Cases & Commentaries
The Forgotten Med
A 78-year-old woman with a history of chronic obstructive pulmonary disease (COPD) came to the hospital with increasing shortness of breath and chest pain. Although she had no history of diabetes, her glucose level on admission was nearly 300 mg/dL. Along with initial treatment, the admitting physician prescribed once-daily insulin (glargine) and ordered finger-stick glucose checks four times daily (QID), with sliding-scale coverage for persistently elevated glucose values.
The patient ruled out for a myocardial infarction, and her COPD exacerbation improved after initial treatment. After 3 days, she was transferred to a skilled nursing facility (SNF) for continued observation and intensive respiratory treatments. Medication orders from the acute care ward were continued at the SNF.
Three days later, a physician evaluated the patient and reviewed her progress. He noted that, while glucose values in the hospital ranged from 150-250 mg/dL, glucose levels currently ranged from 90-140 mg/dL (essentially normal). Therefore, thinking that the patient’s glycemic control had improved, he discontinued the QID glucose checks and insulin sliding-scale orders. However, he failed to notice the existing order for glargine, a long-acting insulin.
Four days later, the patient became unresponsive. The physician ordered stat blood cultures, electrolytes, and a head CT scan. As staff prepared to transport the patient to radiology, the lab called to report a critical value—a glucose level of 22 mg/dL. The physician immediately ordered intravenous dextrose followed by an infusion, which led to a rapid improvement in the patient’s mental status. Luckily, the patient suffered no subsequent events, her glargine was discontinued, and she continued her rehabilitation at the SNF.
The incident led to an internal review of the case. The physician acknowledged that he had seen the glargine order earlier in her SNF stay but had forgotten about it when he discontinued the glucose checks and sliding-scale insulin orders. Also, as in most hospitals, the nursing medication administration record (MAR) listed the once-daily dose of insulin in a different location than the sliding-scale insulin, because one is a regular medication and the other given as needed. This seemingly added to the confusion among the day and night nursing staff—since it was the latter who administered the evening glargine and the former who performed the glucose checks and would have administered the sliding-scale insulin therapy.
While it is tempting to blame the physician for a lack of vigilance or the nurse for failure to recognize discordant orders, this case presents an opportunity to systematically improve safety. Creating a new hospital policy that mandates glucose checks in any patient receiving hypoglycemic agents (either insulin or oral agents) is a simple and common response. Though such policies can help raise awareness, they merely reinforce the providers’ existing clinical practice, which relies (to an unsafe degree) on human vigilance. Although heightened awareness may decrease error in the short term, the benefits extinguish rapidly, and the safety risks may exceed the capacities of human vigilance, regardless of training or reinforcement.(1)
A second common preventive approach aims to redesign workflow and build in safety checks. For example, a hospital might develop specialized documentation that integrates hypoglycemic medication orders, records of administration of those medications, and records of glucose measurements in a uniform place within the patient’s record. Custom documentation is common for higher-risk nursing tasks, such as documenting patient-controlled analgesia or the use of restraints. While these practices are an improvement over strategies that rely on vigilance alone, they increase already burdensome paperwork responsibilities in nursing care (recently estimated at 16% of nursing time ) and either fragment the nursing record or require redundant documentation. Furthermore, although potentially effective for specific problems, the technique scales poorly, since it is impractical to implement on paper task-specific documentation for every potential safety issue involved in medication ordering or administration.
A more powerful solution is a system that actively prevents orders for hypoglycemics from existing on the medication list without the presence of glucose checks. A “forcing function” prevents an action from being performed or allows it only if another specific action is performed first or concurrently. Paper-based order sets employ this strategy when they bundle corollary orders, such as glucose checks with insulin orders, so they are ordered as a unit. However, paper-based systems to bundle corollary orders continue to suffer from poor scalability. Computerized provider order entry (CPOE) systems make implementing forcing functions straightforward and far more scalable. Research demonstrated a 25% improvement in completeness of corollary orders (reduction in errors of omission) with a basic CPOE system.(3) Beyond corollary orders, a second study showed an 81% decrease in all medication errors and an 86% decrease in nonintercepted serious medication errors (harmful errors that reached the patient) (4) in a home-grown CPOE system with integrated decision support.
In human–computer interaction design (the design of the communication and cognitive tasks of working with software, including the visual user interface), forcing functions are divided into triggers and constraints. A trigger is a design element where an initial user action automatically results in another action, such as when closing a software application prompts the user to first save her work. A constraint is a rule stating under what conditions an action is allowed, and prevents actions that would violate a constraint. Requiring a driver to have his foot on the brake before shifting into reverse is a constraint, as is graying out menu items in a software application inapplicable to its current state. Returning to this case, having a new hypoglycemic order (the glargine insulin) prompt a corollary order for glucose checks within a CPOE system (a trigger) is simple. It is more subtle, however, to detect conditions arising other than when an order is placed, such as when an existing order is discontinued or expires, as in this case, when the discontinuation of the glucose checks failed to prompt providers to reconsider the existing insulin order. Building in this type of forcing function requires an if-then rule, stating “if a hypoglycemic is on the medication list, then glucose checks must be on the nurses’ work list.” Each time the orders are altered, the decision support rules must be scanned to see if the new state resulting from the proposed action violates any constraint; if so, an appropriate remedy is suggested to the ordering clinician. Systems that permit an unsafe action under current conditions, such as administering insulin in the absence of glucose checks, are termed underconstrained.
While building a rule-based decision support system might appear straightforward, the number of interactions among rules becomes very large when implemented in systems of even moderate complexity. Unanticipated interactions become difficult to predict and control.(5) The labor of producing and maintaining the rule base is large, and interruptions to clinician workflow become frequent. Interruptions in the form of alerts decrease user acceptance, and “alert fatigue” develops quickly (6), rendering the safety checks ineffective. “Noninterruptive” decision support, which alerts the user about a potential issue without diverting him or her from the current task, may enjoy better user acceptance and cause less alert fatigue. Examples of noninterruptive support are few in currently available clinical software but exist in other applications, such as a word processor that indicates a spelling error with red underlining but does not stop the user from typing. Institutions, however, would face the difficult choice of what constraint violations are sufficiently worrisome to warrant interruption, how to interrupt, and what medicolegal ramifications those decisions might have.
Even when finely honed, a single set of decision support rules will not satisfy all users. Although it is useful to remind many clinicians of the interaction between angiotensin-converting enzyme inhibitors and spironolactone (additive potassium retention)(7), cardiologists who order these medications daily and are intimately familiar with them will find an interruptive alert very unwelcome. Unfortunately, tailoring decision support by department, or even by individual, multiplies the cost and labor.
The benefits of CPOE depend on details of software design, user training, and institutional factors.(8) Weakness in any of these elements can create opportunities for error that did not exist in paper-based systems.(9) While many successful home-grown CPOE systems have highly customized clinician interfaces, interface design for CPOE has received insufficient attention from commercial vendors and lags well behind that found in other software applications, despite the critical human factors requirements on which safe clinical use depends. Beyond software, vendors should develop their expertise in the organizational factors necessary for successful health information technology applications and bring this expertise to customers as part of their sales package. Several studies have demonstrated the power of specific CPOE systems to reduce error at individual institutions, but the benefits may not apply to all CPOE systems in all places. CPOE is undoubtedly among our most powerful tools to enhance patient safety, but its optimal implementation remains an unfinished science, and an active area of research, commercial development, and entrepreneurial activity.
- Policy- and paper-based medication safety improvement strategies, while useful, are limited in their scope, flexibility, scalability, and power.
- Computerized provider order entry (CPOE) is a powerful tool to address a range of patient safety issues and a subject of intense national interest among payors and regulatory agencies.
- Implementing CPOE is complex and resource intensive, holds many challenges outside the technical issues of implementation, and may introduce new risks.
- Software design for CPOE systems is an unfinished science.
- Much of the utility of CPOE comes from clinical decision support, the implementation and maintenance of which is a separate challenge.
Russ Cucina, MD, MS Assistant Professor of Medicine Hospitalist Group & Medical Informatics University of California, San Francisco
1. Wachter RM, Shojania KG. Internal Bleeding. New York, NY: Rugged Land; 2004.
2. Korst LM, Eusebio-Angeja AC, Chamorro T, Aydin CE, Gregory KD. Nursing documentation time during implementation of an electronic medical record. J Nurs Adm. 2003;33:24-30. [ go to PubMed ]
3. Overhage JM, Tierney WM, Zhou XH, McDonald CJ. A randomized trial of "corollary orders" to prevent errors of omission. J Am Med Inform Assoc. 1997;4:364-75. [ go to PubMed ]
4. Bates DW, Teich JM, Lee J, Seger D, Kuperman GJ, Ma'Luf N. The impact of computerized physician order entry on medication error prevention. J Am Med Inform Assoc. 1999;6:313-21. [ go to PubMed ]
5. Clancey WJ. Mycin’s Map. In: Situated Cognition: On Human Knowledge and Computer Representations. Cambridge, UK: Cambridge University Press; 1997:29-45.
6. Payne TH, Nichol WP, Hoey P, Savarino J. Characteristics and override rates of order checks in a practitioner order entry system. Proc AMIA Symp. 2002; 602-606. [ go to PubMed ]
7. Juurlink DN, Mamdani MM, Lee DS, et al. Rates of hyperkalemia after publication of the Randomized Aldactone Evaluation Study. N Engl J Med. 2004;351:543-51. [ go to PubMed ]
8. Ash JS, Berg M, Coiera E. Some unintended consequences of information technology in health care: the nature of patient care information system-related errors. J Am Med Inform Assoc. 2004;11:104-112. [ go to PubMed ]
9. Koppel R, Metlay JP, Cohen A, et al. Role of computerized physician order entry systems in facilitating medication errors. JAMA. 2005;293:1197-203. [ go to PubMed ]