Clinicians are confronted by increasing amounts of clinical data for each patient they treat as well as an exponentially increasing volume of relevant medical research. While electronic health records and databases help physicians manage this rising tide of information, patient-specific recommendations provided by clinical decision support systems can do even more by improving decision making and helping ensure patient safety. Examples of various types of clinical decision support systems include diagnostic support such as MYCIN and QMR, alerts and reminders based on the Arden Syntax, and patient management systems that use computer representations of patient care guidelines. With evidence supporting the effectiveness of all these systems, it is important to plan for decision support when designing and installing modern health information systems.
A broad range of information systems can now generate patient-specific advice to aid clinical decision making.
A large part of any physician’s work, especially in non-procedural disciplines, involves acquiring information and then, aided by evidence and experience, making decisions for the best possible outcome. In earlier days, this whole process could take place in the brain of the practitioner. However, with the burgeoning amount of data now available for each patient and the increasing body of medical evidence, we need tools to help us make rational decisions based on all this information.
Computer technology can assist by generating case-specific advice for clinical decision making. The systems used are usually referred to as clinical decision support systems or CDSS.
The types of CDSS available are as broad as human ingenuity allows: from personal digital assistant applications customized by a single clinician to multihospital mainframe-based surveillance systems meant to assure care for thousands of patients. The systems can be classified by the nature of their interaction with the clinician.
In the first type of interaction, a clinician solicits advice from the CDSS. The two most famous examples are the MYCIN and QMR systems. MYCIN was developed in the 1970s to help clinicians choose antibiotics for bacteremia or meningitis. The clinicians would enter a series of facts about history, physical findings, and laboratory results into the system, which would then give patient-specific recommendations for antibiotic coverage. While in experimental tests the system could be as good as a panel of clinicians, MYCIN was never widely used because of difficulties with maintenance and incorporating the system into a clinician’s workflow.
The QMR system uses an ingenious algorithm modeled on the clinical reasoning of a single University of Pittsburgh internist. The program takes historical and physical findings and generates a differential diagnosis. It does this by using a large database of “evoking strengths,” “importance,” and “frequencies” of findings seen in diseases within its domain. A pathognomonic finding has a high evoking strength while a finding commonly seen in a given disease also increases the weighting given to that disease in the differential diagnosis. An important finding that is absent will downgrade the diagnosis within the differential diagnosis. Like MYCIN, this program was found to function as well as practising clinicians. However, the labor-intensive, often biased process of entering findings limits its utility. Indeed, QMR is now used as an interactive textbook where it is just one of several influences on the final differential diagnosis.
CDSS may have even more potential in applications that deliver unsolicited advice. These applications judiciously deliver information or knowledge that can beneficially alter clinical decision making. The Arden Syntax is a quasi-programming language that allows the encoding of decision rules into a computer-readable format.
Consider the following example from the Brigham and Women’s Hospital: A hospital laboratory technician runs an electrolyte panel on a sample of blood from a patient. The result, a low potassium of 2.9 mmol/L, is entered into the laboratory information system. This act results in a digital message entering the hospital information system. This message contains, among other things, the patient’s hospital ID number and the potassium result. The message goes to a centralized computer application that manages a series of clinical rules. Here the information in the message is compared with all relevant rules. In our example, the medical logic module (MLM) would extract the fact “potassium is 2.9” and search out rules applicable to this fact, one of which might be IF “potassium is <3.0” AND “patient is on digoxin” THEN “stat page MD.” An appropriate message would then be transmitted to the pager of the physician responsible for the patient. These types of systems are in use today.
Computerized physician order entry (CPOE) is another technology that enables the presentation of tailored knowledge or information at the time of a clinical decision. CPOE refers to a variety of computer-based systems for ordering medications or tests. These systems automate and standardize the ordering process. Information gleaned from computer records of these orders can also be used to guide quality improvement efforts.
Consider the following example. A physician orders an X-ray for a patient who has inverted her ankle. When the order is entered into the computer, the physician is reminded of the exact elements of the evidence-based criteria for such an order (the Ottawa Ankle Rules). The order is not registered until the physician acknowledges that the patient either does or does not meet the criteria. Such systems have been shown to decrease inappropriate X-ray orders by as much as 47%.
Similar CPOE systems can be successful in a wide variety of settings, with some of the most compelling uses being the prevention of drug dose and interaction errors. A 1999 report from the Institute of Medicine, revealed the magnitude of ongoing medical error, documenting as many as 7000 annual deaths in the US due to medication errors in hospitals. A more recent Canadian report called for investment in information-technology infrastructures that support the standardized identification, reporting, and tracking of patient safety data. Research that identifies CDSS methods that alert health care providers to errors has become a critical element in improving patient safety.
The preceding examples of solicited and unsolicited CDSS all centre on isolated transactions that each make up only a small fraction of a patient’s care. Disease management systems are specialized CDSS that help clinicians and patients negotiate complex treatment algorithms for conditions such as asthma, hypertension, diabetes, and hyperlipidemia. Reflecting the changing locus of control in the clinician-patient relationship, many of the systems are designed expressly for the patient. Diabetes systems might be the best example of cases where patient-specific data, such as blood glucose measurements and food intake, are used to generate customized educational modules and detailed dietary recommendations.
How to represent complex clinical guidelines in computer applications is an area of considerable research . While implementing the IF-THEN-ELSE rule illustrated by the Brigham and Women’s Hospital hypokalemia/digoxin example is relatively straightforward, the programming of guidelines with their multiple, often subjective decision points tests the limits of scientific disciplines such as decision analysis and knowledge representation. Nonetheless, there are numerous examples of computerized guidelines in use today.
Clearly, to exploit the opportunities for these types of clinical decision support interventions, we must have effective health information systems in place. The Figure provides a blueprint for a health information system that has evolved from one first developed for the US National Library of Medicine in the early 1980s. Note that CDSS is an integral part of this system, alongside laboratory, radiology, and health records. This blueprint can be used not only for large hospitals but also for distributed networks of care sites such as practitioners’ offices.
Do clinical decision support systems improve patient care? The question is surprisingly hard to answer. Designing good evaluative trials is difficult and the sheer variety of systems and functions makes comparison complicated. A recent review of 57 randomized controlled trials and 10 systematic reviews evaluated the effectiveness of computer-based delivery of health evidence, including CDSS. The authors found that clinician or patient compliance with evidence-based recommendations improved only a modest amount: from 52% without to 57% with a system. Seven of eight relevant systematic reviews found a positive effect on provider or patient behavior. It is worth noting, however, that this review excluded computer-generated reminders.
In the coming years, provincial and federal health ministries will invest billions of dollars in new health information systems. Computer decision support systems integrated with computerized physician order entry can help make this investment worthwhile by leading to safer, more efficient, and more effective health care.
It is crucial that clinicians be involved in the development and rigorous scientific evaluation of these systems. Clinicians are also best placed to decide how CDSS should be implemented in local care environments. We urge clinicians to identify opportunities for CDSS and to advocate within their health care settings for the development of systems that bring about meaningful improvement of health outcomes.
1. van der Lei J. Clinical decision support systems. In: van Bemmel J, Musen M (eds). Handbook of Medical Informatics. Heidelberg: Springer-Verlag, 1997:261-276.
2. Shortliffe EH, Davis R, Axline SG, et al. Computer-based consultations in clinical therapeutics: Explanation and rule acquisition capabilities of the MYCIN system. Comput Biomed Res 1975;8:303-320. Abstract Full Text
3. Musen M, Shahar Y, Shortliffe EH. Clinical decision support systems. In: Shortliffe EH, Perrault L, Wiederhold G, et al. (eds). Medical Informatics: Computer Applications in Health Care and Biomedicine. New York: Springer-Verlag, 2001:573-609.
4. Miller RA, Pople HE Jr, Myers JD. Internist-1, an experimental computer-based diagnostic consultant for general internal medicine. N Engl J Med 1982;307:468-476. PubMed Abstract
5. Friedman CP, Elstein AS, Wolf FM, et al. Enhancement of clinicians’ diagnostic reasoning by computer-based consultation: A multisite study of two systems. JAMA 1999;282:1851-1856. PubMed Abstract Full Text
6. Blois MS. Clinical judgment and computers. N Engl J Med 1980;303:192-197. PubMed Abstract
7. Jenders RA, Hripcsak G, Sideli RV, et al. Medical decision support: Experience with implementing the Arden Syntax at the Columbia-Presbyterian Medical Center. Proc Annu Symp Comput Appl Med Care 1995:169-173. PubMed Abstract
8. Doolan DF, Bates DW, James BC. The use of computers for clinical care: A case series of advanced US sites. J Am Med Inform Assoc 2003;10:94-107. PubMed Abstract Full Text
9. Chin H, Wallace P. Embedding guidelines into direct physician order entry: Simple methods, powerful results. Proc AMIA Symp 1999:221-225. PubMed Abstract Full Text
10. Kohn LT, Corrigan JM, Donaldson MS (eds). To Err Is Human: Building a Safer Health System. Washington, DC: National Academy Press, 1999.
11. National Steering Committee on Patient Safety. Building a Safer System: A National Integrated Strategy for Improving Patient Safety in Canadian Health Care. Ottawa: National Steering Committee on Patient Safety, 2002. Full Text
12. Cramer K, Hartling L, Wiebe N, et al. Computer-Based Delivery of Health Evidence: A Systematic Review of Randomised Controlled Clinical Trials and Systematic Reviews of the Effectiveness on the Process of Care and Patient Outcomes. Edmonton: Alberta Heritage Foundation for Medical Research, 2003. Overview Full Text
13. Maviglia SM, Zielstorff RD, Paterno M, et al. Automating complex guidelines for chronic disease: Lessons learned. J Am Med Inform Assoc 2003;10:154-165. PubMed Abstract Full Text
14. Hripcsak GH. IAIMS architecture. J Am Med Inform Assoc 1997;4:S20-S30. PubMed Abstract Full Text
Martin Pusic, MD and Mark Ansermino, MB BCh, FRCPC, FFA
Dr Pusic is assistant professor, Department of Pediatrics, UBC, and research director, Division of Emergency Medicine, Children’s and Women’s Health Centre of BC. Dr Ansermino is a member of the Centre for Health Innovation and Improvement (CHIi) and a pediatric anesthesiologist at BC’s Children’s Hospital.