A Model for Monitoring Spontaneously Reported Medication Errors Using the Adjuvanted Recombinant Zoster Vaccine as an Example

A European legislation was put in place for the reporting of medication errors, and guidelines were drafted to help stakeholders in the reporting, evaluation, and, ultimately, minimization of these errors. As part of pharmacovigilance reporting, a proper classification of medication errors is needed. However, this process can be tedious, time-consuming, and resource-intensive. To fulfill this obligation regarding medication errors, we developed an algorithm that classifies the reported errors in an automated way into four categories: potential medication errors, intercepted medication errors, medication errors without harm (i.e., not associated with adverse reaction(s)), and medication errors with harm (i.e., associated with adverse reaction(s)). A fifth category (“conflicting category”) was created for reported cases that could not be unambiguously classified as either potential or intercepted medication errors. Our algorithm defines medication error categories based on internationally accepted terminology using the Medical Dictionary for Regulatory Activities (MedDRA®) preferred terms. We present the algorithm and the strengths of this automated way of reporting medication errors. We also give examples of visualizations using spontaneously reported vaccination error data associated with the adjuvanted recombinant zoster vaccine. For this purpose, we used a customized web-based platform that uses visualizations to support safety signal detection. The use of the algorithm facilitates and ensures a consistent way of categorizing medication errors with MedDRA® terms, thereby saving time and resources and avoiding the risk of potential mistakes versus manual classification. This allows further assessment and potential prevention of medication errors. In addition, the algorithm is easy to implement and can be used to categorize medication errors from different databases.


Introduction
Medication errors are defned as unintended failures in the drug treatment or vaccination process (during prescription, storage, distribution, preparation, or administration) that lead to, or have the potential to lead to, harm to the patient [1].As it has been estimated that about 18.7%-56.0%of adverse events (AEs) among hospitalized patients are a result of preventable medication errors, they have been recognized as a major public health burden [1].Tese errors can occur with any medicinal product, at any step in patient care, and in any care setting [2].Te most common medication errors in hospitals are prescription and administration errors, which account for about 75% of medication errors in this setting [2].Some examples include prescription for or administration to the wrong patient, failure to prescribe an indicated medication or prescribing without indication, administration of an inappropriate dose or via the wrong route, and failure to administer the medication when due [2,3].Several factors can infuence the incidence of medication errors.Tese factors can be related to the medication itself (e.g., similar sounding names or a low therapeutic index), to the patient (e.g., age, comorbidities, nonadherence to the medication, impaired cognition, or polypharmacy), or to the healthcare professional (e.g., use of abbreviations in prescriptions, unclear handwriting, or lack of up-to-date knowledge) [2,3].
To prevent medication errors, a process of care should be designed to ensure patients are protected against these errors and their potential harm.Tis can be achieved by ensuring eforts are made by regulatory authorities and manufacturers (e.g., clear product label information), by ensuring an upto-date medication list, by developing educational programs or by using information technology (e.g., drug databases or computerized physician order entry) [3].It is important to monitor and analyze medication errors, whether they result in harm or not.Findings must be communicated to improve the process of care and ensure that the risk of such errors can be minimized throughout the product life cycle [2,4].Since 2012, the European Union pharmacovigilance legislation requires the reporting of all AEs resulting from medication errors to EudraVigilance [5].To support stakeholders involved in the reporting, evaluation, and prevention of medication errors with implementing this legislation, the European Medicines Agency (EMA) published a good practice guide in 2015 to improve recording, coding, reporting, and assessment of medication errors, regardless of their association with AEs.Intentional overdose, of-label use, misuse, and abuse are not in the scope of this guide [1].
As part of pharmacovigilance reporting, a proper classifcation of medication errors is needed [1].Tis process can be tedious, time-consuming, and resource-intensive.To facilitate this classifcation, GlaxoSmithKline (GSK) has developed an algorithm that allows automatic categorization of medication errors from its spontaneous report database, based on Medical Dictionary for Regulatory Activities (MedDRA ® ) preferred terms (PTs).
Te current paper describes the algorithm and its advantages and limitations.It also gives examples of possible visualizations of medication errors using a previously described tool [6].As an example, we present vaccination errors associated with the adjuvanted recombinant zoster vaccine (RZV, Shingrix, GSK), which was frst licensed in October 2017 [7][8][9].Tis vaccine was chosen as prior experiences with other vaccines indicated that reports of vaccination errors are highest in the period shortly after licensure [10].RZV entered the market when another vaccine (Zostavax, Merck Sharp, and Dome) requiring diferent storage conditions, preparation, and administration procedures had already been available for a decade [11,12].We previously observed that the lack of familiarity with the RZV vaccine likely contributed to vaccination errors [13].Te use of the visualizations helped us to quickly identify and gain insights into the types of errors reported with this vaccine and to identify potential areas where preventive measures were benefcial [13].

Materials and Methods
GSK collects spontaneous reports of all AEs following immunization with its vaccines in its worldwide safety database as per good pharmacovigilance practices.Tese spontaneous report data from unsolicited communications describing one or more AEs in vaccinated patients are either submitted to GSK directly and voluntarily by individual reporters (e.g., healthcare professionals, regulatory authorities, or consumers, who report for themselves or others) or are identifed in the scientifc literature or interactive digital media as single case reports [6].AEs and medication errors are encoded in the database using the MedDRA ® terminology, the international medical terminology developed under the auspices of the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH), in line with EMA's good practice guide [1].
Medication errors may trigger a series of events, and more than one stage in the treatment process may be afected by an error.For example, a prescription error can lead to a dispensing error and consequently result in an administration error.Terefore, one spontaneous report (further referred to as a "case") can contain more than one medication error term in MedDRA ® .It is important to capture the primary error and any subsequent errors reaching the patient and to assess the clinical consequences for the patient.Likewise, for a given medication (e.g., a vaccine), more than one dose can be recorded in a case, with medication errors reported after each dose.Tis algorithm has the advantage of classifying the same case into diferent categories at the product and dose level.Hence, if more than one product is recorded as suspect (e.g., in the case of vaccine coadministration), the algorithm will distinguish which product has been administered incorrectly.
For a reported case, each dose of a selected product is classifed by the algorithm into one of the four main categories of medication errors, in line with the EMA's good practice guide [1]: potential medication errors, intercepted medication errors, medication errors without harm (i.e., not associated with adverse reaction(s)) and medication errors with harm (i.e., associated with adverse reaction(s)) (Table 1).A ffth category ("conficting category") was created for reported cases that could not be unambiguously classifed as either potential or intercepted medication errors (Table 1).For these cases, corrections to the database entry should be requested to the case management group.A fowchart is presented in Figure 1, and the code can be found in the S1 Supplementary Materials.
Events coreported with a medication error are by default considered harm, regardless of whether they are caused by the medication error, except for the following MedDRA ® PTs (i.e., nonvalid-coreported AE): all MedDRA ® PTs be- longing to the primary system organ class "product issues," the high level group term "of-label uses and intentional product misuses/uses issues," the high level terms (HLT) "adverse efect absent," "exposure associated with pregnancy, delivery and lactation," "normal newborn status" and "normal pregnancy, labor and delivery," or the PT "breastfeeding," which are not adverse reactions/AEs.Te selection of the MedDRA ® PTs can be adapted according to MedDRA ® version updates.
Te automated process allows categorization of new cases as well as retrospective categorization in searches.To validate the algorithm, two methods for categorizing medication errors were compared, one using the algorithm

Visualization of Medication Errors.
Regular and standardized review of safety data is essential.Terefore, GSK routinely performs quantitative signal detection for its products.A quantitative signal of disproportionate reporting at the MedDRA ® PT level for a vaccine-event pair is gen- erated if the lower limit of the 95% confdence interval of the stratifed (by sex, age group, geography, and reporting period) proportional reporting ratio is above 2, and at least 3 cases are reported.Tese quantitative signals, along with relevant visualizations, are made available on a previously described customized web-based platform, the signal mining and management (SMM) tool [7].Diferent algorithms and visualizations are integrated into this tool to ease medical review and data analysis, which includes mining the raw data and signals, looking at trends, testing hypotheses, reviewing clinical details of cases, etc. Visualizations specifc to medication errors have been developed in the SMM tool, as further elaborated below.
To illustrate the strengths and possibilities of developing visualizations on top of the automatic categorization of medication errors by our algorithm, we present and describe the visualizations for the category "medication error without harm," using spontaneous reports associated with RZV as an example.We do not aim to discuss specifc fndings here but merely want to use this example to illustrate the methodology.
Figure 2 shows an overall view of the diferent visualizations as they appear in the SMM tool.Data are shown for a specifc period of time ("period"), which can be selected in the tool (Figure 2, blue box) and can be compared to the cumulative period ("total," from entry of the frst case in the database until the date of freezing).A drop-down menu allows the user to visualize the information for one of the predefned medication error categories (Figure 2, red box).Tis information includes a tabular view of medication errors by MedDRA ® PT (grouped by MedDRA ® HLT) (Figure 2, purple box) and graphs depicting the evolution over time of reported medication errors (Figure 2, green and orange boxes, Figures 3 and 4).To ease the data review and its documentation, the table with medication errors by PT (Table 2) can be exported in a diferent format (i.e., pdf ).
Visualization of the evolution over time of the ratio of medication errors without harm over all spontaneous reports for the selected vaccine in the database (Figure 2, green boxes, Figure 3) allows the identifcation of trends in support of safety signal detection.Proportions rather than absolute numbers are shown because changes in absolute numbers can be the result of varying reporting habits or variations in the number of administered doses.Such changes are expected to have an equivalent impact on the number of medication errors and on the number of all other spontaneous reports and would therefore not have a major impact on their ratio.Te graphical representation shown in Figure 3 enabled the immediate identifcation of a safety signal due to the high proportion Table 1: General principles of the algorithm used to categorize medication errors for a dose of product X (MedDRA ® version 23.1).

Category Algorithm Potential errors
If one of the reported MedDRA ® PTs is "Circumstance or information capable of leading to medication error," the case is categorized as potential error for the dose of product X

Intercepted medication errors
If one of the reported MedDRA ® PTs contains the word "intercepted" and the MedDRA ® PT "Circumstance or information capable of leading to medication error" was not reported after a dose of product X, the case is categorized as intercepted medication error for the dose of product X

Medication errors without harm
If there is no potential error or intercepted medication error and all MedDRA ® PTs reported belong to the SMQ "medication error," the case is categorized as medication error without harm for the dose of product X

Medication errors with harm
If there is no potential error or intercepted medication error, at least one valid event belonging to the SMQ "medication error" and at least one event not belonging to the SMQ "medication error," the case is categorized as medication error with harm for the dose of product X Te following MedDRA ® PTs are not considered valid events to be used for the defnition of medication error with harm: the MedDRA ® PTs belonging to the primary SOC "product issues," the HLGT "of-label uses and intentional product misuses/uses issues," the HLTs "adverse efect absent," "exposure associated with pregnancy, delivery and lactation," "normal newborn status" and "normal pregnancy, labor and delivery" or the PT "breastfeeding"

Conficting category
For a dose of product X, if one of the reported MedDRA ® PTs contains the word "intercepted" and the PT "circumstance or information capable of leading to medication error," the case is categorized under conficting category and should be corrected HLGT, high-level group term; HLT, high-level term; MedDRA ® , medical dictionary for regulatory activities; PT, preferred term; SMQ, standardized MedDRA ® query; SOC, system organ class.
Advances in Pharmacological and Pharmaceutical Sciences of medication error reports shortly after the initial launch of RZV (52% of all spontaneous reports associated with RZV worldwide).Te safety signal was mainly a result of an incorrect route of administration, the wrong reconstitution of the vaccine, or the wrong storage conditions [13].Te identifcation of this signal triggered the introduction of corrective actions (e.g., implementation of educational programs and product label information clarifcation), which led to a decrease in the proportion of medication errors [6,13].To better understand these data and possible safety signals, the data can also be visualized at the country level (Figure 4).
Such visualization, as shown in Figure 3, also allowed the detection of an increase in the reporting of medication errors on two other occasions: the frst one starting in January 2019 and the second one starting more recently, in May 2020.To identify the specifc type of medication error responsible for the observed increases in reports, we looked at the visualization of the diferent types of medication errors by MedDRA ® HLTas a percentage of all medication errors over time (Figure 2, orange box, Figure 5).We saw that the most frequently reported error was mainly linked to product administration errors and issues.Further analyses allowed us to determine that while the reporting of events such as "incorrect   Advances in Pharmacological and Pharmaceutical Sciences route of administration" remained stable (Figure 6), an increase was seen for the MedDRA ® PTs "incomplete course of vaccination" or "inappropriate schedule of product administration" (Figure 7).Tis increase was assumed to result from the product supply issue GSK faced in 2019 and, more recently, from the restrictions imposed by the coronavirus disease 2019 (COVID-19) pandemic that prevented individuals from receiving their second dose.
For the category "medication error with harm," the most frequently coreported AEs can also be visualized (Table 3).A comparison can then be made between the safety profle linked to the medication error and the known safety profle of the product when administered according to the label.

Discussion
Te broadening of the defnition of a signal to include medication errors refects major eforts to reduce the burden of harm from medication errors and protect patient safety.As part of our eforts to fulfll our obligation under the legislation and good practice guide, an algorithm that can be used to categorize medication errors reported to diferent databases was developed.To our knowledge, such an algorithm to categorize medication errors was not available.
Te algorithm allows in-stream and retrospective categorization of medication errors via an automated process, minimizing the risk of mistakes.It has already proven its use as it enabled the identifcation of a safety signal related to medication errors after the launch of RZV, thereby allowing the implementation of measures to minimize the risk of medication errors [6,13].Te algorithm could be useful in the context of the current COVID-19 pandemic, as several vaccines with diferent storage requirements, preparation schedules, and administration schedules are on the market [14].Moreover, while the algorithm is currently being used in the context of vaccines, we strongly believe that its usefulness can also be extended to other medicinal products due to its straightforward implementation.All medication error data are routinely discussed in periodic safety update reports due to regulatory requirements.
Several benefts are associated with the algorithm.Te manual review of medication errors is time-consuming, requires human resources, and is prone to errors, and coding conventions may change over time.Te implementation of this algorithm may help circumvent these issues as it categorizes cases in an automated way.Additionally, it only requires some adaptations that can easily be implemented when there are changes in legislation (e.g., requiring other information on medication errors to be collected, reported, and analyzed) or updates to any MedDRA ® version.As the algorithm is based on internationally accepted terminology using MedDRA ® , the categorization is always done in a consistent way.Tis increases the quality of categorization by decreasing the subjectivity of reviewing and categorizing medication errors by diferent reviewers.Tis automated process can also minimize the risk of mistakes with manual classifcations.Te algorithm allows a continuous categorization of new and existing cases entered in the database, as new information arises.With the implementation of the new ICH-E2B (R3) reporting format in the EudraVigilance system in November 2017 [15], it is also possible to identify which of the reported suspect medications was actually involved in the error, although this type of information is not always available.Before the implementation of the European legislation, the categorization of case reports required a manual review of the cases with no re-evaluation after processing.By implementing this algorithm, the categorization of medication errors is performed in the same way for all case reports.In addition, data integrity issues can be fagged automatically.For example, when intercepted error and potential error are coded for the same case, the case will be categorized in a separate "conficting category," which  Advances in Pharmacological and Pharmaceutical Sciences     MedDRA ® , medical dictionary for regulatory activities; PT, preferred term.
Advances in Pharmacological and Pharmaceutical Sciences Terefore, when multiple medications are involved as suspects in one report of medication error with harm, it is not always possible to identify which of the reported suspect medications was involved in the error.Finally, as no other methods/algorithms are freely available, this algorithm was only compared with the manual method.

Conclusions
A new algorithm to categorize medication errors in an automated way was developed.Tis algorithm can be applied to diferent databases as it is easy to implement and is thought to facilitate the assessment of medication errors.In addition, it has already proven its use, as it enabled the identifcation of a safety signal related to medication errors after the launch of RZV.

Figure 2 :Figure 3 :
Figure 2: Example of visualizations in the SMM tool: overall view.HLT, high-level term; MedDRA ® , medical dictionary for regulatory activities; NEC, not elsewhere classifed; PT, preferred term; SMM, signal mining and management; and SMQ, standardized MedDRA ® query.

Figure 4 :
Figure 4: Evolution of the ratio of medication errors without harm over all spontaneous reports for the selected vaccine in the United States.

Figure 7 :
Figure 7: Evolution of the ratio of case reports with an incomplete schedule over all reported cases.
PTs equal to Circumstance or information capable of leading to medication erroror Circumstance or information capable of leading to device use error?

Table 2 :
Example of table with reported medication errors by MedDRA® HLT and PT for the category medication without harm.
® , medical dictionary for regulatory activities; NEC, not elsewhere classifed; PT, preferred term.aDatafor a chosen period.bDatafrom entry of the frst case in the database until date of database freezing.cTotalnumber of distinct cases; may not correspond to the sum per event reported in the table as diferent MedDRA ® PTs may be reported in the same case.Advances in Pharmacological and Pharmaceutical Sciences will trigger further follow-up and corrections in the database.Finally, the use of predefned visualizations allows for quick identifcation of signals and monitoring of changes over time.Te limitations of the algorithm are mainly linked to coding issues or maintenance of the list of terms.Indeed, the algorithm is based on coding practices, which can contain errors and inconsistencies, while the implementation of the algorithm requires correct and consistent coding.Hence, it is important to have rules in place to fag conficting coding.Standardized MedDRA ® Query.Due to the nature of spontaneous reports, the evaluation of medication errors can be used to quickly identify and gain insights into the types of errors reported and to identify potential areas where preventive measures could be benefcial, rather than to quantify the risks associated with medication errors.Another limitation with the use of this algorithm is that no causality assessment is performed.Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 2018 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 2019 Jan Feb Mar Apr May Jun Jul Aug Sep Oct 2020 Number of medication error reports/total number of reports (%) Figure5: Evolution of the type of medication errors (as a percentage of all reported medication errors) over time.MedDRA ® HLT, medical dictionary for regulatory activities high-level term; NEC, not elsewhere classifed; and PT, preferred term.

Table 3 :
Example of table with most frequent coreported adverse events for the category medication errors with harm.