NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Academy of Engineering (US) and Institute of Medicine (US) Committee on Engineering and the Health Care System; Reid PP, Compton WD, Grossman JH, et al., editors. Building a Better Delivery System: A New Engineering/Health Care Partnership. Washington (DC): National Academies Press (US); 2005.

Cover of Building a Better Delivery System

Building a Better Delivery System: A New Engineering/Health Care Partnership.

Show details

Changing Health Care Delivery Enterprises

Seth Bonder

The Bonder Group

The health care delivery (HCD) system in the United States is in crisis. Access is limited, costs are high and increasing at an unacceptable rate, and concerns are growing about the quality of service. Many, including the Institute of Medicine, believe the system should be changed significantly in two ways: (1) HCD enterprises should be reengineered to make them more productive, efficient, and effective; and (2) substantially more effort should be devoted to a strategy of prevention and management of chronic diseases instead of the current heavy reliance on the treatment of diseases. Although operations research can make substantial contributions to both areas, the focus of this paper is on: (1) reengineering HCD enterprises, particularly areas in which operations research can provide valuable support to senior health care managers; and (2) enterprise-level HCD simulation models to determine the reengi-neering initiatives with the biggest payoffs before implementation.

HCD enterprises are very large, complex operational systems comprised of large numbers of people and machine elements. Tens of thousands of people are involved as providers, patients, support staff, and managers organized into specialties, departments, laboratories, and other organizations that are considered independent service units (“stovepipes”). Machines include durable medical equipment, information technologies, communications equipment, expendable supplies, rehabilitation equipment, and so on. These elements are affected by many clinical and administrative processes (e.g., arrivals, testing, diagnosis, treatment, scheduling, purchasing, billing, recruiting, etc.), most of which are probabilistic (i.e., uncertain) and change significantly over time.

Perhaps most important, these processes involve large numbers of interactions within units, among units, and across processes. Decisions by enterprise managers regarding one unit may have second, third, and fourth order effects, which may be more significant than the first order effect. HCD enterprises are driven by endogenous and exogenous human decisions made by providers, patients, insurers, administrators, politicians, government employees, and others. Demand and supply issues have complex feedback effects. A great many resources are required for the development and operation of an HCD enterprise. For example, the University of Michigan's budget for its HCD enterprise is more than $1 billion; the Henry Ford Health System's budget is $2.5 billion, and these are relatively small HCD enterprises. Billions of dollars have been spent on cost containment initiatives over the past 15 years by the Agency for Healthcare Research and Quality (formerly the Agency for Health Care Policy and Research), the U.S. Department of Defense, the Veterans Administration, National Institutes of Health, foundations, universities, and others to reengineer the HCD system. Nevertheless, costs continue to rise at double-digit rates.

We need better ways of analyzing systems of this magnitude. The operations research community has been involved with HCD enterprises for more than 40 years working on a wide range of problems, such as inventory for perishables; management of intensive care units; laboratory and radiology scheduling; relieving congestion in outpatient clinics; nurse staffing, scheduling, and assignments; and layouts for operating and emergency rooms. These efforts have focused on the small, stovepipe units, referred to by Don Berwick as clinical and support “microsystems,” and have produced some useful information for unit managers but have not addressed enterprise-level reengineering and planning issues (the so-called “macrosystem”). Macrosystem issues have interactive effects across the enterprise and have large cost, access, and effectiveness impacts. Some of these interrelated issues are listed below:

  • the mix of health services necessary to support a given population
  • the staff required (e.g., specialties, numbers, locations) to provide necessary services
  • the impacts of changing demands (e.g., aging populations, effects of preventive measures)
  • the impacts of new HCD models (e.g., home health care, task performance substitution)
  • the effects of centralized radiology services
  • the impacts of primary care outreach
  • facility capacity for the next 20 years and the best way to provide it
  • operational changes to adapt to regulatory changes (e.g., Medicare)

These and other macrosystem issues can be addressed quantitatively using enterprise-level simulation models that represent all of the elements, units, and processes in the enterprise as well as the interactions among them. Because analyses of these issues are necessarily prospective, the models must be structural rather than statistical. Statistical models, which are usually used in economics and the social sciences, use existing system data to develop aggregated statistical relationships between system inputs and outputs (i.e., the model). Statistical models are used primarily retrospectively, that is, for making inferences and evaluations. In contrast, structural models are usually developed in the engineering and physical sciences by modeling the detailed physics of each process and activity. Structural models are used prospectively, that is, for predictions and planning. Statistical models are less appropriate to prospective analyses of future systems because the data used to develop statistical models are intrinsically tied to the existing system.

Figure 1 provides an overview of a particular enterprise-level HCD simulation model. The figure shows the elements in the Healthcare Complex Model (HCM), which was developed seven years ago and has been continually updated in a prototyping process by Vector Research Incorporated (now the Altarum Institute). HCM simulates individual patient episodes in a network of facilities for a population of patients. The network of facilities, with its entities and processes, is referred to as a “complex” (synonymous with an enterprise). Complexes usually have one or two major medical centers (where much of the tertiary care is provided), five to ten hospitals, and many clinics. The model can be adapted to represent specific features of any HCD enterprise.

FIGURE 1. Overview of the Healthcare Complex Model.

FIGURE 1

Overview of the Healthcare Complex Model.

Inputs to the model include demographics of the population that receives care. A model preprocessor converts the demographics into a stream of patients entering the complex; each patient's condition is described by an International Classification of Diseases, ninth edition (ICD-9) code. Patients can enter the enterprise at a clinic, a hospital, or a medical center. They can be referred physically or via telemedicine consults from clinics to hospitals or to a medical center. Providers of various types are located at each facility in the complex. The care protocols represent practice guidelines and patient pathways, define what service patients receive next, where patients receive the service, and the type of personnel who will provide it. The model keeps track of the resources used and estimates costs using related cost models. Each protocol is a tree with many probabilistic branches to simulate that different providers may provide patients having the same condition with different medical services. The care protocols may be tailored for simulations of specific enterprises and facilities. The model represents various ancillary personnel (e.g., nurses, nurse assistants, medical technicians, etc.) and various ancillary resources (e.g., laboratories, pharmacies, beds, CAT scans, MRIs, and durable medical equipment). Finally, the model represents various clinical (e.g., computerized patient record system) and administrative (e.g., billing, scheduling) information technologies and communications systems.

Because the HCM explicitly simulates all of the entities, processes, and activities in the system, any one or combination of them can be changed, and the impact on various output costs and access metrics can be observed. For example, HCM can determine how a change affects the cost of running the enterprise, a hospital, or a particular unit in a hospital. It can calculate the impact on access metrics for the enterprise, a hospital, or a unit in a hospital. Because the model is being enhanced continually via a prototyping process, consideration has been given to simulating false positive and false negative statistical errors and their effects. Although these are not outcomes, they would provide useful quality information about the simulated HCD enterprise.

HCM has reasonable fidelity at this stage in its development. It contains more than 1,200 ICD-9 code conditions (e.g., acute appendicitis, asthma, cellulite, open chest wound, viral hepatitis, low back pain, etc.) and more than 1,500 clinical tasks/procedures (e.g., preoperative anesthesia, computer tomography for staging/radiation, EEG, interpretation of angiogram, administration of antibiotics, etc.). The model simulates 60 different kinds of health care providers, 17 types of ancillary resources (e.g., x-ray, ultrasound, pathology, dialysis unit, etc.), 6 different inpatient beds, and 23 combinations of telemedicine equipment. And its fidelity improves with every study.

The model was tested on one of the smaller regional HCD enterprises in the military health system (MHS). The enterprise has one major medical center, two hospitals, two clinics, and a managed care support contractor that provides additional capacity for the region. Together they handled about 1.6 million outpatient visits in fiscal year 1999. The model was adapted to represent the facilities, workforce, ancillary resources, information technologies, and clinical protocols used by the regional complex. Using population demographics provided by the government, regional operations for the year 1999 were simulated a number of times (because of the probabilistic nature of the protocols) to develop stable average outputs. These were compared to the historical values from the enterprise's 1999 operations with encouraging results. Total outpatient visits differed by 0.11 percent, same-day surgeries by 1.02 percent, inpatient admissions by 2.99 percent, emergency room visits by 6.04 percent, and average length of stay by 0.94 percent. More detailed comparisons of outpatient visits by individual facility and individual specialty all differed by less than 4 percent. Although this was not a true validation study (which would require implementing model-suggested changes and comparing predicted impacts with actual results after the changes), it did show that simulation models can represent the complex dynamics of health care enterprise operations and can generate useful information and insights for enterprise managers.

HCM has been used in a number of other studies including the geographic distribution of primary care providers for a large, dispersed enterprise; telemedicine needs for a MHS regional complex; centralization of radiologists to service a 20-facility enterprise; and determining return-on-investment for information technologies. HCM is currently being used to determine capacity requirements for an enterprise that would experience increased demand following a bioterrorist attack.

Enterprise-level simulation models like HCM can be used to address a broad range of issues facing enterprise executives. Here is one challenge that could be posed: Given a population of patients, how can operations research determine an efficient set of resources to provide an acceptable level of services to that population. Assuming the HCD enterprise is a shell with no existing medical services, models like HCM can be used to address difficult issues, such as designing a system from scratch to serve a given population (sometimes referred to as “zero-based” design). A schematic drawing of the analysis process is shown in Figure 2. For purposes of this discussion, we assume that an acceptable level of service can be defined in terms of some access/quality metrics, cost of enterprise operations, and cost of the resources.

FIGURE 2. Zero-based HCD enterprise design.

FIGURE 2

Zero-based HCD enterprise design.

The resources required to service the specified population depend not only on characteristics of the population (e.g., conditions, prevalence, incidence, etc.), but also on the protocols, as well as the degree to which the enterprise strategy for servicing the population focuses on treatment or prevention/ management of medical conditions. The three-dimensional structure shown on the right side of Figure 2 allows the analysis team to select a population, a protocol set, and a mixed treatment/prevention strategy as input to the analysis process. (The protocols are obviously related to the strategy and designed to reflect the strategy.) Figure 2 shows that input (1, 2, T), representing population 1, protocol set 2, and a treatment-focused strategy is used to begin the analysis.

Regardless of the input set, the enterprise will need a “base structure” consisting of a primary care package, medical records, medical logistics, a medical infrastructure package, and other base resources, as shown in the figure. Enterprise operations with the base-level resources, protocol set 2, and strategy T can then be simulated for a period of time to see if it provides an acceptable level of service to the selected population (#1). If the answer is no (as shown by the decision diamond), the analysis can then try adding individual resource packages to see which provides the most improvement in service capability to the population. Resource packages are designed by the user team (e.g., pediatrician/ internist/obstetrician/ENT package, which can be substituted for a primary care package; a gastroenterologist/orthopedist package; an oncologist/urologist package; a cardiologist/ thoracic surgeon package; an emergency room package; and other resource groupings). Enterprise operations are simulated for each package to determine the improvement in service capability above the base level. The resource package with the most improvement on the margin is added to the enterprise (as shown under the variable structure).

This process is repeated, and resource packages with the most marginal improvement to the enterprise are added until an acceptable level of service is reached. (Mathematical programming techniques would likely make this iterative search process more efficient.) When this process is complete, the sum of the base and variable resources constitute an efficient set of resources that provide an acceptable level of service (measured by access/quality and cost metrics) to the designated population using the specified protocols. The effect of different protocols on the resource requirements, as well as resource requirements for other populations, can be determined in a similar way. This process could be used to design a “versatile” set of resources that would provide a capability to serve multiple populations using different protocols.

Operations research could address some of the important enterprise-level issues but would require cultural changes on the part of enterprise management, as well as the operations research community. Enterprise management would have to encourage centralized planning for enterprise design and resource allocation issues, simultaneously maintaining decentralized operations. Higher order (and usually large) effects of interactions across stovepipes can only be identified at this level. Enterprise management would have to encourage a culture of prospective analyses to identify necessary changes that would be useful and would provide a high return on investment. (Retrospective analysis is an expensive trial-and-error process to learn what doesn't work). Enterprise management would have to establish a “requirements-pull” process for equipment and IT decisions, rather than the existing “technology-push” process, which is based on what is available from industry rather than what is needed. Management would also have to require that processes be reengineered when implementing new technologies (technology changes overlaid on existing processes produce zero value).

The health operations research community would also have to make important cultural changes. It needs to begin addressing enterprise-level issues, which should not remain in the purview of health econometricians who have failed to solve the cost, access, and quality problems that have beleaguered health enterprises and the nation. The operations research community would have to start working with enterprise-level structural models and begin using them for prospective analyses. Health operations research practitioners must become integral partners with senior enterprise managers in their business planning. They should use their 40 years of tactical-level support as an entreé and then demonstrate (and market!) the value of enterprise-level analyses to enterprise managers.

Copyright © 2005, National Academy of Sciences.
Bookshelf ID: NBK22868

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (4.4M)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...