This study focuses on administration error rates and the potential risk factors that can persist in a manual administration process that benefits from automated prescription and dispensing. The study was performed in two units with 10 years of experience using this technology.
The methodology used was direct observation of medication administration, which is the most efficient and practical medication-error-detection method and one that produces valid and reliable results.16–18
Since a common language was necessary to standardize diagnosis and systematize the detection, analysis, and recording of medication errors, we followed the Ruiz Jarabo Group medication-error taxonomy,15
which is an adaptation of the National Coordinating Council for Medication Error Reporting and Prevention taxonomy in the Spanish health system. Widely used by hospitals and other healthcare settings within the external medication errors reporting system of the ISMP-Spain, this taxonomy makes it possible to standardize description of the errors detected, the drugs involved, the cause of error, and the consequences and contributing factors involved.
The total error rate was high, as approximately one in five administrations were imprecise. However, this high incidence was due to wrong-technique errors (dietary restrictions); the incidence of other errors, excluding wrong-time errors, was significantly lower (6.8%). The main reason for such a high error rate was the lack of correct nursing working procedures, which generates three problems. First, the time schedule for medication is defined by nurses who often fail to consult administration guidelines; neither the CPOE nor the ADCs provide information about which medicines need to be administered on an empty stomach. Second, even though staff is aware of dietary restrictions, all the medication needed for the shift is retrieved from the ADCs at the beginning of the shift, without separating them in a different container. Third, although all the medication is removed from the ADCs by nurses using their personal fingerprint, oral medication at 13:00 and 20:00 is administered by nursing assistants, who have less knowledge about dietary restrictions.
Other considerations should be taken into account. Since these errors were not prevented—the observers were instructed to prevent only those errors that could produce an adverse event—the same error was repeated throughout the study; hence such a high error rate (128 times for PPIs and 39 and 27 times for tacrolimus and mycophenolate). In clinical terms, these errors could not be considered severe, as, in the case of PPIs, a reduction in bioavailability or a lack of effectiveness is not expected when they are co-administered with food, even though the summary of product characteristics recommends administration on an empty stomach. With immunosuppressive drugs, the clinical significance would have been significantly higher, but plasma concentrations were monitored in all cases. For these drugs, nursing staff did follow the schedule established (09:00 (administration every 24 h) and 09:00 and 21:00 (administration every 12 h)), as this is the schedule that best adapts to the patient's lifestyle outside the hospital.
In any case, we believe it is necessary to improve training in oral administration techniques and to change the way nurses work in the institution (ie, separating medication with dietary restrictions routinely and ensuring that all medications are administered by the nurse responsible for the patient). As a result of this study, the Pharmacy Department has started to adapt and implement the Guidance on the Interdisciplinary Safe Use of Automated Dispensing Cabinets, elaborated by the ISMP, which includes strict quality monitoring of nursing practice using data from the ADC management software.
Errors that may have been of greater clinical significance were much less frequent. No cases of wrong patient were detected, and the low incidence of wrong drug and dose was related to the introduction of profiled ADCs in the organization. However, despite this barrier control in dispensing, these errors still occur because nurses do not check the electronic prescription just before administration; this could have prevented two drug errors, 18 dose errors, three treatment duration errors, and four omission errors.
The analysis of potential error severity revealed that almost 96% of errors would cause no harm (Ruiz Jarabo 2008 taxonomy category C). Few of the errors were severe, because, as mentioned above, most errors were due to incorrect technique and dietary considerations, and very few errors were due to wrong drug or dose after implementing ADCs. Although this could indicate an excessively precise semantic structure for error reporting, it is the only way to detect non-severe errors that could be indicators of failures in the medication administration process and potentially lead to more severe errors in the long term. Our study is limited in that it did not take adverse events into account.
As for the potential risk factors involved, other than the positive correlation for PPIs, immunosuppressants, and prokinetics, a correlation was found for antibiotics due to problems with the reconstitution and dilution technique; this also explains the positive correlation for the parenteral route. We did not find any correlation between administration errors and nurses' age, category (career or not), or experience, or between the number of medicines per shift or number of beds under charge. Such a correlation could have been associated with understaffing. Finally, although a statistical association could not be found between organizational characteristics and error rate, we believe it is necessary to improve working procedures in units with automated prescription and dispensing (eg, the expiry date had been checked in only 0.7% of administrations, and 53% of drugs were not retrieved from the ADC just before administration). Another critical point was the poor communication between the physician and the nurse when treatment was modified in the CPOE. This could explain the positive correlation between the morning shift and the evening shift—when more treatment modifications were made—and the error rate. Also important is the fact that nurses do not cross-check the medication prepared with the prescription online just before administration.
Our results cannot generally be compared with those of other studies, mainly because of the error-detection method used (ie, direct observation versus voluntary reporting or medical chart review). Even with direct observation, reported error rates differ. These differences could be related to how medication errors are defined (eg, many studies define the error in relation to food intake as wrong time error, which is then excluded from the overall analysis due to its scant clinical relevance), the denominator used to calculate the error rate (eg, total doses administered vs 1000 patient-days), the type of medication use process (manual or automated), and the specific population evaluated (eg, adults, children, medical patients, surgical patients, ICU patients).
In a multicenter study in six hospitals in Catalonia,19
only 2% of the 1500 observed administrations involved an error. Omission was the most common, representing 40% of the total, followed by wrong time and wrong frequency. Only nine wrong infusion speed errors were detected, and wrong technique errors were not mentioned. These results differ considerably from ours, possibly because all the errors found were prevented—making it difficult to repeat the same type of error—and its undisguised methodology and multicenter design could have made it difficult to observe and detect the errors consistently in different hospitals. However, other studies show a closer error rate: Barker et al20
reported a 19% administration error rate in 36 healthcare facilities in Georgia and Colorado (USA). This rate is very similar to ours, except for the 43% in cases of omission, which was an uncommon error in our study. Fontan et al21
reported a 23% error rate after implementing electronic prescribing and ADC in a pediatric nephrology unit, and Bruce and Wong22
reported a 25% rate of parenteral drug administration errors by nursing staff on an acute medical admissions unit. Three studies analyzing the effect of implementing ADCs showed a 10.4%, 10.6%, and 13.5% error rate after implementation.6
The optimal solution for ensuring safety in the administration process today seems to be implementation of a BCMA system, which makes it possible to check the five rights at the bedside (right patient, right drug, right dose, right route, and right time).8–13
However, this technology cannot be implemented in Spain in the short term, not only because of its high cost, but also because of its important infrastructure requirements, as pharmaceutical manufacturers do not provide medications in unit dose packages with symbols that are readily deciphered by commonly used scanning equipment. Furthermore, as is the case with ADCs, BCMA will not prevent all types of preparation and administration error, such as wrong reconstitution or wrong technique, which are the most common errors detected in this study.
Although not connected to this BCMA system, the implementation of the e-MAR itself could reduce the error rate dramatically, as documenting administration in real time requires the prescription to be checked at the bedside and the medication to be administered by the nurse responsible for the patient (instead of a nursing assistant). Furthermore, a Clinical Decision Support tool can be incorporated to generate recommendations for nurses about preparation of medication before parenteral administration, adequate infusion rates, and scheduling according to dietary restrictions. Even though we are aware that the utility of the software depends on use at the bedside and adherence of nurses to recommendations, we expect to reduce administration errors by 80% with this technology. However, it will not prevent 100% of wrong drug and dose errors, or potential wrong patient errors, which are probably the most dangerous.
The limitations of this study are those associated with direct observation and the possibility of the Hawthorne effect. However, previous studies have demonstrated a negligible effect on the observed party through direct observation,16
and we did not find any differences between the error rate on the first and last day of data collection per nurse. Due to the nature and objectives of the study, adverse events were not taken into account. Finally, the study is also limited by its single-department design and the applicability to other institutions that have different processes for medication prescription and delivery or do not have a clinical pharmacist available.