Medical Interpretation Research Articles

40 Trust, belonging and understanding - The experiences of families who use languages other than English (LOE) that are hospitalized on general paediatric inpatient units

Abstract Background Despite recognition that social determinants are critical to health outcomes, many groups, including those with lower incomes, racial/ethnic minorities, and other historically marginalized populations, continue to experience significant social and structural barriers to care. There is need for research that incorporates patient and stakeholder engagement in developing, testing, and disseminating interventions to improve outcomes for patient populations that have been marginalized in the health system. Patients who use a language(s) other than English (LOE) for healthcare communication are at increased risk for experiencing adverse events and worse health outcomes in healthcare settings, including in paediatric hospitals. Canadian census data indicates 12.7% of Canadians predominantly speak a language other than French or English. This work aims to generate knowledge to improve health outcomes for children and families who speak LOE. Objectives (1) To understand the lived experience of families who speak LOE around the hospitalization of their child and (2) To understand the perspectives of patients and families who speak LOE on how the healthcare system can improve care during hospitalization. Design/Methods The premise of this study is participatory research. We designed a qualitative study involving children hospitalized in the General Paediatric Inpatient Unit at the Hospital for Sick Children. The study includes family caregivers of children age 0 to 18 years who identify as speaking LOE. Children and families could speak any language that we were able to obtain interpreter support for. We excluded patients that spoke a traditional Indigenous language as we noted this work required a separate study in partnership with the Indigenous community. We conducted semi-structured virtual individual interviews with children and families post-discharge with a medical interpreter. An interview guide was developed and transcripts were analyzed using thematic analysis. We took several steps to ensure an effective study design given our goals of participatory research including ensuring patient and family engagement throughout the design process. Results 19 families participated in the interviews and 16 different languages were represented amongst our participant groups. Our study is unique in especially unique in it's ability to recruit and interview individuals with so many different primary languages. Our study provides rich insight into opportunities within the paediatric inpatient setting for quality improvement to address the health disparities experienced by patients and families who use LOE for healthcare communication. Our analysis of the interviews identify several themes that provide insight into the lived experience of patients and families who use LOE for healthcare communication. 1) Inconsistent interpreter use affects communication “in the moment” and the ability to develop an “overall understanding” of the current clinical situation and broader healthcare context. Our participants note a "snowball effect" when interpreters are not consistently utilized. 2) In addition to language, multiple additional social factors contribute to an increased sense of vulnerability and modulates how families “trust” the healthcare system. 3) Despite families expressing positive experiences, they noted that they did not, nor expect to feel like they “belonged” within the healthcare system. It is critical to recognize the complex interplay between many factors surrounding our participants’ identities. Though they use a LOE for healthcare, the intersectional marginalization they experience was thoroughly represented during our conversations Conclusion Our research aimed to take a participatory approach to consider the lived experience of patients and families who use LOE for healthcare communication. This work builds on the current literature on patient and family centered care, experiences of families who speak LOE and participatory community-based research in the paediatric hospital settings. Our study's findings support the development of a research and intervention agenda in Canada to improve health outcomes for patients and families with LOE.

34 Improving language concordant care in the paediatric emergency department

Abstract Background Paediatric patients and families with limited English proficiency (LEP) experience worse health outcomes. Use of professional medical interpreter services (IS) helps ensure that all patients receive high quality care, regardless of their preferred language of communication. Objectives The aim of this quality improvement (QI) study was to improve language concordant care within a paediatric emergency department (ED) in downtown Toronto by increasing the rate of over-the-phone interpreter use by 50% by July 2024 and to understand common reasons for interpreter non-use. Design/Methods Using the model for improvement, we assembled an interdisciplinary team that included physicians, ED registration clerks, triage nurses, and language services staff to understand current workflow and develop targeted interventions aimed at addressing key drivers through multiple plan-do-study-act (PDSA) cycles. Interventions included introduction of a standardized script to capture language preference at ED registration, EMR-integrated preferred language notification and smart text template in provider notes for standardized IS use or non-use documentation, and departmental education. Our primary outcome measure was rate of over-the-phone IS use for all ED visits. Process measures were rate of EMR documentation of patients and caregivers with non-English language preference (NELP) and rate of provider documentation of IS use or non-use. Balancing measures included time to provider initial assessment (PIA) and ED length of stay. Results The baseline period (November 2022-April 2023) revealed an over-the-phone IS rate of 5.2% for all ED visits. The initial results from the first 6 months of the intervention period demonstrated an increase in physician documentation of IS use or non-use in patients with NELP, from 10% to 76%. Additionally, 18.6% of patients or caregivers with NELP had IS use per provider notes. Over the next 9 months of the intervention period, special cause variation is anticipated in the rate of IS use and rate of EMR documentation of NELP. The most common reasons for IS non-use were: ‘deemed unnecessary by provider’, ‘declined by family’ and ‘use of family member/friend as interpreter’. There was no change in PIA or ED length of stay for patients with NELP. Conclusion This QI initiative increased consistent provider documentation of IS use in ED patient records in one of the most linguistically diverse cities in the world. Future efforts will focus on exploring additional ways to capture language preference beyond registration, facilitating IS use in the ED, and addressing top reasons for non-use of IS in patients and caregivers with NELP.

Medical Team Practices and Interpreter Alterations on Family-Centered Rounds.

Robust evidence demonstrates inequities in communication during family-centered rounds for families who use a language other than English (LOE) for health care. This study aimed to characterize the type of interpreter alterations occurring on family-centered rounds and identify medical team communication practices associated with alterations. In this observational study of interpreter-supported family-centered rounds, we recorded and transcribed family-centered rounds encounters for Spanish-speaking families. We assessed measures of medical team communication behaviors and interpreter alterations (omissions, additions, and substitutions) using previously described instruments. We used a content analysis approach to apply defined codes to each interpreted segment and to characterize the nature of interpreter alterations. We assessed the association between medical team communication behaviors and interpreter alterations using χ2 tests. We recorded, transcribed, and coded 529 interpreted segments of 10 family-centered rounds encounters. At least 1 alteration was present in 72% (n = 382/529) of interpreted segments. Omissions were the most common alteration (n = 242/529, 46%) followed by substitutions (n = 177/529, 34%) and additions (n = 71/529, 13%). Interpretation resulted in a potentially negative alteration in 29% (n = 155/529) and a positive alteration in 9% (n = 45/529) of segments. Greater number of sentences in the segment preceding interpretation was associated with an increase in loss of information (P < .001), loss of social support (P = .003), and loss of partnership (P = .020). To improve communication with families that use an LOE, medical teams must abide by best practices for using an interpreter such as frequent pausing to prevent loss of both biomedical and psychosocial information.

Spanish Terms for Cleft Lip Within the United States.

To examine Spanish terms used for “cleft lip” within the United States. Perceived acceptability of these terms was also considered. An online survey was distributed to Spanish-speaking medical interpreters, clinicians, and parents of children with cleft lip. Survey questions probed demographics, language use, preferred term for “cleft lip,” and perceived acceptability of terms for cleft lip. There were 158 respondents. Significant differences were observed with profession ( P &lt; .001) as 62.5% of medical providers used labio hendido, while 71.8% of interpreters preferred labio leporino. Perceived acceptability of terms was consistent with the term use patterns reported by respondent groups. Significant differences were also observed with geographical region ( P = .001) as 90% of respondents from the Northeast United States employed labio leporino, compared to 60% to 70% of respondents from other regions. Selection and perception of Spanish terms for “cleft lip” may differ across regions and professions. Cleft teams should invite stakeholders to discuss what terms best suit their community.

Bridging the Language Gap in Healthcare: Implementing a Qualified Medical Interpreter Program for Lesser-Spoken Languages

Linguistic inequity drives systemic disparities in healthcare for non-native English speakers. This study evaluates a project to train and provide qualified medical interpreters (QMI) to assist volunteer and safety-net clinics and community-based organizations in supporting healthcare for immigrants and refugees. We provided scholarships to bilingual community members to take a medical interpreter training course and developed a workforce for those who passed the training course. We focused on lesser-spoken foreign languages such as Arabic, Amharic, Pashto, Dari, and Burmese. Those who passed the course participated in a semi-structured interview to learn about their experiences in the training program, as well as barriers and facilitators to becoming a QMI. To date, 23 people have passed the training and are part of the QMI workforce program that has provided 94 h of interpreter services over four months, serving 66 individual patients. The evaluation showed that community members have interest in becoming QMIs and many have the required language proficiency to enroll and pass training. Finding full-time employment for less spoken languages has proven to be challenging.

Communication Skills of Medical Interpreters: A Qualitative Explanatory Study of Healthcare Professionals' and Medical Interpreters' Perceptions.

Background/Objectives: Medical interpreters support communication between medical professionals and foreign patients. However, the communication skills required of medical interpreters in the field are currently unclear. The purpose of this study was to investigate what medical professionals and medical interpreters consider to be most important communication skills of medical interpreters, and whether there are differences in perceptions between medical professionals and medical interpreters. Methods: From December 2023 to February 2024, we interviewed 10 medical professionals and medical interpreters (five medical professionals and five medical interpreters) working throughout Japan and in the United States. The narratives collected in the interviews were analyzed and evaluated in a conceptual framework. Results: Healthcare professionals and medical interpreters exhibited differences in perceptions regarding the most important communication skills for medical interpreters. The narratives of medical professionals and medical interpreters regarding medical interpreters' communication skills were divided into a 2 × 2 grid consisting of four quadrants with two axes (intervention and perspective). Regarding the communication skills of medical interpreters, medical professionals were primarily concerned with the transmission of medical information and felt that it was necessary for interpreters to understand medical professionals' perspectives and what they wanted to say in order to convey it accurately. In contrast, medical interpreters emphasized the importance of empathy and communication skills for respecting the other person and understanding the other person's opinion and perspective. Additionally, to understand the patient's perspective, medical interpreters reported that it was important to understand the cultural background of the patient and to maintain good relationships with all of the people around the patient. Conclusions: We explored the communication skills of medical interpreters and the characteristics of the perceptions and narratives of medical professionals and medical interpreters. Concepts regarding the communication skills of medical interpreters were divided into two axes (intervention and perspective) and presented as a 2 × 2 grid with four quadrants. The results of our study suggested that medical interpreters should actively intervene in interpretation situations involving foreign patients.

Novel SERPINC1 Mutation in a Brazilian Patient with Sudden-onset Sagittal Sinus Thrombosis Caused by Congenital Antithrombin Deficiency

A 19-year-old Brazilian man visited the emergency department of our hospital complaining of nausea and fever. He presented with left hemispatial neglect. Magnetic resonance imaging led to a diagnosis of sagittal sinus thrombosis. Blood tests showed decreased antithrombin (ATIII) activity, and ATIII deficiency was diagnosed. He was treated with heparin sodium and warfarin. Genetic counseling through medical interpreters was recommended, and the patient and his mother consented to genetic testing, which revealed a novel mutation variant of the SERPINC1 gene. Genetic counseling is important to connect patients to genetic testing, even if language barriers are present.

Applying queueing theory to evaluate wait-time-savings of triage algorithms

In the past decade, artificial intelligence (AI) algorithms have made promising impacts in many areas of healthcare. One application is AI-enabled prioritization software known as computer-aided triage and notification (CADt). This type of software as a medical device is intended to prioritize reviews of radiological images with time-sensitive findings, thus shortening the waiting time for patients with these findings. While many CADt devices have been deployed into clinical workflows and have been shown to improve patient treatment and clinical outcomes, quantitative methods to evaluate the wait-time-savings from their deployment are not yet available. In this paper, we apply queueing theory methods to evaluate the wait-time-savings of a CADt by calculating the average waiting time per patient image without and with a CADt device being deployed. We study two workflow models with one or multiple radiologists (servers) for a range of AI diagnostic performances, radiologist’s reading rates, and patient image (customer) arrival rates. To evaluate the time-saving performance of a CADt, we use the difference in the mean waiting time between the diseased patient images in the with-CADt scenario and that in the without-CADt scenario as our performance metric. As part of this effort, we have developed and also share a software tool to simulate the radiology workflow around medical image interpretation, to verify theoretical results, and to provide confidence intervals for the performance metric we defined. We show quantitatively that a CADt triage device is more effective in a busy, short-staffed reading setting, which is consistent with our clinical intuition and simulation results. Although this work is motivated by the need for evaluating CADt devices, the evaluation methodology presented in this paper can be applied to assess the time-saving performance of other types of algorithms that prioritize a subset of customers based on binary outputs.

Perforator Selection with Computed Tomography Angiography for Unilateral Breast Reconstruction: A Clinical Multicentre Analysis.

Background and Objectives: Despite CTAs being critical for preoperative planning in autologous breast reconstruction, experienced plastic surgeons may have differing preferences for which side of the abdomen to use for unilateral breast reconstruction. Large language models (LLMs) have the potential to assist medical imaging interpretation. This study compares the perforator selection preferences of experienced plastic surgeons with four popular LLMs based on CTA images for breast reconstruction. Materials and Methods: Six experienced plastic surgeons from Australia, the US, Italy, Denmark, and Argentina reviewed ten CTA images, indicated their preferred side of the abdomen for unilateral breast reconstruction and recommended the type of autologous reconstruction. The LLMs were prompted to do the same. The average decisions were calculated, recorded in suitable tables, and compared. Results: The six consultants predominantly recommend the DIEP procedure (83%). This suggests experienced surgeons feel more comfortable raising DIEP than TRAM flaps, which they recommended only 3% of the time. They also favoured MS TRAM and SIEA less frequently (11% and 2%, respectively). Three LLMs-ChatGPT-4o, ChatGPT-4, and Bing CoPilot-exclusively recommended DIEP (100%), while Claude suggested DIEP 90% and MS TRAM 10%. Despite minor variations in side recommendations, consultants and AI models clearly preferred DIEP. Conclusions: Consultants and LLMs consistently preferred DIEP procedures, indicating strong confidence among experienced surgeons, though LLMs occasionally deviated in recommendations, highlighting limitations in their image interpretation capabilities. This emphasises the need for ongoing refinement of AI-assisted decision support systems to ensure they align more closely with expert clinical judgment and enhance their reliability in clinical practice.

Prediction of early recurrence of adult-type diffuse gliomas following radiotherapy using multi-modal magnetic resonance images.

Adult-type diffuse gliomas are among the central nervous system's most aggressive malignant primary neoplasms. Despite advancements in systemic therapies and technological improvements in radiation oncology treatment delivery, the survival outcome for these patients remains poor. Fast and accurate assessment of tumor response to oncologic treatments is crucial, as it can enable the early detection of recurrent or refractory gliomas, thereby allowing timely intervention with life-prolonging salvage therapies. Radiomics is a developing field with great potential to improve medical image interpretation. This study aims to apply a radiomics-based predictive model for classifying response to radiotherapy within the first 3 months post-treatment. Ninety-five patients were selected from the Burdenko Glioblastoma Progression Dataset. Tumor regions were delineated in the axial plane on contrast-enhanced T1(CE T1W) and T2 fluid-attenuated inversion recovery (T2_FLAIR) magnetic resonance imaging (MRI). Hand-crafted radiomic (HCR) features, including first- and second-order features, were extracted using PyRadiomics (3.7.6) in Python (3.10). Then, recursive feature elimination with a random forest (RF) classifier was applied for feature dimensionality reduction. RF and support vector machine (SVM) classifiers were built to predict treatment outcomes using the selected features. Leave-one-out cross-validation was employed to tune hyperparameters and evaluate the models. For each segmented target, 186 HCR features were extracted from the MRI sequence. Using the top-ranked radiomic features from a combination of CE T1W and T2_FLAIR, an optimized classifier achieved the highest averaged area under the curve (AUC) of 0.829±0.075 using the RF classifier. The HCR features of CE T1W produced the worst outcomes among all models (0.603±0.024 and 0.615±0.075 for RF and SVM classifiers, respectively). We developed and evaluated a radiomics-based predictive model for early tumor response to radiotherapy, demonstrating excellent performance supported by high AUC values. This model, harnessing radiomic features from multi-modal MRI, showed superior predictive performance compared to single-modal MRI approaches. These results underscore the potential of radiomics in clinical decision support for this disease process.

Computational imaging for rapid detection of grade-I cerebral small vessel disease (cSVD)

An early identification and subsequent management of cerebral small vessel disease (cSVD) grade 1 can delay progression into grades II and III. Machine learning algorithms have shown considerable promise in medical image interpretation automation. An experimental cross-sectional study aimed to develop an automated computer-aided diagnostic system based on AI (artificial intelligence) tools to detect grade 1-cSVD with improved accuracy. Patients with Fazekas grade 1 cSVD on Non-Contrast Magnetic Resonance Imaging (MRI) Brain of age >40 years of both genders were included. The dataset was pre-processed to be fed into a 3D convolutional neural network (CNN) model. A 3D stack with the shape (120, 128, 128, 1) containing axial slices from the brain magnetic resonance image was created. The model was created from scratch and contained four convolutional and three fully connected (FC) layers. The dataset was preprocessed by making a 3D stack, and normalizing, resizing, and completing the stack was performed. A 3D-CNN model architecture was designed to train and test preprocessed images. We achieved an accuracy of 93.12 % when 2D axial slices were used. When the 2D slices of a patient were stacked to form a 3D image, an accuracy of 85.71 % was achieved on the test set. Overall, the 3D-CNN model performed very well on the test set. The earliest and the most accurate diagnosis from computational imaging methods can help reduce the huge burden of cSVD and its associated morbidity in the form of vascular dementia.

Adaptively triggered comparisons enhance perceptual category learning: evidence from face learning

Categorical learning is important and often challenging in both specialized domains, such as medical image interpretation, and commonplace ones, such as face recognition. Research has shown that comparing items from different categories can enhance the learning of perceptual classifications, particularly when those categories appear highly similar. Here, we developed and tested novel adaptively triggered comparisons (ATCs), in which errors produced during interactive learning dynamically prompted the presentation of active comparison trials. In a facial identity paradigm, undergraduate participants learned to recognize and name varying views of 22 unknown people. In Experiment 1, single-item classification trials were compared to a condition in which ATC trials were generated whenever a participant repeatedly confused two faces. Comparison trials required discrimination between simultaneously presented exemplars from the confused categories. In Experiment 2, an ATC condition was compared to a non-adaptive comparison condition. Participants learned to accuracy and speed criteria, and completed immediate and delayed posttests. ATCs substantially enhanced learning efficiency in both experiments. These studies, using a novel adaptive procedure guided by each learner’s performance, show that adaptively triggered comparisons improve category learning.

Hospital commitments to address diagnostic errors: An assessment of 95 US hospitals.

Diagnostic errors are a leading cause of patient harm. In 2022, the Leapfrog Group published a report containing 29 evidence-based practices that hospitals can adopt to reduce diagnostic errors. To understand the extent to which US hospitals have already implemented these practices, we conducted a national pilot survey of Leapfrog-participating hospitals. To reduce respondent burden, we divided the 29 practices across two surveys: one focused on organizational culture and structure (Domain 1), and the second focused on the diagnostic process itself (Domain 2). A total of 95 hospitals from 23 states responded to one or both surveys. On average, hospitals reported implementing 9 of the 16 practices (56%) in Domain 1 and 8 of the 13 practices (62%) in Domain 2. The rate of practice implementation varied greatly, with some hospitals implementing as few as threepractices in their domain. The most commonly implemented practices were ensuring access to medical interpreters, continuous access to radiologists, ensuring staff and patients can report diagnostic errors and concerns, and having a formal process to identify and notify patients when diagnostic errors occur. The least implemented practices included convening a multidisciplinary team focused on diagnostic safety and quality, a CEO commitment to diagnostic excellence, conducting diagnosis-focused risk assessments, and training clinicians to optimize clinical reasoning in the diagnostic process. The findings suggest large and important implementation gaps for practices related to diagnostic excellence and can inform new initiatives to promote diagnostic excellence in US hospitals.

From Photography to Radiology: How Physicians Leveraged Early Hospital X-ray Machines to Supplant Photographers.

At the end of the nineteenth century, the advent of x-ray machines fueled American medicine's reliance on technology, transforming hospitals and the medical profession. X-ray manufacturers pursued the nascent hospital market as competition and patent feuds accelerated x-ray machine modifications. Hospitals incorporated clunky new machines and employed x-ray photographers, but as the unruly apparatus stabilized, physicians joining the new specialty of radiology discounted the toils of machine troubleshooting and promoted their medically qualified x-ray interpretations. This article frames early medical radiography in terms of boundary work, highlighting how discourse among physicians, x-ray photographers, and hospital administrators vied to establish a privileged demarcation between radiological science and photographic craft. Ultimately, radiologists supplanted x-ray photographers by leveraging the automation of x-ray machines and capitalizing on the epistemic shift from photographic objectivity to qualified interpretations. By focusing on this overlooked aspect of x-ray incorporation into hospitals, this work provides a unique perspective on how harnessing mechanization and authoritative medical interpretations can shift professional boundaries.

Medical Interpretation for Emerging Populations: Building a Multi-Modal Interpretation Program for Pashto-Speakers at a Public Safety-Net Health System

Summary: Limited English proficient patients are prone to adverse health care effects compared to English proficient patients, including decreased access to care, lower health literacy, and worse clinical outcomes. This report describes a multi-modal medical interpreter program at a safety-net health system designed for the emerging Afghan population in San Antonio, Texas.

“They really trust us!”: medical interpreter’s roles and experiences in an integrated primary care clinic

ABSTRACT This study describes medical interpreters’ experiences with behavioral health (BH) services in a primary care clinic. Focus group data with medical interpreters representing multiple languages was analyzed using hybrid inductive-deductive thematic analysis. Themes related to interpreter roles were: (1) case management, (2) patient-interpreter relationship, and (3) patient-provider liaison. Themes related to barriers and facilitators to interpreter-mediated BH care were: (1) cultural factors, (2) patient-provider interactions, (3) BH-specific considerations, and (4) clinic factors. Results illustrate ways that interpreters directly (e.g. interpreter-mediated services) and indirectly (e.g. relationship building) support care. The interpreter-patient relationship reportedly helped improve patient attitudes and buy-in for BH.

Evaluating AI Proficiency in Nuclear Cardiology: Large Language Models take on the Board Preparation Exam.

Previous studies evaluated the ability of large language models (LLMs) in medical disciplines; however, few have focused on image analysis, and none specifically on cardiovascular imaging or nuclear cardiology. This study assesses four LLMs - GPT-4, GPT-4 Turbo, GPT-4omni (GPT-4o) (Open AI), and Gemini (Google Inc.) - in responding to questions from the 2023 American Society of Nuclear Cardiology Board Preparation Exam, reflecting the scope of the Certification Board of Nuclear Cardiology (CBNC) examination. We used 168 questions: 141 text-only and 27 image-based, categorized into four sections mirroring the CBNC exam. Each LLM was presented with the same standardized prompt and applied to each section 30 times to account for stochasticity. Performance over six weeks was assessed for all models except GPT-4o. McNemar's test compared correct response proportions. GPT-4, Gemini, GPT4-Turbo, and GPT-4o correctly answered median percentiles of 56.8% (95% confidence interval 55.4% - 58.0%), 40.5% (39.9% - 42.9%), 60.7% (59.9% - 61.3%) and 63.1% (62.5 - 64.3%) of questions, respectively. GPT4o significantly outperformed other models (p=0.007 vs. GPT-4Turbo, p<0.001 vs. GPT-4 and Gemini). GPT-4o excelled on text-only questions compared to GPT-4, Gemini, and GPT-4 Turbo (p<0.001, p<0.001, and p=0.001), while Gemini performed worse on image-based questions (p<0.001 for all). GPT-4o demonstrated superior performance among the four LLMs, achieving scores likely within or just outside the range required to pass a test akin to the CBNC examination. Although improvements in medical image interpretation are needed, GPT-4o shows potential to support physicians in answering text-based clinical questions.

Evaluating the efficacy of few-shot learning for GPT-4Vision in neurodegenerative disease histopathology: A comparative analysis with convolutional neural network model.

Recent advances in artificial intelligence, particularly with large language models like GPT-4Vision (GPT-4V)-a derivative feature of ChatGPT-have expanded the potential for medical image interpretation. This study evaluates the accuracy of GPT-4V in image classification tasks of histopathological images and compares its performance with a traditional convolutional neural network (CNN). We utilised 1520 images, including haematoxylin and eosin staining and tau immunohistochemistry, from patients with various neurodegenerative diseases, such as Alzheimer's disease (AD), progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD). We assessed GPT-4V's performance using multi-step prompts to determine how textual context influences image interpretation. We also employed few-shot learning to enhance improvements in GPT-4V's diagnostic performance in classifying three specific tau lesions-astrocytic plaques, neuritic plaques and tufted astrocytes-and compared the outcomes with the CNN model YOLOv8. GPT-4V accurately recognised staining techniques and tissue origin but struggled with specific lesion identification. The interpretation of images was notably influenced by the provided textual context, which sometimes led to diagnostic inaccuracies. For instance, when presented with images of the motor cortex, the diagnosis shifted inappropriately from AD to CBD or PSP. However, few-shot learning markedly improved GPT-4V's diagnostic capabilities, enhancing accuracy from 40% in zero-shot learning to 90% with 20-shot learning, matching the performance of YOLOv8, which required 100-shot learning to achieve the same accuracy. Although GPT-4V faces challenges in independently interpreting histopathological images, few-shot learning significantly improves its performance. This approach is especially promising for neuropathology, where acquiring extensive labelled datasets is often challenging.

Robust, credible, and interpretable AI-based histopathological prostate cancer grading.

Prostate cancer (PCa) is among the most common cancers in men and its diagnosis requires the histopathological evaluation of biopsies by human experts. While several recent artificial intelligence-based (AI) approaches have reached human expert-level PCa grading, they often display significantly reduced performance on external datasets. This reduced performance can be caused by variations in sample preparation, for instance the staining protocol, section thickness, or scanner used. Another limiting factor of contemporary AI-based PCa grading is the prediction of ISUP grades, which leads to the perpetuation of human annotation errors. We developed the prostate cancer aggressiveness index (PCAI), an AI-based PCa detection and grading framework that is trained on objective patient outcome, rather than subjective ISUP grades. We designed PCAI as a clinical application, containing algorithmic modules that offer robustness to data variation, medical interpretability, and a measure of prediction confidence. To train and evaluate PCAI, we generated a multicentric, retrospective, observational trial consisting of six cohorts with 25,591 patients, 83,864 images, and 5 years of median follow-up from 5 different centers and 3 countries. This includes a high-variance dataset of 8,157 patients and 28,236 images with variations in sample thickness, staining protocol, and scanner, allowing for the systematic evaluation and optimization of model robustness to data variation. The performance of PCAI was assessed on three external test cohorts from two countries, comprising 2,255 patients and 9,437 images. Using our high-variance datasets, we show how differences in sample processing, particularly slide thickness and staining time, significantly reduce the performance of AI-based PCa grading by up to 6.2 percentage points in the concordance index (C-index). We show how a select set of algorithmic improvements, including domain adversarial training, conferred robustness to data variation, interpretability, and a measure of credibility to PCAI. These changes lead to significant prediction improvement across two biopsy cohorts and one TMA cohort, systematically exceeding expert ISUP grading in C-index and AUROC by up to 22 percentage points. Data variation poses serious risks for AI-based histopathological PCa grading, even when models are trained on large datasets. Algorithmic improvements for model robustness, interpretability, credibility, and training on high-variance data as well as outcome-based severity prediction gives rise to robust models with above ISUP-level PCa grading performance.

Interpreting Difficult Conversations-Evaluating How to Support Medical Interpreters in the Delivery of Serious News.

Background: Despite their essential role in language concordant patient care, medical interpreters do not routinely receive training focused on difficult conversations and may not feel comfortable interpreting these encounters. Previous studies, while acknowledging the need for increased support, have provided limited strategies targeted at enhancing interpreter training and improving interpreter comfort levels in difficult conversations. Methods: Fifty-seven in-person medical interpreters providing services at our quaternary and community hospitals completed a 21-question mixed-methods survey regarding their comfort levels and experiences surrounding serious illness conversations. Results: Most medical interpreters reported being uncomfortable interpreting conversations surrounding difficult diagnosis, poor prognosis, and/or end-of-life. Nearly all respondents (98%) indicated that pre-meetings and/or debriefings with the medical team are helpful, yet only 25% reported frequent participation in these meetings. Conclusions: Our study highlighted the significant variability in medical interpreter training as well as ranging comfort levels in interpreting difficult conversations. Medical providers should not presume that interpreters are instantly prepared for these encounters. Current findings call for novel training opportunities specific to medical interpreters and difficult dialogues, as well as improved adherence of interprofessional pre-meeting/debriefings when serious news is discussed.