MedExpQA: Multilingual benchmarking of Large Language Models for Medical Question Answering

Large Language Models (LLMs) hold the potential for clinical task-shifting by processing unstructured clinical text, enabling tasks such as clinical concept extraction and medical question answering from electronic health records. If implemented reliably, such approaches may benefit over-burdened healthcare systems, particularly in resource-limited settings and for traditionally overlooked populations, provided that local fine-tuning is supported by appropriate clinical and technical expertise. However, this powerful technology remains largely understudied in real-world contexts, particularly in the Global South. This study aims to assess whether openly available LLMs can be used reliably for processing medical notes in real-world settings in South Asia. We used publicly available LLMs to parse de-identified clinical notes from a large electronic health records (EHR) database in Pakistan, containing hospital records for 8.2 million patients. ChatGPT (GPT-3.5) as a general-purpose LLM, and GatorTron (base), BioMegatron, BioBert and ClinicalBERT as medical LLMs were evaluated when applied to these data, after fine-tuning them with (a) publicly available clinical datasets namely Informatics for Integrating Biology & the Bedside (I2B2) and National NLP Clinical Challenges (N2C2) for medical concept extraction (MCE) and emrQA for medical question answering (MQA), and (b) the local Pakistani de-identified EHR dataset, which includes inpatient Discharge Summaries (DS) and Subjective, Objective, Assessment, and Plan (SOAP) notes, as detailed in this paper. MCE models were applied to these clinical notes using both 3-label and 9-label formats, while MQA models were applied to medical questions. Internal and external validation performance was measured for (a) and (b) using F1 score, precision, recall, and accuracy for MCE and BLEU and ROUGE-L, which measure lexical and sequence similarity, for MQA. When clinical LLMs were not fine-tuned on the local EHR dataset, their performance during external validation on local data was notably poorer compared to internal validation on the dataset used for fine-tuning, with reductions of at least 15% in F1 scores for MCE and 35% in ROUGE-L and BLEU scores for MQA tasks. This suggests potential bias and highlights the inability of the medical LLMs to reliably handle the data distribution of the local population without further fine-tuning and adaptation. This trend persisted across two distinct natural language processing tasks: concept extraction and question answering, spanning a spectrum of task complexities. However, fine-tuning the LLMs with local EHR data significantly improved model performance across both tasks, yielding a 7.5% to 15% increase in the F1 score for MCE and a 27% to 53% increase in ROUGE-L and BLEU scores for MQA. Notably, ChatGPT, as a general-purpose LLM, stood out as an exception, demonstrating superior performance across all measured metrics on the local dataset compared to the publicly available dataset, with improvements ranging from 3% to 17% on the local EHR dataset, even without fine-tuning on the local data. Publicly available LLMs, predominantly trained on data from high-income regions, were found to be unreliable when applied in a real-world clinical setting in Pakistan. Fine-tuning them with local EHR data and regional clinical contexts improved their reliability, demonstrating a feasible adaptation strategy that is substantially less resource-intensive than training large language models from scratch. Close collaboration between local clinical and technical experts to curate and leverage more representative, inclusive, and unbiased medical datasets, can play a crucial role in further ensuring reliability of LLMs for resource-limited, overburdened settings, to be used in ways that are safe, fair, and beneficial for all.

Large language models (LLMs) are increasingly adopted in medical question answering (QA) scenarios. However, LLMs have been proven to generate hallucinations and nonfactual information, undermining their trustworthiness in high-stakes medical tasks. Conformal Prediction (CP) is now recognized as a robust framework within the broader domain of machine learning, offering statistically rigorous guarantees of marginal (average) coverage for prediction sets. However, the applicability of CP in medical QA remains to be explored. To address this limitation, this study proposes an enhanced CP framework for medical multiple-choice question answering (MCQA) tasks. The enhanced CP framework associates the non-conformance score with the frequency score of the correct option. The framework generates multiple outputs for the same medical query by leveraging self-consistency theory. The proposed framework calculates the frequency score of each option to address the issue of limited access to the model’s internal information. Furthermore, a risk control framework is incorporated into the enhanced CP framework to manage task-specific metrics through a monotonically decreasing loss function. The enhanced CP framework is evaluated on three popular MCQA datasets using off-the-shelf LLMs. Empirical results demonstrate that the enhanced CP framework achieves user-specified average (or marginal) error rates on the test set. Moreover, the results show that the test set’s average prediction set size (APSS) decreases as the risk level increases. It is concluded that it is a promising evaluation metric for the uncertainty of LLMs.

While large language models (LLMs) have shown promise in medical text analysis, their application in automated medical billing code extraction remains underexplored, particularly for the German medical fee schedule system (GOÄ). Therefore, an LLM was fine-tuned to perform multi-label classification of GOÄ codes from radiology reports automatically, and its performance was compared with state-of-the-art commercial and open-source LLMs. Following ethics committee approval, we analyzed 499,601 radiology reports from 124,497 patients, containing 1,799,971 manually identified GOÄ codes as ground truth. The MediPhi-Instruct 4B model was fine-tuned using five-fold cross-validation. Performance was evaluated on the hold-out test set and compared against GPT-5, GPT-4.1, GPT-oss, Kimi-K2, Deepseek-R1, Deepseek-V3, Gemini 2.5, Llama-70B, and Qwen-3 LLMs on a subset of 500 anonymized and 350 cleaned reports using zero-shot and few-shot prompting techniques. The fine-tuned model achieved an accuracy of 77.15% ± 0.47% and a micro-average F1-score of 87.79% ± 0.31% on the hold-out test set. On a subset of 500 real-world samples, our models outperformed the best-performing LLM, Gemini 2.5 Flash, with an F1-score of 70.32% ± 1.54% compared to 58.22% ± 1.50% (p < 0.001). For the cleaned dataset of 350 samples, GPT-5 achieved the best F1-score of 89.51 ± 1.52% and outperformed the fine-tuned models (p < 0.001). Fine-tuned LLMs can effectively automate GOÄ code classification from radiology reports, with the potential of outperforming commercial LLMs. This approach shows promise for improving billing efficiency and accuracy in healthcare settings, though manual verification is still recommended. Question LLMs with high parameters possess medical knowledge, but how effective are they at predicting billing codes from radiology reports compared to smaller, fine-tuned models? Finidngs A fine-tuned ensemble model achieved competitive results and can outperform larger, proprietary LLMs. Clinical relevance Smaller, fine-tuned models offer an efficient alternative to proprietary LLMs in generating billing codes and can be integrated to assist clinical coding. This technology has the potential to transform clinical billing procedures, but its use should be overseen by qualified professional personnel.

Evaluating gpt-4 for zero-shot classification of bleeding and clotting events: Can large language models serve as second reviewers?

The integration of large language models (LLMs) into medical diagnostics has garnered substantial attention due to their potential to enhance diagnostic accuracy, streamline clinical workflows, and address health care disparities. However, the rapid evolution of LLM research necessitates a comprehensive synthesis of their applications, challenges, and future directions. This scoping review aimed to provide an overview of the current state of research regarding the use of LLMs in medical diagnostics. The study sought to answer four primary subquestions, as follows: (1) Which LLMs are commonly used? (2) How are LLMs assessed in diagnosis? (3) What is the current performance of LLMs in diagnosing diseases? (4) Which medical domains are investigating the application of LLMs? This scoping review was conducted according to the Joanna Briggs Institute Manual for Evidence Synthesis and adheres to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews). Relevant literature was searched from the Web of Science, PubMed, Embase, IEEE Xplore, and ACM Digital Library databases from 2022 to 2025. Articles were screened and selected based on predefined inclusion and exclusion criteria. Bibliometric analysis was performed using VOSviewer to identify major research clusters and trends. Data extraction included details on LLM types, application domains, and performance metrics. The field is rapidly expanding, with a surge in publications after 2023. GPT-4 and its variants dominated research (70/95, 74% of studies), followed by GPT-3.5 (34/95, 36%). Key applications included disease classification (text or image-based), medical question answering, and diagnostic content generation. LLMs demonstrated high accuracy in specialties like radiology, psychiatry, and neurology but exhibited biases in race, gender, and cost predictions. Ethical concerns, including privacy risks and model hallucination, alongside regulatory fragmentation, were critical barriers to clinical adoption. LLMs hold transformative potential for medical diagnostics but require rigorous validation, bias mitigation, and multimodal integration to address real-world complexities. Future research should prioritize explainable artificial intelligence frameworks, specialty-specific optimization, and international regulatory harmonization to ensure equitable and safe clinical deployment.

The rapid growth of biomedical literature poses challenges for manual knowledge curation and synthesis. Biomedical Natural Language Processing (BioNLP) automates the process. While Large Language Models (LLMs) have shown promise in general domains, their effectiveness in BioNLP tasks remains unclear due to limited benchmarks and practical guidelines. We perform a systematic evaluation of four LLMs—GPT and LLaMA representatives—on 12 BioNLP benchmarks across six applications. We compare their zero-shot, few-shot, and fine-tuning performance with the traditional fine-tuning of BERT or BART models. We examine inconsistencies, missing information, hallucinations, and perform cost analysis. Here, we show that traditional fine-tuning outperforms zero- or few-shot LLMs in most tasks. However, closed-source LLMs like GPT-4 excel in reasoning-related tasks such as medical question answering. Open-source LLMs still require fine-tuning to close performance gaps. We find issues like missing information and hallucinations in LLM outputs. These results offer practical insights for applying LLMs in BioNLP.

MedSumGraph: enhancing GraphRAG for medical QA with summarization and optimized prompts.

BackgroundDisease name recognition is a fundamental task in clinical natural language processing, enabling the extraction of critical patient information from electronic health records. While recent advances in large language models (LLMs) have shown promise, most evaluations have focused on English, and little is known about their robustness in low-resource languages such as Japanese. In particular, whether these models can perform reliably on previously unseen in-hospital data, which differs from training data in writing styles and clinical contexts, has not been thoroughly investigated.ObjectiveThis study evaluated the robustness of fine-tuned LLMs for disease name recognition in Japanese clinical notes, with a particular focus on their performance on in-hospital data that was not included during training.MethodsWe used two corpora for this study: (1) a publicly available set of Japanese case reports denoted as CR, and (2) a newly constructed corpus of progress notes, denoted as PN, written by ten physicians to capture stylistic variations of in-hospital clinical notes. To reflect real-world deployment scenarios, we first fine-tuned models on CR. Specifically, we compared a LLM and a baseline-masked language model (MLM). These models were then evaluated under two conditions: (1) on CR, representing the in-domain (ID) setting with the same document type, similar to training, and (2) on PN, representing the out-of-domain (OOD) setting with a different document type. Robustness was assessed by calculating the performance gap (ie, the performance drop from in-domain to out-of-domain settings).ResultsThe LLM demonstrated greater robustness, with a smaller performance gap in F1-scores (ID–OOD = −8.6) compared to the MLM baseline performance (ID–OOD = −13.9). This indicated more stable performance across ID and OOD settings, highlighting the effectiveness of fine-tuned LLMs for reliable use in diverse clinical settings.ConclusionsFine-tuned LLMs demonstrate superior robustness for disease name recognition in Japanese clinical notes, with a smaller performance gap. These findings highlight the potential of LLMs as reliable tools for clinical natural language processing in low-resource language settings and support their deployment in real-world health care applications, where diversity in documentation is inevitable.

PurposeCurrent benchmarks evaluating large language models (LLMs) in ophthalmology are narrow and disproportionately prioritize accuracy. We introduce BEnchmarking LLMs for Ophthalmology (BELO), a standardized evaluation benchmark developed through multiple rounds of expert checking by 13 ophthalmologists. BEnchmarking LLMs for Ophthalmology assesses ophthalmology-related knowledge and reasoning quality.SubjectsThis study did not involve human participation.DesignCross-sectional study.MethodsUsing keyword matching and a fine-tuned PubMed Bidirectional Encoder Representations from Transformers model, we curated ophthalmology-specific multiple-choice questions (MCQs) from diverse medical data sets (Basic and Clinical Science Course [BCSC], Multi-Subject Multi-Choice Dataset for Medical domain [MedMCQA], Medical Question Answering [MedQA], Biomedical Semantic Indexing and Question Answering [BioASQ], and PubMed Question Answering [PubMedQA]). The data set underwent multiple rounds of expert checking. Duplicate and substandard questions were systematically removed. Ten ophthalmologists refined the explanations of each MCQ's correct answer. This was further adjudicated by 3 senior ophthalmologists. To illustrate BELO's utility, we evaluated 8 LLMs (OpenAI o1, o3-mini, GPT-5, GPT-4o, DeepSeek-R1, MedGemma-4B, Llama-3-8B, and Gemini 1.5 Pro).Main Outcome MeasuresThe 8 LLMs were evaluated in terms using accuracy, macro-F1, and 5 text-generation metrics (Recall-Oriented Understudy for Gisting Evaluation, BERTScore, BARTScore, Metric for Evaluation of Translation with Explicit Ordering, and AlignScore). In a further evaluation involving human experts, 2 ophthalmologists qualitatively reviewed 50 randomly selected outputs for accuracy, comprehensiveness, and completeness.ResultsBEnchmarking LLMs for Ophthalmology consists of 900 high-quality, expert-reviewed questions aggregated from 5 sources: BCSC (260), BioASQ (10), MedMCQA (572), MedQA (40), and PubMedQA (18). To demonstrate BELO's utility, we conducted a series of benchmarking exercises. In the quantitative evaluation, GPT-5 achieved the highest accuracy (0.90, 95% confidence interval [CI]: 0.89–0.92) and macro-F1 score (0.91, 95% CI: 0.89–0.93). On the other hand, the models' performance on text-generation metrics varied and were generally suboptimal, with scores ranging from 20.4 to 72.0 (out of 100, excluding the BARTScore metric), indicating room for improvement in clinical reasoning. In expert evaluations, GPT-4o was rated highest for accuracy and readability, while Gemini 1.5 Pro scored highest for completeness. A public leaderboard has been established to promote transparent evaluation and reporting. Importantly, the BELO data set will remain a hold-out, evaluation-only benchmark to ensure fair and reproducible comparisons of future models.ConclusionsBEnchmarking LLMs for Ophthalmology provides a robust clinically relevant benchmark for evaluating both the accuracy and reasoning capabilities of current and emerging LLMs in ophthalmology. Future BELO benchmarking efforts will expand to include vision-based question answering and clinical scenario management tasks.Financial Disclosure(s)The authors have no proprietary or commercial interest in any materials discussed in this article.

A survey for large language models in biomedicine.

Large language models (LLMs) are rapidly advancing natural language processing and driving increasing interest across biomedical research, clinical care, and healthcare operations. This systematic review synthesizes current empirical evidence on LLM applications in healthcare and biomedical informatics, focusing on their capabilities, evaluation practices, and reported outcomes. Existing studies highlight promising uses in research, where LLMs assist literature-informed genomic interpretation, functional annotation, and biological hypothesis generation, as well as support tasks related to protein and multiomics analysis. In clinical contexts, LLMs are primarily evaluated for natural language-driven tasks, including electronic health record summarization, clinical documentation support, medical question answering, and information extraction, rather than autonomous diagnostic or therapeutic decision-making. Early investigations also describe potential value in healthcare operations, patient communication, clinician education, and drug discovery workflows largely through knowledge retrieval, text generation, and semantic search. However, current evidence remains preliminary: most studies are retrospective, benchmark-based, simulation-driven, or limited to controlled research settings. Key challenges include hallucination and unreliable reasoning, bias and inequitable performance across populations, data privacy and security constraints, reproducibility limitations, and lack of prospective clinical validation and regulatory guidance. Emerging strategies, such as domain-specific pretraining, retrieval augmentation, multi-modal architectures, and privacy-preserving learning, aim to improve reliability, safety, and real-world applicability. This review concludes by outlining methodological, infrastructural, and governance requirements needed to responsibly integrate LLMs into biomedical workflows, emphasizing that clinical deployment remains exploratory and must be rigorously evaluated.

Model tuning or prompt Tuning? a study of large language models for clinical concept and relation extraction

Mental health diagnoses possess unique challenges that often lead to nuanced difficulties in managing an individual’s well-being and daily functioning. Self-report questionnaires are a common practice in clinical settings to help mitigate the challenges involved in mental health disorder screening. However, these questionnaires rely on an individual’s subjective response which can be influenced by various factors. Despite the advancements of Large Language Models (LLMs), quantifying self-reported experiences with natural language processing has resulted in imperfect accuracy. This project aims to demonstrate the effectiveness of zero-shot learning LLMs for screening and assessing item scales for depression using LLMs. The DAIC-WOZ is a publicly available mental health dataset that contains textual data from clinical interviews and self-report questionnaires with relevant mental health disorder labels. The RISEN prompt engineering framework was utilized to evaluate LLMs’ effectiveness in predicting depression symptoms based on individual PHQ-8 items. Various LLMs, including GPT models, Llama3_8B, Cohere, and Gemini were assessed based on performance. The GPT models, especially GPT-4o, were consistently better than other LLMs (Llama3_8B, Cohere, Gemini) across all eight items of the PHQ-8 scale in accuracy (M = 75.9%), and F1 score (0.74). GPT models were able to predict PHQ-8 items related to emotional and cognitive states. Llama 3_8B demonstrated superior detection of anhedonia-related symptoms and the Cohere LLM’s strength was identifying and predicting psychomotor activity symptoms. This study provides a novel outlook on the potential of LLMs for predicting self-reported questionnaire scores from textual interview data. The promising preliminary performance of the various models indicates there is potential that these models could effectively assist in the screening of depression. Further research is needed to establish a framework for which LLM can be used for specific mental health symptoms and other disorders. As well, analysis of additional datasets while fine-tuning models should be explored.

With the rapid development of large language models (LLMs) in natural language processing and generation tasks, researchers are continuously validating their application potential in the field of medical artificial intelligence (AI). Generative Pre-trained Transformer 5 (GPT-5), Pathways Language Model 2 (PaLM-2), Large Language Model Meta AI 3 (Llama-3), Gemini-2.5, and various medical vertical models have demonstrated excellent performance in many medical and health-related tasks. However, LLMs still face key bottlenecks in real-world medical scenarios, including “illusions” slow knowledge updates, and a lack of interpretability. These problems severely restrict the safe deployment of large language models in high-risk medical scenarios and have become a major obstacle to the transition of medical artificial intelligence from research to clinical application. Retrieval-augmented generation (RAG) technology has therefore become an important solution for improving the credibility of medical LLMs. RAG significantly reduces the risk of information errors by retrieving authoritative medical knowledge before generation while maintaining knowledge updates without retraining. A review of research progress on RAG in the medical field is of great theoretical value and practical significance for promoting the design, evaluation, and standardized application of trustworthy medical AI. This review aims to systematically summarize the research progress of RAG in medical scenarios, including the technical frameworks, typical applications, and development trends of three types of methods—naive RAG, advanced RAG, and modular RAG—and further discuss their significance in medical reliability, health equity, and personalized medicine. The naive RAG architecture, through a standard “index-retrieval-generation” process, transforms external knowledge bases into a structured vector space that can be queried by an LLM, achieving positive results in electronic medical record (EMR) summary, preoperative assessment, and medical question answering (QA). However, naive RAG still has limitations in retrieval accuracy and cross-modal processing. To address these shortcomings, advanced RAG methods have been devised that significantly improve the model’s decision-making capabilities by improving retrieval strategies, enhancing inference chains, and introducing self-reflection mechanisms. Furthermore, modular RAG design offers composable, multi-module systems to support multi-source knowledge integration and complex task decomposition. In terms of application value, RAG’s contributions to the healthcare field are primarily reflected in these aspects. On the one hand, it significantly improves the reliability of medical AI by introducing traceable knowledge to reduce illusions. On the other hand, it helps promote healthcare equity by driving a shift toward a patient-centered healthcare model through localized knowledge bases and multilingual support. However, the deployment of RAG in medical scenarios still faces many challenges, such as data privacy risks. Future research should focus on improving self-supervised reflection capabilities and developing cross-modal knowledge fusion technologies. As research develops, RAG is expected to become a core foundational capability of future intelligent healthcare systems, further promoting the development of safe and efficient intelligent healthcare.

In this study, we perform a comprehensive evaluation of sentiment classification for German language data using three different approaches: (1) dictionary-based methods, (2) fine-tuned transformer models such as BERT and XLM-T and (3) various large language models (LLMs) with zero-shot capabilities, including natural language inference models, Siamese models and dialog-based models. The evaluation considers a variety of German language datasets, including contemporary social media texts, product reviews and humanities datasets. Our results confirm that dictionary-based methods, while computationally efficient and interpretable, fall short in classification accuracy. Fine-tuned models offer strong performance, but require significant training data and computational resources. LLMs with zero-shot capabilities, particularly dialog-based models, demonstrate competitive performance, often rivaling fine-tuned models, while eliminating the need for task-specific training. However, challenges remain regarding non-determinism, prompt sensitivity and the high resource requirements of large LLMs. The results suggest that for sentiment analysis in the computational humanities, where non-English and historical language data are common, LLM-based zero-shot classification is a viable alternative to fine-tuned models and dictionaries. Nevertheless, model selection remains highly context-dependent, requiring careful consideration of trade-offs between accuracy, resource efficiency and transparency.