2024 Volume 47 Issue 8 Pages 1460-1466
Dispensing errors pose a significant health risk, with drug name similarity being a potential contributory factor. To determine the impact of drug name similarity on dispensing errors within clinical settings, we analyzed 563 dispensing errors at an acute hospital in Japan from April 2015 to June 2018. Drug name similarity between two drugs was classified into Name-Similar and Name-Dissimilar groups using the m2-vwhtfrag index, the value of the drug name similarity. Drug efficacy similarity was categorized into Efficacy-Same, Efficacy-Close, and Efficacy-Far. The drug name similarity and drug efficacy similarity of all possible pair combinations were obtained and similarly classified. The proportion of the number of pairs with dispensing errors per the total number of drug pairs in the hospital’s drug formulary in each category was calculated. The highest proportion of the number of pairs with dispensing errors was 36% for the Efficacy-Same and Name-Similar group, and the lowest proportion was 0.022% for the Efficacy-Far and Name-Dissimilar group. The proportion of the number of pairs with dispensing errors was significantly higher in the Name-Similar category than in the Name-Dissimilar category for all drug efficacy categories. Our results indicate that drug name similarity increases the risk of dispensing errors, and that m2-vwhtfrag is a useful indicator to assess dispensing errors in clinical practice. Such drug name and efficacy similarity evaluations can help identify factors causing dispensing errors, and predict the risk of dispensing errors for newly adopted drugs, considering the relationship with the whole drug formulary in the hospital dispensary.
Medication errors are a very serious safety issue. For example, in Japan, the Pharmaceuticals and Medical Devices Agency (PMDA) and pharmaceutical companies issued an alert after a fatal accident involving the mistaken prescription and administration of the muscle relaxant Saxin (Hepburn romanization: sakushin) instead of the hydrocortisone drug Saxizone® (Hepburn romanization: sakushizon) Injection Solution.1) In the U.S., the name of the antidepressant Brintellix was changed to Trintellix in 2016, following numerous cases of medication errors involving confusion with the anticoagulant Brilinta.2) Drug name similarity is considered one of the main causes of such medication errors. Therefore, it is important to predict the effect of drug name similarity on medication errors and to take measures to prevent such errors before they occur.
Various indices have been developed to prevent drug misidentification due to drug name similarity. In the United States, the phonetic and orthographic computer analysis (POCA) score has been developed as an index that reflects sound and letter similarities between drug names to prevent the creation of similar drug names.3) However, it would be difficult to use such indices directly for drug name similarity in Japanese, where letters and pronunciation are different.
In Japan, various indices have been developed to evaluate the similarity between two drug names: 1) cos1 evaluates the degree of agreement of the letters in 2 names, 2) head-tail-cosine (htco) evaluates the degree of agreement between the letters at the beginning and the end of the name, 3) dlen evaluates the length difference of the two drug names, and 4) edit evaluates the number of steps required to edit one drug name to match the other drug name. All of these indicators were developed with a focus on specific similarity factors. On the other hand, Ohtani et al. introduced a fragment pattern indicator, vwhtfrag (Visually Weighted Head and Tail-weighted Fragmentary Pattern Based Measure), which is a drug name similarity indicator that can evaluate these drug name similarity factors with a single indicator,4) providing greater accuracy. In addition, Tamaki et al. introduced an enhanced version of vwhtfrag, termed vwhtfrag+, which assigns greater weight to the initial characters of the drug names.5) We also developed m2-vwhtfrag, which takes into account the difference in length between the two drug names. This index, m2-vwhtfrag, has been shown to correlate well with subjective similarity.6)
Thus, a variety of indices have been developed in Japan to evaluate the degree of drug name similarity between drug names based on Japanese “katakana” characters. However, the impact of drug name similarity on actual dispensing errors has not been evaluated using these indices. It is also unclear how many similar drug names there are, or to what extent there is drug name similarity among all drug pairs in a hospital dispensary. We considered that by evaluating the impact of drug name similarity on dispensing errors using the indices developed in Japan, it should be possible to implement countermeasures that would reduce dispensing errors in clinical settings in Japan.
Therefore, the purpose of this study was to evaluate the impact of drug name similarity on actual dispensing errors using the m2-vwhtfrag index, which can take account of multiple aspects of drug name similarity.
A retrospective study was conducted on dispensing errors recorded in the hospital dispensary of an acute hospital in Japan from April 2015 to June 2018. In the hospital, there was a rule to record all dispensing errors uncovered during the dispensing audit and subsequent stages, and all errors are considered to be accumulated. Only errors involving drugs with different product names, including both the brand and generic names, were included. Errors in dosage strength, dosage forms, or quantities were excluded as long as the product name was the same. Furthermore, Chinese herbal medicines, medicines containing both Kanji and alphabetic characters, medical devices (e.g., injection needles), mixtures of three or more drugs, self-injection medicines, and medicines with Kanji characters between Katakana characters were excluded from the analysis. In addition, a list of drugs (oral and topical) stored in the hospital dispensary during a given period was used. In the dispensary, there was no fixed standard for the arrangement of drugs on the shelves, and drugs were randomly arranged except for areas divided by the general classification of dosage forms, such as topical and oral use, liquids, and sprays. Shelf numbers were listed next to the drug name on the prescription, and the dispensing pharmacist can use the shelf number as an additional reference during dispensing.
Calculation of Drug Efficacy Similarity and Drug Name SimilarityIn actual dispensing situations, drug efficacy similarity can contribute to drug name confusion,7) and so it is necessary to evaluate the influence of drug efficacy similarity when assessing drug name similarity. The degree of drug efficacy similarity was calculated based on the index used in a study7) that investigated the influence of drug efficacy similarity on dispensing errors. To evaluate the drug efficacy similarity, we used a four-digit number, called the “drug efficacy classification number,” which excludes the first two digits of the Japanese Standard Commodity Classification number. Drug efficacy similarity was defined as ‘Efficacy-Far’ when all or the last three digits of the drug efficacy classification number were different, ‘Efficacy-Close’ when the last one or two digits were different, and ‘Efficacy-Same’ when all digits were the same (Fig. 1).
The drug efficacy similarity was established based on the drug efficacy classification of the Japanese National Health Insurance Drug Price Standard. It categorizes similarity into three levels: ‘Efficacy-Far,’ defined when either all digits or just the last three digits of the therapeutic category number differ; ‘Efficacy-Close,’ when only the final one or two digits vary; and ‘Efficacy-Same,’ when all digits are identical.
The evaluation of drug name similarity is based on the product name under which the dispensing errors occurred (the product name refers to the katakana part, excluding modifiers indicating dosage form such as “spray,” “dry powder,” “cream,” etc.). M2-vwhtfrag is a similarity index that evaluates the similarity of katakana characters and is calculated on the basis of head matches, common strings, look-alike katakana characters, and differences in string length (Fig. 2). Examples of calculation of m2-vwhtfrag are shown in Supplementary Fig. 1. It has been shown that similar and dissimilar pairs can be best classified when the threshold value of this index is set to 0.456 by Tamaki et al.6) In our analysis, the pairs were classified into “Name-Similar,” when the index value exceeded 0.456, and “Name-Dissimilar,” when the index value was below 0.456. Correspondence between Japanese katakana and the Roman alphabet is shown in Supplementary Table 1.
An example of the calculation of m2-vwhtfrag between “Hepburn romanization: a-i-u-e-ka-ku-ki” and “Hepburn romanization: a-i-e-o-sa-ka-ke-ki” written in Japanese katakana is shown in the figure. Katakana characters for voiced and semi-voiced sounds are calculated as unvoiced sounds. Five factors, denoted as A through E, have been established to optimize the quantification of corresponding characters for each fragment sequence. Parameter A is a factor that reduces the prominence of matches outside the initial and terminal positions. This approach diminishes the weight of non-extreme matches. Parameter B is deducted from relevant equations to elevate the score of consecutive matches relative to non-continuous string matches. Parameter C represents the character match between visually similar katakana. Parameter D is a factor that increases the contribution of the initial position of the match. Parameter E decreases the score when a discrepancy in string length is observed. The m2-vwhtfrag is calculated using these parameters by summing the calculated values for: a) the match of the head, b) the match of neither head nor tail, c) the match of the tail, and d) the difference in string length, then dividing by the average string length. These parameters were optimized by Tamaki et al.6) to classify positive and negative cases with the highest accuracy. The positive cases were actual pairs with drug dispensing errors, most likely caused by drug name similarities, and the negative cases were randomly generated drug pairs.
Programs created in Python 3.8 were used to classify drug efficacy similarity, to calculate drug name similarity, and to process drug names and create drug name pairs.
AnalysisThe m2-vwhtfrag index values were calculated for all drug pairs involved in dispensing errors during the period of interest. Drug pairs in which dispensing errors occurred based on drug efficacy similarity and drug name similarity were grouped into 6 groups: Name-Similar/Name-Dissimilar × Efficacy-Same/Efficacy-Close/Efficacy-Far, and the number of error pairs in each group was calculated.
The m2-vwhtfrag index values were also calculated for all possible pairs of all drugs stocked in the hospital dispensary. For oral and topical drugs stored in the hospital dispensary, only product name was taken to create all possible drug pairs, and the drug efficacy similarity and drug name similarity of these pairs were calculated and classified into 6 groups described above.
In each group, the number of dispensing error pairs per the number of drug name pairs generated from the drug formulary was calculated. Within the same drug similarity group, the proportion of pairs with dispensing errors was compared between the dissimilar and similar groups using the χ-square test. The significance level was set at 5%, and adjusted p-values were calculated using the Bonferroni correction. The analysis was performed using Python 3.8.
Ethical ConsiderationThis study is an analytical study using existing information and does not contain personally identifiable information. This study was approved by the Ethics Committee for Research Involving Human Subjects, Keio University Faculty of Pharmacy (No. 190422-1), and was conducted in accordance with the ethical guidelines for medical research involving human subjects.
Six hundred and forty-one dispensing errors occurred during the relevant period. After excluding drug name pairs that fell under the exclusion criterion, 563 pairs of dispensing errors were used for analysis. For all drug combinations in which dispensing errors occurred, the distribution of drug name similarity for all dispensing error drug pairs and the distribution of drug name similarity by drug efficacy similarity are shown in Figs. 3A–D. The combinations with m2-vwhtfrag = 0 had the highest number of dispensing errors, with another peak between 0.95 and 1.0. In the “Efficacy-Same” group, there was also a peak at m2-vwhtfrag = 0.95–1.0.
Relative frequencies were calculated by dividing the number of dispensing errors within each m2-vwhtfrag range, with increments of 0.05, by the overall count of such errors.
The number of pairs created from the drug formulary in the dispensary was compared by classifying them into six groups: Name-Similar/Name-Dissimilar × Efficacy-Same/Efficacy-Close/Efficacy-Far (Table 1). In terms of drug name similarity, the number of drug pairs was higher in “Name-Dissimilar” than in “Name-Similar.” In terms of drug efficacy similarity, the number of drug pairs was the highest for drug pairs with “Efficacy-Far” and lowest for drug pairs with “Efficacy-Same.”
| Drug name | |||
|---|---|---|---|
| Dissimilar | Similar | ||
| Drug efficacy | Same | 7289 | 349 |
| Close | 39500 | 1723 | |
| Far | 533523 | 20961 | |
For each drug pair involved in the dispensing errors, the drug efficacy similarity and drug name similarity were calculated and classified into Name-Similar/Name-Dissimilar × Efficacy-Same/Efficacy-Close/Efficacy-Far, and the proportion of the number of pairs with dispensing errors for each group was calculated (Table 2). In the same drug efficacy similarity group, the proportion of the number of pairs with dispensing errors was significantly higher in the “Name-Similar” group than in the “Name-Dissimilar” group. In addition, in the same drug name similarity group, the higher the drug efficacy similarity, the higher the proportion of the number of pairs with dispensing errors.
| Drug name | p-Value | |||||
|---|---|---|---|---|---|---|
| Dissimilar | Similar | |||||
| Number of pairs with dispensing errors | % | Number of pairs with dispensing errors | % | |||
| Drug efficacy | Same | 139 | 1.9 | 126 | 36 | <.0001 |
| Close | 95 | 0.24 | 34 | 2.0 | <.0001 | |
| Far | 119 | 0.022 | 50 | 0.24 | <.0001 | |
The number of all drug pairs for all drugs in the dispensary is taken as 100%.
Table 3 shows the drug name pairs that had four or more dispensing errors for each group. There were five pairs of drug names that were mistaken four or more times in Name-Dissimilar × Efficacy-Far.
| A. Name-Dissimilar Group | ||||
|---|---|---|---|---|
| Group | Name 1 | Name 2 | m2-vwhtfrag | Number of incidents |
| Efficacy-Far × Name-Dissimilar | カロナール karonaru (Acetaminophen) | クエン kuen (Ferrous Citrate) | 0 | 4 |
| ジャヌビア janubia (Sitagliptin) | フェキソフェナジン fuekisofuenajin (Fexofenadine) | 0.009 | 4 | |
| エリキュース erikyusu (Apixaban) | ビソプロロール bisopurororu (Bisoprolol) | 0.079 | 4 | |
| ワーファリン wafuarin (Warfarin) | フロセミド furosemido (Furosemide) | 0.094 | 5 | |
| ノイロトロピン noirotoropin (Neurotropin) | ピドキサール pidokisaru (Pyridoxal) | 0.182 | 4 | |
| Efficacy-Close × Name-Dissimilar | ブロムフェナク buromufuenaku (Bromfenac) | サンベタゾン sambetazon (Sanbetason) | 0 | 6 |
| ゾルピデム zorupidemu (Zolpidem) | ブロチゾラム burochizoramu (Brotizolam) | 0.266 | 18 | |
| アスピリン asupirin (Aspirin) | アセトアミノフェン asetoaminofuen (Acetaminophen) | 0.389 | 4 | |
| リドメックス ridomekkusu (Prednisolone) | リンデロン rinderon (Betamethasone) | 0.404 | 8 | |
| Efficacy-Same × Name-Dissimilar | セルタッチ serutatchi (Felbinac) | アドフィード adofuido (Flurbiprofen) | 0 | 6 |
| トラゼンタ torazenta (Linagliptin) | ジャヌビア janubia (Sitagliptin) | 0 | 6 | |
| フェルム fuerumu (Ferrous Fumarate) | クエン kuen (Ferrous Citrate) | 0 | 4 | |
| モーラス morasu (Ketoprofen) | ロキソプロフェン rokisopurofuen (Loxoprofen) | 0 | 13 | |
| レバミピド rebamipido (Rebamipide) | テプレノン tepurenon (Teprenone) | 0.133 | 18 | |
| ニフレック nifurekku (Sodium Chloride, Potassium Chloride, Sodium Bicarbonate, Anhydrous Sodium Sulfate combination powder) | モビプレップ mobipureppu (Sodium Chloride, Potassium Chloride, Sodium Sulfate Anhydrous combination powder) | 0.348 | 7 | |
| B. Name-Similar Group | ||||
| Group | Name 1 | Name 2 | m2-vwhtfrag | Number of incidents |
| Efficacy-Far × Name-Similar | ユリノーム yurinomu (Benzbromarone) | ユリーフ yurifu (Silodosin) | 1.197 | 6 |
| Efficacy-Close × Name-Similar | パンデル panderu (Hydrocortisone) | パスタロン pasutaron (Urea) | 0.642 | 4 |
| テルペラン teruperan (Metoclopramide) | テプレノン tepurenon (Teprenone) | 0.665 | 6 | |
| Efficacy-Same × Name-Similar | エイゾプト eizoputo (Brinzolamide) | コソプト kosoputo (Dorzolamide) | 0.622 | 4 |
| プラバスタチン purabasutachin (Pravastatin) | ピタバスタチン pitabasutachin (Pitavastatin) | 0.707 | 25 | |
| ジャディアンス jadeiansu (Empagliflozin) | ジャヌビア janubia (Sitagliptin) | 0.873 | 4 | |
| ラベプラゾール rabepurazoru (Rabeprazole) | ランソプラゾール ransopurazoru (Lansoprazole) | 0.957 | 18 | |
| セフジニル sefujiniru (Cefdinir) | セフカペン sefukapen (Cefcapene) | 0.975 | 32 | |
Name 1 and Name 2 are structured in two layers. The top layer displays the drug’s stem in katakana. The bottom layer displays the stem using Hepburn Romanization, with the drug’s active ingredients shown in parentheses.
This is the first study to show that drug name similarity increases the risk of dispensing errors, as determined by classifying all the drugs stored in the dispensary into six groups in terms of drug name similarity and drug efficacy similarity, and calculating the proportion of the number of pairs with dispensing errors in each group.
In this study, we calculated the proportion of the number of drug name pairs in which dispensing errors occurred, taking into account all the drug name pairs that could be generated from all the drugs stored in the dispensary. Pairs in which dispensing errors occurred included many with small m2-vwhtfrag values. The reason for this could be that the number of drugs with similar names in the hospital’s drug formulary was small, whereas the number of drugs with dissimilar names was overwhelmingly large. By considering the distribution of drug name similarity among drugs stored in the hospital, the impact of drug name similarity on drug name confusion could be properly evaluated. It is also considered that many drugs with high efficacy similarity have matching stems, and as a result, the number of matching letters tends to increase, and so the drug name similarity also increases. In this study, the analysis stratified by drug efficacy similarity suggested that the more similar the drug names are, the more often dispensing errors occur, regardless of the drug efficacy similarity.
Many of the drug name pairs that were misidentified in clinical practice had small m2-vwhtfrag values, which means that they were dissimilar, suggesting that factors other than drug name similarity had a significant impact on actual dispensing errors. In the group with the drug efficacy similarity of “Efficacy-Same”/“Efficacy-Close” and m2-vwhtfrag less than 0.456 (i.e., many drug names were not considered similar), these errors were considered to have been strongly influenced by drug efficacy similarity. On the other hand, dispensing errors occurred even in the drug name pair group where the drug efficacy similarity was “Efficacy-Far” and m2-vwhtfrag was less than 0.456, i.e., neither the drug name nor the drug efficacy seemed to be similar. In addition to drug efficacy and drug name, it has been reported that appearance similarity, shelf placement, and psychological status of the worker affect drug misidentification.8–10) It may be important to consider factors other than drug efficacy and drug name in analyzing incidents of dispensing errors among drugs in these categories. For example, although warfarin (Hepburn romanization: warufarin) and furosemide (Hepburn romanization: furosemido) are not similar in either efficacy or drug name, five incidents of dispensing errors involving this pair were included in this study. These drugs are very similar in the appearance of their press-through pack (PTP) sheets sold in Japan, and it is thought that the similarity in appearance may have led to dispensing errors. As described above, the analytical method used in this study, which classifies drug name pairs involved in incidents into six groups based on drug name similarity and drug efficacy similarity, can be applied to assist risk management of dispensing errors due to the influence of drug name and drug efficacy similarity.
Furthermore, the analytical method in this study can also be applied to predict the risk of dispensing errors in advance in clinical practice. By calculating the similarity of drug name and drug efficacy for all pairs between a drug and all drugs used in clinical practice, it is possible to predict which drug name pairs are likely to be mistaken for each other. Based on these predictions, measures can be taken to prevent the occurrence of dispensing errors. Moreover, the analytical method used in this study may also be useful because it provides basic information on drug pairs that should be taken into consideration in deciding whether or not to adopt a new drug in a hospital.
There are some limitations in our study. First, it was conducted in only one acute care hospital in Japan, which may limit the range of results. The generalizability of the findings needs to be examined in other settings. Second, a limitation concerns the validation of the drug efficacy similarity assessment. The efficacy similarity index, based on the drug efficacy classification number used by Suzuki et al. in the previous study,7) has not undergone thorough validation to confirm its effectiveness as a similarity index. Nevertheless, the observed correlation between high drug efficacy similarity and increased dispensing errors suggests that the index is reasonable and implies that drugs with greater efficacy similarity are more prone to confusion. However, further research should be conducted to validate the drug efficacy similarity index. Finally, the dispensing errors covered in this study were those discovered at or after the dispensing audit stage; dispensing errors noticed before the dispensing audit were not included.
This is the first study to demonstrate that drug name similarity increases the risk of dispensing errors in clinical practice. Our results suggest that m2-vwhtfrag can be used as a similarity index to assess drug name similarity in clinical practice. Such an evaluation of drug pairs by name and efficacy similarity could be useful not only for a detailed study of actual incidents, but also for predicting the risk of drug name confusion involving newly adopted drugs, taking into account all the drugs contained in a dispensary.
This study was supported by JSPS KAKENHI Grant Number: JP22K19657.
H Sagawa, H Kizaki, and S Hori carried out the study design planning. T Kamimura, S Suzuki, and S Hayashi collected the dispensing errors and processed the data for the analysis. H Sagawa and K Yoshikawa conducted the whole analyses of drug name similarity. H Tamaki, H Satoh, and Y Sawada advised on the study’s concepts and processes. S Hori supervised this research. All authors read and approved the final manuscript.
The authors declare no conflict of interest.
The data that support the findings of this study are available from the corresponding author upon reasonable request, due to privacy or ethical restrictions.
This article contains supplementary materials.
