Listen carefully: The risk of error in spoken medication orders

https://doi.org/10.1016/j.socscimed.2010.01.042Get rights and content

Abstract

Clinicians and patients often confuse drug names that sound alike. We conducted auditory perception experiments in the United States to assess the impact of similarity, familiarity, background noise and other factors on clinicians' (physicians, family pharmacists, nurses) and laypersons' ability to identify spoken drug names. We found that accuracy increased significantly as the signal-to-noise (S/N) ratio increased, as subjective familiarity with the name increased and as the national prescribing frequency of the name increased. For clinicians only, similarity to other drug names reduced identification accuracy, especially when the neighboring names were frequently prescribed. When one name was substituted for another, the substituted name was almost always a more frequently prescribed drug. Objectively measurable properties of drug names can be used to predict confusability. The magnitude of the noise and familiarity effects suggests that they may be important targets for intervention. We conclude that the ability of clinicians and lay people to identify spoken drug names is influenced by signal-to-noise ratio, subjective familiarity, prescribing frequency, and the similarity neighborhoods of drug names.

Introduction

In clinical medicine, the risks of misinterpretation of telephone orders are widely recognized (Koczmara et al., 2006, Pennsylvania Patient Safety Authority, 2006, The Joint Commission, 2008). The use of the telephone to communicate medication orders leads to error because of both ambient noise and the limited bandwidth of most telephones (Aronson, 2004, Hoffman and Proulx, 2003, Lambert, 2008, Rodman, 2003, Wiener et al., 2004). Telephones typically carry signals between 300 Hz and 3 kHz, a much narrower bandwidth than that of FM radio (30 Hz–15 kHz) or CD audio (20 Hz–20 kHz), whereas much of the important acoustic information that allows people to distinguish between similar consonant sounds lies above 3 kHz and is missing entirely from the telephone signal (Rodman, 2003). There are 3.8 billion prescriptions dispensed in outpatient pharmacies annually in the United States (IMS Health, 2008). Telephone orders account for 3–4% of retail prescription volume. This translates to 114 million telephone prescriptions annually, or 312,000 per day. One study of 813 telephone orders to two chain pharmacies found that the wrong medication name was transcribed in 1.4% of the orders (Camp, Hailemeskel, & Rogers, 2003). The 1.4% rate may not be a generalizable estimate, but given the number of opportunities, even a very low error rate would translate into a large number of errors.

Spoken orders were once common in inpatient settings also, although less so after accrediting agencies pressed for their elimination. One 346-bed hospital counted 4197 medication-related verbal orders in a seven day period (Wakefield et al., 2008). Hospital pharmacists reported 35 min of every 8 h shift were spent resolving problems with spoken orders (Allinson, Szeinbach, & Schneider, 2005). Respondents identified “people talking in the background” and “background noise” as the greatest barriers to the correct processing of spoken orders. Other factors included lack of familiarity with the patient's clinical condition or the medication, bad connections and excessively rapid speech (Allinson et al., 2005).

The use of cell phones and voicemail and the noisy environments in which orders are sent and received increase the risk of spoken prescription orders being misperceived. There are many examples of auditory perception errors, some with fatal consequences (e.g., Liquibid vs. Lithobid, Cardene vs. codeine, Lopid vs. Slobid, erythromycin vs. azithromycin, Klonopin vs. clonidine, Viscerol vs. vistaril, Orgaran vs. argatroban) (Allinson et al., 2005, Dr, 2004, Koczmara et al., 2006, Pennsylvania Patient Safety Authority, 2006, Vivian, 2004).

Identifying the factors that influence accuracy in the perception of spoken drug names may facilitate interventions designed to make telephone orders safer. The U.S. Food and Drug Administration and the pharmaceutical industry have struggled to develop methods for evaluating the confusability of new drug names (U. S. Food and Drug Administration, 2008). Objective measures of similarity and prescribing frequency can reliably predict the probability that two names will be confused in visual perception and short term memory (Lambert, 1997, Lambert et al., 2003, Lambert et al., 2001b, Lambert et al., 2002, Lambert et al., 1999, Lambert et al., 2004), and processes have been described for designing safer drug names (Lambert, Lin, & Tan, 2005). One important part of that process is to use established experimental paradigms from psycholinguistics to evaluate the confusability of proposed drug names in relevant tasks (e.g., auditory perception, visual perception, and short term memory). An earlier study of noise and pharmacy dispensing errors found, counter-intuitively, that noise improved performance but recommended that more controlled experiments be done to clarify the relationship between noise and error rates (Flynn et al., 1996). In this study, we sought to demonstrate how this type of experimentation could shed light on the factors that influence auditory perception of drug names. Thus, one of the key challenges we addressed was to translate the basic science of auditory word perception into the applied domain of drug name confusion. The work was designed to determine how and to what extent characteristics of drug names, order takers, and practice environments affect a listener's ability to identify spoken drug names.

To explain auditory perceptual confusions, we used Luce's Neighborhood Activation Model (NAM) (Grossberg, 1986, Luce et al., 2000, Luce and Pisoni, 1998, Vitevitch and Luce, 1999). According to this model, stimulus input activates a set of similar sounding acoustic-phonetic patterns in memory. The activation levels of the acoustic-phonetic patterns are a function of their degree of match with the input. In turn, these patterns activate a set of word decision units tuned to the acoustic-phonetic patterns. The word decision units compute probabilities for each pattern based on the intelligibility and frequency of occurrence of the word to which the pattern corresponds and the activation levels and frequencies of occurrence of all other similar sounding words in the system. The word decision unit that computes the highest probability wins, and its word is what is heard. In short, word decision units compute probability values based on the acoustic-phonetic similarity of the word to the input, the frequency of the word, and the activation levels and frequencies of all other similar words activated in memory.

The NAM predicts that multiple activation has consequences: Spoken words with many similar-sounding, higher frequency (or more commonly occurring) neighbors will be processed more slowly and less accurately than words with few neighbors. These predictions have been confirmed in many studies: Words in densely populated, high frequency similarity neighborhoods are indeed processed less quickly and less accurately than words in low density, lower frequency neighborhoods, and words with higher frequency of occurrence are processed more rapidly and accurately than lower frequency words (Jusczyk and Luce, 2002, Lambert et al., 2003).

The NAM employs an explicit mathematical function that attempts to predict auditory perceptual errors based on the intelligibility of stimulus word, the frequency of occurrence of the stimulus word, and the similarity and frequency of neighboring words. This function is known as frequency-weighted neighborhood probability (FWNP, or neighborhood probability). Detailed mathematical descriptions of the function used to compute neighborhood probabilities for each name are given elsewhere (Jusczyk and Luce, 2002, Lambert, Lin, Toh et al., 2005). Other things being equal, neighborhood probabilities will increase as the number, similarity, and prescribing frequency of neighbors decrease. The Neighborhood Activation Model provided the framework for the development of several hypotheses about auditory perception: (1) Accuracy will increase as neighborhood probability increases; (2) Accuracy will increase as the signal-to-noise (S/N) ratio increases; (3) Accuracy will increase as objective prescribing frequency of the target name increases; and (4) Accuracy will increase as subjective familiarity with the target name increases.

Although frequency and neighborhood effects are well established in the study of ordinary words, we sought to extend this understanding in three ways. First, we planned to study proper names from a large, finite set of words, namely the closed-set lexicon of drug names, given that most previous work has been done with open-set word lists (Clopper et al., 2006, Sommers et al., 1997). Second, we wished to study multisyllabic words rather than the monosyllabic (often consonant-vowel-consonant) words that have typically been used in studies of neighborhood effects. That required us to develop measures of similarity and new measures of neighborhood probabilities for multisyllabic words (Lambert, Lin, Toh et al., 2005). Third, we wished to see whether the neighborhood effects would be present in both experts (clinicians) and novices (laypeople). We expected neighborhood effects in experts because, presumably, they would possess lexical representations (i.e., mental word lists) of large numbers of drug names, and these representations would compete in the manner described above. We thought neighborhood effects might be attenuated or absent in laypeople, who may lack representations for most drug names and hence would not experience the competition that causes neighborhood effects.

Even if we were not making any original contribution to the basic science of auditory perception, we believe that the model provides a powerful conceptual and experimental framework for understanding drug name confusion, one that could advance worldwide efforts to predict and prevent such errors. Thus, the present paper is offered as an example of translational research, where concepts well-known to one community are applied to problems in a different domain (Woolf, 2008).

We attempted to control for a large set of factors that might be associated with name confusion error rates. Among these was the type of name (brand or generic). One might expect brand names to be more confusing since they are typically shorter (Lambert, Chang, & Lin, 2001a), and shorter names tend to have more neighbors (Andrews, 1997, Luce and Pisoni, 1998, Storker, 2004). Conversely, generic names use a common system of stems (i.e., suffixes) which tends to increase their average similarity to one another, thereby increasing their confusability (Lambert et al., 2001a). Either way, the distinction between brand and generic names is an important one in practice, so we designed our experiments to take it into account.

Section snippets

Design

We used a within-participants design to examine the effect of signal-to-noise (S/N) ratio, prescribing frequency, subjective familiarity, and similarity neighborhoods on the ability to correctly identify spoken drug names. All participants heard an equal number of drug names at all three S/N ratios and all of the drug names appeared at all three S/N ratios. Three versions of the experiment ensured that drug names and S/N ratios were counterbalanced across participants and that participants only

Results

Each of the 239 participants responded to 197 stimuli, producing 47,083 total responses. Mean accuracy on the perception task was 32.27% (s.d. = 7.94, range = 12.18%–49.75%, also see Table 5). Roughly 77% of correct responses were identified by verbatim matching, 9.5% by computer scoring of alternative pronunciations and 13.5% by expert scoring of alternatives. This error rate is roughly two orders of magnitude greater than what one would expect in real-world practice (Flynn, Barker, &

Discussion

Our goal was to identify factors that affect accuracy in auditory perception of drug names. Results of our experiments supported our three specific hypotheses and our general model of auditory perception: (1) Accuracy (for clinicians) was influenced by the similarity neighborhood of each drug name, i.e., by the similarity and prescribing frequencies of neighboring names. Names with less similar and less commonly prescribed neighbors were more accurately perceived than names with more similar

Conclusion

Objectively measurable properties of drug names can be used to predict their confusability. The ability to accurately identify spoken drug names is influenced by signal-to-noise ratio, subjective familiarity, prescribing frequency, and the similarity neighborhoods of drug names. To minimize errors, order takers should be able to increase the source volume, and should have noise-cancelling headphones or quiet areas where they can take orders. Although not directly supported by these experiments,

Acknowledgements

This research was supported in part by grant #1 R01 HS011609 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. The authors acknowledge assistance of David Lewis, David Tcheng, Robert LaVigne, Darren Toh, Teresa Markis, and Meghana Desai. We are also indebted to the study participants from the American Association of Family

References (52)

  • S.C. Camp et al.

    Telephone prescription errors in two community pharmacies

    American Journal of Health-System Pharmacy

    (2003)
  • C. Clopper et al.

    Effects of open-set and closed-set task demands on spoken word recognition

    Journal of the American Academy of Audiology

    (2006)
  • J.D. Cohen et al.

    PsyScope: a new graphic interactive environment for designing psychology experiments

    Behavioral Research Methods, Instruments, and Computers

    (1993)
  • Dr. orders liquibid–lithobid dispensed – death results. Case on point: Clifford v. Geritom Med., Inc., 681 N.W.2d 680-MN (2004)

    Nursing Law's Regan Report

    (2004)
  • E.A. Flynn et al.

    Relationships between ambient sounds and accuracy of pharmacists' prescription-filling performance

    Human Factors

    (1996)
  • M.A. Gernsbacher

    Resolving 20 years of inconsistent interactions between lexical familiarity and orthography, concreteness, and polysemy

    Journal of Experimental Psychology: General

    (1984)
  • J. Gonzalez et al.

    Hemispheric differences in indexical specificity effects in spoken word recognition

    Journal of Experimental Psychology: Human Perception and Performance

    (2007)
  • P. Hagmann et al.

    Hand preference and sex shape the architecture of language networks

    Human Brain Mapping

    (2006)
  • D. Hedeker et al.

    SuperMix mixed effects models

  • J.M. Hoffman et al.

    Medication errors caused by confusion of drug names

    Drug Safety

    (2003)
  • D.W. Hosmer et al.

    Applied logistic regression

    (1989)
  • D.H. Howes

    On the relation between the intelligibility and frequency of occurrence of English words

    Journal of the Acoustical Society of America

    (1957)
  • IMS Health

    2007 channel distribution by U.S. dispensed prescriptions

  • I. IMS

    National prescription audit plus

    (2001)
  • D. Jurafsky et al.

    Speech and language processing: An introduction to natural language processing, computational linguistics, and speech recognition

    (2000)
  • P.W. Jusczyk et al.

    Speech perception

  • Cited by (33)

    • Categorizing and understanding medication errors in hospital pharmacy in relation to human factors

      2020, Saudi Pharmaceutical Journal
      Citation Excerpt :

      Guérin et al. study revealed that most machines in the pharmacy produce a high level of noise such as ventilation systems, pneumatic tube systems, and refrigerators (Guérin et al., 2013). High noise levels are linked to affect verbal recognition of drug names (Lambert et al., 2010) and increase medication errors (Beso et al., 2005; Westbrook et al., 2010). Also, lacking sufficient facilities and narrow places were discussed in the study, a participant revealed how the IV room was designed inappropriately and how he and his colleagues were facing problems working there.

    • The similarity of drug names as a possible cause of confusion: Analysis of data from outpatient care

      2020, Zeitschrift fur Evidenz, Fortbildung und Qualitat im Gesundheitswesen
    • Reducing avoidable medication-related harm: What will it take?

      2019, Research in Social and Administrative Pharmacy
    • Making long-distance relationships work: Quantifying lexical competition with Hidden Markov Models

      2016, Journal of Memory and Language
      Citation Excerpt :

      Indeed, priming a target word (e.g., “bull”) with the competitor with the highest NWP that does not share phonemes (e.g., “veer”) reduced identification accuracy (Goldinger, Luce, & Pisoni, 1989) and increased shadowing latencies (Luce et al., 2000) for the target. In addition, FWNPs can be predictive in situations that cause listeners to activate a particular set of words; FWNPs calculated for the names of drugs, using a lexicon of drug names rather than all English words, significantly predict clinicians’ identification accuracy in noise (Lambert et al., 2005, 2010). The mechanisms of activation and competition described by FWNPs are not limited to auditory speech; when the perceptual input is visual (lipread), FWNPs derived from visual confusion matrices predict lipread word-identification accuracy (Auer, 2002; Feld & Sommers, 2011).

    • The therapeutic impacts of environmental design interventions on wellness in clinical settings: A narrative review

      2016, Complementary Therapies in Clinical Practice
      Citation Excerpt :

      There have been claims in animal experimentations that loud noises and other associated stressors such as cold and heat can delay the rate of wounds healing in rats [28,132,191], although this has not been attested by hospitalised patients. Researchers have reported that noise can affect pharmaceutical name recognition, and suggested that areas where drugs are prescribed and administered should be as quiet as possible, well-illuminated, and free from distractions that include background music and ringing phones [91–94,136]. According to Frumkin and Louv [57] it could be argued that people are closely attached to the natural world, which suggests that contact with nature is beneficial to health and wellbeing.

    View all citing articles on Scopus
    View full text