Computerized Prescribing: Title and subTitle BreakBuilding the Electronic Infrastructure for Better Medication Usage

PHYSICIANS should never again write a prescription. Given the explosion of scientific information and advances in computer technology, prescribing medications on a blank piece of paper will soon seem as antiquated as ordering tinctures of botanicals in Latin. Instead, all medications should be prescribed on a computer interacting with 3 databases: (1) the patient's drug history, (2) a scientific drug information reference and guideline database, and (3) patient-specific information, including age, weight, allergies, diagnoses, and relevant laboratory results.

Computerized prescribing represents a long-overdue change in the way physicians practice medicine.^{1 - 3} Prescribing the old-fashioned way with pen and pad is so prone to mistakes that leaders in medicine and pharmacy urge converting to electronic prescribing as a top priority to address the leading cause of iatrogenic illness—medication errors, the majority of which could be prevented with computerized systems.^{4 - 6} Because of the increasing number of drugs, regimen complexity, continuously changing drug indications, and adverse effects, physician memory can no longer serve as a reliable bridge between research advances and clinical practice.⁷ Despite this need, less than 1% of the 3 billion prescriptions in the United States this year will be "written" on a computer (William Lockwood, Jr, American Society for Automation in Pharmacy, Blue Bell, Pa, written communication, November 1, 1997). Furthermore, few systems fulfill even our minimal criteria for system design.

There has been growing interest in direct clinician order entry into pharmacy computers in both inpatient and outpatient settings. Ending pharmacists' telephone tag when they cannot decipher illegible handwriting, or need to ensure formulary compliance or corroborate benefit eligibility promises substantial efficiencies for the pharmacist, physician, and patient.^{8 - 9} Transmission of electronic prescriptions directly to the pharmacy speeds inpatient delivery time or ensures that medications are ready when an ambulatory patient arrives at the pharmacy.^{10 - 11} Moreover, the convenience of renewing medications via a few mouse clicks instead of manually recopying multiple medications would save physicians countless hours.

While these efficiencies associated with computerized prescribing portend gains in cost-effectiveness, they pale in comparison to its more important potential to transform the entire medication use process.^{2 - 3 ,11 - 13} We fear that this profound transformative potential is receiving insufficient attention as vendors scramble to place their software products and as hospitals, managed care and pharmacy benefit managers, and pharmaceutical manufacturers focus on the economic advantages.

Clinical issues, such as those illustrated in Table 1, should provide the rationale and foundation for electronic solutions. While an aura of inevitability characterizes the expansion of computerization, without a guiding clinical vision there is no guarantee that the needs of patients and clinicians will be met optimally. Lest this larger goal of improving patient care be lost in the drive to automate, we outline 8 key areas in which computerized prescribing might transform care processes and patient outcomes. In each area, we identify existing problems and explore ways computerized prescribing could improve these functions.

Currently, a physician treating a patient with gonorrhea must either memorize or manually look up the latest Centers for Disease Control and Prevention (CDC) recommendations. It would be easier to type "GC" and have the regimen flow onto the prescribing screen. The clinician could either accept this guideline or choose an alternate regimen when indicated (however, the computer would already "know" patient allergies, and this information would be incorporated into the recommendation). Using instantaneously updated CDC guidelines in this way epitomizes expert guidance based on a drugs-of-choice model.^{1 ,14}

Such a powerful feature raises critical questions related to software design, clinical decision making, and guideline development. Recommendations need to be flexible. Early experience with the PRODIGY demonstration project under way in England suggests that rigid protocols requiring prescribers to first enter a diagnosis, while useful, can also lead to frustrations and delays.^{15 - 16} Software design should also recognize that physicians often deviate from recommendations for good clinical reasons—information worth capturing by a computer prompt, enabling subsequent evaluation and revision of guidelines.

This advance turns on its head the prevailing drug utilization review paradigm, with its after-the-fact inspection, second guessing, and outlier targeting.^{17 - 19} Because the criteria are embedded in the prescribing software, these dynamic guidelines are simultaneously integrated into care and field-tested rather than mechanically applied weeks later when they are of little benefit to the patient.^{20 - 21}

Evidence suggests that physicians will significantly improve their ordering practices in response to electronically delivered recommendations.^{22 - 27} Most physicians welcome guidelines when they are practical, publicly developed, evidence- and consensus-based, unbiased by commercial considerations, specific but flexible, locally modifiable, and continually updated.^{28 - 30} Unfortunately, currently marketed software does not meet these expectations. Moreover no mechanism exists currently to develop such optimized, dynamic prescribing guidelines.^{31 - 33}

The authoritarian model, in which the physician, with exclusive comprehensive medical knowledge, pens an order and the patient is expected to comply, is dead.^{7 ,34 - 36} Making the patient an active partner in both selection and implementation of therapy represents a cornerstone of effective treatment.^{36 - 38} This requires a dialogue between physician and patient. Adding the computer makes 3.

The computer ought not to sit between the patient and provider as another barrier to effective communication, further mystifying the physician's knowledge. Instead, patient and provider should sit on the same side of the screen, with the computer interactively mediating the knowledge possessed by the patient, provider, and medical literature (Figure 1).

The physician can review with the patient the indications, possible adverse effects, cost, and alternatives (including nondrug therapy). The computer's unique capabilities to display cascading levels of detail should be exploited. Time permitting, providers and patients can zoom in on more detailed information and/or zoom out for a summary of key take-home messages. Patient concerns are thus addressed up front, not after the patient goes home and consults the Physicians' Desk Reference or the Internet. Application of this technology transforms prescribing into shared decision making, as patient and provider negotiate the best therapy via joint review of information in the computer, which then records their decision.^{39 - 40}

In 1990, Healthy People 2000 set the goal of electronically screening 75% of prescriptions for drug interactions and allergies.⁴¹ While most pharmacies now use computers with such capabilities, they generally lack reliable software and patient-specific information (eg, allergies and diagnosis) to carry out this function dependably.

A recent investigation of 245 pharmacies found that more than 30% filled simultaneous prescriptions for potentially lethal combinations, such as terfenadine and erythromycin.^{42 - 43} Various drug-drug interaction screening software programs demonstrate poor agreement, and often flag numerous insignificant interactions that can overwhelm pharmacists, leading them to ignore significant interactions.¹⁷ Our outpatient pharmacy computer lacked allergy information on 72% of patients whose clinic charts clearly documented a drug allergy.⁴⁴ As long as there is no linkage between pharmacy and clinical data, such as allergies or diagnoses, automated screening will perform suboptimally.

Reengineering screening for drug-drug or drug-disease (β-blockers in asthma) interactions by moving it upstream from the pharmacy to the moment the prescription is written not only allows more timely detection and prevention of problems but also redefines the roles of physician, pharmacist, and pharmacy database. Pharmacists no longer function mainly as data transcribers. The pharmacy database now becomes an interactive component in a system automatically overseeing the patient's care, rather than a passive repository sitting at the end of the dispensing process.⁴⁵

Pharmacy and laboratory functions are intimately related. Abnormal renal or hepatic laboratory results often necessitate altering drug regimens. Many medications require laboratory monitoring, the results of which have an impact on subsequent prescriptions.⁴⁶ The notion that the process of putting chemicals into humans' bodies should be disconnected from the systems that sample their biochemical (or hematologic) effects is difficult to understand, yet a recent survey of 10 Chicago hospitals disclosed that none had laboratory and pharmacy computer systems that interacted directly with each other (G.D.S., unpublished results, September 1996).

As a result of this disconnectedness, for example, our hospital fills several hundred oral potassium prescriptions annually for outpatients who have a concurrent elevated potassium level recorded in the laboratory computer,⁴⁷ and patients with toxic reactions to theophylline continue to receive theophylline infusions for hours, even days, after an elevated level is recorded in the laboratory.⁴⁸ Clinical pharmacists spend much of their time working to bridge this gap between laboratory and pharmacy, overseeing therapeutic drug monitoring or dosing.

Systems such as those developed at Regenstrief Institute, Indianapolis, Ind; LDS Hospital, Salt Lake City, Utah; and Brigham and Women's Hospital, Boston, Mass, have demonstrated the cost-saving and lifesaving benefits of such drug-laboratory checks.^{3 ,21 ,49} Several commercial vendors have also developed systems in which laboratory and pharmacy data interact.⁵⁰ Using rule-writing software, the computer is programmed, for example, to check the most recent digoxin and potassium levels when a prescription order is placed for these drugs, or to question a vancomycin prescription for patients who have no blood cultures or one positive for a methicillin-sensitive organism.^{51 - 53} Such linkages not only help prevent gross errors in drug choice and dosing but could also help fine-tune dosing for drugs such as anticonvulsants or anticoagulants.⁵⁴ The potential for improving utilization of both the laboratory and pharmacy is enormous.⁵⁵

Dosing mistakes are the most common and preventable type of major medication errors.^{6 ,56 - 58} Software programs can perform dosing calculations and automated checks, thereby helping to prevent egregious mishaps, such as lethal chemotherapy overdoses.^{2 ,59 - 60}

Adding a time dimension, including linkages to outpatient scheduling software, creates a powerful tool for calculating dosing schedules, proper pill quantities, and monitoring compliance. Because these 2 functions—prescribing and scheduling—are so integral to outpatient visits, automating them together represents a logical and timesaving (but rarely considered) first step toward computerizing a physician's office.

Electronic prescribing not only saves time, it captures it, enabling the creation of graphic flow sheets. Longitudinal medication summaries can be displayed along with clinical and laboratory events, which can be related to drug and dosage. Relationships previously buried in the handwritten chart are uncovered.⁶¹ Aggregating treatment data for multiple patients over time allows clinicians to better project the natural history of disease and response to medication, thereby mapping the natural history of a drug for individual patients as well as population outcomes.^{62 - 63}

Despite legislation mandating patient medication information and counseling, pharmacists' compliance has been suboptimal and physicians' improvement undetectable.^{64 - 67} Even when teaching is provided, it is often poorly documented, and the efforts of pharmacists, physicians, and nurses remain uncoordinated.

Education should go beyond a one-time, one-size-fits-all leaflet shoved at a worried patient by a busy pharmacist.^{65 ,68} Education should be a multipronged, multiprovider, multimedia process that begins the moment drug therapy is being considered. Using the computer, the clinician can walk the patient through the relevant information during drug selection, simultaneously formulating both the prescription and the written education message from materials in the database. Information particularly relevant for that patient can be highlighted on the screen and the individualized printout.

Nurses and pharmacists could build on this physician-patient encounter, reenforcing key messages, using the leaflet and the same database to document not only that education was done but also what was taught. This redefined counseling role must recognize that communication now occurs across a continuum of settings, including the home via telephone. While many modern service industries provide telephone support based on entries in a customer-specific database, medicine rarely does so. Pharmacists, if liberated from their transcriptionist tasks, can play an expanded role in monitoring and interactive patient education, responding to and documenting medication questions and concerns that arise between visits.^{5 ,69 - 71}

What questions regarding adverse effects should be asked of a patient taking a drug? Beyond the obvious, such as whether there is bleeding if taking an anticoagulant or hypoglycemic symptoms if taking insulin, the medical literature has not delineated the symptoms and signs that need to be checked for a particular medication. Fears that querying patients about known reactions might evoke false-positive (ie, imagined) adverse effects have proven unfounded.⁷² Rather than false positives, the problem has been poor sensitivity of adverse effect screening, leading to wholesale failure of early detection, recording, and reporting of adverse drug reactions.^{73 - 74}

Just as pilots and mechanics are expected to check a list of potential problems before taking a plane back into the sky, it seems reasonable that physicians should solicit for known drug problems before renewing a prescription. The computer can facilitate adverse effect documentation, possibly via a checklist the patient completes in the waiting room or online. This necessitates prioritizing key effects to avoid overwhelming the patient with queries about hundreds of reported symptoms, as well as prompts for possible novel effects. With proper design, this provides the foundation for a comprehensive postmarketing surveillance system.

When a drug treatment is discontinued there is usually a reason. While hospitals do somersaults trying to "comply with JCAHO [Joint Commission on Accreditation of Healthcare Organizations]" requirements for reporting adverse drug reactions, the process of capturing this and other important outcome information could be simplified.^{75 - 76} In addition to flagging "alerting orders" (for example, stat diphenhydramine or vitamin K), any order to discontinue a medication might prompt the clinician to indicate why: (1) adverse reaction, (2) symptom/disease resolved, (3) failure to achieve desired therapeutic response, or (4) more desirable alternative.

By combining drug indication data (automatically recorded when the prescription is written) with adverse effect data, usage patterns, and the reasons for discontinuing therapy, we have the potential to create a postmarketing surveillance system that has eluded researchers and the Food and Drug Administration for decades.^{13 ,75 ,77 - 78} Linkage of this information with serial functional status, mortality, hospital admission, and other new-diagnosis data would open significant opportunities for outcomes research.^{79 - 80}

Thus, the greatest potential of computerized prescribing lies in its power to help create the infrastructure for a longitudinal patient record and a knowledge-generating database. The current emphasis on profiling prescribing patterns and targeting outliers needs to be replaced by an enhanced database supporting continuous improvement oriented toward learning rather than judgment and surveillance.⁸¹ Only then can practicing physicians honestly, accurately, and confidentially record the clinical data they alone can supply, and only then can society feel assured that the care received by each patient reflects state-of-the-art medicine and that every patient experience is contributing to the care of those who follow.⁸²

Computerized prescribing restructures the way scientific and treatment information is used and the way health professionals work.⁸³ A key component of physician order entry, electronic prescribing of medications represents an important step both in the creation of the computerized patient record and toward the more rational use of drugs.^{2 ,21 ,84}

While the voyage toward increased automation appears inexorable, historical impediments (Table 2) and potential problems need to be understood and overcome. Just as barnacles dragging on a ship impede its mobility and maneuverability, these constraints interfere with the speed and direction of progress. Overcoming these limitations will require thoughtful innovation and testing. Were the task simple, such a system would have been implemented long ago. Nonetheless, a number of other countries have made considerable progress, while the United States lags behind.^{2 ,16 ,80 ,99}

Vigilance will be required to anticipate and address inherent limitations and new problems raised in the implementation of computerized prescribing. Accumulating experience shows that automation can introduce unforeseen errors, such as dependence on alarms that fail.^{56 ,100} Likewise, it has been shown that physicians may change their behavior in response to computerized alerts but often retain scant knowledge of the educational message, so that their behavior reverts to baseline when the alerts are removed.²⁷ What happens when residents who are trained using computer order entry then go out to practice in an environment lacking such support? More critically, how can the technology be reliably safeguarded against crashes and security breeches?

A more fundamental question is whether electronic prescribing will be implemented primarily to support physicians and patients who seek to optimize pharmacotherapy or will be driven more by marketing and economic considerations.^{21 ,96 - 97 ,101 - 103} Commercial interests, particularly pharmaceutical manufacturers and pharmacy benefit management firms, see the computer as a tool to help shape the drug information clinicians receive and to obtain data on the medications they order.^{102 ,104 - 106} Lest the computer become a Trojan horse delivering an army of extraneous concerns to examination room desktops or merely a way to automate the inefficiencies of dozens of conflicting formularies, clinicians must provide scientific vision and professional leadership to guide the development and implementation of computerized prescribing.

Developing computerized prescribing as a clinical rather than a commercial tool will require a consortium of health professionals to proactively create validated software and care guidelines in the public domain, while simultaneously ensuring patient confidentiality, professional autonomy, and accountability (Table 3). Making prescribing a collaborative venture rather than a contested terrain¹⁰⁷ necessitates rapid sharing of the best practices and outcomes, made possible only by shared, standardized electronic data.^{108 - 109}

Although the task of guiding this proposal from the conceptual level through the thickets of comprehensive design, economic feasibility, and practitioner acceptance is formidable, the justification for action was sounded by Relman¹¹⁰ 15 years ago: "It is the cost of ignorance that has now become too steep for us to bear." Electronic prescribing, supported by the appropriately designed electronic medical record, constitutes a foundation for overcoming this constraint.