Search

Research on inclusion in early childhood educational (ECE) settings points to the need for a deeper understanding of professionals' perceptions of the dilemmas and challenges situated in the varying contexts of these settings. Viewing inclusion as a complex phenomenon, this study uses the Deweyan theory of democracy as a lived experience to describe how professionals in three ECE settings understand the values of inclusion and democracy and how these values are enacted in their everyday practices of creating democratic communities. Three case studies were conducted, including individual interviews with head teachers, focus-group interviews with professionals and semi-structured observations in three ECE settings. The results show that the three settings present unique prerequisites and challenges for the building of democratic communities related to the varying views on inclusion and democracy. The understandings of values of inclusion and democracy and how these are enacted in the three ECE settings are discussed with regard to the possibilities and obstacles involved in the creation of democratic communities.

Abstract Background Large language models are increasingly evaluated for use in healthcare. However, concerns about their impact on disparities persist. This study reviews current research on demographic biases in large language models to identify prevalent bias types, assess measurement methods, and evaluate mitigation strategies. Methods We conducted a systematic review, searching publications from January 2018 to July 2024 across five databases. We included peer-reviewed studies evaluating demographic biases in large language models, focusing on gender, race, ethnicity, age, and other factors. Study quality was assessed using the Joanna Briggs Institute Critical Appraisal Tools. Results Our review included 24 studies. Of these, 22 (91.7%) identified biases. Gender bias was the most prevalent, reported in 15 of 16 studies (93.7%). Racial or ethnic biases were observed in 10 of 11 studies (90.9%). Only two studies found minimal or no bias in certain contexts. Mitigation strategies mainly included prompt engineering, with varying effectiveness. However, these findings are tempered by a potential publication bias, as studies with negative results are less frequently published. Conclusion Biases are observed in large language models across various medical domains. While bias detection is improving, effective mitigation strategies are still developing. As LLMs increasingly influence critical decisions, addressing these biases and their resultant disparities is essential for ensuring fair artificial intelligence systems. Future research should focus on a wider range of demographic factors, intersectional analyses, and non-Western cultural contexts. Graphic Abstract

Abstract Recent advancements in generative artificial intelligence have enabled analysis of text with visual data, which could have important implications in healthcare. Diagnosis in ophthalmology is often based on a combination of ocular examination, and clinical context. The aim of this study was to evaluate the performance of multimodal GPT-4 (GPT-4 V) in an integrated analysis of ocular images and clinical text. This retrospective study included 40 patients seen in our institution with images of their ocular examinations. Cases were selected by a board-certified ophthalmologist, to represent various pathologies. We provided the model with each patient image, without and then with the clinical context. We also asked two non-ophthalmology physicians to write diagnoses for each image, without and then with the clinical context. Answers for both GPT-4 V and the non-ophthalmologists were evaluated by two board-certified ophthalmologists. Performance accuracies were calculated and compared. GPT-4 V provided the correct diagnosis in 19/40 (47.5%) cases based on images without clinical context, and in 27/40 (67.5%) cases when clinical context was provided. Non-ophthalmologist physicians provided the correct diagnoses in 24/40 (60.0%), and 23/40 (57.5%) of cases without clinical context, and in 29/40 (72.5%) and 27/40 (67.5%) with clinical context. For all study participants adding context improved accuracy (p = 0.033). GPT-4 V is currently able to simultaneously analyze and integrate visual and textual data, and arrive at accurate clinical diagnoses in the majority of cases. Multimodal large language models like GPT-4 V have significant potential to advance both patient care and research in ophthalmology.

Background and aim: Visual data from images is essential for many medical diagnoses. This study evaluates the performance of multimodal Large Language Models (LLMs) in integrating textual and visual information for diagnostic purposes. Methods: We tested GPT-4o and Claude Sonnet 3.5 on 120 clinical vignettes with and without accompanying images. Each vignette included patient demographics, a chief concern, and relevant medical history. Vignettes were paired with either clinical or radiological images from two sources: 100 images from the OPENi database and 20 images from recent NEJM challenges, ensuring they were not in the LLMs' training sets. Three primary care physicians served as a human benchmark. We analyzed diagnostic accuracy and the models' explanations for a subset of cases. Results: LLMs outperformed physicians in text-only scenarios (GPT-4o: 70.8 %, Claude Sonnet 3.5: 59.5 %, Physicians: 39.5 %, p

Abstract Large language models (LLMs) can optimize clinical workflows; however, the economic and computational challenges of their utilization at the health system scale are underexplored. We evaluated how concatenating queries with multiple clinical notes and tasks simultaneously affects model performance under increasing computational loads. We assessed ten LLMs of different capacities and sizes utilizing real-world patient data. We conducted >300,000 experiments of various task sizes and configurations, measuring accuracy in question-answering and the ability to properly format outputs. Performance deteriorated as the number of questions and notes increased. High-capacity models, like Llama-3–70b, had low failure rates and high accuracies. GPT-4-turbo-128k was similarly resilient across task burdens, but performance deteriorated after 50 tasks at large prompt sizes. After addressing mitigable failures, these two models can concatenate up to 50 simultaneous tasks effectively, with validation on a public medical question-answering dataset. An economic analysis demonstrated up to a 17-fold cost reduction at 50 tasks using concatenation. These results identify the limits of LLMs for effective utilization and highlight avenues for cost-efficiency at the enterprise scale.

Abstract Background Septic encephalopathy is frequent but its pathophysiology is enigmatic. We studied expression of neurotransmitters, inflammation and integrity of the blood–brain barrier (BBB) in several brain regions during abdominal sepsis. We compared mice with either lethal or surviving phenotype in the first 4 sepsis days. Mature CD-1 females underwent cecal ligation and puncture (CLP). Body temperature (BT) was measured daily and predicted-to-die (within 24 h) mice (for P-DIE; BT 35 °C), and healthy controls (CON). Brains were dissected into neocortex, cerebellum, midbrain, medulla, striatum, hypothalamus and hippocampus. Results CLP mice showed an up to threefold rise of serotonin in the hippocampus, 5-hydroxyindoleacetic and homovanillic acid (HVA) in nearly all regions vs. CON. Compared to P-SUR, P-DIE mice showed a 1.7 to twofold rise of HVA (386 ng/g of tissue), dopamine (265 ng/g) and 3,4-Dihydroxyphenylacetic acid (DOPAC; 140 ng/g) in the hippocampus, hypothalamus and medulla (174, 156, 82 ng/g of tissue, respectively). CLP increased expression of TNFα, IL-1β and IL-6 mRNA by several folds in the midbrain, cerebellum and hippocampus versus CON. The same cytokines were further elevated in P-DIE vs P-SUR in the midbrain and cerebellum. Activation of astrocytes and microglia was robust across regions but remained typically phenotype independent. There was a similar influx of sodium fluorescein across the BBB in both P-DIE and P-SUR mice. Conclusions Compared to survivors, the lethal phenotype induced a stronger deregulation of amine metabolism and cytokine expression in selected brain regions, but the BBB permeability remained similar regardless of the predicted outcome.

Abstract Background While positive school climate is important for students’ well-being and mental health, school personnel may experience challenges in creating a nurturing school climate. School-Wide Positive Behavioral Interventions and Supports (SW-PBIS) has shown positive effects on school climate and children’s prosocial behaviors, but fewer studies have been conducted in a European context. Aim This project aims to investigate the effectiveness of SW-PBIS program for students’ social-emotional skills and academic achievement as well as teachers’ and students’ perceptions of classroom learning environment. Furthermore, the study intends to evaluate how school-level factors mediate or moderate the effects of the intervention. In addition, the study includes a qualitative evaluation of the dynamic interaction processes that occur during program implementation in local school contexts. Methods Data on school- and individual-level measures are collected in intervention and control schools. With regard to challenges in retaining control groups over extended time periods, two waves of recruitment are used. In the first wave, an active control group is used and data are collected during three time points. In the second wave, a wait-list control group will be used and data will be collected during two time points during one school year. Hierarchical regression analyses will be conducted to explore the effects of SW-PBIS on the outcomes of the study. An ethno-methodological approach will be applied to provide a detailed examination of the social interactional and meaning-making practices of different school implementation teams, and the negotiation of normative expectations and rules of conduct in peer-teacher-student interactions in different classrooms. Discussion The study is expected to contribute to knowledge on the effects of the SW-PBIS program and how these effects may be mediated or moderated by school-level factors. Combining quantitative and qualitative methods to explore the significance of school contexts in the implementation of the SW-PBIS program constitutes the strength of the study. The challenge in the study is the extended period of implementation of SW-PBIS, which entails difficulties in retaining a control group over the required time period. Therefore, two waves of recruitment are used, encompassing different procedures of allocation to intervention or control group. Trial registration ClinicalTrials.gov Identifier: NCT06270914 on the 22nd of February, 2024 (retrospectively registered).

Education in segregated settings for pupils with intellectual disability (ID) has often been portrayed as a unified form of schooling. There is a risk of providing a simplified picture of these settings and of what pupils with ID need to learn. This is generally stipulated in policy documents, leaving room for staff in school to interpret curricula and form learning environments. The aim of this study is to explore how four different learning environments are constructed in segregated settings where pupils with ID are educated in Sweden. The study was conducted as a multiple case study, using observations and interviews with teachers and pupils in four special classes for pupils with ID (SCIDs). Drawing on curriculum theory, the four learning environments, in the results named as Alpha, Beta, Gamma and Delta class, respectively, appeared to be characterized by different ideas of schooling, labelled as knowledge-mediating, socializing, functional life skills and caring as well as teacher-centred classrooms. The study contributes to a deepened understanding of the complexity of education for pupils with ID in segregated settings.

Abstract Nano‐patterned magnetic materials have opened new venues for the investigation of strongly correlated phenomena including artificial spin‐ice systems, geometric frustration, and magnetic monopoles, for technologically important applications such as reconfigurable ferromagnetism. With the advent of atomically thin 2D van der Waals (vdW) magnets, a pertinent question is whether such compounds could make their way into this realm where interactions can be tailored so that unconventional states of matter can be assessed. Here, it is shown that square islands of CrGeTe3 vdW ferromagnets distributed in a grid manifest antiferromagnetic correlations, essential to enable frustration resulting in an artificial spin‐ice. By using a combination of SQUID‐on‐tip microscopy, focused ion beam lithography, and atomistic spin dynamic simulations, it is shown that a square array of CGT island as small as 150 × 150 × 60 nm3 have tunable dipole–dipole interactions, which can be precisely controlled by their lateral spacing. There is a crossover between non‐interacting islands and significant inter‐island anticorrelation depending on how they are spatially distributed allowing the creation of complex magnetic patterns not observable at the isolated flakes. These findings suggest that the cross‐talk between the nano‐patterned magnets can be explored in the generation of even more complex spin configurations where exotic interactions may be manipulated in an unprecedented way.

Background and aimIn the last years, natural language processing (NLP) has transformed significantly with the introduction of large language models (LLM). This review updates on NLP and LLM applications and challenges in gastroenterology and hepatology.MethodsRegistered with PROSPERO (CRD42024542275) and adhering to PRISMA guidelines, we searched six databases for relevant studies published from 2003 to 2024, ultimately including 57 studies.ResultsOur review of 57 studies notes an increase in relevant publications in 2023–2024 compared to previous years, reflecting growing interest in newer models such as GPT-3 and GPT-4. The results demonstrate that NLP models have enhanced data extraction from electronic health records and other unstructured medical data sources. Key findings include high precision in identifying disease characteristics from unstructured reports and ongoing improvement in clinical decision-making. Risk of bias assessments using ROBINS-I, QUADAS-2, and PROBAST tools confirmed the methodological robustness of the included studies.ConclusionNLP and LLMs can enhance diagnosis and treatment in gastroenterology and hepatology. They enable extraction of data from unstructured medical records, such as endoscopy reports and patient notes, and for enhancing clinical decision-making. Despite these advancements, integrating these tools into routine practice is still challenging. Future work should prospectively demonstrate real-world value.

Abstract This study was designed to assess how different prompt engineering techniques, specifically direct prompts, Chain of Thought (CoT), and a modified CoT approach, influence the ability of GPT-3.5 to answer clinical and calculation-based medical questions, particularly those styled like the USMLE Step 1 exams. To achieve this, we analyzed the responses of GPT-3.5 to two distinct sets of questions: a batch of 1000 questions generated by GPT-4, and another set comprising 95 real USMLE Step 1 questions. These questions spanned a range of medical calculations and clinical scenarios across various fields and difficulty levels. Our analysis revealed that there were no significant differences in the accuracy of GPT-3.5's responses when using direct prompts, CoT, or modified CoT methods. For instance, in the USMLE sample, the success rates were 61.7% for direct prompts, 62.8% for CoT, and 57.4% for modified CoT, with a p-value of 0.734. Similar trends were observed in the responses to GPT-4 generated questions, both clinical and calculation-based, with p-values above 0.05 indicating no significant difference between the prompt types. The conclusion drawn from this study is that the use of CoT prompt engineering does not significantly alter GPT-3.5's effectiveness in handling medical calculations or clinical scenario questions styled like those in USMLE exams. This finding is crucial as it suggests that performance of ChatGPT remains consistent regardless of whether a CoT technique is used instead of direct prompts. This consistency could be instrumental in simplifying the integration of AI tools like ChatGPT into medical education, enabling healthcare professionals to utilize these tools with ease, without the necessity for complex prompt engineering.

Background: Coarctation of the aorta (CoA) is a relatively common congenital heart defect. The underlying causes are not known, but a combination of genetic factors and abnormalities linked to embryonic development is suspected. There are only a few studies of the underlying molecular mechanisms in CoA. The aim of the current study was to expand our understanding of the pathogenesis of CoA by characterizing the transcriptome of the coarctation area. Methods: Tissue samples from 21 pediatric patients operated for CoA were dissected into separate biopsies consisting of the localized coarctation itself, proximal/distal tissue and ductus. RNA was sequenced to evaluate gene expression in the different biopsies. Results: We observed an activation of acute phase response in samples from the localized coarctation compared to samples from distal or proximal tissue. However, we observed even bigger differences for patient age and sex than compared to biopsy location. A cluster of genes located at 1q21, including the S100 gene family, displayed contrasting expression depending on patient sex, and appeared to affect the balance between inflammatory and interferon pathways. Biopsies from patients

BackgroundEmpathy, a fundamental aspect of human interaction, is characterized as the ability to experience another being’s emotions within oneself. In health care, empathy is a fundamental for health care professionals and patients’ interaction. It is a unique quality to humans that large language models (LLMs) are believed to lack. ObjectiveWe aimed to review the literature on the capacity of LLMs in demonstrating empathy. MethodsWe conducted a literature search on MEDLINE, Google Scholar, PsyArXiv, medRxiv, and arXiv between December 2022 and February 2024. We included English-language full-length publications that evaluated empathy in LLMs’ outputs. We excluded papers evaluating other topics related to emotional intelligence that were not specifically empathy. The included studies’ results, including the LLMs used, performance in empathy tasks, and limitations of the models, along with studies’ metadata were summarized. ResultsA total of 12 studies published in 2023 met the inclusion criteria. ChatGPT-3.5 (OpenAI) was evaluated in all studies, with 6 studies comparing it with other LLMs such GPT-4, LLaMA (Meta), and fine-tuned chatbots. Seven studies focused on empathy within a medical context. The studies reported LLMs to exhibit elements of empathy, including emotions recognition and emotional support in diverse contexts. Evaluation metric included automatic metrics such as Recall-Oriented Understudy for Gisting Evaluation and Bilingual Evaluation Understudy, and human subjective evaluation. Some studies compared performance on empathy with humans, while others compared between different models. In some cases, LLMs were observed to outperform humans in empathy-related tasks. For example, ChatGPT-3.5 was evaluated for its responses to patients’ questions from social media, where ChatGPT’s responses were preferred over those of humans in 78.6% of cases. Other studies used subjective readers’ assigned scores. One study reported a mean empathy score of 1.84-1.9 (scale 0-2) for their fine-tuned LLM, while a different study evaluating ChatGPT-based chatbots reported a mean human rating of 3.43 out of 4 for empathetic responses. Other evaluations were based on the level of the emotional awareness scale, which was reported to be higher for ChatGPT-3.5 than for humans. Another study evaluated ChatGPT and GPT-4 on soft-skills questions in the United States Medical Licensing Examination, where GPT-4 answered 90% of questions correctly. Limitations were noted, including repetitive use of empathic phrases, difficulty following initial instructions, overly lengthy responses, sensitivity to prompts, and overall subjective evaluation metrics influenced by the evaluator’s background. ConclusionsLLMs exhibit elements of cognitive empathy, recognizing emotions and providing emotionally supportive responses in various contexts. Since social skills are an integral part of intelligence, these advancements bring LLMs closer to human-like interactions and expand their potential use in applications requiring emotional intelligence. However, there remains room for improvement in both the performance of these models and the evaluation strategies used for assessing soft skills.

This review analyzes current clinical trials investigating large language models' (LLMs) applications in healthcare. We identified 27 trials (5 published and 22 ongoing) across 4 main clinical applications: patient care, data handling, decision support, and research assistance. Our analysis reveals diverse LLM uses, from clinical documentation to medical decision-making. Published trials show promise but highlight accuracy concerns. Ongoing studies explore novel applications like patient education and informed consent. Most trials occur in the United States of America and China. We discuss the challenges of evaluating rapidly evolving LLMs through clinical trials and identify gaps in current research. This review aims to inform future studies and guide the integration of LLMs into clinical practice.

This article presents findings from an ongoing project focused on teaching in Schools for Students with Intellectual Disabilities (SSIDs) in Sweden. Methodologically, the project employs professional development circles as a collaborative working method, aiming to bridge the gap between research and practice by facilitating dialog between teachers and researchers. Through qualitative content analysis of data collected from these circles, the study delves into professionals’ instructional priorities and didactic choices, offering insights into how teachers adapt instruction to meet students’ diverse needs and prerequisites. By fostering knowledge exchange and shared knowledge production, the project aims to support school development and enhance teacher professionalism in SSIDs.

Traditional methods for measuring chemical exposure have challenges in terms of obtaining sufficient data; therefore, improved methods for better assessing occupational exposure are needed. One possible approach to mitigate these challenges is to use self-monitoring methods such as sensors, diaries, or biomarkers. In the present study, a self-monitored method for measuring soot exposure, which included real-time air monitoring, a work diary, and the collection of urine samples, was evaluated. To validate the method, exposure measurements during the workday and diary entries were compared with velocities calculated from GPS tracking and the expected polycyclic aromatic hydrocarbon (PAH) metabolite patterns in urine. The method was applied with chimney sweeps, an occupational group at a high risk of many severe health outcomes and for whom effective control measures for reducing exposure are needed. In the study, 20 chimney sweeps followed a self-monitoring protocol for 8 consecutive workdays. Personal exposure to soot was measured as black carbon (BC) using micro-aethalometers. A diary was used to record the work tasks performed, and urine samples were collected and analysed for PAH metabolites. From the expected 160 full day measurements, 146 (91%) BC measurements and 149 (93%) diaries were collected. From the expected 320 urine samples, 304 (95%) were collected. The tasks noted in the diaries overlapped with information obtained from the GPS tracking of the chimney sweeps, which covered 96% of the measurement time. The PAH metabolites in urine increased during the work week. Factors believed to have positively influenced the sample collection and task documentation were the highly motivated participants and the continuous presence of trained occupational hygiene professionals during the planning of the study and throughout the measurement stage, during which they were available to inform, instruct, and address questions. In conclusion, the self-monitored protocol used in this study with chimney sweeps is a valuable and valid method that can be used to collect larger numbers of samples. This is especially valuable for occupations in which the employees are working independently and the exposure is difficult to monitor with traditional occupational hygiene methods.

Children with severe disabilities are often excluded from educational opportunities due to inter alia attitudinal barriers and a lack of teacher training. This scoping review paper assessed intervention studies focused on inclusive education for children with severe disabilities. It was found that a total of 13 studies met the review's inclusion criteria. The results revealed a paucity of literature and highlighted five main areas of focus in the available intervention studies. The majority of intervention studies were found to be focused on increasing engagement, reducing behaviours and improving social inclusion as outcomes. There was limited focus on studies focusing on the improvement of academic outcomes.

Author(s): Dana Lew [sup.1], Eyal Klang [sup.2], Shelly Soffer [sup.3], Adam S. Morgenthau [sup.4] Author Affiliations: (1) https://ror.org/04a9tmd77, grid.59734.3c, 0000 0001 0670 2351, Division of Internal Medicine, Icahn School of [...], Purpose Sarcoidosis is a complex disease which can affect nearly every organ system with manifestations ranging from asymptomatic imaging findings to sudden cardiac death. As such, diagnosis and prognostication are topics of continued investigation. Recent technological advancements have introduced multiple modalities of artificial intelligence (AI) to the study of sarcoidosis. Machine learning, deep learning, and radiomics have predominantly been used to study sarcoidosis. Methods Articles were collected by searching online databases using keywords such as sarcoid, machine learning, artificial intelligence, radiomics, and deep learning. Article titles and abstracts were reviewed for relevance by a single reviewer. Articles written in languages other than English were excluded. Conclusions Machine learning may be used to help diagnose pulmonary sarcoidosis and prognosticate in cardiac sarcoidosis. Deep learning is most comprehensively studied for diagnosis of pulmonary sarcoidosis and has less frequently been applied to prognostication in cardiac sarcoidosis. Radiomics has primarily been used to differentiate sarcoidosis from malignancy. To date, the use of AI in sarcoidosis is limited by the rarity of this disease, leading to small, suboptimal training sets. Nevertheless, there are applications of AI that have been used to study other systemic diseases, which may be adapted for use in sarcoidosis. These applications include discovery of new disease phenotypes, discovery of biomarkers of disease onset and activity, and treatment optimization.

Background: In several cities of Cameroon, the consumption of soy cheese skewers commonly known as “soybean soya” is becoming very common in the streets. Its consumption is increasing thereby encouraging its local production using household methods. Food contamination from processing to consumption is becoming a serious health concern. Therefore, the purpose of this study was to ascertain the microbiological quality of soy cheese skewers sold in some markets and the Critical Control Points (CCPs) in the production process. Methods: Samples were collected in June and August 2022 from different vendors, stored in sterile plastic laminas and placed in a pack containing ice and taken to the laboratory for the determination of the microbiological quality. In order to determine CCPs based on microbiological hazard, samples of slurry, soymilk, fresh soy cheese, stewed soy cheese, and stewed soy cheese skewers were collected from different producers at different production steps and analyzed for microbial contaminations. Results: Results of samples collected from the markets indicated Total Mesophilic Aerobic Bacteria between 4.87 to 7.79 log Colony Forming Unit (CFU)/g, where 86% of the samples were higher than the normal; yeast and mold counts ranged from 2.84 to 5.91 CFU/g; staphylococci, and total coliforms ranged from 0 to 5.17 and 0 to 2.37 log CFU/g, present in 59 and 72% of samples, respectively. Conclusion: Most commercial soy cheese skewer samples showed poor microbiological quality. To reduce these contaminations, preventive measures should be put in place at CCPs (stewing and packaging on skewers) to ensure the safety of food. DOI: 10.18502/jfqhc.11.1.14993

Abstract Background Inclusive learning environments are considered as crucial for children’s engagement with learning and participation in school. Partnering for change (P4C) is a collaborative school-based service delivery model where services are provided at three levels of intensity based on children’s needs (class, group-, individual interventions). Interventions in P4C are provided universally to support all children with learning, not only children with special education needs (SEN), and as such are expected to be health-promoting. Aim The aim of the study is to evaluate the effectiveness and cost-effectiveness of P4C as well as school staff members’ and children’s experiences after P4C. Methods In a parallel, non-randomised controlled intervention design, 400 children, aged 6–12 years, and their teachers, will be recruited to either intervention classes, working according to the P4C, or to control classes (allocation ratio 1:1). Data will be collected at baseline, post-intervention (4 months), and 11 months follow-up post baseline. The primary outcome is children’s engagement with learning in school. Secondary outcomes include for example children’s health-related quality of life and wellbeing, occupational performance in school, attendance, and special educational needs. The difference-in-differences method using regression modelling will be applied to evaluate any potential changes following P4C. Focus group interviews focusing on children, and professionals’ experiences will be performed after P4C. A health economic evaluation of P4C will be performed, both in the short term (post intervention) and the long term (11-month follow-up). This study will provide knowledge about the effectiveness of P4C on children’s engagement with learning, mental health, and wellbeing, when creating inclusive learning environments using a combination of class-, group- and individual-level interventions. Trial registration number NCT05435937.

Abstract The United States Medical Licensing Examination (USMLE) has been a subject of performance study for artificial intelligence (AI) models. However, their performance on questions involving USMLE soft skills remains unexplored. This study aimed to evaluate ChatGPT and GPT-4 on USMLE questions involving communication skills, ethics, empathy, and professionalism. We used 80 USMLE-style questions involving soft skills, taken from the USMLE website and the AMBOSS question bank. A follow-up query was used to assess the models’ consistency. The performance of the AI models was compared to that of previous AMBOSS users. GPT-4 outperformed ChatGPT, correctly answering 90% compared to ChatGPT’s 62.5%. GPT-4 showed more confidence, not revising any responses, while ChatGPT modified its original answers 82.5% of the time. The performance of GPT-4 was higher than that of AMBOSS's past users. Both AI models, notably GPT-4, showed capacity for empathy, indicating AI's potential to meet the complex interpersonal, ethical, and professional demands intrinsic to the practice of medicine.

Abstract Background The task of writing multiple choice question examinations for medical students is complex, timely and requires significant efforts from clinical staff and faculty. Applying artificial intelligence algorithms in this field of medical education may be advisable. Methods During March to April 2023, we utilized GPT-4, an OpenAI application, to write a 210 multi choice questions-MCQs examination based on an existing exam template and thoroughly investigated the output by specialist physicians who were blinded to the source of the questions. Algorithm mistakes and inaccuracies, as identified by specialists were classified as stemming from age, gender or geographical insensitivities. Results After inputting a detailed prompt, GPT-4 produced the test rapidly and effectively. Only 1 question (0.5%) was defined as false; 15% of questions necessitated revisions. Errors in the AI-generated questions included: the use of outdated or inaccurate terminology, age-sensitive inaccuracies, gender-sensitive inaccuracies, and geographically sensitive inaccuracies. Questions that were disqualified due to flawed methodology basis included elimination-based questions and questions that did not include elements of integrating knowledge with clinical reasoning. Conclusion GPT-4 can be used as an adjunctive tool in creating multi-choice question medical examinations yet rigorous inspection by specialist physicians remains pivotal.