Study Site and Setting

The present study was conducted in two surgical ICUs (with 10 and 14 beds) of an acute care tertiary teaching hospital in Korea with 950 total beds, covering general medical, surgical, and specialty care, including oncology. The two ICUs had a total of 41 registered nurses working three shifts per day, a staff physician as manager, and three internists. Patient severity levels were measured using the Patient Severity Classification developed by the Korean Association of Critical Care Nurses in 1992. Past research has demonstrated the Patient Severity Classification scale has adequate validity and reliability based on the Acute Physiology and Chronic Health Evaluation score [8].

Medication orders in the ICU were entered by the physicians taking charge of a patient's care (mainly residents) in each medical department or internists of the ICU. The hospital had a pharmacist prescription review process for inpatients' prescriptions. Medication administration was documented in an ICU flow chart as a log of all medication-related activities. The current study was conducted between May 1, 2008 and February 30, 2009.

Characteristics of the CPOE system

The hospital installed the first CPOE system of Medtrak (Sydney, Australia) for inpatient, outpatient, and ICU settings in 1996. It was revised and customized through partnership with a local IT solution company in 2001. Since then, the revised system was used. With the CPOE system, the user group had defined a medication administration protocol for drugs that are frequently prescribed in the ICU; this protocol was used in the training of in-house staff and nurses. The CPOE system had a function for checking drug–drug interactions based on recommendations of the Korean government for drug utilization review purposes (Health Review and Assessment Service) [9]. The CPOE system also displayed the patients' drug and food allergy information captured by nurse history-taking. During the study period, other medication-related decision support functions and electronic health records (EHRs) were not implemented. The CPOE system was checked regularly for new or modified medication orders. Nurses were made aware of any changes via notification pop-up windows on the order-retrieval screen. High-volume medications were administered at 0300, 0900, 1500, and 2100 hours (mostly at 0900 hours).

Measurements

The primary outcome measure was the error incidence of error-prone medications in the prescription, administration, and documentation phases. The secondary outcome measure was the rate of nurses' interception through compliance with the local medication administration protocols regarding prescription errors.

The error-prone medications were defined from a literature review and the site's local experience. We adopted the “high-alert medication” list generated by the Institute for Safe Medication Practices as medications associated with an increased likelihood of errors [4]. This list was compared with the local use patterns identified by analyzing medication orders prescribed in the ICUs for the past year. The top ten most frequently used medications via intravenous (IV), intramuscular (IM), and subcutaneous (SC) routes were chosen, of which eight drug categories were high-alert medications and two were diuretics and corticosteroids. The eight high-alert drug categories were adrenergic agents, calcium, digoxin, insulin, lidocaine, heparin, magnesium, and potassium chloride. Only these targeted medications were analyzed.

Observations and Data Collection

Observations were conducted on three randomly selected days per week between 0700 and 1800 hours. Six research nurses including authors (I.C., Y.J.C., and M.H.) were involved; each was assigned one of four different roles: direct observations (two nurses), order review (one nurse), chart review (one nurse), and error judgment and classification (two senior nurses). Data were collected using the MedObs data collector, a database program developed by the authors with Microsoft Access 2007. MedObs was structured into four sections that outlined what data should be captured for each role, the reference criteria for each item, and how to use the program (Fig. 1). These functions are described in detail elsewhere [10]. For retrospective medical chart reviews, patient prescriptions were anonymized and de-identified prior to analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Sample screenshot of the MedObs data collector (a data input form for overt observation).

https://doi.org/10.1371/journal.pone.0114243.g001

The research nurses participated in an on-site training session involving a description and demonstration of the direct observation technique, simulated medication administration scenarios, and ten real medication doses with practice data collection. A high level of agreement was reached for observational data (κ = 0.90). Written informed consent to participate in this study was obtained from 38 nurses after they were informed about the purpose of the study. The MedObs collector approached one nurse preselected randomly and observed her medication. Observations were conducted between May 1 and August 30, 2008, and the prescription data and medication administration records (MARs) were collected between May 1, 2008 and February 30, 2009. MedObs was used in all of the observation and chart review procedures. The data collected from observations and chart reviews were examined independently by two senior researchers using predefined error categories. Any disagreements regarding the classification were resolved by discussion.

Definition and Classification of Medication Errors

Prescription, administration, and documentation errors were defined by adapting previously reported definitions [11]–[13]. Fig. 2 shows the analysis scheme used in the present study for three phases. A prescription error referred to omitted information, unclear information, or conflicting information. Omitted information included incomplete prescriptions such as omission of diluent (IV mix fluid), dose, route, frequency, or instruction. Unclear information included duplications of the same order in a particular day or not specifying the total dose. Conflicting prescriptions included mismatches of a drug form and route, such as furosemide (20 mg) 3 ampules QD (quaque die, every day) ordered with normal saline (500 ml), but with the route given as IM. Administration errors were defined as any discrepancy between the prescriber's error-free medication order and what was actually administered to a patient. Administration errors were divided into the following categories: incorrect drug, route, dose, and instruction. Documentation errors referred to the discrepancy between what was observed and the MAR in the ICU flow chart.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. The scheme indicating the definition and classification of medication errors by clinical phase.

https://doi.org/10.1371/journal.pone.0114243.g002

Analysis

We compared the incidence of medication errors by error category for each phase and conducted a stratified analysis of our sample for prescription and administration errors to determine relevant causes. Incidence and comparison are presented as counts with percentages and rates.

Prescription Errors

More than 50% of the prescriptions included at least one error, despite the use of a CPOE system, which is higher than the rate of 12% found in an outpatient setting [13]. There are two possible explanations for this high error rate in the present study. First, although the CPOE system had several medication decision support functions, such as drug–drug interactions and drug allergies, the system might be more vulnerable to information omissions and insufficient for basic functions such as consistency checking for prescription attributes, which was the main reason for the high rate of prescription errors. These errors were intercepted frequently and had a relatively low potential for harm. However, instruction omission such as IV infusion rate or duration of high-risk medication has great potential for harm. Additionally, more than half of the unclear information on dosages was not intercepted.

CPOE systems have been widely viewed as being crucial for reducing the rate of prescription errors, which is the largest identified cause of preventable adverse drug events [14]–[16]. However, mixed results have been demonstrated for CPOE systems. One study that tested the CPOE systems of 62 hospitals in the U.S. identified a very large variation (10–82%) in the detection rate of medication orders judged likely to cause serious harm to patients [17]. It is important to ascertain whether the actual CPOE systems implementations are achieving goals. Currently in the U.S., thanks to the Federal public policy efforts to incentivize health IT use, renewed attention and focus on health IT-related errors, such as malfunctions during use, incorrect use by someone, or incorrectly entered, displayed, or transmitted errors, has been highlighted. The health IT industry has been encouraged to fix and improve them [18]. End-users are also recommended to monitor and review how the systems are used internally and to measure patient-safety sensitive adverse events and medical errors regularly, related to information technology. This present study was a useful opportunity to review and improve the hospital's system.

The second possible explanation for the high error rate found in the present study is that the physicians may not have sufficient knowledge of the correct use of the system for the medication protocols. Consistent with this notion, the present study found that diluent fluid omission was the second most frequently omitted information. The most frequent omission was route attribute, which could be associated with taking shortcuts, using default selections of drug attributes, or habitual omission assuming that the route attribute is self-explicit with drugs. Furthermore, the physicians who entered the orders were mainly in-house interns or junior residents; thus, although it was assumed they were knowledgeable regarding the use of CPOE system and medication protocol, this might not have been the case. To decrease these errors, an educational approach and a better system are needed, with the goal of increased attention to issues of patient-safety sensitive functions of CPOE. Health informatics academics argue that the health care industry is relatively early in its evolution but far less than 50% of users are familiar with EHR technologies [18].

ICU systems are clearly much more complex than inpatient and outpatient systems since more drugs are used and most of those drugs are administered intravenously. Cross-referencing between prescriptions and administrations in the present study indicated that nurses played a crucial role, intercepting 93% of the prescription errors, which is higher than the rate of 86% reported by Leape et al [19]. Our results show that most of the prescriptions with omitted information such as diluent or route were corrected by nurses ensuring compliance with the hospital's ICU medication protocol. Considering that CPOE is a communication tool to deliver information between a physician and nurses administering medication, any information omission could cause misunderstanding and mistakes. A prescription should be complete and explicit for nurses as well as prescribers.

To prevent prescription errors, our findings support the need for qualified and trained nurses, more staff physicians (intensivists) working in an ICU, and a better CPOE system. A better system could integrate medication protocols into the CPOE system as a form of clinical decision support to prevent prescription errors.

Administration Errors

The rate of administration errors found in the current study (19.0%) was higher than rates found in previous research [1], [20]. Specifically, one study in which a pharmacist was closely involved in the process of drug administration in a U. S. hospital reported a rate of 3.3% [21]. Another study found that the rate of medication errors in a medical ICU were 19.7% before and 8.7% after the implementation of bar-code-assisted medication administration [22]. This wide variation in medication error rates is due to the ratio of IV to PO (per os, by way of the mouth) doses and the use of diverse methods of data expression within the literature, such as the percentage of total opportunities for error or separate error rates for each phase of the medication process. The inclusion of only error-prone medications and the exclusion of PO medication in the current study might mean that the administration error rate of 19% is not actually much higher than rates found in previous research. However, given that the alert level and frequency of use of the medications were categorized as high, such errors could be quite influential.

Our examination of the order types of administration errors revealed that approximately two-thirds (37/58) of the errors occurred in association with verbal orders, with incorrect route errors being particularly prevalent. This implies that verbal orders, which constitute a relatively common part of the workflow in the ICU setting, are associated with a higher risk of administration errors. Specifically, those drugs that are in the form of liquids that can be administered either IV or SC/IM (e.g., furosemide, heparin, and methylprednisolone) were commonly associated with route errors. In the current study, verbal orders were communicated between clinicians in person or by telephone, and the prescriptions were entered into the CPOE system by the in-house staff or other delegated internists after the administration had been performed. Thus, it was unclear whether the recorded errors were due to miscommunication or inaccurate recollection by the recording physician. However, in the current study, all observers were asked to collect clinical-context data related to communications pertaining to medication between clinicians, and they reported no discrepancies between them except in one case. Furthermore, one interesting finding was that the documentation pattern differed between the corrected administrations of prescription errors and administration errors. More than 60% of the corrected administrations were recorded as prescribed rather than as administered, which were explicit documentation errors. For a prescription with omitted or incorrect information, that information was more likely to be omitted or incorrect in the administration document, even though the administration was performed correctly. It is not clear whether this pattern was relevant to the staff being concerned about leaving discrepant information between prescriptions and MARs or simply copying the prescription, but it resulted in no one knowing the correct dosage or route for a particular drug being given to a patient. In contrast, only 10% of administration errors were recorded as prescribed rather than administered, whereas 62% of administration errors were recorded as administered. In turn, this means that the nurses were more confident on medications classified as administration errors rather than those classified as prescription errors. These findings imply that the prescriptions entered later into the system were more likely to be incorrect than prescriptions before the administration phase, which were regarded as definite administration errors in the MAR.

Koppel and colleagues [23] are the only other authors to have discussed the immediate verbal order problem. The authors discussed the “now” (i.e., immediate) order as an example of medication errors caused by CPOE systems. These problems can be eliminated by clearly defining the verbal ordering process, integrating it into the CPOE system, and implementing consistency checking between the CPOE and the MAR. These functions alone could have contributed to a 64% decrease in the administration errors observed in the present study.

Documentation Errors

Documentation errors have received less attention from researchers than prescription and administration errors. This lack of attention may be due to medication documentation errors being regarded as unimportant or not causing direct harm to patients. Few studies have explored medication documentation errors by nurses despite the introduction of a sixth right (“right documentation”); hence, little is known about this type of error. However, maintaining an accurate MAR is essential for both the safety and quality of care, since this record serves as a log of all medication-related activities and as a reference for the team of health providers [24]. Documentation errors can also increase the rate of false-positive administration errors when only employing a chart review method in studies of adverse drug events.

A study found that the documentation error rate increased from 13.1% to 26.7% after an electronic MAR (eMAR) intervention and most of errors were recording omissions [22]. In our sample of handwritten records, recording omissions represented only 3.5% of more than 80% of documentation errors that included a high rate of omitted information. This finding implies that although the simple adoption of a system may reduce documentation errors, the incidence of new types of documentation errors may increase. According to a study by Moreland et al., the level of satisfaction with the idea of eMARs was lowest among nurses in ICUs, although that satisfaction improved over time after they were actually implemented [25]. Those authors noted that the eMAR system did not consider the workflow for specific medication orders that occur frequently among critically ill patients, such as numerous immediate, one-time, and as-needed medication orders. An eMAR system should be able to handle verbal orders effectively so as to reduce the rate of documentation errors.

Limitations

This study had several limitations. First, we did not obtain information about whether the medication errors caused actual harm to patients or about error severity, because we focused on the errors occurring during the medication-use process and the nurses' responses to prescription errors. Second, a single-site design was used; thus, it may not be possible to generalize the findings to other institutions that employ different processes for medication delivery, that use other forms of technology to prevent medication errors, or that have a clinical pharmacist available for consultation. The current research setting employed a pharmacist prescription review process, but it did not observe pharmacy intervention due to a shortage of clinical pharmacists. Continuing advances in technology mean that ICUs are moving away from handwritten documentation systems and toward eMAR systems. However, many ICUs continue to use paper-based flow charts and specialized medication charts for intermittent and continuous medications. Our findings could therefore benefit hospitals using CPOE systems with handwritten MARs or those considering adopting an eMAR system. The third limitation was that the night shift was not included in the observation time frame, which restricted the medication doses observed. Furthermore, the category of clinical appropriateness of drug doses in the prescription phase was not used in this study (with the exception of some clear-cut cases) because many cases were controversial due to the high severity of the patients' conditions. This might have contributed to a reduced rate of prescription errors. Finally, this study may have been susceptible to the limitations associated with direct observation (e.g., the observation effect). However, discussions during the observers' training sessions indicated that by the fifth day into the observation period, nurses were no longer affected by being observed.