PaperlayerView exact revision on arXiv

Source-linked AI summary

"I just assumed that it would translate": examining MT risk awareness among healthcare staff with abbreviations as a use case

Eleanor Taylor-Stilgoe, Félix do Carmo, Constantin Orăsan

arXiv:2610.02496v1cs.AI

TL;DR

This study addresses limited evidence on healthcare staff awareness of MT risks in patient medical records, particularly when abbreviations are involved. It uses contextualised French and Spanish clinical examples translated by GT to examine staff responses, finding that participants sometimes overlooked or misinterpreted abbreviations and identified possible harms extending to patient death.

  • Problem

    Research on MT risks in medical-record translation and staff awareness remains limited, while abbreviation misuse or misinterpretation can cause patient harm up to death.

    Source: §1 · pp. 1–2

  • Method

    The study presented contextualised French and Spanish clinical examples translated with GT to healthcare staff to examine abbreviation-related risks and risk awareness.

    Source: §1 · p. 2, §3.3 · p. 3

  • Results

    Participants across roles and backgrounds sometimes overlooked or misinterpreted abbreviations, and identified potential consequences ranging from inappropriate medication or treatment delays to patient death.

    Source: §5.5 · p. 7, §5.6 · p. 7

  • Takeaways & Limitations

    Healthcare professionals are advised to use caution when translating medical records with MT and include source texts during checks for abbreviation-related or other translation issues.

    Source: §6 · p. 8

    The small, snowball-sampled participant group restricts generalisability and risks sample homogeneity.

    Source: §8 · p. 8

Abstract

from arXiv · show

In the UK, public healthcare staff report turning to machine translation (MT) - predominantly Google Translate (GT) - to communicate with patients across language barriers. Though intended to support their duty of care, potentially uninformed reliance on MT in such contexts could have serious consequences for patient safety. Research nonetheless remains limited on staff awareness of the possible risks posed by higher-stakes MT use in general and with patient medical records in particular, most existing literature instead examining its use in interpersonal situations or with patient-oriented documentation. Moreover, medical abbreviations are well-documented as increasing patient risk even monolingually, with outcomes from their misuse and/or misinterpretation ranging from temporary harm to the death of the patient. Abbreviations were therefore selected as a use case for identifying the potential risks posed by their translation with MT. Contextualised French and Spanish data examples drawn from authoritative clinical corpora and translated via GT were presented during semi-structured interviews to 21 healthcare staff participants in diverse roles and specialties. The results were then subject to qualitative analysis and cross-analysis.

1 Introduction

UK healthcare staff use MT, predominantly Google Translate, amid language barriers and a duty of care, while research on risks in medical-record translation remains limited. This study examines staff awareness and abbreviation-related risks using French and Spanish clinical examples translated with GT.

1 Introduction

  • Staff use MT, predominantly Google Translate, to bridge language barriers despite NHS guidance cautioning that translation quality is not assured.Time, cost, and resourcing constraints contribute to this practice.Source: §1 · p. 1
  • Research on generic MT risks in medical records remains limited; most prior work examines interpersonal interpreting or patient-oriented documents, including discharge instructions.Recent evaluations of discharge-instruction MT have urged caution over abbreviations, jargon, and readability; other cited work largely focuses on interpersonal encounters or patient-facing documents.Source: §1 · p. 1
  • Medical abbreviation misuse or misinterpretation can harm patients even monolingually, with documented consequences ranging from temporary harm to death.The combination of abbreviations, a second language, and generic MT remains underexplored, particularly for medical records (Institute for Safe Medication Practices, 2001; Otero López et al., 2004; Brunetti et al., 2007; BBC News, 2008; Sheppard et al., 2008; The Joint Commission, 2026).Source: §1 · pp. 1–2
  • The study examines healthcare staff awareness of patient-safety risks from translating medical records with MT and risks involving translated medical abbreviations.It uses contextualised French and Spanish clinical-corpus examples translated with GT to investigate abbreviation-related risks and staff understanding.Source: §1 · p. 2

2 Related work

Prior work documents healthcare staff using MT amid service constraints, while research on abbreviation translation in medical records remains limited. Existing research also raises concerns about risk awareness and abbreviation-related patient harm.

2 Related work

  • A survey of 1,200 UK residents found MT use was primarily for leisure (80.1%), and only 5.2% of those users expressed medical-context risk concerns.Prior work also reported a healthcare professional using GT for medication administration despite being unable to recognise errors because of unfamiliarity with the source language.Source: §2 · p. 2
  • Healthcare staff reportedly turn to MT, predominantly GT, amid inadequate language-service access, time pressures, or cost constraints, viewing it as a “better than nothing” option.Reported practice also reflects a gap between institutional ideals and clinicians’ working conditions (Royal College of Midwives, 2017; Moberly, 2018; Bell et al., 2020; Vieira et al., 2020; Vieira, 2024); international research describes routine-care use, not complex scenarios, alongside absent policy or training.Source: §2 · p. 2
  • Medical abbreviations are error-prone, and inconsistent standardisation and variation in professional knowledge contribute to concerns about misinterpretation and patient harm.While discharge-instruction MT evaluations have highlighted abbreviation risks, translation of abbreviations in medical records remains an overlooked subdomain.Source: §2 · p. 2

3 Data research methodology

The study used authoritative French and Spanish clinical corpora to identify abbreviation examples, preserve realistic source context, and select two cases for interviews.

3.1 Data collection

  • The study treated shortened forms, acronyms, contractions, and initialisms as medical abbreviations and selected French CAS and Spanish CodiEsp clinical corpora for their authoritative data and clinical breadth.CAS contains 4,900 cases; CodiEsp includes development, test, training, and background sets, and both corpora cover multiple specialties and comorbidities.Source: §3.1 · p. 3

3.2 Data preparation and processing

  • A custom tool filtered source-text matches, while spelling and grammar noise and each match’s paragraph context were retained for realistic translation assessment.Selected matches were translated with GT and analysed in Sketch Engine.Source: §3.2 · p. 3

3.3 Data example selection

  • Two salient cases were selected: Spanish ophthalmology with four untranslated instances of “AO” and French cardiology with one contextually key instance of “DRS”.The examples represented a clinically normal examination and ongoing retrosternal pain, respectively.Source: §3.3 · p. 3

4 Interview methodology

Twenty-one healthcare staff in varied roles took part in remote, staged interviews using blind, contextualised French and Spanish examples. The study elicited baseline views, responses to possible translation errors, and post-discussion confidence and harm assessments.

4.1 Participant recruitment

  • Snowball sampling recruited staff across clinical, clinical-support, and non-clinical roles without restricting language knowledge, experience, or specialty.Of 21 participants, 15 were clinical, 2 clinical support, and 4 non-clinical; all but one worked in NHS settings.Source: §4.1 · p. 3

4.2 Study design and implementation

  • Participants first discussed their background and MT use in lower- and higher-stakes settings before seeing the examples, to reduce priming and social-desirability bias.Interviews were remote and averaged 90 minutes.Source: §4.2 · pp. 3–4
  • Blind, contextualised Spanish “AO” and French “DRS” examples were presented in sequence, with participants asked to assess source text, respond to possible errors, and rate confidence and potential harm.The sequence escalated from ophthalmology with a clinically normal outcome to cardiology with ongoing symptoms; ratings were repeated after discussion of the abbreviations and safety implications.Source: §4.2 · p. 4

4.3 Data analysis framework

  • Researchers used question-based hierarchical codes for qualitative content analysis, then thematically coded data-driven insights relevant to the study aims.Higher-stakes attitudes also received a “pragmatic” code for “better than nothing” workplace MT use, regardless of personal attitude.Source: §4.3 · p. 4

5 Findings and discussion

Staff attitudes toward MT shifted negatively in higher-stakes settings, while discussion of untranslated abbreviations changed some confidence and harm assessments. Participants described safeguards and risks ranging from treatment delays to death, and many reported reflecting on their practices after the interviews.

5.1 Stage 1: Attitude towards MT use across lower- and higher-stakes settings (n=15)

  • 53% of participants with comparable responses reported negative or negative pragmatic attitudes toward higher-stakes MT use, versus 40% positive attitudes in lower-stakes settings.Negative pragmatic participants said they might still use GT as a last resort for lack of a viable alternative; two clinical participants used MT only as guidance within a wider clinical assessment.Source: §5.1 · pp. 4–5, Figure 1 · p. 5

5.2 Stage 2: Impact of ‘AO’ on general confidence scoring (n=21)

  • After discussion, 19% cited ‘AO’ as influencing general confidence, a 14% increase; initially, no participants mentioned it despite 48% mentioning other abbreviations.One nurse lowered confidence after learning ‘AO’ was untranslated in four places and could leave key information missing, including for a specialist.Source: Figure 2 · p. 5, §5.2 · p. 5
  • After discussion, 17% said learning ‘AO’ negatively affected their perception of potential harm, an overall influence increase of 17%.One clinician raised their harm score after reconsidering the assumption that someone in the specialty would recognize the abbreviation.Source: Stage 2: Impact of ‘AO’ on patient… · pp. 5–6, Figure 3 · p. 6

5.3 Stage 3: Impact of ‘DRS’ on general confidence scoring (n=21)

  • After discussion, 52% cited risks from untranslated ‘DRS’ as contributing to a general-confidence score change, a 47% increase in influence.Some participants had initially interpreted ‘DRS’ as ‘doctors’, ‘[health] services’, or ‘Emergency Department’; one participant’s confidence fell from 4 to 2 after discussion.Source: Figure 4 · p. 6, §5.3 · p. 6

5.4 Stage 3: Impact of ‘DRS’ on patient harm scoring (n=18)

  • 33% cited ‘DRS’ as influencing their patient-harm assessment after discussion, compared with 22% before, an influence increase of 11%.One time-critical-care clinician raised their harm score over concerns that obtaining symptom information by other means would delay examination.Source: Figure 5 · p. 6, §5.4 · p. 6

5.5 Summary of key findings: healthcare staff risk awareness in using MT with patient medical record documentation

  • Participants reported not relying on MT alone, but some conflated fluency or readability with translation quality and bilingualism with domain-specific translation competence.Reported safeguards included wider clinical assessment, checking with bilingual speakers or specialists, online cross-checking, and patient confirmation; overlooked or misread abbreviations could remove clinically significant information.Source: §5.5 · p. 7

5.6 Summary of key findings: potential risks identified through the use of abbreviations with MT in healthcare

  • Unresolved or misinterpreted abbreviations were linked to risks from inappropriate medication or treatment to missed necessary treatment, with outcomes ranging from temporary harm to death.Participants identified failure to provide the full term at least once in the source text as a risk contributor because it deprives MT of early contextual clarity.Source: §5.6 · p. 7

5.7 Feedback review: impact of the study on participants’ views (n=18)

  • After the interviews, 50% of commenters reported increased MT risk awareness, 39% reflected on professional practices, and 22% raised concerns about language barriers in patient care.One participant said the study prompted reflection on abbreviation use despite time constraints; another questioned trust in information relayed through colleagues and MT.Source: §5.7 · pp. 7–8

6 Conclusions and recommendations

Staff showed awareness of many general MT limitations, but subtler abbreviation errors could be overlooked; missing source terms can leave clinically important abbreviations untranslated. The authors recommend caution, including full terms in records, checking source text alongside MT, and improving staff literacy.

  • Untranslated abbreviations may be overlooked or misread, and missing full terms at source can make translation less successful, potentially allowing small errors to cascade into clinical consequences.Staff sometimes assumed untranslated abbreviations belonged to another specialty or represented another concept, or glossed over them in favor of more transparent elements.Source: §6 · p. 8
  • Healthcare professionals are advised to use caution when translating patient medical records with MT and to check the source text alongside the output.The source text can help identify abbreviation-related and other translation issues.Source: §6 · p. 8
  • Including the full term at least once in the source is recommended to reduce downstream translation errors.The authors note that omitting full terms may contribute to risk, particularly alongside insufficient MT risk awareness.Source: §6 · p. 8
  • A context-sensitive abbreviation checker integrated into EHR systems is proposed as a way to safeguard record accuracy without unduly increasing manual reporting burdens.Source: §6 · p. 8
  • Targeted training in MT literacy and risk awareness is recommended to address gaps in understanding translation quality, competence, and MT-related risks.The authors state that this could help staff understand MT strengths, weaknesses, and risks at work.Source: §6 · p. 8

7 Positionality statement

The first author’s healthcare and translation experience informed the study’s use case, understanding of staff pressures, and interview-scenario design; controls were used to maintain neutrality and mitigate potential bias.

  • The first author’s four years in healthcare and eight years in freelance translation informed the abbreviation use case, understanding of staff pressures, and design of contextualised, role-specific scenarios.The authors state that strict controls were used to maintain researcher neutrality and mitigate potential bias.Source: §7 · p. 8

8 Limitations of the study

The study’s language scope and sampling constrain how broadly its findings can be generalised. The authors note that the constraints reflect recruitment and validation challenges, while still yielding contextually grounded insights.

  • The study was restricted to French and Spanish, since limited domain-specific data and resources to validate MT outputs made broader language coverage unfeasible.Including scripts relevant to major UK migrant communities could have enhanced the analysis.Source: §8 · p. 8
  • The small sample and snowball sampling restrict generalisability and risk sample homogeneity.The authors describe these constraints as reflecting the challenges of recruiting healthcare professionals with heavy workloads.Source: §8 · p. 8

‘AO’ and ‘DRS’ data examples

Two clinical examples show GT leaving key abbreviations untranslated: Spanish ophthalmology’s ‘AO’ and French cardiology’s ‘DRS’. In both, the missing abbreviation translations affect clinically meaningful information.

Example 1: ‘AO’

  • The Spanish ophthalmology example contains four untranslated instances of ‘AO’, omitting that the examination findings apply to both eyes.The abbreviation means ‘ambos ojos’; the passage gives ‘OU’ or ‘BE’ as correct translations.Source: Example 1: ‘AO’ · p. 11

Example 2: ‘DRS’

  • In the French clinical example, GT leaves ‘DRS’ untranslated, although it denotes the patient’s ongoing retrosternal pain.The example is largely cardiological and shifts to psychiatry in its penultimate sentence.Source: Example 2: ‘DRS’ · p. 11

B Appendix

The appendix lists participant identifiers and professional roles, then provides interview prompts on machine translation use and perceived risks in healthcare settings. The prompts move from personal use to patient-record scenarios, reference materials, confidence, and potential patient harm.

B Appendix

  • The participant list spans clinical, support, non-clinical, and former roles; IV4 and IV14 were unable to attend.Source: B Appendix · p. 12

Lower-stakes settings

  • Participants were asked whether they use machine translation in personal life, which tools they use, and what they think of them.Source: Lower-stakes settings · p. 13

Higher stakes settings

  • Prompts covered role tenure, experiences with foreign-language patient records, contact frequency and purpose, and willingness to use machine translation.Source: Higher stakes settings · p. 13

Presentation of the ST only

Presentation of the MT with ST available for reference

Confidence scoring requested prior to and following the discussion stage

Potential for patient harm scoring requested prior to and following the discussion stage

Loading 2610.02496v1…