Context Hospital computerized physician order entry (CPOE) systems are widely regarded as the technical solution to medication ordering errors, the largest identified source of preventable hospital medical error. Published studies report that CPOE reduces medication errors up to 81%. Few researchers, however, have focused on the existence or types of medication errors facilitated by CPOE.
Objective To identify and quantify the role of CPOE in facilitating prescription error risks.
Design, Setting, and Participants We performed a qualitative and quantitative study of house staff interaction with a CPOE system at a tertiary-care teaching hospital (2002-2004). We surveyed house staff (N = 261; 88% of CPOE users); conducted 5 focus groups and 32 intensive one-on-one interviews with house staff, information technology leaders, pharmacy leaders, attending physicians, and nurses; shadowed house staff and nurses; and observed them using CPOE. Participants included house staff,nurses, and hospital leaders.
Main Outcome Measure Examples of medication errors caused or exacerbated by the CPOE system.
Results We found that a widely used CPOE system facilitated 22 types of medication error risks. Examples include fragmentedCPOE displays that prevent a coherent view of patients’ medications, pharmacy inventory displays mistaken for dosage guidelines, ignored antibiotic renewal notices placed on paper charts rather than in the CPOE system, separation of functions that facilitate double dosing and incompatible orders, and inflexible ordering formats generating wrong orders. Three quarters of the house staff reported observing each of these error risks, indicating that they occur weekly or more often. Use of multiple qualitative and survey methods identified and quantified error risks not previously considered, offering many opportunities for error reduction.
Conclusions In this study, we found that a leading CPOE system often facilitated medication error risks, with many reported to occur frequently. As CPOE systems are implemented, clinicians and hospitals must attend to errors that these systems cause in addition to errors that they prevent.
Adverse drug events (ADEs) are estimated to injure or kill more than 770 000 people in hospitals annually.1 Prescribing errors are the most frequent source.2–5Computerized physician order entry (CPOE) systems are widely viewed as crucial for reducing prescribing errors2–3,6–17 and saving hundreds of billions in annual costs.18–19 Computerized physician order entry system advocates include researchers, clinicians, hospital administrators, pharmacists, business councils, the Institute of Medicine, state legislatures, health care agencies, and the lay public.2–3,6–10,12, 14–17,20–22 These systems are expected to become more prevalent in response to resident working-hour limitations and related care discontinuities23 and will supposedly offset causes (eg,job dissatisfaction) and effects (eg, ADEs) of nursing shortages.24–25 Such a system is increasingly recommended for outpatient practices (Box).
Adoption of CPOE perhaps gathered such strong support because its promise is so great, effects of medication error so distressing,circumstances of medication error so preventable, and studies of CPOE preliminary yet so positive.21, 26–28 Studies of CPOE,however, are constrained by its comparative youth, continuing evolution, need to focus on potential rather than actual errors, and limited dissemination (in 5% to 9% of US hospitals).29–36 Two critical studies21, 30 examined distinctions between reductions in possible ADEs vs actual reductions in ADEs; the former are well documented and often cited, but the latter are largelyundocumented and unknown. Studies of CPOE efficacy (17% to 81% error reduction) usually focus on its advantages2–3,6–11,14–16and are generally limited to single outcomes, potential error reduction, or physician satisfaction.28, 30, 34–40 Often studies combine CPOE and clinical support systems in their analyses.30, 40–41
In the past 3 years, though, a few studies21, 26–28,30–31,33, 42–46 suggested some ways that CPOE might contribute to medicationerrors (eg, ignored false alarms, computer crashes, orders in the wrong medical records). Several decades of human-factorsresearch, moreover, highlighted unintended consequences of technologic solutions, with recent discussions on hospitals.32–33,42–44,47–52
We undertook a comprehensive, multimethod study of CPOE-related factors that enhance risk of prescription errors.
We performed a quantitative and qualitative study incorporating structured interviews with house staff, pharmacists, nurses, nurse-managers, attending physicians, and information technology managers; real-time observations of house staff writing orders, nurses charting medications, and hospital pharmacists reviewing orders; focus groups with house staff; and written questionnaires administered to house staff. Qualitative research was iterative and interactive (ie, interview responses generated new focus group questions; focus group responses targeted issues for observations).
We studied a major urban tertiary-care teaching hospital with 750 beds, 39 000 annual discharges, and a widely used CPOE system (TDS) operational there from 1997 to 2004. Screens were usually monochromatic with pre-Windows interfaces (Eclipsys Corp, Boca Raton, Fla). The system was used on almost all services and integrated with the pharmacy’s and nurses’ medication lists.
This study was approved by the University of Pennsylvania institutional review board. The researchers were not involved in CPOE system design, installation, or operation.
Intensive One-on-One House Staff Interviews. To develop our initial questions, we conducted 14 one-on-one house staff interviews. An experienced sociologist (R.K.) conducted the open-ended interviews, focusing on stressors and other prescribing-error sources (mean interview time, 26 minutes; range, 14-66 minutes).
Focus Groups. We conducted 5 focus groups with house staff on sources of stress and prescribing errors, moderated by an experienced sociologist (R.K.) and audiorecorded. Participants were reimbursed $40 (average group size, 10; range, 7-18; and average length, 1.75 hours; range, 1.4-2 hours).
Expert Interviews. We interviewed the surgery chair, pharmacy and technology directors, clinical nursing director, 4 nurse-managers, 5 nurses, an infectious disease fellow, and 5 attending physicians. All interviews, except 1, were privately conducted by the same investigator (R.K.).
Shadowing and Observation. During a discontinuous 4-month period (2002-2003), we shadowed 4 house staff, 3 attending physicians, and 9 nurses engaged in patient care and CPOE use. We observed 3 pharmacists reviewing orders. The researcher (R.K.) wore a faculty identification badge. Observation notes were freehand but guided by the interview findings.
Survey. From 2002 to the present, we distributed structured, self-administered questionnaires to house staff who order medicationsvia CPOE. The 71-item questionnaire focused on working conditions and sources of error and stress. We report here on 10 CPOE-related questions. We constructed the survey after our interviews and focus groups, leading us to provide separate answer optionsabout sources of error and sources of stress; add questions on CPOE as a possible source of error risk, an issue that emerged in our qualitative research; and quantify the frequency of these error risks. Not all CPOE-related error risks are amenable to survey questions. We have robust survey results on 10 of the 22 identified error risks; these findings are presented with the qualitative findings.
The sampled population (N = 291) included house staff who typically enter more than 9 medication orders per month. The target study population excluded 648 residents in services that seldom use CPOE: pathology, podiatry, occupational medicine, anesthesia, radiology, radiation oncology, ophthalmology, and dermatology.
More than 70% of the questionnaires were administered at routine house staff meetings. Other house staffwere located via departmental coordinators or pagers. Participants received $5 coupons for local coffee shops. Two hundred sixty-one house staff (88% of the target population) completed the questionnaire.
Characteristics of the house staff were as follows. Of 94 interns contacted, 85 (90.4%) participated; of 96 second-year residents, 84 (87.5%) participated; and of 107 third- through fifth-year residents, 92 (85.9%) participated. The participating sample was 44.8% female, 66.3% white, and 32.5% were interns. Participants’ mean age was 29.6 years. These data did not differ significantly from characteristics of nonparticipants.
Our qualitative and quantitative research identified 22 previously unexplored medication-error sources that users report to befacilitated by CPOE. We group these as (1) information errors generated by fragmentation of data and failure to integrate the hospital’s several computer and information systems and (2) human-machine interface flaws reflecting machine rules that do not correspond to work organization or usual behaviors.
Information Errors: Fragmentation and Systems Integration Failure
Assumed Dose Information. House staff often rely on CPOE displays to determine minimal effective or usual doses. The dosages listed in the CPOE display, however, are based on the pharmacy’s warehousing and purchasing decisions, not clinical guidelines.For example, if usual dosages are 20 or 30 mg, the pharmacy might stock only 10-mg doses, so 10-mg units are displayed on the CPOE screen. Consequently, some house staff order 10-mg doses as the usual or “minimally effective” dose. Similarly, house staff often rely on CPOE displays for normal dosage ranges.
House staff regularly use CPOE to determine dosages (Table). In the last 3 months, 73% of house staff reported using CPOEdisplays to determine low doses for medications they did not usually prescribe; 82% used CPOE displays to determine range of doses (Table). Two fifths (38%-41%) used CPOE displays to determine dosages at least a few times weekly; 10% to 14% usedCPOE displays in this misleading way daily.
Medication Discontinuation Failures. Ordering new or modifying existing medications is usually a separate process from canceling(“discontinuing”) an existing medication. Without discontinuing the current dose, physicians can increase or decrease medication(giving a “double” total dose, eg, every 6 hours and every 8 hours), add new but duplicative medication, and add conflictingmedication. Medication-canceling ambiguities are exacerbated by the computer interface and multiple-screen displays of medications; as discussed below, viewing 1 patient’s medications may require 20 screens.
Discontinuation failures “for at least several hours” from not seeing patients’ complete medication records were reported by 51% (Table). Twenty-two percent indicated that this failure occurs a few times weekly, daily, or more frequently.
Procedure-Linked Medication Discontinuation Faults. Procedures and certain tests are often accompanied by medications. If procedures are canceled or postponed, no software link automatically cancels medications.
Immediate Orders and Give-as-Needed Medication Discontinuation Faults. NOW (immediate) and PRN (give as needed) orders may not enter the usual medication schedule and are seldom discussed at handoffs. Also, because medication charting is so cumbersome and displays so fragmented, NOW and PRN orders are less certain to be charted or canceled as directed. Failure to chart or cancel can result in unintended medications on subsequent days or reordering (duplications) on the same day.
Antibiotic Renewal Failure. To maximize appropriate antibiotic prescribing, house staff are required to obtain approval by infectious disease fellows or specialist pharmacists. Lack of coordination among information systems, however, can produce gaps in therapy because antibiotics are generally approved for 3 days. Before the third day, house staff should request continuation or modification. To aid this process, reapproval stickers are placed on paper charts on the second day. However, when house staff order medications, they primarily use electronic charts, thus missing warning stickers. No warning is integrated into the CPOE system, and ordering gaps expand until noticed. Some unintentional “gaps” continue indefinitely because it is unknown whether antibiotics were intentionally halted. In the last 3 months, 83% of house staff observed gaps in antibiotic therapy because of unintended delays in reapproval. Twenty-seven percent reported this occurrence a few times weekly; 13%, once daily or more frequently (Table).
Diluent Options and Errors. A recent CPOE innovation requires house staff to specify diluents (eg, saline solution) for administeringantibiotics. A few diluents interact with antibiotics, generating precipitates or other problems. Many house staff are unaware of impermissible combinations. Pharmacists catch many such errors, but their interventions are time-consuming and not ensured.
Allergy Information Delay. CPOE provides feedback on drug allergies, but only after medications are ordered. Some house staff ignored allergy notices because of rapid scrolling through screens, the need to order many medications, difficulties discontinuingand reordering medications, possibility of false allergy information, and, most important, post hoc timing of allergy information. House staff claimed post hoc alerts unintentionally encourage house staff to rely on pharmacists for drug-allergy checks, implicitly shifting responsibility to pharmacists.
Conflicting or Duplicative Medications. The CPOE system does not display information available on other hospital systems. For example, only the pharmacy’s computer provides drug interaction and lifetime limit warnings. Pharmacists call house staff to clarify questionable orders, but this additional step costs time and increases error potential. House staff and pharmacists reported that this method generates tension.
Human-Machine Interface Flaws: Machine Rules That Do Not Correspond to Work Organization or Usual Behaviors
Patient Selection. It is easy to select the wrong patient file because names and drugs are close together, the font is small, and, most critical here, patients’ names do not appear on all screens. Different CPOE computer screens offer differing colors and typefaces for the same information, enhancing misinterpretation as physicians switch among screens. Patients’ names are grouped alphabetically rather than by house staff teams or rooms. Thus, similar names (combined with small fonts, hectic workstations, and interruptions) are easily confused.
Fifty-five percent of house staff reported difficulty identifying the patient they were ordering for because of fragmented CPOE displays; 23% reported that this happened a few times weekly or more frequently (Table).
Wrong Medication Selection. A patient’s medication information is seldom synthesized on 1 screen. Up to 20 screens might beneeded to see all of a patient’s medications, increasing the likelihood of selecting a wrong medication.
Seventy-two percent of house staff reported that they were often uncertain about medications and dosages because of “difficulty in viewing all the medications on 1 screen.”
Unclear Log On/Log Off. Physicians can order medications at computer terminals not yet “logged out” by the previous physician,which can result in either unintended patients receiving medication or patients not receiving the intended medication.
Failure to Provide Medications After Surgery. When patients undergo surgery, CPOE cancels their previous medications. Whensurgeons order new or renewed medications, however, the orders are “suspended” (not sent to the pharmacy) until “activated” by postanesthesia-care nurses. But these “activations” still do not dispense medications. Physicians must reenter CPOE and reactivate each previously ordered medication. Surgery residents reported that they sometimes overlooked this extra process.
Postsurgery “Suspended” Medications. Physicians ordering medications for postoperative patients whom they actually observe on hospital floors can be de