Improving patient safety is high on the agenda in healthcare since, in 1999, the Institute of Medicine published its report “To Err is Human”, which highlighted the magnitude of the problem of patients being harmed during medical care.1 It showed that as many as 98 000 people die each year from medical errors in hospitals. A large proportion of these errors concern medication errors at different stages of the medication use process, including prescribing, transcribing, dispensing and administering of drugs,2 with prescribing errors being the most common.3^–5 The frequency of the different types of errors varies across settings, but the differences may also be explained by the differences in definitions of medication errors and differences in the methodology of determining medication errors.

Next to defining the prevalence of these different types of medication errors, some studies have determined the potential harm these medication errors could cause without determining whether that harm actually occurred.56 Therefore, only an estimation of the clinical impact of medication errors can be made as occurrence of real harm related to medication errors is only inferred. Increasingly, prospective cohort and observational studies screen for injuries caused by medication, that is adverse drug events (ADEs).347^–11 These studies have also assessed whether an ADE that occurred was associated with a medication error and therefore was considered to be preventable. About 26% to 42% of the ADEs were preventable, and these preventable adverse drug events (pADEs) seemed to be mainly caused by prescribing and transcribing errors.34811

To minimise pADEs, it is important to know the relation between the subtypes of prescribing and transcribing errors and the risk of pADEs. Unfortunately, little is known about this relation. Yet, such information would provide important clues as to which type of errors cause most harm and should therefore be the primary focus for intervention and prevention programmes. Therefore, we performed a study to investigate the impact of the various types of prescribing and transcribing errors on pADEs in hospitalised patients.

METHODS

Setting, design and patients

This study is conducted in the framework of a study on the effect of a computerised Physician Order Entry system on Medication Safety and associated costs (POEMS study). The study was performed in three medical wards of the 1300 bed University Medical Center in Groningen (two general internal medicine wards and one gastroenterology/rheumatology ward) and in two medical wards (one geriatric and one general internal medicine ward) of the 600 bed teaching hospital “TweeSteden” in Tilburg and Waalwijk, The Netherlands. In these medical wards, the process of medication ordering and administration consisted of a hand-written system: physicians prescribe medication orders on charts and nurses transcribe these medication orders on administration charts.

The occurrence of prescribing and transcribing errors, and related harm was determined in patients hospitalised on these five medical wards using a prospective cohort design.

During a 5-month period, from July 2005 through November 2005, all patients admitted for more than 24 h to the study wards were included. Patients received a letter with information about the study, and they could object to inclusion.

Data collection

During daily ward visits, the investigators collected data on patients’ characteristics (sex, age, length and weight), diseases (reasons for admission and diagnoses), medication (medication orders during hospital stay) and adverse events (any untoward medical occurrences during stay) which consisted of newly upcoming symptoms or increasing of consisted symptoms. These data were prospectively extracted from the medical records, the medication order and administration charts. When the investigators noticed a potentially life-threatening error related to a medication order during the process of data collection, they intervened in the prescribing process for ethical reasons. Such errors were not excluded from this study.

Classification of prescribing and transcribing errors

Medication errors were categorised according to the classification scheme for medication errors developed by The Netherlands Association of Hospital Pharmacists.12 In this scheme a distinction is made between prescribing, transcribing, dispensing, administering and “across setting” errors. In this study, only prescribing and transcribing errors were recorded. Prescribing errors are subdivided into administrative errors (errors on readability, patient data, ward and prescriber data, drug name, dosage form and route of administration), dosing errors (errors on strength, frequency, dosage, length of therapy and directions for use) and therapeutic errors (interactions, contra-indications, incorrect monotherapy, duplicate therapy and errors on therapeutic drug monitoring or laboratory monitoring). Transcribing errors were classified as errors in the process of interpreting, verifying and transcribing of medication orders.

Inappropriate drug choices were not actively assessed and were only taken into account when these were obvious.

Classification of severity of errors

All medication errors were classified according to the severity of the consequence of the error using the scheme of the National Coordinating Council for Medication Error Reporting and Prevention (NCC MERP).13 The severity of the consequence of the medication errors could range from a medication error that did not reach the patient (B) up to a medication error that reached the patient and led to the death (I). In the NCC MERP classification, category A is meant for situations that can lead to a medication error. In this study this category is not used. This classification is illustrated in table 1.

Causality assessment

For all medication errors made while a patient was admitted, all adverse events that were extracted from the charts were assessed for a causal relationship. The relationship between a medication error and an adverse event only was assessed. In other words, we did not assess the relationship between prescribed drugs and adverse events (including non-preventable ADEs). This causality assessment was carried out by five pharmacists. In aid of that assessment, an algorithm was developed, based on both the NCC MERP scheme and the Yale algorithm, an algorithm for the causality assessment between a drug and an event.14 The first three items of the Yale algorithm were used; determination whether the event has been widely known to occur as a consequence of the drug’s administration, whether there might be underlying clinical conditions which are responsible for the event, and whether the timing of the event is as expected in case of an ADE.

After individual assessment by the pharmacists, four meetings took place, where consensus was reached for all cases on both causality and severity. Because we expected a priori that the reliability between the individual pharmacists would be low,1516 we made use of this consensus procedure. Other studies into medication safety applied a consensus procedure as well.61718 Although this method has its limitations too, such as dominant raters having an unproportionally large influence on outcome assessments, it seems to be the best practical solution.

Outcome

The measure of outcome was the prevalence of pADEs. A pADE was defined as an ADE that occurred due to a medication error and where the causality assessment procedure indicated that there was at least a possible relationship between ADE and medication error. In this paper, the term pADE is similar to preventable harm.

Determinants

The different types of prescribing and transcribing errors as determinants for pADEs were studied. Administrative errors were not taken into account in the analysis because there were no pADEs related to this type of error. Patients’ age, patients’ sex, number of errors related to one patient, number of medication orders related to one patient and drug groups, associated with high risk on preventable harm, were considered as potential confounders. Relevant drug groups were selected from both the literature19 and the associations with pADEs in our own data (percentage of pADEs per prescribed drug group). These were antidiabetics, anticoagulants, drugs for anaemia, corticosteroids, antibiotics for systemic use, anti-inflammatory drugs, analgesics, antiepileptics, psycholeptics and drugs for gout.

Data analysis

All data were processed with MS Access 2003. SPSS version 12 and the SAS statistical package version 9.1 were used for analysis. The association between type of medication error and pADEs was determined by logistic regression analysis with the patient as unit of analysis. Patients were included only when a medication error had occurred. Potential confounders were taken into account in a univariate analysis. Potential confounders from the univariate analysis (p<0.05) were included into a multivariate logistic regression model.

RESULTS

During the 5-month period of data collection, 558 patients with 592 hospital admissions were included in this study (28 patients were re-admitted once, and three patients were re-admitted twice). Four patients did not provide consent and were excluded from the study. Table 2 summarises the characteristics of all study patients.

A total of 7286 medication orders were written, of which 4369 (60%) contained at least one prescribing or transcribing error. A total number of 5725 prescribing or transcribing errors were identified, of which 103 (1.8%) resulted in preventable harm (fig 1). In nine cases, the study investigators intervened to preclude unacceptable patient harm: four dosing errors and five transcribing errors. These errors were classified as D: errors that required an intervention to preclude harm.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Frequency of patients, medication orders and medication errors. *Percentage of total number of medication errors.

Table 1 shows the frequency of each type of error classified within a severity category. The most commonly identified type of medication error was an administrative error, but this type caused no pADEs (fig 1, table 1). In contrast, 56 (16%) of a total of 340 therapeutic errors led to a pADEs. Examples of prescribing and transcribing errors related to pADEs are given in table 3.

Most prescribing and transcribing errors did not reach patients (severity category B). Still a substantial number of the prescribing and transcribing errors reached patients but did not lead to an adverse drug event during their hospital stay (665 errors). The majority of errors that caused pADEs concerned temporary harm (category E)—for example constipation.

Preventable harm was caused by 1.8% of all medication errors or 1.4% of all medication orders. A total of 14.8% patients suffered from preventable harm related to medication errors in the prescribing or transcribing process. (fig 1)

After adjusting for confounding factors in a multivariate analysis, therapeutic errors were more strongly associated with pADEs (odds ratio (OR) = 1.98; 95% CI 1.53 to 2.56) than transcribing errors (OR = 1.12; 95% CI 1.01 to 1.25) (table 4). In the multivariate analysis, dosing errors were not significantly associated with pADEs.

DISCUSSION

Our study shows that approximately 2% of the medication errors related to the prescribing and transcribing process lead to pADEs. Of these errors, therapeutic errors are more strongly associated with pADEs than transcribing errors. Although we expected a relation between dosing errors and pADEs, a significant association could not be demonstrated. Dosing errors were errors like dosing too high or too low, but also errors like unclear or incomplete dosages. This last category of errors was usually corrected before reaching the patient and did not lead to preventable harm. This could explain the absence of a significant association between dosing errors and preventable harm. Though administrative errors are the most common errors made, no related harm could be detected. Most administrative errors did not reach patients, probably because they were intercepted by nurses or pharmacy staff through the various checking procedures in the prescribing process.

Earlier studies have already shown that prescribing errors are the most responsible for preventable harm.392021 Our study shows that this is particularly the result of therapeutic errors. By calculating odds ratios, this study provides an estimate of the magnitude of the risk, which builds on the previously conducted descriptive studies.

No other determinants of pADEs were found after adjustment for type of error, suggesting that the impact of the other included factors, such as high age (⩾80), is “mediated” through type of error.

In our study, the frequency of errors is extremely high; more than half of the medication orders contained one or more errors. Although high frequencies have been mentioned in literature,2223 the number of medication errors in our study is higher. Differences in definitions and methodology are possible explanations for these findings. In our opinion, the very high error rates are primarily caused by the detailed classification scheme for medication errors. All kinds of administrative aspects (no administration route, missing of start date, etc.) were taken into account as errors. We defined these items as errors, while other studies did not. Our conclusion is, however, that these administrative errors do not (often) lead to patients being harmed. Nevertheless, removing these errors is still important for patient safety. Correction of administrative errors during the medication process can consume a substantial amount of time and effort, and patient safety may be compromised indirectly because less time remains for identifying and correcting those types of errors that do result in patient harm. Furthermore, the correction procedures are at risk from human failures or weaknesses in the system.24

This study has several potential limitations. A weak point of this study is that the causality assessment was made by pharmacists only. No other healthcare professionals (eg, physicians) were involved in the assessment procedure. Although the pharmacists in our study had broad clinical experience, a different clinical view from physicians could be expected. Nevertheless contrary to this expectation, Dean and Barber25 showed that the reliability of the assessment of medication errors’ severity was not affected by a rater’s professional background, when enough assessors were included in the procedure (three or more). However, a group of raters including different professions might increase the acceptance of our findings by the different healthcare professionals.

Another limitation of our study is the lack of data on reliability between the investigators who collected data (eg, kappa values). However, we think that this issue may have had a limited effect on our study findings as we used the classification scheme for medication errors developed by The Netherlands Association of Hospital Pharmacists12 that precisely defines specific types and subtypes of medication errors. The investigators were extensively trained in using this classification scheme at the start of the study. The investigators individually assessed the medication of the same 10 patients and then discussed differences in classification. Furthermore, during the whole study period, the investigators discussed on a regular basis how to collect and interpret data, and any extraordinary cases were classified in mutual agreement. This approach should have limited variability between the different investigators, but we have no objective means to tell if this assumption holds true. In our study the number of pADEs may be over-rated because all adverse events with a possible relationship to a medication error have been taken into account. However, the results may also underestimate the extent of medication errors, because only adverse events occurring during the hospital stay were retrieved and not those occurring after a patient was discharged. Medication errors which reached patients during the hospital stay and did not immediately lead to patient harm could have the potential to do so in the future. For example, no gastric protection during the use of a combination of a NSAID and prednisolone usually does not lead immediately to a gastrointestinal bleeding, but could do so in the future.

Furthermore, only internal medicine, gastroenterology, rheumatology and geriatric medical units in two hospitals were studied, so the results may not apply to other medical specialities or to other hospitals.

Finally, this study considered medication errors in the process of prescribing and transcribing only. To provide a full overview of the extent to which medication errors can lead to harm, administration errors should be studied as well. Although studies have been conducted into the incidence and the potential severity of administration errors,2627 more research is needed to explore the association of administration errors with the occurrence of pADEs, specifically in comparison with the other types of medication errors.

One of the main strengths of this study is that it provides information not only on error frequencies, as many other studies do, but also on the associated frequency of actual patient harm. Another strength of this study is the prospective nature and the use of an epidemiological design to determine potential associations between error type and ADEs.

To conclude, the findings indicate that a substantial percentage of the hospitalised patients suffer from pADEs due to prescribing and transcribing errors; in particular therapeutic errors (interactions, contra-indications, incorrect monotherapy, duplicate therapy and errors on monitoring) are clinically relevant. Intervention and prevention programmes should focus on these medication errors. A Computerised Physician Order Entry System especially with a clinical decision support system could be a possible solution to reduce these types of medication errors.28^–30 Future research is needed to determine the impact of such interventions on the reduction in therapeutic errors and preventable patient harm as well as the cost-effectiveness of these interventions.