What if the biggest threat to sustainable improvement in healthcare is not failing to learn—but failing to forget?

Sustainability is not static maintenance; it is evolution. Health systems that can adapt, discard and relearn, retaining what matters and deliberately forgetting what no longer serves patients, can sustain high-value care and continually move beyond low-value practices.

When learning is not enough: why health systems must learn to forget

Healthcare prides itself on being a learning system. We collect data, consider the context, analyse outcomes and iterate improvement cycles. Yet, true sustainability—the ability to embed and maintain improvement over time—requires more than learning: it demands deliberate unlearning. Outdated order sets, decision rules and routines have the potential to shape care long after evidence has evolved.1 A paradox of improvement is that systems cannot sustain what is new until they have let go of what is old....

Patient safety huddles have been employed across healthcare settings to boost safety culture and improve patient outcomes. However, there is a dearth of literature pertaining to physicians’ levels of interest in participating in these huddles, as well as the impact of physician presence on patient care dynamics. Multidisciplinary huddles aimed at identifying and addressing patient safety issues related to the electronic health record (EHR), for example, were helpful for promoting discussion of EHR-related safety concerns.1 Furthermore, there is a robust literature demonstrating the patient safety benefits of perioperative time-outs in the surgical literature, as well as structured handoffs at the time of care transfer.2–5 Yet hospitalist physicians were not always routinely a key part of these huddles, even though representatives from the medical and surgical services were among the 40 roles that attended some huddles in these studies.

As people age, so do their kidneys.1 The average decline in renal function is approximately 1 mL/min/year after 30 years old.2 3 Over 40% of people >70 years meet the Kidney Disease Improving Global Outcomes definition of chronic kidney disease (CKD), reflecting the increased prevalence of this condition in older people.4 5 Renal function decline is further accelerated through conditions such as hypertension, atherosclerosis, diabetes and cardiovascular disease, which are common comorbidities in older people.6 Safe prescribing of medication for older people with reduced kidney function is an ongoing challenge. Reduced renal excretion of medications eliminated through the kidney exposes people to increased drug plasma levels and increases the risk of medication-related harm leading to adverse drug events, increased hospitalisations and increased mortality.1 7 The presence of polypharmacy, among other issues, in this age...

Background

In 2017, the PREDICT (Paediatric Research in Emergency Departments International Collaborative) network conducted a cluster randomised controlled trial (cRCT) at 26 Australian and New Zealand hospitals to improve bronchiolitis care. Findings demonstrated that targeted interventions significantly improved adherence with five evidence-based low-value bronchiolitis practices (no chest radiography, salbutamol, glucocorticoids, antibiotics and epinephrine) in the first 24 hours of hospitalisation (adjusted risk difference, 14.1%; 95% CI: 6.5% to 21.7%; p<0.001). During the intervention year (2017), intervention hospital (n=13) compliance was 85.1% (95% CI: 82.6% to 89.7%). This study aimed to determine if improvements in bronchiolitis management were sustained at intervention hospitals 2 years post-trial completion.

Methods

International, multicentre follow-up study of hospitals in Australia and New Zealand that participated in a cRCT of de-implementation of low-value bronchiolitis practices, 1 year (2018) and 2 years (2019) post-trial completion, obtained retrospectively from medical audits. Sustainability was defined a priori as no more than a <7% decrease to any level of improvement in adherence for all five low-value practices (composite outcome) from the cRCT intervention year.

Results

Of the 26 hospitals, 11 intervention and 10 control hospitals agreed to participate in the follow-up study. Data were collected on 3299 infants with bronchiolitis 1 year (intervention and control hospitals) and 1689 infants 2 years post-trial (intervention hospitals). Adherence with no use of the five low-value practices 2 years post-trial completion was 80.9% (adjusted predicted adherence, 80.8%, 95% CI: 77.4% to 84.2%; estimated risk difference from cRCT outcome –3.9%, 95% CI: –8.6% to 0.8%) at intervention hospitals, fulfilling the a priori definition of sustainability.

Discussion

Targeted interventions, delivered over one bronchiolitis season, resulted in sustained improvements in bronchiolitis management in infants 2 years later. This follow-up study provides evidence for sustainability in de-implementing low-value care in bronchiolitis management.

Trial registration details

Australian and New Zealand Clinical Trials Registry No: ACTRN12621001287820.

Background

The 2017 Paediatric Research in Emergency Departments International Collaborative (PREDICT) Bronchiolitis Knowledge Translation (KT) Study, a cluster randomised trial in 26 Australasian hospitals, found targeted interventions provided over one bronchiolitis season effectively de-implemented five low-value practices (salbutamol, glucocorticoids, chest radiography, antibiotics and epinephrine) by 14.1% (adjusted risk difference, 95% CI 6.5% to 21.7%; p<0.001). A 2-year follow-up study found de-implementation was sustained. This process evaluation aimed to identify factors that influenced sustainability of de-implementation of these five low-value practices in PREDICT Bronchiolitis KT Study intervention hospitals and examine fidelity and/or adaptation of the targeted interventions over 4 years post intervention delivery (sustainment).

Methods

Semistructured qualitative interviews were conducted, over 2021 and 2022, with a purposive sample of emergency department (ED) and paediatric inpatient clinicians. Data were analysed thematically into facilitators and barriers using the Consolidated Framework for Sustainability Constructs in Healthcare (CFSCH). The Framework for Reporting Adaptations and Modifications to Evidence-based Implementation Strategies was used to explore fidelity and adaptation.

Results

50 clinicians (nurses: n=26; doctors: n=24) from 12 intervention hospitals were interviewed. Eight themes were identified and mapped to three CFSCH domains: (1) organisational setting; (2) initiative design and delivery and (3) people involved. Facilitators were a culture of evidence-based practice, ongoing multimodal education, strong clinical leadership as unofficial champions and the previous effectiveness of the PREDICT Bronchiolitis KT Study interventions. Barriers were lack of paediatric trained ED staff, assumptions by senior clinicians that junior doctors can provide evidence-based bronchiolitis management, bronchiolitis not a current improvement priority and lack of bronchiolitis education sessions. Use of the targeted interventions reduced over time and, when used, was adapted locally.

Conclusion

This study provides insights into factors influencing the sustainability of de-implementation of low-value care in acute care settings. Fostering an evidence-based practice culture, supported by senior leadership and ongoing multimodal education, supports sustainability of improvements in this setting.

Trial registration number

Australian and New Zealand Clinical Trials Registry No: ACTRN12621001287820.

Background

Medical safety huddles are short, structured meetings for physicians to proactively discuss and respond to profession-specific patient safety concerns, with the goal of decreasing future adverse events. Prior observational studies found associations with improved patient safety outcomes, but no randomised controlled studies have been conducted.

Objective

The primary objective was to determine the impact of medical safety huddles on adverse events. Secondary objectives included the fidelity of huddle implementation and the impact on patient safety culture among physicians.

Design

Stepped-wedge cluster randomised trial with four sequences, and each hospital site was a cluster.

Setting

Inpatient oncology, surgery and rehabilitation programmes in four academic hospitals.

Participants

Physicians in participating programmes.

Intervention

Medical safety huddles were adapted for local context and implemented sequentially based on a computer-generated random sequence every 2 months after a 4-month control period. All sites remained in the intervention phase for at least 9 months.

Main outcome and measures

The primary outcome was the rate of adverse events, as determined through blinded chart audits of 912 randomly selected patients. The fidelity of implementation was assessed through the huddle attendance rate, number of safety issues raised in the huddles and number of actions taken in response. Patient safety culture was assessed using the Agency for Healthcare Research and Quality Hospital Survey on Patient Safety.

Results

The adjusted rate of adverse events (per 1000 patient days) in the postintervention phase was 12% lower compared with preintervention (RR: 0.88; 95% CI: 0.80 to 0.98; p=0.016). The odds of having adverse events posthuddle implementation were 17% lower in the postintervention period compared with preintervention (OR intervention vs control: 0.83; 95% CI: 0.80 to 0.87; p<0.001). The mean huddle attendance rate at each site ranged from 30% to 85%, and the mean number of issues raised per huddle and the mean number of actions taken per huddle ranged from 1.6 to 3.1. The mean (SD) overall patient safety rating increased from 2.3 (0.53) to 2.8 (0.88), p=0.010. The mean per cent (SD) positive score for the composite measures of ‘Organisational learning’ increased significantly from 35% (26%) to 54% (23%), p=0.00, ‘Response to error’ 37% (24%) to 52% (22%), p=0.025 and ‘Communication about error’ 36% (28%) to 64% (42%), p=0.016 after implementation.

Conclusions and relevance

Medical safety huddles decreased adverse events and may improve patient safety culture through engaging physicians.

Trial registration number

NCT05365516.

Objectives

To capture experiences of people self-reporting harm and contrast responses and actions between those who do or do not take formal action.

Design

Semi-structured qualitative interview study.

Setting

People self-reporting harm experienced in the National Health Service (NHS) or their family/friends identified from a general Great British population survey.

Participants

49 participants.

Results

There were commonalities in experiences after harm whether formal action (including making a formal complaint or litigation) was taken or not. Many participants reported raising concerns informally with NHS staff, trying to access explanations or support, but were usually unsuccessful. Decision-making on action was complex. There were multiple reasons for not pursuing formal action, including fears of damaging relationships with clinicians, being occupied coping with the consequences of the harm or not wanting to take action against the NHS. NHS advocacy services were not regarded as helpful. Knowledge of how to proceed and feeling entitled to do so, along with proactive social networks, could facilitate action, but often only after people were spurred on by anger and frustration about not receiving an explanation, apology or support for recovery from the NHS. Those from marginalised groups were more likely to feel disempowered to act or be discouraged by family or social contacts, which could lead to self-distancing and reduced trust in services.

Conclusions

People actively seek resolution and recovery after harm but often face multiple barriers in having their needs for explanations, apologies and support addressed. Open and compassionate engagement, especially with those from more marginalised communities, plus tailored support to address needs, could promote recovery, decrease compounded harm and reduce use of grievance services where other provision may be more helpful.

Background

Increasing age is associated with reductions in kidney function and increasing polypharmacy. Most medicines are eliminated through the kidney, meaning older patients are at risk of medication accumulation and toxicity. This scoping review synthesised: (1) the prevalence at which older patients with reduced kidney function in primary care are exposed to inappropriate prescribing; (2) its associated harms; (3) the reasons for this occurring; and (4) the interventions used to improve prescribing practices.

Methods

This scoping review searched ‘Medline’, ‘Embase’, ‘PsycINFO’, ‘CINAHL’ and ‘Web of Science’ for publications before October 2024. References were managed on EndNote V.X5 and thematic data analysis was undertaken on Microsoft Excel. Common themes were identified, summary statistics were calculated and insights were summarised through a narrative technique.

Results

43 relevant studies explored the scale of inappropriate prescribing, estimating prevalences of patient exposure ranging from 0.6% to 49.1% (median 24.9%). Five studies explored the associated harm from inappropriate prescribing, but only one study assessed harm as a primary outcome. Eight studies that assessed difficulties in following prescribing guidelines in reduced kidney function suggested that a lack of awareness and trusted guidelines are fundamental problems. While 13 studies evaluated interventions for improving prescribing in reduced kidney function, only two demonstrated evidence of effectiveness and only one intervention was theoretically informed.

Conclusions

Despite significant heterogeneity in study characteristics, it is clear that the prevalence of inappropriate prescribing for older people is uncomfortably high. There is a lack of evidence linking this to associated adverse outcomes, as well as identifying the causative issues driving this behaviour and the preventative interventions that could prevent harm.

I have heard faculty members complain, on occasion, that students develop bad habits at these clinics because of inadequate supervision. Certainly the quality of care and the ethics of students ‘practicing’ on those who cannot afford other care should be reviewed.

-E. Poulsen, JAMA (1995)1

Introduction

Student-run clinics (SRCs), in which medical and health professions students take responsibility for operational and logistics management of charitable clinics,2 are a powerful expression of service-based learning: students hone clinical and administrative skills while communities receive essential medical services that might otherwise be unavailable. Yet over the past 20 years, these clinics have begun globalising2–5 and increasing in complexity.5 This is happening within a landscape of limited evidence,5 6 growing concerns about ethics and substandard care,7–13

Healthcare contributes nearly 5% of global greenhouse gas (GHG) emissions, along with significant waste, air pollution and water use.1 The production, transport and use of pharmaceuticals, chemicals, medical devices and medical supplies, as well as testing and procedures involved with healthcare delivery, carry a substantial environmental footprint.2 Given that climate change is the defining health challenge of this century, health systems have a moral and professional responsibility not only to provide high-quality care and ensure the best possible patient outcomes but also to minimise environmental harm and protect future generations. Environmentally sustainable healthcare is consistent with high-quality care, especially when framed in terms of stewardship,3 reducing low-value care and waste and improving efficiency and resilience. Additionally, interventions to achieve sustainable healthcare and reduce pollution must ensure that high-quality care is maintained. The study by Spoyalo et al4 is a fine example...

Thirty years ago, Sue Sheridan welcomed her first child, Cal, into the world. At 16 hours of age, a clinician observed that he was jaundiced and entered this assessment into his medical record. But Sheridan didn’t know that. She sensed something was wrong and asked repeatedly about her concerns. She was pegged as a ‘nervous mother’ and reassured. Upon discharge, the nurse’s note again described neonatal jaundice ‘from head to toe’, signalling the need for monitoring and testing. Sheridan was not informed of the potential seriousness of this condition or what symptoms to look for. On day 3, a paediatrician assessed a limp and lethargic infant. On day 4, Sheridan—still concerned—returned to the hospital. There, the bilirubin level returned at 34.6 mg/dL. Cal incurred severe brain damage from kernicterus, diagnosed months later.

Today, Sheridan—and nearly all parents in the USA—would have access to their child’s electronic medical record, including...

Background

Children represent fewer than 10% of emergency medical services (EMS) encounters in the USA. We evaluated whether agency-level paediatric volume is associated with the quality of prehospital care provided.

Methods

We conducted a retrospective analysis of 7104 agencies that contributed data consistently to the 2022–2023 National Emergency Medical Services Information System database, including children (<18 years) from an out-of-hospital EMS encounter. We assessed outcomes based on adherence to paediatric-specific quality benchmarks using mixed-effects models.

Results

We identified 3 403 925 paediatric encounters (median age 10 years; IQR 3–15). The annual paediatric volumes serviced by the study agencies per year ranged from 0.5 to 62 443. Six measures had a positive association with EMS volume, one measure had a negative association with EMS volume and four measures had no association with EMS volume. Higher volumes were associated with beta agonist administration for asthma/wheeze (adjusted OR (aOR) 1.08 per twofold increase in volume, 95% CI 1.06 to 1.11), epinephrine for anaphylaxis (aOR 1.09, 95% CI 1.05 to 1.08), vital signs assessment in trauma (aOR 1.05, 95% CI 1.04 to 1.07), benzodiazepines for status epilepticus (aOR 1.21, 95% CI 1.17 to 1.25), oxygen or positive pressure ventilation for hypoxia (aOR 1.06, 95% CI 1.04 to 1.09) and naloxone for opioid overdose (aOR 1.08, 95% CI 1.02 to 1.14). Higher paediatric volume was negatively associated with improvement of pain status in trauma (aOR 0.96, 95% CI 0.95 to 0.97). Paediatric volume was not associated with management of hypoglycaemia (aOR 1.01, 95% CI 0.97 to 1.06) or hypotension (aOR 0.98, 95% CI 0.92 to 1.04), or analgesia (0.99, 95% CI 0.97 to 1.01) and pain assessment (aOR 1.01, 95% CI 0.99 to 1.04) in trauma.

Conclusion

Higher paediatric volume EMS agencies had better adherence to some paediatric care quality measures but showed no association or an inverse association with others. Efforts to improve prehospital paediatric care quality should pay special attention to low-volume agencies.

Background

Guideline concordance is associated with improved patient outcomes. Accurately quantifying the concordance between provided care and guideline recommendations offers valuable insights into the alignment of care with established guidelines and supports proactive approaches for improving the quality of care. Traditional models for calculating guideline concordance are effective in assessing clinical performance via cohort averages. However, these models fail at the individual patient level by not accounting for past clinical activities and their timing, which may give a distorted impression of the actual alignment between guideline recommendations and received care.

Objectives

To develop a model for evaluating guideline concordance that provides accurate concordance scores at the individual patient level.

Methods

The newly developed ratio model incorporates past clinical activities and their timing (ie, past clinical trajectories), resulting in accurate, patient-centred concordance scores. We discuss its advantages and limitations and showcase its performance using clinical indicators for patients with type 2 diabetes mellitus.

Results

The ratio model demonstrates enhanced precision in evaluating guideline concordance at the individual level and better reflects the clinical trajectory of individual patients. While primarily designed to produce accurate individual patient scores, the model is also effective for assessing clinical performance through cohort averages. The ratio model is adaptable to diverse clinical contexts requiring regular follow-up, including chronic disease management, vaccination programmes, cancer surveillance and routine health screenings.

Conclusions

The ratio model provides accurate and patient-centred guideline concordance scores. The model’s enhanced precision at the individual level creates opportunities for research and clinical applications, including integration into clinical decision support systems.

Objectives

Many hospitals use fall prevention alarms, despite the limited evidence of effectiveness. The objectives of this study were (1) to identify, conceptualise and select strategies to deimplement fall prevention alarms and (2) to obtain feedback from key stakeholders on tailoring selected deimplementation strategies for the local hospital context.

Methods

Hospital staff working on fall prevention participated in group concept mapping (GCM) to brainstorm strategies that could be used for fall prevention alarm deimplementation, sort statements into conceptually similar categories and rate statements based on importance and current use. Hospital staff also participated in site-specific focus groups to discuss current fall prevention practices, strategies prioritised through GCM and theory-informed strategies recommended by the study team, and potential barriers/facilitators to deimplementing fall prevention alarms.

Results

90 hospital staff across 13 hospitals brainstormed, rated and sorted strategies for alarm deimplementation. Strategies that were rated as highly important but underutilised included creating/revising staff roles to support fall prevention (eg, hiring or designating mobility technicians) and revising policies and procedures to encourage tailored rather than universal fall precautions. 192 hospital staff across 22 hospitals participated in site-specific focus groups. Participants provided feedback on each strategy’s relevance for their site (eg, if site currently has a mobility technician) and local barriers or facilitators (eg, importance of having separate champions for day and night shift). Findings were used to develop a tailored implementation package for each site that included a core set of strategies (eg, external facilitation, education, audit-and-feedback, champions), a select set of site-specific strategies (eg, designating a mobility technician to support fall prevention) and guidance for how to operationalise and implement each strategy given local barriers and facilitators.

Conclusion

Findings from this study can be used to inform future programmes and policies aimed at deimplementing fall prevention alarms in hospitals.

Purpose

Misgendering of transgender and non-binary (TGNB) individuals in healthcare settings can lead to worsened mental and physical health outcomes and decreased utilisation of care. Few studies have investigated the factors that contribute to this phenomenon. The purpose of this study was to apply qualitative methods to explore sources of misgendering, its perceived impact, prevention strategies and clinician responses to accidentally misgendering a patient, as identified by TGNB patients and gender-affirming care clinicians.

Methods

Between April and June 2022, 20 semi-structured interviews were performed at an academic medical centre in Southern California. Participants were recruited via purposive sampling and included: (1) TGNB patients (n=8) recruited from an interdisciplinary gender-affirming urological practice and (2) gender-affirming care clinicians (n=12) recruited from a regional interdisciplinary Gender Health conference, three of whom identified as TGNB. Interviews were conducted in person or virtually using an open-ended topic guide, audio recorded and transcribed verbatim. Inductive thematic analysis was performed by two independent study personnel who hand-coded the transcripts.

Results

Four overarching themes were identified: (1) misgendering originates from multiple sources, (2) misgendering discourages individual access to healthcare, creates community hesitation and its perceived impact is modified by setting and intentionality, (3) building a gender-affirming healthcare system requires integration of behaviour, policy and technology and (4) clinicians respond to accidental misgendering by acknowledging, apologising, advancing and acting.

Conclusion

Our data suggest that misgendering arises from both interpersonal communication and structural factors within healthcare systems, leading to perceived harm and diminished TGNB access to health services. Any potential solution to reduce this phenomenon will require a multifaceted approach integrating behavioural, technological and institutional policy strategies with system-level implementation efforts.

Background

Access to electronic health records (EHRs) has the potential to improve the quality of care. Clinical notes, free-text entries documenting clinicians’ observations and decisions, are central to EHRs. Sharing these notes may reduce information asymmetry, enhance transparency and empower patients. However, their impact on care quality remains unclear.

Aim

To assess the impact of sharing clinical notes online with patients on the domains of quality as defined by the Institute of Medicine (ie, patient-centredness, effectiveness, efficiency, safety, timeliness and equity).

Methodology

A systematic review was conducted with no time limit, using CINAHL, Cochrane, OVID Embase, HMIC, Medline/PubMed and PsycINFO. A narrative synthesis method was employed to extract the study characteristics, and reported outcomes were organised using the six IOM quality domains. The risk of bias of included studies was assessed using the Risk of Bias in Non-randomised Studies of Interventions (ROBINS-I) tool.

Results

Nineteen studies involving 203 152 participants met inclusion criteria. Outcomes included patient-centredness (n=16), patient safety (n=14), equity (n=6), efficiency (n=4), timeliness (n=0) and effectiveness (n=0). Patient-centredness studies reported high satisfaction (n=6), increased engagement (n=11) and stronger patient–provider trust (n=7). Patient safety studies noted improvements in medication adherence (n=4) and note accuracy (n=5), alongside privacy concerns (n=5). Equity studies found benefits for minority (n=3) and less-educated patients (n=2), with one reporting equitable outcomes (n=1). No significant changes in efficiency were observed (n=4).

Discussion

Online sharing of clinical notes with patients positively impacted self-reported patient-centredness and patient safety, particularly benefiting underserved populations. However, privacy concerns must be effectively addressed, and robust safeguarding is essential to mitigate confidentiality issues. Further research is needed to evaluate the long-term impact on timeliness, effectiveness and efficiency of care.

Background

Intraoperative hypothermia can lead to adverse clinical outcomes and avoidable financial and environmental costs. Environmentally preferable warming practices have been identified, including using reusable resistive blankets, extending the life cycle of forced air warming (FAW) garments and minimising flannel blanket use. This study integrates existing environmental data with best practices and quality improvement methodology to develop an optimised patient warming pathway (OPWP). This pathway was adapted to our local context, implemented and evaluated.

Methods

The OPWP was developed using a scoping review, prior environmental impact assessment and root cause analysis. It was tailored to the workflows, patient population and warming practices at a tertiary care hospital and implemented using a multifaceted approach encompassing nine PDSA (Plan-Do-Study-Act) cycles. Major interventions included expanding pre-warming criteria to meet best practice guidelines, preserving the FAW Flex Gown, staff education and training, behaviourally informed strategies, gamification and policy development. Pre-intervention and post-intervention audits assessed environmental and financial savings, incidence of hypothermia and patient-reported outcomes (PROs).

Results

The OPWP recommends preferential use of the resistive blanket for intraoperative warming, preservation of the Flex Gown for postoperative use when warming with FAW and minimising flannel blanket use. A modified pathway was implemented using FAW with preservation of a single Flex Gown throughout the perioperative journey. From pre-intervention (N=51) to post-intervention (N=64), flannel blanket use decreased from an average of 6 to 3 per patient (p<0.01). Active warming increased from 55% to 80% (p=0.04) preoperatively and from 0% to 55% (p<0.01) postoperatively. There was no significant change in the incidence of hypothermia (18% to 15%, p=0.77) and PROs remained favourable. Implementation of this pathway could lead to annual environmental savings of 940 339 kg of carbon dioxide equivalents and cost savings of $C117 978.

Conclusions

This study demonstrates the successful implementation of an evidence-based and environmentally sustainable perioperative warming pathway to achieve low-carbon, high-quality patient care.

Artificial intelligence (AI) has great potential to assist healthcare staff and organisations in maintaining and improving the quality and safety of healthcare1 in the face of workforce shortages, rising service demand and escalating costs. Despite hundreds of regulator-approved AI-enabled tools internationally, relatively few feature in routine clinical care,2 in part due to inattention to how AI tools integrate into sociotechnical healthcare environments.3 In this Viewpoint, based on our experience as AI implementation researchers, we discuss what we see as seven key barriers to the adoption of AI in healthcare and offer some solutions.

AI literacy and engagementUnderdeveloped professional skills and consumer understanding

AI will never be adopted at scale unless health professionals better understand AI and its limitations, acquire competencies in co-designing, co-evaluating and effectively using AI tools, undertake continual vigilance of AI tool performance and avoid over-reliance on AI with...

Language barriers between healthcare professionals and their patients remain a persistent challenge. Patients with limited proficiency in the primary language of the country where they receive care face higher risks of adverse events, misdiagnosis and unplanned readmissions.1 2 In linguistically diverse countries, services often fall short of meeting the needs of patients who speak minority languages. This leads to inequities in care across inpatient, outpatient and emergency settings. While concern for language discordance in healthcare is by no means a novel development, guidelines have rarely progressed beyond recommending implementation of professional interpreter services.3 In-person interpretation is generally considered the gold standard for addressing language barriers during direct care delivery4 5 and is in line with regulatory and ethical standards.6 Other modalities of interpretation, including telephone and video, are alternative options, although the feasibility of providing timely written...

Introduction

Machine translation of patient-specific information could mitigate language barriers if sufficiently accurate and non-harmful and may be particularly useful in healthcare encounters when professional translators are not readily available. We evaluated the translation accuracy and potential for harm of ChatGPT-4 and Google Translate in translating from English to Spanish, Chinese and Russian.

Methods

We used ChatGPT-4 and Google Translate to translate 50 sets (316 sentences) of deidentified, patient-specific, clinician free-text emergency department instructions into Spanish, Chinese and Russian. These were then back-translated into English by professional translators and double-coded by physicians for accuracy and potential for clinical harm.

Results

At the sentence level, we found that both tools were ≥90% accurate in translating English to Spanish (accuracy: GPT 97%, Google Translate 96%) and English to Chinese (accuracy: GPT 95%; Google Translate 90%); neither tool performed as well in translating English to Russian (accuracy: GPT 89%; Google Translate 80%). At the instruction set level, 16%, 24% and 56% of Spanish, Chinese and Russian GPT-translated instruction sets contained at least one inaccuracy. For Google Translate, 24%, 56% and 66% of Spanish, Chinese and Russian translations contained at least one inaccuracy. The potential for harm due to inaccurate translations was ≤1% for both tools in all languages at the sentence level and ≤6% at the instruction set level. GPT was significantly more accurate than Google Translate in Chinese and Russian at the sentence level; the potential for harm was similar.

Conclusion

These results support the potential of machine translation tools to mitigate gaps in translation services for low-stakes written communication from English to Spanish, while also strengthening the case for caution and for professional oversight in non-low-risk communication. Further research is needed to evaluate machine translation for other languages and more technical content.