Computer Decision Support Systems*
Thomas H. Payne, MD
Computer decision support systems are computer applications designed to aid clinicians in making diagnostic and therapeutic decisions in patient care. They can simplify access to data needed to make decisions, provide reminders and prompts at the time of a patient encounter, assist in establishing a diagnosis and in entering appropriate orders, and alert clinicians when new patterns in patient data are recognized. Decision support systems that present patient-specific recommendations in a form that can save clinicians time have been shown to be highly effective, sustainable tools for changing clinician behavior. Designing and implementing such systems is challenging because of the computing infrastructure required, the need for patient data in a machine-processible form, and the changes to existing workflow that may result. Despite these difficulties, there is substantial evidence from trials in a wide range of clinical settings that computer decision support systems help clinicians do a better job caring for patients. As computer-based records and order-entry systems become more common, automated decision support systems will be used more broadly. (CHEST 2000; 118:47S–52S)
Key words:decision support systems, clinical; drug therapy, computer-assisted; hospital information systems; medical informatics; medical records systems, computerized; practice guidelines; therapy, computer-assisted
Abbreviations:CPRS⫽Computerized Patient Record System; VA⫽Veterans Affairs
O
ver the last 20 years, computing systems have become increasingly common in health-care set-tings. Initially they were used for administrative and financial purposes—a role that continues today— but the use of computer systems to aid clinical decision making is growing. Computer decision support sys-tems are computer applications designed to aid clinicians in making diagnostic and therapeutic deci-sions in patient care. There is substantial evidence from trials in a wide range of clinical settings that computer decision support systems help clinicians do a better job caring for patients.Computers are designed to follow predefined instructions provided to them. In health-care deci-sion support systems, these instructions range from simple statements of the formIF this has occurred, THEN do the following pertaining to a specific laboratory result, to highly complex clinical guide-lines that include hundreds of interconnected rules. The early use of computers in clinical decision making—and one of the most successful uses to-day—involved use of simple rules governing the
myriad of small decisions clinicians make every day. For example, IF the patient has the diagnosis of obstructive lung disease, THEN an influenza vaccine should be given annually. As one of the pioneers in computing decision support systems has stated, “Careful attention to mundane and tedious detail can be more important than brilliance in the day-to-day care of patients . . . the kind of work that humans neither relish nor reliably perform.”1
How Can Computers Aid Decision Making? There are several ways that computers can help clinicians make better decisions.
Simplify Access to Data Needed To Make Decisions
Most practitioners use computing systems, either directly or indirectly, to gather laboratory results, radiology reports, or the narrative text of notes or consultations. This is because laboratories and tran-scription services have long used computing systems to report these data. Reporting of results and the creation of a customized report or graph can make patterns more apparent, leading to faster decision making. The value of flow sheets for following chronic conditions has long been recognized.2
*From the VA Puget Sound Health Care System, University of Washington, Seattle, WA.
Correspondence to: Thomas H. Payne, MD, VA Puget Sound Health Care System, 1660 South Columbian Way, Mail Stop S-007-CIM, Seattle, WA 98108; e-mail: [email protected]
Graphic display of laboratory data can make patterns rapidly apparent; if combined with display of medi-cations or other interventions, a better understand-ing of the course of disease may result.3This is one
of the simplest forms of decision support, and it is highly popular with clinicians because it does not require data entry and saves time.
Provide Reminders and Prompts
One of the most powerful tools in the field of clinical computing is the capability to generate re-minders and prompts to clinicians. As several reviews have shown, reminders change clinician behavior to improve delivery of chronic, acute, and preventive medical care.4 – 6 Reminders can be brought to the
attention of clinicians in a variety of ways: printed sheets can be affixed to a chart before a visit,7
windows can appear on a screen, or a list of remind-ers can appear on an electronic “cover sheet.” Usu-ally, reminders include a short message recommend-ing some action be taken along with the rationale for the reminder appearing on that particular patient (Fig 1). Methods of creating, editing, and using rules to trigger reminders vary greatly among computer decision support systems.8
Assist in Order Entry
Among the most successful uses of clinical com-puting applications is to check orders directly en-tered by a clinician in real time. Feedback to the
clinician through screen dialog boxes can alert the clinician to drug sensitivity, drug-allergy, drug-drug, drug-disease, and drug-laboratory interactions, and potential duplication of services. Orders that should be considered when one order is placed (“corollary” orders, such as blood levels when an aminoglycoside is prescribed) are much more likely to be ordered when presented at the time of order entry.9 If
order-entry screens are designed to display the results of previously ordered tests of the type being ordered, test ordering has been shown to be reduced by 13%.10 Applications to allow direct entry of
medication orders are among the most difficult clinical computing applications to develop, yet they have been demonstrated to dramatically reduce se-rious medication errors.11
Assist in Diagnosis
An early goal of computing systems used in clinical care was to help the physician establish a diagnosis. Programs such as Internist 1, Quick Medical Refer-ence (First Data Bank; San Bruno, CA), DXplain (Laboratory of Computer Science; Boston, MA), and Iliad (Applied Medical Informatics; Salt Lake City, UT) were designed to consider historical and physi-cal examination findings, laboratory and test results, and create a list of diagnoses to explain those find-ings.12,13These systems were based on large
collec-tion of rules and tables that related the presence or absence of findings with diseases and other condi-tions. Though they performed remarkably well, the
requirement that large amounts of data be entered limited their broad use in clinical care. Freestanding applications are now less common than applications that are tightly integrated with patient data in a repository or computer-based medical record sys-tem.14Moreover, much of the information needed to
use these applications—for example the presence or absence of symptoms or physical examination find-ings—is not routinely captured in computing systems in a form that can be processed by decision support systems.
Review New Clinical Data; Alert When Important Patterns Are Recognized
Reminders and order checks are useful methods for drawing clinicians’ attention to important occur-rences when the clinician is viewing a computer screen or paper chart, or is in the process of ordering. However, in some cases, there is a need to bring clinical events such as a new or changed laboratory result, hospital discharge, or combination of events to the attention of the clinician at the moment the event occurs. Event monitors are appli-cations that can “eavesdrop” on newly available data or the occurrence of events (hospital admission, discharge, etc) by receiving electronic messages from computing systems when specified events occur.15
When an electronic message is received, a specified rule can then be run to determine if there is a need to notify the clinician or take other action. An example of the use of an event monitor is in the handling of culture and sensitivity results. An event monitor can notify the clinician when there is a mismatch between newly available sensitivity results and antimicrobials being given to the patient.
Characteristics of Successful Computer Decision Support Systems
There is now substantial literature describing the design, implementation, and evaluation of computer decision support systems. From this literature, and from our own experience, we can describe several characteristics of successful computer decision sup-port systems.
They Give Patient-Specific Recommendations
When a computing system provides advice based on a guideline, it is most useful if the recommenda-tion is based on that patient’s data. Of course, this requires that data needed to make the recommen-dation be available to the decision support system in machine-processible form. Collecting these data from their source (laboratory, pharmacy, and other systems) is preferable to requiring data entry by the clinician.16
Passive display of guideline documents in the literature, on the World Wide Web, or in other electronic media is not a reliable method for improv-ing compliance or changimprov-ing practitioner behavior.17
There are several reasons for this. First, there are myriad guidelines available on the World Wide Web, and finding credible guideline documents quickly can take substantial time.18 Second, no matter how
convenient access to the documents becomes, simply viewing a document describing a guideline will be less effective than making it easy to follow that guideline.
They Save Time
Most clinicians note that they have less time available than in the past, because of increased patient volumes, greater demands for documenta-tion,19 and the complexity of modern practice. An
extremely effective method for changing clinician behavior is to make it as fast or faster to comply with a recommendation or guideline than not to comply. There are many strategies for helping clinicians save time in the process of complying with a guideline.
One approach is to save the time required for visit documentation. If applications can guide clinicians through a guideline and at the same time speed documentation of the visit, then there is an addi-tional incentive to use the application, thereby in-creasing compliance with the guideline.20If the same
application allows simple and fast ordering of ser-vices, there is yet another incentive to use the application.
Some guidelines can be implemented in part or in full by creating collections of orders that can be selected in particular clinical situations. For example, when patients are admitted to the hospital for treat-ment of community-acquired pneumonia, guidelines cover the ordering of cultures and other diagnostic tests, and empirical selection of antibiotics. Collec-tions of orders— often referred to as order sets or order templates— can be created and offered to the clinician either in paper form or in an order-entry application. Since the orders are conveniently grouped to save time during the ordering process, this is a potentially useful method of implementing some types of guidelines.
They Are Incorporated Into Workflow of Clinic, Office, or Hospital
Health-care delivery is a complex effort with labor divided among many professions: physicians, nurses, pharmacists, other professionals, and support staff. Computing systems should be designed to fit into this workflow as smoothly as possible, because changing the workflow of large numbers of
profes-sionals is difficult. A successfully designed computer decision support system must either fit in well with this workflow, or those implementing it must be prepared to change the workflow.
Examples of Successful Computer Decision Support Systems
There are numerous examples of highly successful computer decision support systems in use today. Though these systems are rarely broadly exported beyond the organization in which they were devel-oped, they serve as examples of what is possible with the technology of today, in real-world settings.
HIV Guidelines at Boston Beth Israel
Management of HIV infection is a complex, rap-idly evolving field with a large number of guidelines available to practitioners. Safran and colleagues21
hypothesized that compliance with HIV care guide-lines would be greater if the guideguide-lines were incor-porated into the clinical computing applications in use at Boston Beth Israel Hospital. They demon-strated that clinicians who received alerts and re-minders containing patient-specific recommenda-tions instituted appropriate treatment far more rapidly than clinicians who did not. One of the most dramatic effects was in the time between the avail-ability of confirmed T-cell counts ⬍200 cells per mL3and institution of prophylaxis against Pneumo-cystis carinii pneumonia. In the control group, the average elapsed time was 122 days, while in patients treated by physicians who received prompts to pre-scribe prophylaxis, the average elapsed time was 11 days. An important reason for this success was the provision of screens that both notified the clinician of the need for prophylaxis and greatly simplified the process of ordering it.
Antimicrobial Use in the ICU
In a trial conducted in the critical care unit, researchers at the LDS Hospital in Salt Lake City, Utah, studied outcomes for patients for whom anti-microbials were ordered using a computer decision support system.22 This application considers patient
allergies, likely pathogens, local patterns of antimi-crobial resistance, antimiantimi-crobials on the formulary, hepatic and renal function, the results of cultures, and other factors when recommending therapy. The authors also noted that manually gathering data needed to prescribe antimicrobials would take up to 25 min. Patients treated using this computer-assisted management program for antimicrobials received fewer doses of antimicrobials, had fewer days of
excessive drug dosage, fewer prescriptions for drugs to which the patient was allergic, shorter length of hospital stay, and lower hospital costs, compared to patients treated without this program. There were several other improvements beyond these, leaving little question of the advantage of such an application over the traditional methods for ordering antimicro-bials.
The Veterans Affairs Computerized Patient Record System
Much of the evidence that computer decision support systems are effective derives from studies performed in institutions that have had clinical com-puting systems in use for a decade or more. These studies have been a model for organizations that are considering implementing systems of their own. The Veterans Health Administration has recently em-barked on the development and large-scale imple-mentation of the Computerized Patient Record Sys-tem (CPRS) in all Veterans Affairs (VA) medical centers, including VA Puget Sound in Washington State. The CPRS is a package developed by the VA to allow clinician order entry, note entry, and to provide results reporting and order checks.23
CPRS includes numerous decision support fea-tures: on-screen reminders for a wide variety of acute, chronic, and preventive care topics; order checks for drug-drug, drug-food, drug-allergy inter-actions, and for duplicate laboratory testing; order sets; graphic display of laboratory results; and tem-plates to guide and simplify document creation. Since all orders entered on hospital wards using CPRS are entered directly by practitioners in our hospital, there is great potential to influence compli-ance with organizational guidelines and recommen-dations for management of patients with specific clinical conditions. The following section describes our early efforts to take advantage of one of these decision support features: order sets.
The choice of diagnostic tests and antimicrobial selection can influence the effectiveness and costs of therapy. Our Pulmonary and Infectious Diseases sections have developed collections of orders for practitioners to consider when writing hospital ad-mission orders for patients with community-acquired pneumonia. These orders are easily selectable from hospital admission order screens used whenever patients are hospitalized on the Medical Service. Additions or changes to these screens are immedi-ately available on the⬎2,000 workstations used by physicians, nurses, and other staff at VA Puget Sound.
community-acquired pneumonia—and for hundreds of screens containing other orders—is completely within the control of the staff of our organization. This is both a great advantage and a challenge, since these orders must be regularly reviewed to remain current, and there are inevitably debates on what form the orders should take. Our solution is to document both the content and review process for these order screens on our internal Web site, using a format inspired by pioneering work at LDS Hospital. Each order col-lection has a primary author, a primary reviewing group, and is reviewed by other organizational groups such as nursing and pharmacy. The rationale for the orders and references supporting them are also included.
We are only 8 months into our use of the CPRS on our busiest wards, and we have not yet studied the effect of these decision support features on practi-tioner behavior or on clinical or financial outcomes. However, we have established the foundation on which our organization can implement computer-based decision support. The components of that foundation are described in the next section.
Requirements for Implementing Computer Decision Support Systems
Implementing computer decision support systems, particularly order-entry systems, is a difficult under-taking that requires an organizational will to change, and to be patient with the process of change. It requires substantial financial and human resources to introduce computing systems into hospitals and clinics. Even when there is organizational commit-ment to implecommit-ment computing systems, it may take months or more for physicians and other staff to feel the change was worth it.24
Next, there must be a computing infrastructure in place: databases containing patient data, worksta-tions near or at the point of care, and reliable networks to connect them. Printed reports can be used for many purposes, and to reduce the need for workstations in examination rooms, but order-entry applications with immediate feedback depend on the direct use of mobile or desktop workstations by clinicians.
There must be guidelines or algorithms incorpo-rated into the clinical computing system, and there must be patient data in processible form. The latter requirement is difficult to achieve quickly, and rep-resents one of the greatest hurdles for the broader use of computer decision support systems. Much of the content of the patient record is in narrative text form, though there has been recent progress in unlocking this information using natural language processing techniques.25
Clinical guidelines vary greatly in their level of precision. Some contain clear descriptions that can be readily interpreted and acted on by the physician, while others include more general statements of the approach to a particular problem. In order for computer decision support systems to be effective, there must be a trusted, tested, precise rule or guideline to be implemented in the automated sys-tem.26,27There is no substitute for testing of decision
support systems in real-world settings, with serial editing and improvement of the rule until it per-forms as expected.
A commonly voiced concern of physicians about computer systems in clinical care is whether using them will take more time, which may be the case, at least initially.28,29There is also concern both among
health-care providers and the public that health information remains confidential. If computer deci-sion support systems use identified patient data, they must be designed to protect the confidentiality of those data. Combinations of policies and technical approaches can greatly reduce the risk that confiden-tial information will be inappropriately disclosed.30 Exchange of Rules Used by Computer Decision Support Systems
Some organizations have created large collections of rules that could potentially be exported to other settings where the computing infrastructure neces-sary to use them is in place. Exchanging rules would allow the organization to benefit from the iterative refinement of rules achieved in the site where they were developed, and allow organizations to concen-trate their effort on developing rules in other areas. There have been several efforts to allow exchange of decision support rules: (1) The Arden Syntax, a standard language for expressing rules used to gen-erate alerts and reminders,31has been used in some
organizations to encode rules, and to exchange them with other organizations. (2) More recently, the GuideLine Interchange Format has been proposed to allow exchange clinical practice guidelines among institutions and computer-based applications.32 (3)
There is a utility within the VA CPRS to allow exchange of the code required to trigger reminders between the ⬎100 sites using CPRS. Today, these tools for exchange of decision support rules are used by only a small number of organizations, but as the number of organizations utilizing computer decision support systems rises, these tools could increase sharing of automated clinical rules.
Summary
Computer decision support systems that present patient-specific recommendations in a form that can
save clinicians time have been shown to be highly effective, sustainable tools for changing clinician behavior. Designing and implementing such systems is challenging because of the computing infrastruc-ture required, the need for patient data in a ma-chine-processible form, and the changes to existing workflow that may result. As computer-based records and order-entry systems become more com-mon, automated decision support systems will be used more broadly.
References
1 McDonald CJ, Tierney WM. Computer-stored medical records: their future role in medical practice. JAMA 1988; 259:3433–3440
2 Whiting-O’Keefe QE, Simborg DW, Epstein WV, et al. A computerized summary medical record system can provide more information than the standard medical record. JAMA 1985; 254:1185–1192
3 Powsner SM, Tufte ER. Graphical summary of patient status. Lancet 1994 344:386 –389
4 Hunt DL, Haynes RB, Hanna SE, et al. Effects of computer-based clinical decision support systems on physician perfor-mance and patient outcomes: a systematic review. JAMA 1998; 280:1339 –1346
5 Johnston ME, Langton KB, Haynes RB, et al. Effects of computer-based clinical decision support systems on clinician performance and patient outcome: a critical appraisal of research. Ann Intern Med 1994; 120:135–142
6 Shea S, DuMouchel W, Bahamonde L. A meta-analysis of 16 randomized controlled trials to evaluate computer-based clinical reminder systems for preventive care in the ambula-tory setting. J Am Med Inform Assoc 1996; 3:399 – 409 7 McDonald CJ, Hui SL, Smith DM, et al. Reminders to
physicians from an introspective computer medical record: a two-year randomized trial. Ann Intern Med 1984; 100:130 – 138
8 Reggia JA, Tuhrim S. An overview of methods for computer-assisted medical decision making. In: Reggia JA, Tuhrim S, eds. Computer-assisted medical decision making. New York, NY: Springer-Verlag, 1985; 3– 45
9 Overhage JM, Tierney WM, Zhou XH, et al. A randomized trial of “corollary orders” to prevent errors of omission. J Am Med Inform Assoc 1997; 4:364 –375
10 Tierney WM, McDonald CJ, Martin DK, et al. Computerized display of past test results: effect on outpatient testing. Ann Intern Med 1987; 107:569 –574
11 Bates DW, Leape LL, Cullen DJ, et al. Effect of computer-ized physician order entry and a team intervention on pre-vention of serious medication errors. JAMA 1998; 280:1311– 1316
12 Berner ES, Webster GD, Shugerman AA, et al. Performance of four computer-based diagnostic systems. N Engl J Med 1994; 330:1792–1796
13 Berner ES, Jackson JR, Algina J. Relationships among per-formance scores of four diagnostic decision support systems. J Am Med Inform Assoc 1996; 3:208 –215
14 Miller RA, Masarie FE Jr. The demise of the “Greek Oracle”
model for medical diagnostic systems. Methods Inf Med 1990; 291:1–2
15 Hripcsak G, Cimino JJ, Johnson SB, et al. The Columbia-Presbyterian Medical Center decision-support system as a model for implementing the Arden Syntax. Proc Annu Symp Comput Appl Med Care 1991; 1:248 –252
16 Payne TH, Sengupta S, Sittig DF. Electronic exchange of patient information in health care organizations: the infra-structure for electronic health records. In: Murphy G, Han-ken M, Waters K, eds. Electronic health records: changing the vision. Philadelphia, PA: WB Saunders, 1999; 129 –142 17 US Congress, Office of Technology Assessment. Identifying
health technologies that work: searching for evidence. Wash-ington, DC: US Government Printing Office; September 1997; OTA-H-608
18 Owens DK. Use of medical informatics to implement and develop clinical practice guidelines. West J Med 1998; 168: 166 –175
19 Iezzoni LI. The demand for documentation for Medicare payment. N Engl J Med 1999; 341:365–367
20 Schriger DL, Baraff LJ, Rogers WH, et al. Implementation of clinical guidelines using a computer charting system: effect on the initial care of health care workers exposed to body fluids. JAMA 1997; 278:1585–1590
21 Safran C, Rind DM, Davis RB, et al. Guidelines for manage-ment of HIV infection with computer-based patient’s record. Lancet 1995; 346:341–346
22 Evans RS, Pestotnik SL, Classen DC, et al. A computer-assisted management program for antibiotics and other anti-infective agents. N Engl J Med 1998; 338:232–238
23 Payne TH. The transition to automated practitioner order entry in a teaching hospital: the VA Puget Sound experience. Proc AMIA Annu Fall Symp 1999; 589 –593
24 Chin HL, McClure P. Evaluating a comprehensive outpatient clinical information system: a case study and model for system evaluation. Proc Annu Symp Comput Appl Med Care 1995; 1:717–721
25 Hripcsak G, Kuperman GJ, Friedman C, et al. A reliability study for evaluating information extraction from radiology reports. J Am Med Inform Assoc 1999; 6:143–150
26 East TD, Bohm SH, Wallace CJ, et al. A successful comput-erized protocol for clinical management of pressure control inverse ratio ventilation in ARDS patients. Chest 1992; 101:697–710
27 McDonald CJ, Overhage JM. Guidelines you can follow and trust: an ideal and an example. JAMA 1994; 271:872– 873 28 Bates DW, Boyle DL, Teich JM. Impact of computerized
physician order entry on physician time. Proc Annu Symp Comput Appl Med Care 1994; 1:996
29 Sittig DF, Stead WW. Computer-based physician order entry: the state of the art. J Am Med Inform Assoc 1994; 1:108 –123 30 Committee on Maintaining Privacy and Security in Health Care. Applications of the national information infrastructure. For the record. Washington, DC: National Academy Press, 1997
31 Hripcsak G, Cimino JJ, Johnson SB, et al. The Columbia-Presbyterian Medical Center decision-support system as a model for implementing the Arden Syntax. Proc Annu Symp Comput Appl Med Care 1991; 1:248 –252
32 Ohno-Machado L, Gennari JH, Murphy SN, et al. The guideline interchange format: a model for representing guidelines. J Am Med Inform Assoc 1998; 5:357–372
