (ACR) Class III osteo- arthritis. At that point, a decision must be made by patient and doctor as to the appropriate therapy. Following the therapy, sooner or later, the patient will either transition to one of the other three ACR classes, will generally deteriorate, or will die (Box 4-1). The transition to another ACR class or to death is modeled based on certain probabilities, depending on the characteristics of the patient

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE Background Quality mortality of life ACR THA vs Initial THA functional conservative outcome status management Revision count OA progression under conservative management Prosthesis status after THA Infection failure count Aseptic failure count Infection failure rate Aseptic failure rate THA = total hip arthroplasty Identity of ACR = American College of Rheumatology last surgery OA = osteoarthritis FIguRE 4-1 A dynamic influence diagram for a model of the choice between THA and conservative management. Source: Adapted from Hazen, 2004. and, possibly, the characteristics of the surgical procedure, such as the Figure 4-1.eps hospital where it was performed and the experience of the surgeon. Although the characteristics of the surgery are not shown in this model, they could easily be included. After being reclassified in the new ACR class, the patient may re- main in that class for some time and then, later, transition to another class via infective or aseptic failure of the hip prosthesis or to death from another cause or to a general deterioration of his or her health. Considering all of these possibilities, which can happen randomly over time with certain probabilities, the prognosis of the patient’s future quality of life is determined by a reverse analysis from the last point in the patient’s life to the beginning point where the decision must be made as to surgery or conservative management. Also using reverse analysis,

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS BOx 4-1 American College of Rheumatology Functional Classifications for Hip Osteoarthritis Class Description I Complete ability to carry on all usual duties without handicap II Adequate for normal activities despite handicap of discomfort or limited motion in the hip III Limited to little or none of duties of usual occupation or self-care IV Incapacitated, largely or wholly bedridden or confined to wheelchair, little or no self-care Source: Hazen, 2004. the total lifetime costs of THA and conservative management can be calculated and minimized. Therefore, a patient with given characteristics has an expected quality of life depending on the decision made at the beginning point. Using the model to work back to that point, the best (optimal) decision can be chosen to maximize the remaining quality of life. If other types of therapy are available or become available in the future, they can be included in the model and analyzed as to their outcomes for expected quality of life. For example, cortisone shots or cartilage replacement might be considered in the model. In Hazen’s example, he used Cox stochastic trees to predict what the Cox progression to death and the stages would be for an 85-year- old white male and for a 60-year-old white female. A stochastic tree is a graphical modeling tool that allows the explicit factoring of temporal uncertainty—for example, age-dependent mortality rates—in a decision tree analysis. An 85-year-old white male has to choose one of two options— THA or conservative management. Using the model, if he chooses THA, his quality of life would be in ACR Class I for about 2.9 years on average. If Class II were the result of the THA, he would have 1.4 years in this class. If Class III were the result, he would have about

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE .06 years in this class. If he chose conservative management, he could expect to be in Class III for 3.9 years and then proceed to Class IV until death. THA would not give this patient more years of expected life than conservative management because of his advanced age—his life expec- tancy is about 4.4 years with either decision. However, the patient’s mean quality of life, when adjusted for Class I for quite a long time, Class II for a fair amount of time, and so on, would be significantly better than if he opted for conservative management. Hazen estimated that THA at the time this research was conducted (in the mid-1990s) would cost the 85-year-old $25,000 in fixed costs plus another $5,000 for costs associated with possible revision surgery, for a total of $30,000. With conservative management, he would spend $20,000 for home and/or nursing care prior to dying. By these criteria, for an additional $10,000, the patient would buy a better quality of life for about $5,000 per additional quality-adjusted life year (QALY), far less than the amount a person would spend for additional QALYs with a coronary artery bypass or a stent implant. So, the model shows that for this person, at age 85, THA at $5,000 per QALY is a good deal. Applying the same analysis for a 60-year-old white female consider- ing THA, her mean quality of life would be 9 years in Class I and 5 years in Class II (a total of 14 years of high quality) as opposed to 7 years in Class III. The total cost for THA would be about $54,000. However, if the patient decided to forego the surgery, the estimated cost, taking into account the attendant problems with her hip, medication, and long- term care, would be about $174,000 for care over her lifetime. In this case, she would actually save money by undergoing a THA. Relevance to TBI Care Management For an mTBI patient, decisions for therapies and the timing of treatments might be considered as a flow of the patient through differ- ent mTBI states. The transitions over time (considered probabilistically) could be based on repeated traumas in the field and by the characteristics of the treatment provided. Mild TBI, however, is more complex than THA; each injury is dif- ferent and results in different physical, psychological, and functional effects. The additional complexity can be indicated in the model with more branching nodes and arcs. Decisions and outcomes are expressed

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS stochastically. Patients and providers must then make continuing decisions about when, how, and what to do next. The model could begin with the field incident, the decisions made at that time, subsequent transfers and decisions, subsequent discharge and decisions, and so forth over time. Or the model could begin with a later discovery of mTBI and the flows of decisions, therapies, and outcomes (again probabilistically) from that point on. The model of- fers many combinations of timing and choices of therapies, locations, personnel, and modes of care delivery. It is also possible to use different or multiple criteria to model the decisions that would optimize a patient’s predicted quality of life. In the THA example, the criteria were (1) to maximize the quality of lifetime remaining and (2) to minimize the costs to the individual and/or society. Similar or different criteria could be used to optimize mTBI treatment plans, locations, and/or personnel. ExAMPLE 2 SCREENINg BLOOD FOR THE HuMAN IMMuNODEFICIENCy vIRuS ANTIBODy The development and application of a decision analytic model to examine alternative strategies for screening blood for the HIV antibody and making decisions affecting blood donors was presented as having strong parallels to testing for mTBI and making decisions about next steps for wounded soldiers (Schwartz et al., 1990). At the time this work was done, in the mid 1980s, limited knowl- edge was available about the biology, epidemiology, and early blood manifestations of HIV. Furthermore, the initial and conditional sen- sitivities and specificities of enzyme immunoassays (EIA) and western blot (WB) tests had wide ranges of error. Finally, nothing was then known of the effectiveness of registries, counseling of donors, self- reporting of donors’ sexual and drug-injection activities, or related educational programs. The purpose of the model was to determine which screening tests to use, and in what sequence, to minimize the number of HIV-infected units of blood and blood products entering the nation’s blood supply at an acceptable cost. The model is a Bayesian decision-tree model, and the decisions are probabilistically based.

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE The variables in the model are: available screening tests, the initial and conditional ranges of sensitivities and specificities of the tests at different levels of HIV (based on time since incidence), the costs of administering tests, the costs of informing donors of positive results, the costs of maintaining a donor registry of positives, and the prevalence and incidence of HIV in different donor populations (Figure 4-2). Given a unit of blood to test, a decision must be made as to which screening test to use initially (an EIA or the WB test). Based on the results of the first test, a decision must then be made to accept the unit into the blood supply (test negative) or to conduct another test (initial test positive) to determine (still probabilistically) if the positive result is correct. Following the second test, subsequent decisions are made: con- duct a third test; accept the unit into the blood supply; reject the unit but do not inform the donor; reject the unit and inform the donor; enter the donor in a registry of individuals whose blood will not be accepted into the blood supply in the future. Although more than three tests might be conducted, the first two provided enough information to make informed decisions based on the conditional probabilities of the presence or absence of HIV. The model provides the following outputs: • the expected number of infected units entering the blood supply during a specific period of time (e.g., a year) • the expected number of uninfected units (good) discarded (i.e., wasted) during that period of time Donor Test Notification subgroup 1 1 1 Transfuse 2 2 2 Discard Removed Retest from Donor 3 3 3 donor pool pool 4 5 8 15 FIguRE 4-2 The decision support model for HIV antibody testing of blood and plasma donors. Source: Schwartz et al., 1990. Copyright © 1990 American Medical Association. All rights reserved. Figure 4-2.eps

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS • the expected number of uninfected donors falsely notified • the expected cost per donated unit for the particular screening regimen • a wasted-unit index (the expected number of uninfected units discarded divided by the number of infected units discarded) • the incremental cost and incremental number of wasted units for different screening regimens This model was used at a meeting of an expert panel of the National Heart Lung Blood Institute to inform panelists who were deciding which blood-screening regimen to use at all blood-screening centers in the United States. The model provided the panelists with information and a basis for comparing screening regimens. A collection of 15 screening tests was pared down to eight tests that could be used to detect HIV. The goal was to identify a sequence of tests that could reliably differentiate a unit of blood with HIV from a unit of blood without HIV. Given the results of a test, a decision was made to use, discard, or retest the unit. Decisions about notifying the donor were made next. If the donor was not notified, then he or she would be put back into the eligible donor pool. If the donor was notified of a possible HIV infection, he or she would be asked not to donate again, and his or her name might be placed in a donor registry of individuals who might have HIV infections. The prevalence of the disease in high-prevalence areas, mainly in very large cities, was about 2.9 carriers in every 10,000 donors. In low- prevalence areas, the rate was about 1.6 carriers in every 10,000 donors. With an EIA test, initial sensitivity was about 0.98, with a range of 0.94 to 0.995. This range was entered into the model to find the scenarios that yielded the most reliable results. During the first few weeks of HIV infection, it is very difficult to detect the virus, and it rarely shows up on tests. None of the tests modeled was good at detecting the very early stage of the disease. The most effective test had a sensitivity of 0.6, with a huge range of 0.11 to 0.83 and a negativity of 0.2, with another huge range. The specificity for this test was fairly good, but the range was 0.96 to 0.996. The model also provided estimated costs for performing different sequences of tests and for donor notification. The model was used to examine various testing strategies with the objective of minimizing the risk per million donated units of accepting an HIV-contaminated unit of blood (Box 4-2).

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE BOx 4-2 Strategies for Screening Donated Blood for HIv Strategy 1: All units are tested with an EIA: EIA-negative units are transfused; EIA-positive units are retested twice with the same EIA. If both subsequent EIAs are negative, the unit is transfused. If either of the subsequent EIAs is positive, the unit is discarded and a WB is performed. Donors with WB-positive results are informed of the results and placed on a registry. Strategy 2: All units are tested with an EIA: EIA-negative units are transfused; EIA-positive units are tested with a WB. WB-negative units are retested with an EIA from a different cell line. All units posi- tive by the first EIA are discarded, but only donors who test positive by WB or negative by WB but positive by the second EIA are informed of the test results. Strategy 3: All units are tested with an EIA: All units are retested with the same EIA, regardless of the result of the first EIA. Units nega- tive by both EIAs are transfused. If either EIA is positive, the same EIA is performed a third time. Units negative by two of the three EIAs are transfused. Units positive by two EIAs are discarded and tested with a WB. WB-positive donors are informed of the test results. Strategy 4A: All units are tested with an EIA: EIA-negative units are transfused; EIA-positive units are retested with an EIA from a different cell line. If the second EIA is negative, the unit is retested again Under the strategy used at the time of the study (in the 1980s), about 20.5 units per million in high-prevalence areas would be con- taminated. In low-prevalence areas, about 4.7 units per million would be contaminated. In addition, almost 2,000 units of good blood from both areas would be thrown away because of false positives. The best strategy for reducing risk to the population was Strategy 5; in high-prevalence areas 2.4 units of contaminated blood would be accepted in every million units, and in low-prevalence areas there would be 0.3 units per million. However, this strategy would mean that a large volume of good blood would be wasted.

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS with the second EIA. Units negative by two EIAs are transfused. Units with two positive EIAs are discarded and tested with a WB. Donors with WB-positive results are informed of the results and placed on a registry. Strategy 4B: All units are tested with an EIA: EIA-negative units are transfused; EIA-positive units are retested with an EIA from a differ- ent cell line. If the second EIA is negative, the unit is retested again with the first EIA. Units negative by two EIAs are transfused. Units with two positive EIAs are discarded and tested with a WB. Donors with WB- positive results are informed of the results and placed on a registry. Strategy 5: All units are tested with EIAs from two different cell lines. Units negative by both EIAs are transfused. All units testing posi- tive by either EIA are discarded and tested with a WB. Donors with WB-positive results are informed of the results and placed on a registry. Strategy 6: All units are tested with EIAs from two different cell lines. Units negative by both EIAs are transfused. Units positive on only one EIA are retested with the EIA that was positive and tested with a WB. WB-positive units are discarded and the donor is informed of the test results. Units positive by two EIAs but negative by a WB are discarded, but the donor is not informed of the results. Units positive by only one EIA and WB-negative are transfused. Source: Schwartz et al., 1990. Copyright © 1990 American Medical Associa- tion. All rights reserved. Relevance to TBI Management An mTBI version of this model could begin with field incidence, decisions then made, transfers, decisions then made, discharge, deci- sions then made, and so forth over time. Or it could begin with a later discovery of mTBI and the flows of decisions and therapies and proba- bilistic results. To determine if a patient has an mTBI, he or she undergoes tests for, or may self-report, one or more of the following conditions or symptoms:

OCR for page 49
0 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE • a period of loss or altered level of consciousness • a loss of memory of events immediately before or after the injury • a change in mental state at the time of the injury (confusion, disorientation, slowed thinking, etc.) • transient or permanent neurological deficit(s) (e.g., weakness, loss of balance, change in vision, praxis, paresis/plegia, sensory loss, aphasia, etc.) • an intracranial lesion Although tests may indicate mTBI, each test has initial and condi- tional sensitivities and specificities. Moreover, because there is no “gold standard” or best-practice diagnostic test, false positives and false nega- tives are common. Follow-on tests, such as magnetic resonance imaging (MRI), functional MRI, and other noninvasive procedures, also have initial and conditional sensitivities and specificities. Because there is no best-practice therapy for the various levels and causes of mTBI, therapies also have a probabilistic chance of success or failure. In addition, the patient and family’s involvement and compli- ance in therapies may modify the probability of a successful outcome. In the framework of the model, decisions about therapies or treatments are like the staged decisions described above about accepting or rejecting a unit of blood and whether or not to inform the donor. The model is rich in possibilities for a decision-tree analysis of the timing and choices of tests, therapies, locations, personnel involved, and modes of care delivery (Figure 4-3). The model can also simulate different or multiple criteria to maximize the long-term quality of life for the patient and family and/or to minimize the long-term costs of treatment. ExAMPLE 3 POLICy DECISION MODELINg OF THE COSTS AND RESuLTS OF MEDICAL SCHOOL EDuCATION The purpose of this policy decision model was to support the deci- sion of a board of regents and state legislature on financing state-funded medical education and meeting the state’s long-term needs for physi- cians in rural and urban areas (Lee et al., 1987). Funding for medical

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS Testing and treatment decisions for mTBI patients Type and Tests to Choice of treatment magnitude of determine locations IED or incident mTBI Choice of therapy: Return to combat Removed from Hold for observation treatment pool: with modest Success therapy Persons involved in Failure Evacuate an IED or Self-removed Long-term stateside - other therapies incident FIguRE 4-3 The decision support model for testing and treatment decisions for mTBI patients. Source: Adapted from Schwartz4-3.eps Figure et al., 1990. bitmap with some vector boxed matter education in the state—which is not identified in the paper—was characterized as one of the highest in the country, although the state’s per capita income was relatively low. In addition, most graduating physi- cians and medical residents left the state for more lucrative employment elsewhere. The variables and parameters in the model were in-state, out-of- state, urban, and rural students and residents entering medical schools and hospitals in the state over time; all teaching resources necessary to educate medical students and residents (faculty, technology, facilities, and programs); financial costs of education; the practice locations and specialties of physicians who had completed their studies; the needs of citizens for physicians by location and specialty; and the goals and objec- tives of the regents and legislature for medical education in the state. Possible policy scenarios were developed from input received from the state’s board of regents (Box 4-3). The scenarios were evaluated with regard to how well each met the regents’ goals and objectives. Each evaluation was done using a decision-support model with four interrelated modules: (1) a teaching-resource model, (2) a financial-cost model, (3) a physician-output model from medical schools and resi- dency programs, and (4) a physician practice location and specialties model (Figure 4-4).

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE BOx 4-3 System goals Developed using Nominal group Technique with State Board of Regents 1. Develop an organized strategy for providing medical education in the state that is responsive to state needs. 2. Provide quality medical education within available resources. 3. Provide quality medical education in an efficient and effective manner. 4. Emphasize primary care training. 5. Provide a reasonable opportunity for state residents to obtain quality medical education. 6. Provide the appropriate number and type of physicians needed in the state, and encourage an appropriate demographic distri- bution of physicians. 7. Increase cooperation of M.D.’s and D.O.’s in education and services. 8. Improve the health of state residents. Source: Lee et al. 1987. Reprinted with permission from INFORMS. The teaching-resource model was a mathematical programming model. The financial-cost model was a detailed cost-accounting model. The physician-output model was a Markov-chain model. The physician practice location and specialties model was a stochastic forecasting model. Based on the analyses, the board of regents and the legislature passed laws and implemented the following decisions: • One of the three medical schools was privatized, and its state funding would be significantly reduced over time. • Tuition was increased for all students (a larger increase for out- of-state students). In addition, a no-interest, revolving loan fund was established for students who practiced medicine in the state for a certain number of years, and a loan-forgiveness program was

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS Goals and objectives Policy Evaluate scenario(s) implications Assumptions DECISION SUPPORT SYSTEM DATA INPUTS • Students Teaching Physician • Faculty resource output • Physicians • Programs OUTPUTS • Costs • Total costs Physician • Facilities Financial • State costs projection • Physician outputs FIguRE 4-4 Strategy development and evaluation process. Source: Lee et al., 1987. Reprinted with permission from Figure 4-4.eps INFORMS. established for students who remained in the state and practiced in rural areas for a certain number of years. Relevance to TBI Care Management The purpose of this model in the context of TBI care management is to illustrate that a modular OSE model can be used to evaluate vari- ous scenarios for achieving overall goals and objectives set by decision makers, even in complex health-delivery situations, such as those of the U.S. Department of Defense (DOD) and VA. The modular, interconnected model might explore, for example, how TBI care should be organized on a regional basis. TBI care involves resources and resource planning, finances and costs, people (caregivers, patients, and families), and movement and availabilities, as well as

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE forecasting of future needs and outputs. In short, TBI care involves many of the same elements as in the medical education model described above, including data needs of people, technologies, facilities, costs (fixed, semi- fixed, and variable), goals, objectives, and scenarios of interest. The overall TBI model might be laid out in much the same way as the education model (Figure 4-5) but with names, flows, and data specific to TBI. ExAMPLE 4 THE HEALTHCARE COMPLEx MODEL The Healthcare Complex Model was developed for MHS in the 1990s and has been continually updated by Vector Research Inc. (now the Altarum Institute) in a prototyping process. The purpose of the model was to help MHS address a large number of capacity, organiza- tional, resource allocation, and process change issues. Goals and objectives for TBI patients Policy and Evaluate strategy Implications scenarios Assumptions DECISION SUPPORT SYSTEM DATA INPUTS •TBI patients OUTPUTS In-field Out-of-field DOD •Physicians diagnoses and treatment and •Treatment best practices treatment model locations model •Nurses •Treatment locations •Therapists •TBI patients treated Financial/cost Forecasting model •Families model for future TBI •Resources needed patients •Finances •Costs •Technologies •Facilities •Locations FIguRE 4-5 Decision-support model for developing a strategy and evaluation process Figure 4-5.eps for TBI care management. Source: Adapted from Lee et al., 1987.

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS As shown in Figure 4-6, the simulation represents a patient episode, from initial diagnosis through further diagnoses, treatments, monitor- ing, rehabilitation, and so on. The model shows the network of facilities at one or two major medical centers, a group of 5 to 10 hospitals, and multiple clinics. It also shows the dynamic input of patients who enter the system through clinics, hospitals, or major medical centers. Patients can be referred anywhere in the system and receive care by visiting a facility or care provider or by telemedicine. The model simulates the dynamic flow of patients through a health care system as they are being diagnosed, treated, and rehabilitated. Their movement through the system, and even outside the system, is moni- tored throughout the episode. The simulation includes about 60 providers and is driven by care protocols, in this case more than 1,200 ICD-9 code conditions (e.g., asthma, open-chest wound, low back pain) and roughly 1,400 to 1,500 peacetime disease-condition protocols. About 400 to 500 differ- ent protocols for deployment medicine, and several for dental care, move patients through the system as they receive treatment via the protocols and exit the system. The protocols involve care providers, ancillary personnel, HCM SIMULATES PATIENT EPISODES IN A NETWORK OF FACILITIES... Healthcare “Complex” Pa t ie nt s Pat Population s ie rn nt s e at t lP rra fe Re Telem Population Population Patients e dici ne • Providers Pat ie nt s • Care Protocols • Ancillary Personnel • Ancillary Resources • Information Technology Population ...TO EXPLORE THE IMPACTS ON COST, ACCESS, AND QUALITY FIguRE 4-6 Overview of the Healthcare Complex Model. Source: Bonder, 2005. Figure 4-6.eps

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE and other resources and services, such as laboratories, x-rays, ambulatory surgery, pathology, pharmacy, and information technology. This kind of model can address many strategic and tactical ques- tions. For example, strategic questions might range from how many hospitals and/or clinics are required to the details of the kinds of pro- tocols that should be tested. With a model of this kind, one can experi- ment with changes to a particular health care delivery system, as was done when it was applied to the Madigan Army Medical Center health care enterprise, which involved a simulation of a full year of health care delivery services at the facility (Chin et al., 2007). When the results were tested against the actual data, the dynamics were very, very close to those of the real-world system. This simulation demonstrated that OSE techniques can, in fact, simulate a large-scale enterprise of health care delivery to address a whole spectrum of issues. ExAMPLE 5 A MIxED-INTEgER PROgRAMMINg MODEL TO LOCATE TRAuMATIC BRAIN INJuRy TREATMENT uNITS IN THE vA Although the model described in this section was only mentioned briefly at the workshop, it represents a direct application of OSE to the management of TBI patients in the VA medical system. It is summarized here for illustrative purposes. Côté and colleagues (2007) developed a mixed-integer program- ming “optimization” model to help the VA decide where to locate TBI treatment units in the existing system of VA medical centers. The goals were to minimize total patient costs for treatment, lodging, and travel, as well as costs associated with missing targets for the use of the services provided by the facilities. The many variables and parameters in the model include the loca- tion of medical centers, patient locations and travel distances to medical centers, patient attendance attrition due to the distance and cost of travel and lodging, TBI severity levels, lengths of stay by severity, the size of medical centers, as well as their capacities and capabilities for TBI treat- ment, and the costs associated with these variables. The model was designed to help with treatment location decisions in the six Florida-based VA medical centers in Veterans Integrated Services

OCR for page 49
 ExAmPlES of oPErAtioNAl SyStEmS ENgiNEEriNg APPliCAtioNS Network Region 8 (VISN 8). Many scenarios were evaluated, such as opening one, some, or all six of the medical centers for TBI treatment at various levels. Each scenario was evaluated on the basis of admission retention rates, severity levels, and numbers of TBI patients treated at each center, expected treatment costs, expected lost-admission penalty costs, expected capacity penalty costs, and expected lodging and travel costs. The results show that this model was useful for locating TBI treat- ment units, as well as for planning the development or consolidation of rehabilitation programs. The results and system outputs were extremely sensitive to the management structure and environment of the TBI treatment units in VISN 8 and suggested that careful consideration be given to the centralization of health care facilities and admission retention rates, as well as to interactions among these factors, when making decisions concerning the location of treatment facilities. VA management could use the results to make informed decisions about the number, location, capacities, personnel, and costs of opening facilities throughout the region to treat TBI patients, with or without family members in attendance. CONCLuSION The five examples of OSE modeling discussed in this chapter illus- trate the types of problems and objectives that quantitative OSE tools and methods have been used to address in a variety of health care-related settings. The examples showcase a number of technical approaches and how they are structured and illustrate data needs, critical assumptions, constraints, and metrics for evaluating performance using an OSE analysis. The following chapter provides a detailed case study of the develop- ment, implementation, and sustainability of “system-supported” clinical practice at a major academic medical center—an approach that involves the integration of OSE tools and methods with information technology and a team-based approach to care delivery. REFERENCES Bonder, S. 2005. Changing Health Care Delivery Enterprises. Pp. 149–152 in Building a Better Delivery System: A New Engineering/Health Care Partnership, edited by

OCR for page 49
 SyStEmS ENgiNEEriNg to imProvE trAumAtiC BrAiN iNjury CArE P.P. Reid, W.D. Compton, J.H. Grossman, and G. Fanjiang. Washington, D.C.: The National Academies Press. Chin, S., R. Bannick, and A. Kress. 2007. Military Health System (MHS) Beneficiary Ac- cess to Trauma Centers. Available online at http://gis.esri.com/liBrAry/uSErCoNf/ HEAltH0/DoCS/militAry_HEAltH.PDf. Côté, M.J., S.S. Siddhartha, W.B. Vogel, and D.C. Cowper. 2007. A mixed integer program- ming model to locate traumatic brain injury treatment units in the Department of Veterans Affairs: a case study. Health Care Management Science 10(3): 253–267. Hazen, G.B. 2004. Dynamic Influence Diagrams: Applications to Medical Decision Mak- ing. Chapter 24 in Operations Research and Health Care, edited by M. Brandeau, F. Sainfort, and W. Pierskalla. Boston, Mass.: Kluwer Academic Publishers. Lee, H., W.P. Pierskalla, W.L. Kissick, J.H. Levy, H.A. Glick, and B.S. Bloom. 1987. Policy decision modeling of the costs and outputs of education in medical schools. Operations Research 35(5): 667–683. NAE/IOM (National Academy of Engineering/Institute of Medicine). 2005. Building a Bet- ter Delivery System: A New Engineering/Health Care Partnership, edited by P.P. Reid, W.D. Compton, J.H. Grossman, and G. Fanjiang. Washington, D.C.: The National Academies Press. Schwartz, J.S., B.P. Kinosian, W.P. Pierskalla, and H. Lee. 1990. Strategies for screening blood for human immunodeficiency virus antibody. Use of a decision support system. Journal of the American Medical Association 264(13): 1704–1710.