Computer-based Guideline Implementation Systems
A Systematic Review of Functionality and Effectiveness
- Correspondence and reprints: Richard N. Shiffman, MD, Center for Medical Informatics, Yale School of Medicine, P.O. Box 208009, New Haven, CT 06520-8009. e-mail: 〈 〉
- Received 14 August 1998
- Accepted 13 November 1998
In this systematic review, the authors analyze the functionality provided by recent computer-based guideline implementation systems and characterize the effectiveness of the systems. Twenty-five studies published between 1992 and January 1998 were identified. Articles were included if the authors indicated an intent to implement guideline recommendations for clinicians and if the effectiveness of the system was evaluated. Provision of eight information management services and effects on guideline adherence, documentation, user satisfaction, and patient outcome were noted.
All systems provided patient-specific recommendations. In 19, recommendations were available concurrently with care. Explanation services were described for nine systems. Nine systems allowed interactive documentation, and 17 produced paper-based output. Communication services were present most often in systems integrated with electronic medical records. Registration, calculation, and aggregation services were infrequently reported. There were 10 controlled trials (9 randomized) and 10 time-series correlational studies. Guideline adherence improved in 14 of 18 systems in which it was measured. Documentation improved in 4 of 4 studies.
Despite the considerable effort and resources that have been invested in the development and dissemination of clinical practice guidelines, there continues to be considerable variation in the effectiveness of guidelines to bring about changes in the behavior of clinicians. A number of studies have found that, despite serious initiatives on the part of national organizations to develop and disseminate guidelines, practitioners may still ignore them.1 2 3 4 5 6.
Greco and Eisenberg7 devised a general taxonomy of methods that may be used to influence clinician behavior. These include education, feedback, participation by physicians in efforts to bring about change, administrative rules, financial incentives, and penalties. Several investigators have attempted to identify which factors in guideline implementation strategies are most efficacious. Davis and Taylor-Vaisey8 found that reminder systems, academic detailing, and the use of combined interventions were most effective.
Grimshaw and Russell9 found that the guideline implementation strategies most likely to be effective were those that delivered patient-specific advice at the time and place of a consultation. Computers can provide, concurrent with care, advice that is tailored to the needs of individual patients. A systematic review by Johnston et al.10 found that computer-based decision support can improve clinician performance.
Any computer-based tool is more likely to work if it is integrated with clinical activities. Elson11 pointed out the critical role of workflow integration for effective guideline implementation. To be accepted, guideline implementation applications should give back to the user something of value to offset the inconvenience of using the system.
In the course of developing a computer-based guideline implementation system to assist in the management of childhood asthma and in creating additional tools, we identified eight information management services that promote workflow integration—recommendation, documentation, registration, communication, calculation, explanation, presentation, and aggregation. Each service adds value to a computer application that should translate to an increased probability for success. In addition, the services provide a structure for comparison and evaluation of dissimilar implementations.
Because both successful and unsuccessful strategies have provided many of these services, we sought to perform a detailed analysis of the functionality delivered by current computer-based implementation systems. In this paper, we analyze which information management services have been delivered by recently described guideline implementations. We also review the effectiveness of the computer-based interventions in influencing clinicians' behavior and changing patient outcomes.
Using the OVID search engine, the MEDLINE and CINAHL databases from 1992 through January 1998 were searched. Search terms included the following MeSH headings—algorithm, computer-assisted decision making, computer-assisted therapy, consensus statement, guideline adherence, health planning guidelines, health services research, medical audit, practice guideline, process and outcome evaluation, quality assurance, quality of health care, and reference standard—and the following text words—remind$, alert$, guideline$, implement$, and computer$. We also reviewed books and bibliographies of primary and review articles.
We limited the review to papers published in or after 1992 for three reasons: 1) The U.S. Agency for Health Care Policy and Research (AHCPR) began its dissemination of evidence-based clinical practice guidelines in 1992, leading a vanguard of new interest in guideline implementation. 2) Several information management services are delivered optimally with a graphical user interface. The first broadly successful operating system that supported the interface (Windows 3.1, Microsoft Inc., Redmond, Washington) was released early in 1992. 3) Two influential systematic literature reviews on similar topics covered papers that had been published prior to 1992.9 10
We selected studies for review if computers were used as part of an implementation strategy for clinical practice guidelines; the authors specifically indicated an intent to implement guideline recommendations—not simply to provide computer-based decision support; the implementations were intended to influence health care providers (thereby excluding systems that provided recommendations directly to patients); and the studies included an evaluation component that objectively studied some aspect of the effectiveness of the system in a practice setting.
Information Management Services Model
We have devised an information management services model for the implementation of clinical practice guidelines. Briefly, the model comprises eight components, defined as follows:
Recommendation: the determination of appropriate, guideline-specified activities that should occur under specific clinical circumstances
Documentation: the collection, recording, and storage of observations, assessments, and interventions related to clinical care
Explanation: the provision of background information on decision variables and guideline-specified actions (e.g., definitions, measures of quality or cost) and the rationale that supports guideline recommendations, including evidence and literature citations
Presentation: the creation of useful output from internal data stores
Registration: the recording and storage of administrative and demographic data to uniquely identify the patient, provider(s), and encounter
Communication: the transmission and receipt of electronic messages between the clinician and other information providers
Calculation: the manipulation of numeric or temporal data, or both, to derive required information
Aggregation: the derivation of population-based information from individual patient data
▪ A variety of methods may be used to incorporate each service in a computer-based guideline implementation. Although their inclusion should result in a more comprehensive, workflow-integrated system, individual services may be excluded from any specific application.
▪ Each pertinent article was scrutinized by at least two of the authors for evidence that the system described there did or did not provide each information management service. Disagreements between the authors were settled by discussion. In addition, evidence of effectiveness was extracted and summarized. We determined the types of studies that were performed to evaluate each system (using the classification system used by Grimshaw and Russell9) and then ascertained the effectiveness of the system with regard to guideline adherence and other process measures (such as documentation and user satisfaction) and patient outcome measures when available. Because of the wide variety of study types and outcome variables, no quantitative meta-analysis of the results was possible.
A total of 25 papers that described 20 discrete systems were identified. We included more than one report on a single system if descriptions and evaluations of a single system were segregated into more than one report or if the authors investigated more than one guideline implementation strategy on the same system. Features of the guideline implementation systems are summarized in Table 1.
Eleven systems were based on national guidelines, including those published by AHCPR; the American Diabetes Association; the National Cholesterol Education Program; the Joint National Committee on Detection, Evaluation, and Treatment of High Blood Pressure (JNC V); and the U.S. Preventative Services Task Force (USPSTF). In several cases, the authors commented on the need for local modifications of the guidelines. Four systems implemented locally developed guidelines. In five systems, the guideline source was not described.
Thirteen of the guideline implementations addressed patient management issues and therapy, including one system that provided pre-authorization for surgical procedures12 and another system that was intended to improve discussion of advance directives.16 The other seven systems provided guidance with screening and health maintenance activities.
All systems provided patient-specific recommendations. The scope of the recommendations encompassed a broad range of clinical activities, including appropriate tests and treatments, alerts about at-risk states, and reminders of appropriate physical assessments and screening activities. With one exception, all systems provided recommendations concurrently with care.12 A variable number of factors were evaluated by the systems to determine appropriate intervention recommendations.
Nine reports documented that some explanation functionality was provided. These services provided background information, definitions, and risks as well as the rationale that supported specific recommendations. One system offered literature citations.
Most systems provided prompts for documentation of relevant findings that served a reminder function for the clinician-user. In many cases, these data were supplemented by complete medical record capabilities. Documentation services were provided in a variety of ways. Several systems relied on paper-based recording of clinician observations, which were later entered into the computer by clerical personnel. Others made use of online data entry, particularly those that were part of larger electronic medical record (EMR) systems. In nine systems the documentation process was interactive.*
Several of the reports described systems that were integrated with institutional EMR systems.† Documentation services for these systems tended to make use of the functionality of the EMR. Other systems were essentially stand-alone systems and not integrated with an EMR.
Like the data entry services, presentation services also varied considerably. Presentation modalities included paper-based display of reminders that were attached to patient charts, on-screen reminders and alerts, onscreen display of algorithms, patient summaries, customizable after-care instructions, and annual birthday letters to patients regarding appropriate preventive services. Seventeen of the systems made use of paper-based output of some kind.
Electronic communication services most frequently provided interfaces to the EMR and to order-entry functions. Interfaces existed in some systems to pharmacy, scheduling, and laboratory results reporting. Standalone systems, by definition, offered no electronic communication services.
None of the reports on stand-alone systems described provision of registration processes. Some mechanism for identification of patient demographics, provider, and the encounter was assumed to have been present in all cases, but specific mechanisms were not described. The EMR-related systems presumably have the capability to integrate demographic and administrative data.
Calculation services were used to calculate patient ages and intervals between tests and to trigger rules related to preventive services. The Lipid Management Program calculated lipid fractions.27
Aggregation services were described for only a few systems. In many cases, it is clear that database capabilities would allow aggregation of individual patient data, but only two reports explicitly described aggregation services. Schriger et al.30 noted that the database could be used to calculate deviation (non-adherence) rates by physicians. The Beth Israel Clinical Care System captured data about laboratory tests, demographics, dates of admission and discharge, and response to alerts that were used to generate aggregate reports.29
The methodologies used to evaluate the effectiveness of the 20 guideline implementations included ten controlled trials (nine of which were randomized) and ten time-series studies (none of which incorporated external controls; one applied a switchback design). The outcome variables that were measured also varied considerably and are summarized in Table 2.
Four studies looked at documentation and found improvement in each case. The average number of relevant data items for surgical pre-authorization increased from 4.0 to 28.812; the mean percentage increase for documentation of common pediatric problems was 58 percent,21 for management of back pain 30.2 percent,15 and for management of exposure to body fluids 42 percent.30
Eighteen of the 20 studies evaluated provider adherence to the guidelines. In 14 of the 18, some level of improved adherence was described. In several reports, adherence improvements occurred for some of the measured outcomes but not for all.
Failure to improve adherence using computer-based strategies was reported in four studies. An attempt to improve preventive care guideline adherence for hospitalized patients failed because of functional and systemic barriers that interfered with providing preventive care to inpatients.26 One study of prevention and management of pressure ulcers was unable to show any effect of the computer-based intervention on nursing decision making.36 In that case, the authors concluded that there was not enough gain for the effort of data entry. A system designed to influence decision making in emergency room patients with back pain failed because of general confusion regarding the utility of plain x-rays in these patients and the fact that recommendations were not enforced.15 Finally, in a study of diabetes management guidelines, compliance improved to the same degree in both control and intervention groups; the authors questioned study design issues.23
Clinician satisfaction was addressed in four studies. Two investigators found that users were satisfied with computer-based guideline interventions.12 36 On the other hand, physician-users of a clinical algorithm system found data entry so tedious that they refused to continue,21 and Nilasena et al.22 found that 70 percent of users complained that data entry forms were difficult to use and inefficient.
Eight studies examined patient outcomes. A study of an intervention for low-back pain found no effect on cost,15 whereas costs increased in both a system for management of health care workers exposed to body fluids and another that pre-authorized surgery.12 30 Use of a lipid tracking system was associated with improvements in patients' cholesterol and lipid fractions.27 A system for prevention of pressure ulcers was associated with a decreased incidence of decubiti,34 and Dexter et al.16 reported a significant improvement in the completion of advance directives (15 percent vs 4 percent for a control group) using a computer-based reminder system. An intervention to substitute appropriate antihypertensives for calcium channel blockers did not have any effect on patients' blood pressure,28 and alerts about appropriate HIV management did not change admission rates, emergency department visits, survival, or pneumocystis admissions.29
To better understand the design factors responsible for the success or failure of computer-based guideline intervention strategies, we analyzed reports on 20 systems that were intended to implement guideline recommendations in clinical practice. Specifically, we assessed the use of eight information management services, which we believe may be useful in integrating computerized systems into clinical workflow. Many reports failed to describe the systems in sufficient detail to ascertain the presence or absence of some of these services. Therefore, we were unable to create meaningful summary ratings of individual systems that might correlate with the outcomes described. However, we were able to describe qualitatively many aspects of the reported design of current computer applications used as guideline intervention tools and to summarize measures of their effectiveness.
All systems delivered patient-specific recommendations, and in most cases the advice was made available concurrently with care, thus meeting Grimshaw and Russell's criteria for implementations with a high probability of success.9 However, providing recommendations in this manner was neither necessary nor sufficient to ensure adherence. Several authors were unable to influence guideline adherence with concurrent reminders. Even providing delayed feedback was associated in one case with increased procedure authorization rates, although this system's influence may have been related to financial incentives and disincentives.12
The level of specificity of the advice varied considerably, as evidenced by the number of factors that were weighed by the programs to trigger relevant recommendations. Some systems simply checked a patient's age and gender to discern appropriate preventive interventions, whereas others monitored ongoing clinical transactions and considered multiple factors (e.g., diagnoses, laboratory results, and medications) in arriving at recommendations for changing medications or dosages and for planning treatment.
Somewhat surprisingly, fewer than half the reports documented provision of explanation services. More than 15 years ago, Teach and Shortliffe37 showed the importance of providing explanation for computer-based advisories.37 One noteworthy benefit of the use of computers for implementation of guideline recommendations is their capability to link recommendations dynamically to the evidence that supports them.
Most reports described the use of on-screen and paper-based prompts to remind users of critical information that should be documented. Clinicians entered data into computers directly and interactively in fewer than half the systems. Even some long-established EMR systems depended on completion of paper forms with subsequent data entry by clerical personnel. Likewise, paper-based output was described for 17 of the 20 systems. It seems clear that the paperless office remains a vision of the future.
Registration, calculation, communication, and aggregation services were infrequently described. These components offer tremendous potential benefit for well-designed computer-based guideline implementation. Providing communication services requires networked systems. Registration services may seem mundane, but an interface to an administrative database that contains this information may be vital to the success of a computer-based initiative by diminishing the clerical workload for clinicians. Calculation and aggregation services are basic functions of many computer systems that were rarely reported in these guideline implementation systems.
The evaluations of system effectiveness varied markedly in design, implementation, and level of description. In many, the evaluations of effectiveness were methodologically weak. In addition, the guidelines that were implemented differed considerably in content, from health maintenance reminders to alerts for active management of specific disease states. There were also notable variations in clinical settings—inpatient, emergency room, ambulatory clinic, private office, and public health department—and in evaluation methodology.
Fourteen studies reported some improvement in adherence to guidelines, seemingly independent of the information management services provided. Clearly, adherence to guideline recommendations can be improved in many cases using computer-based interventions. Likewise, documentation is regularly assisted with computers, but user satisfaction may be affected adversely by tedious data entry requirements in the absence of offsetting system benefits. In both studies with negative evaluations of user satisfaction,21 23 arduous data entry was suggested as a reason for poor system acceptance. Few studies examined patient outcomes to validate the effectiveness of the systems.
Many factors influence the success or failure of guideline implementation systems. While provision of a wide array of information management services may be important, it may not be sufficient to ensure success. To adequately evaluate the effect of those services on the success or failure of a computer-based guideline implementation, more of the confounding variables need to be controlled. In the studies described here, different types of guidelines, different settings, and different system implementations make conclusions about the relationship between information management services and outcomes difficult. In addition, a component of publication bias is likely to be present, in that the generally favorable results may represent a biased subset of system implementations.
Our information management services model was designed to provide a checklist for providing solutions that maximize workflow integration. Although this model may not cover exhaustively all factors responsible for implementation acceptance, we believe that it can be used profitably for the design of computer-based guideline implementation strategies and can serve as a framework for system evaluation. Future system developers should learn from the successes and failures of past systems.
The authors thank the members of the Guidelines Review Group at the Yale Center for Medical Informatics, who were instrumental in the conceptualization of the information management services model.